Whamcloud - gitweb
LU-5099 api: transfer object type via dt_insert API
[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, 2013, 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
48 /* LUSTRE_VERSION_CODE */
49 #include <lustre_ver.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/types.h>
52 /* prerequisite for linux/xattr.h */
53 #include <linux/fs.h>
54 /* XATTR_{REPLACE,CREATE} */
55 #include <linux/xattr.h>
56
57 /*
58  * struct OBD_{ALLOC,FREE}*()
59  * OBD_FAIL_CHECK
60  */
61 #include <obd_support.h>
62 /* struct ptlrpc_thread */
63 #include <lustre_net.h>
64 #include <lustre_fid.h>
65 /* process_config */
66 #include <lustre_param.h>
67
68 #include "osd_internal.h"
69 #include "osd_dynlocks.h"
70
71 /* llo_* api support */
72 #include <md_object.h>
73 #include <lustre_quota.h>
74
75 #include <ldiskfs/xattr.h>
76
77 int ldiskfs_pdo = 1;
78 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
79                 "ldiskfs with parallel directory operations");
80
81 int ldiskfs_track_declares_assert;
82 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
83                 "LBUG during tracking of declares");
84
85 /* Slab to allocate dynlocks */
86 struct kmem_cache *dynlock_cachep;
87
88 static struct lu_kmem_descr ldiskfs_caches[] = {
89         {
90                 .ckd_cache = &dynlock_cachep,
91                 .ckd_name  = "dynlock_cache",
92                 .ckd_size  = sizeof(struct dynlock_handle)
93         },
94         {
95                 .ckd_cache = NULL
96         }
97 };
98
99 static const char dot[] = ".";
100 static const char dotdot[] = "..";
101 static const char remote_obj_dir[] = "REM_OBJ_DIR";
102
103 static const struct lu_object_operations      osd_lu_obj_ops;
104 static const struct dt_object_operations      osd_obj_ops;
105 static const struct dt_object_operations      osd_obj_ea_ops;
106 static const struct dt_object_operations      osd_obj_otable_it_ops;
107 static const struct dt_index_operations       osd_index_iam_ops;
108 static const struct dt_index_operations       osd_index_ea_ops;
109
110 int osd_trans_declare_op2rb[] = {
111         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
112         [OSD_OT_PUNCH]          = OSD_OT_MAX,
113         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
114         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
115         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
116         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
117         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
118         [OSD_OT_WRITE]          = OSD_OT_WRITE,
119         [OSD_OT_INSERT]         = OSD_OT_DELETE,
120         [OSD_OT_DELETE]         = OSD_OT_INSERT,
121         [OSD_OT_UPDATE]         = OSD_OT_MAX,
122         [OSD_OT_QUOTA]          = OSD_OT_MAX,
123 };
124
125 static int osd_has_index(const struct osd_object *obj)
126 {
127         return obj->oo_dt.do_index_ops != NULL;
128 }
129
130 static int osd_object_invariant(const struct lu_object *l)
131 {
132         return osd_invariant(osd_obj(l));
133 }
134
135 /*
136  * Concurrency: doesn't matter
137  */
138
139 /*
140  * Concurrency: doesn't matter
141  */
142 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
143 {
144         struct osd_thread_info *oti = osd_oti_get(env);
145         return oti->oti_w_locks > 0 && o->oo_owner == env;
146 }
147
148 /*
149  * Concurrency: doesn't access mutable data
150  */
151 static int osd_root_get(const struct lu_env *env,
152                         struct dt_device *dev, struct lu_fid *f)
153 {
154         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
155         return 0;
156 }
157
158 /*
159  * OSD object methods.
160  */
161
162 /*
163  * Concurrency: no concurrent access is possible that early in object
164  * life-cycle.
165  */
166 static struct lu_object *osd_object_alloc(const struct lu_env *env,
167                                           const struct lu_object_header *hdr,
168                                           struct lu_device *d)
169 {
170         struct osd_object *mo;
171
172         OBD_ALLOC_PTR(mo);
173         if (mo != NULL) {
174                 struct lu_object *l;
175
176                 l = &mo->oo_dt.do_lu;
177                 dt_object_init(&mo->oo_dt, NULL, d);
178                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
179                 l->lo_ops = &osd_lu_obj_ops;
180                 init_rwsem(&mo->oo_sem);
181                 init_rwsem(&mo->oo_ext_idx_sem);
182                 spin_lock_init(&mo->oo_guard);
183                 return l;
184         } else {
185                 return NULL;
186         }
187 }
188
189 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
190                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
191 {
192         int rc;
193
194         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
195         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
196                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
197         if (rc > 0) {
198                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
199                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
200                 rc = 0;
201                 lustre_lma_swab(lma);
202                 /* Check LMA compatibility */
203                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
204                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
205                               "for fid = "DFID", ino = %lu\n",
206                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
207                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
208                               PFID(&lma->lma_self_fid), inode->i_ino);
209                         rc = -EOPNOTSUPP;
210                 }
211         } else if (rc == 0) {
212                 rc = -ENODATA;
213         }
214
215         return rc;
216 }
217
218 /*
219  * retrieve object from backend ext fs.
220  **/
221 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
222                        struct osd_inode_id *id)
223 {
224         struct inode *inode = NULL;
225
226         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
227         if (IS_ERR(inode)) {
228                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
229                        id->oii_ino, PTR_ERR(inode));
230         } else if (id->oii_gen != OSD_OII_NOGEN &&
231                    inode->i_generation != id->oii_gen) {
232                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
233                        "i_generation = %u\n",
234                        id->oii_ino, id->oii_gen, inode->i_generation);
235                 iput(inode);
236                 inode = ERR_PTR(-ESTALE);
237         } else if (inode->i_nlink == 0) {
238                 /* due to parallel readdir and unlink,
239                 * we can have dead inode here. */
240                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
241                 iput(inode);
242                 inode = ERR_PTR(-ESTALE);
243         } else if (is_bad_inode(inode)) {
244                 CWARN("%.16s: bad inode: ino = %u\n",
245                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
246                 iput(inode);
247                 inode = ERR_PTR(-ENOENT);
248         } else {
249                 if (id->oii_gen == OSD_OII_NOGEN)
250                         osd_id_gen(id, inode->i_ino, inode->i_generation);
251
252                 /* Do not update file c/mtime in ldiskfs.
253                  * NB: we don't have any lock to protect this because we don't
254                  * have reference on osd_object now, but contention with
255                  * another lookup + attr_set can't happen in the tiny window
256                  * between if (...) and set S_NOCMTIME. */
257                 if (!(inode->i_flags & S_NOCMTIME))
258                         inode->i_flags |= S_NOCMTIME;
259         }
260         return inode;
261 }
262
263 static struct inode *
264 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
265              struct osd_inode_id *id, struct lu_fid *fid)
266 {
267         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
268         struct inode            *inode;
269         int                      rc;
270
271         inode = osd_iget(info, dev, id);
272         if (IS_ERR(inode))
273                 return inode;
274
275         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
276         if (rc == 0) {
277                 *fid = lma->lma_self_fid;
278         } else if (rc == -ENODATA) {
279                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
280                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
281                 else
282                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
283         } else {
284                 iput(inode);
285                 inode = ERR_PTR(rc);
286         }
287         return inode;
288 }
289
290 static struct inode *osd_iget_check(struct osd_thread_info *info,
291                                     struct osd_device *dev,
292                                     const struct lu_fid *fid,
293                                     struct osd_inode_id *id,
294                                     bool in_oi)
295 {
296         struct inode    *inode;
297         int              rc     = 0;
298         ENTRY;
299
300         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
301         if (IS_ERR(inode)) {
302                 rc = PTR_ERR(inode);
303                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
304                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
305                                id->oii_ino, rc);
306
307                         GOTO(put, rc);
308                 }
309
310                 goto check_oi;
311         }
312
313         if (is_bad_inode(inode)) {
314                 rc = -ENOENT;
315                 if (!in_oi) {
316                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
317
318                         GOTO(put, rc);
319                 }
320
321                 goto check_oi;
322         }
323
324         if (id->oii_gen != OSD_OII_NOGEN &&
325             inode->i_generation != id->oii_gen) {
326                 rc = -ESTALE;
327                 if (!in_oi) {
328                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
329                                "oii_gen = %u, i_generation = %u\n",
330                                id->oii_ino, id->oii_gen, inode->i_generation);
331
332                         GOTO(put, rc);
333                 }
334
335                 goto check_oi;
336         }
337
338         if (inode->i_nlink == 0) {
339                 rc = -ENOENT;
340                 if (!in_oi) {
341                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
342
343                         GOTO(put, rc);
344                 }
345
346                 goto check_oi;
347         }
348
349 check_oi:
350         if (rc != 0) {
351                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
352
353                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
354                 /* XXX: There are some possible cases:
355                  *      1. rc = 0.
356                  *         Backup/restore caused the OI invalid.
357                  *      2. rc = 0.
358                  *         Someone unlinked the object but NOT removed
359                  *         the OI mapping, such as mount target device
360                  *         as ldiskfs, and modify something directly.
361                  *      3. rc = -ENOENT.
362                  *         Someone just removed the object between the
363                  *         former oi_lookup and the iget. It is normal.
364                  *      4. Other failure cases.
365                  *
366                  *      Generally, when the device is mounted, it will
367                  *      auto check whether the system is restored from
368                  *      file-level backup or not. We trust such detect
369                  *      to distinguish the 1st case from the 2nd case. */
370                 if (rc == 0) {
371                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
372                             inode->i_generation == id->oii_gen)
373                                 rc = -ENOENT;
374                         else
375                                 rc = -EREMCHG;
376                 }
377         } else {
378                 if (id->oii_gen == OSD_OII_NOGEN)
379                         osd_id_gen(id, inode->i_ino, inode->i_generation);
380
381                 /* Do not update file c/mtime in ldiskfs.
382                  * NB: we don't have any lock to protect this because we don't
383                  * have reference on osd_object now, but contention with
384                  * another lookup + attr_set can't happen in the tiny window
385                  * between if (...) and set S_NOCMTIME. */
386                 if (!(inode->i_flags & S_NOCMTIME))
387                         inode->i_flags |= S_NOCMTIME;
388         }
389
390         GOTO(put, rc);
391
392 put:
393         if (rc != 0) {
394                 if (!IS_ERR(inode))
395                         iput(inode);
396
397                 inode = ERR_PTR(rc);
398         }
399
400         return inode;
401 }
402
403 /**
404  * \retval +v: new filter_fid, does not contain self-fid
405  * \retval 0:  filter_fid_old, contains self-fid
406  * \retval -v: other failure cases
407  */
408 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
409                  struct dentry *dentry, struct lu_fid *fid)
410 {
411         struct filter_fid_old   *ff     = &info->oti_ff;
412         struct ost_id           *ostid  = &info->oti_ostid;
413         int                      rc;
414
415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
416         if (rc == sizeof(*ff)) {
417                 rc = 0;
418                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
419                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
420                 /* XXX: should use real OST index in the future. LU-3569 */
421                 ostid_to_fid(fid, ostid, 0);
422         } else if (rc == sizeof(struct filter_fid)) {
423                 rc = 1;
424         } else if (rc >= 0) {
425                 rc = -EINVAL;
426         }
427
428         return rc;
429 }
430
431 static int osd_lma_self_repair(struct osd_thread_info *info,
432                                struct osd_device *osd, struct inode *inode,
433                                const struct lu_fid *fid, __u32 compat)
434 {
435         handle_t *jh;
436         int       rc;
437
438         LASSERT(current->journal_info == NULL);
439
440         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
441                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
442         if (IS_ERR(jh)) {
443                 rc = PTR_ERR(jh);
444                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
445                       osd_name(osd), rc);
446                 return rc;
447         }
448
449         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
450         if (rc != 0)
451                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
452                       osd_name(osd), rc);
453         ldiskfs_journal_stop(jh);
454         return rc;
455 }
456
457 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
458 {
459         struct osd_thread_info  *info   = osd_oti_get(env);
460         struct osd_device       *osd    = osd_obj2dev(obj);
461         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
462         struct inode            *inode  = obj->oo_inode;
463         struct dentry           *dentry = &info->oti_obj_dentry;
464         struct lu_fid           *fid    = NULL;
465         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
466         int                      rc;
467         ENTRY;
468
469         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
470         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
471                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
472         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
473                 fid = &lma->lma_self_fid;
474                 rc = osd_get_idif(info, inode, dentry, fid);
475                 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
476                         /* For the given OST-object, if it has neither LMA nor
477                          * FID in XATTR_NAME_FID, then the given FID (which is
478                          * contained in the @obj, from client RPC for locating
479                          * the OST-object) is trusted. We use it to generate
480                          * the LMA. */
481                         osd_lma_self_repair(info, osd, inode, rfid,
482                                 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
483                                 LMAC_FID_ON_OST : 0);
484                         RETURN(0);
485                 }
486         }
487
488         if (unlikely(rc == -ENODATA))
489                 RETURN(0);
490
491         if (rc < 0)
492                 RETURN(rc);
493
494         if (rc > 0) {
495                 rc = 0;
496                 lustre_lma_swab(lma);
497                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
498                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
499                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
500                               "fid = "DFID", ino = %lu\n", osd_name(osd),
501                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
502                               PFID(rfid), inode->i_ino);
503                         rc = -EOPNOTSUPP;
504                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
505                         fid = &lma->lma_self_fid;
506                 }
507         }
508
509         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
510                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
511                         struct ost_id   *oi   = &info->oti_ostid;
512                         struct lu_fid   *fid1 = &info->oti_fid3;
513                         __u32            idx  = fid_idif_ost_idx(rfid);
514
515                         /* For old IDIF, the OST index is not part of the IDIF,
516                          * Means that different OSTs may have the same IDIFs.
517                          * Under such case, we need to make some compatible
518                          * check to make sure to trigger OI scrub properly. */
519                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
520                                 /* Given @rfid is new, LMA is old. */
521                                 fid_to_ostid(fid, oi);
522                                 ostid_to_fid(fid1, oi, idx);
523                                 if (lu_fid_eq(fid1, rfid)) {
524                                         if (osd->od_lma_self_repair)
525                                                 osd_lma_self_repair(info, osd,
526                                                         inode, rfid,
527                                                         LMAC_FID_ON_OST);
528                                         RETURN(0);
529                                 }
530                         }
531                 }
532
533                 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
534                        osd_name(osd), PFID(rfid), PFID(fid));
535                 rc = -EREMCHG;
536         }
537
538         RETURN(rc);
539 }
540
541 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
542                           const struct lu_fid *fid,
543                           const struct lu_object_conf *conf)
544 {
545         struct osd_thread_info *info;
546         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
547         struct osd_device      *dev;
548         struct osd_idmap_cache *oic;
549         struct osd_inode_id    *id;
550         struct inode           *inode;
551         struct osd_scrub       *scrub;
552         struct scrub_file      *sf;
553         int                     result;
554         int                     saved  = 0;
555         bool                    in_oi  = false;
556         bool                    triggered = false;
557         ENTRY;
558
559         LINVRNT(osd_invariant(obj));
560         LASSERT(obj->oo_inode == NULL);
561         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
562
563         dev = osd_dev(ldev);
564         scrub = &dev->od_scrub;
565         sf = &scrub->os_file;
566         info = osd_oti_get(env);
567         LASSERT(info);
568         oic = &info->oti_cache;
569
570         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
571                 RETURN(-ENOENT);
572
573         /* For the object is created as locking anchor, or for the object to
574          * be created on disk. No need to osd_oi_lookup() at here because FID
575          * shouldn't never be re-used, if it's really a duplicate FID from
576          * unexpected reason, we should be able to detect it later by calling
577          * do_create->osd_oi_insert(). */
578         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
579                 GOTO(out, result = 0);
580
581         /* Search order: 1. per-thread cache. */
582         if (lu_fid_eq(fid, &oic->oic_fid) &&
583             likely(oic->oic_dev == dev)) {
584                 id = &oic->oic_lid;
585                 goto iget;
586         }
587
588         id = &info->oti_id;
589         if (!list_empty(&scrub->os_inconsistent_items)) {
590                 /* Search order: 2. OI scrub pending list. */
591                 result = osd_oii_lookup(dev, fid, id);
592                 if (result == 0)
593                         goto iget;
594         }
595
596         /* Search order: 3. OI files. */
597         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
598         if (result == -ENOENT) {
599                 if (!fid_is_norm(fid) ||
600                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
601                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
602                                       sf->sf_oi_bitmap))
603                         GOTO(out, result = 0);
604
605                 goto trigger;
606         }
607
608         if (result != 0)
609                 GOTO(out, result);
610
611         in_oi = true;
612
613 iget:
614         inode = osd_iget_check(info, dev, fid, id, in_oi);
615         if (IS_ERR(inode)) {
616                 result = PTR_ERR(inode);
617                 if (result == -ENOENT || result == -ESTALE) {
618                         if (!in_oi)
619                                 fid_zero(&oic->oic_fid);
620
621                         GOTO(out, result = -ENOENT);
622                 } else if (result == -EREMCHG) {
623
624 trigger:
625                         if (!in_oi)
626                                 fid_zero(&oic->oic_fid);
627
628                         if (unlikely(triggered))
629                                 GOTO(out, result = saved);
630
631                         triggered = true;
632                         if (thread_is_running(&scrub->os_thread)) {
633                                 result = -EINPROGRESS;
634                         } else if (!dev->od_noscrub) {
635                                 result = osd_scrub_start(dev);
636                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
637                                               "for "DFID", rc = %d [1]\n",
638                                               osd_name(dev), PFID(fid),result);
639                                 if (result == 0 || result == -EALREADY)
640                                         result = -EINPROGRESS;
641                                 else
642                                         result = -EREMCHG;
643                         }
644
645                         /* We still have chance to get the valid inode: for the
646                          * object which is referenced by remote name entry, the
647                          * object on the local MDT will be linked under the dir
648                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
649                          *
650                          * We do not know whether the object for the given FID
651                          * is referenced by some remote name entry or not, and
652                          * especially for DNE II, a multiple-linked object may
653                          * have many name entries reside on many MDTs.
654                          *
655                          * To simplify the operation, OSD will not distinguish
656                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
657                          * only happened for the RPC from other MDT during the
658                          * OI scrub, or for the client side RPC with FID only,
659                          * such as FID to path, or from old connected client. */
660                         saved = result;
661                         result = osd_lookup_in_remote_parent(info, dev,
662                                                              fid, id);
663                         if (result == 0) {
664                                 in_oi = false;
665                                 goto iget;
666                         }
667
668                         result = saved;
669                 }
670
671                 GOTO(out, result);
672         }
673
674         obj->oo_inode = inode;
675         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
676
677         result = osd_check_lma(env, obj);
678         if (result != 0) {
679                 iput(inode);
680                 obj->oo_inode = NULL;
681                 if (result == -EREMCHG) {
682                         if (!in_oi) {
683                                 result = osd_oi_lookup(info, dev, fid, id,
684                                                        OI_CHECK_FLD);
685                                 if (result != 0) {
686                                         fid_zero(&oic->oic_fid);
687                                         GOTO(out, result);
688                                 }
689                         }
690
691                         goto trigger;
692                 }
693
694                 GOTO(out, result);
695         }
696
697         obj->oo_compat_dot_created = 1;
698         obj->oo_compat_dotdot_created = 1;
699
700         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
701                 GOTO(out, result = 0);
702
703         LASSERT(obj->oo_hl_head == NULL);
704         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
705         if (obj->oo_hl_head == NULL) {
706                 obj->oo_inode = NULL;
707                 iput(inode);
708                 GOTO(out, result = -ENOMEM);
709         }
710         GOTO(out, result = 0);
711
712 out:
713         LINVRNT(osd_invariant(obj));
714         return result;
715 }
716
717 /*
718  * Concurrency: shouldn't matter.
719  */
720 static void osd_object_init0(struct osd_object *obj)
721 {
722         LASSERT(obj->oo_inode != NULL);
723         obj->oo_dt.do_body_ops = &osd_body_ops;
724         obj->oo_dt.do_lu.lo_header->loh_attr |=
725                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
726 }
727
728 /*
729  * Concurrency: no concurrent access is possible that early in object
730  * life-cycle.
731  */
732 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
733                            const struct lu_object_conf *conf)
734 {
735         struct osd_object *obj = osd_obj(l);
736         int result;
737
738         LINVRNT(osd_invariant(obj));
739
740         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
741                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
742                 l->lo_header->loh_attr |= LOHA_EXISTS;
743                 return 0;
744         }
745
746         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
747         obj->oo_dt.do_body_ops = &osd_body_ops_new;
748         if (result == 0 && obj->oo_inode != NULL)
749                 osd_object_init0(obj);
750
751         LINVRNT(osd_invariant(obj));
752         return result;
753 }
754
755 /*
756  * Concurrency: no concurrent access is possible that late in object
757  * life-cycle.
758  */
759 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
760 {
761         struct osd_object *obj = osd_obj(l);
762
763         LINVRNT(osd_invariant(obj));
764
765         dt_object_fini(&obj->oo_dt);
766         if (obj->oo_hl_head != NULL)
767                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
768         OBD_FREE_PTR(obj);
769 }
770
771 /*
772  * Concurrency: no concurrent access is possible that late in object
773  * life-cycle.
774  */
775 static void osd_index_fini(struct osd_object *o)
776 {
777         struct iam_container *bag;
778
779         if (o->oo_dir != NULL) {
780                 bag = &o->oo_dir->od_container;
781                 if (o->oo_inode != NULL) {
782                         if (bag->ic_object == o->oo_inode)
783                                 iam_container_fini(bag);
784                 }
785                 OBD_FREE_PTR(o->oo_dir);
786                 o->oo_dir = NULL;
787         }
788 }
789
790 /*
791  * Concurrency: no concurrent access is possible that late in object
792  * life-cycle (for all existing callers, that is. New callers have to provide
793  * their own locking.)
794  */
795 static int osd_inode_unlinked(const struct inode *inode)
796 {
797         return inode->i_nlink == 0;
798 }
799
800 enum {
801         OSD_TXN_OI_DELETE_CREDITS    = 20,
802         OSD_TXN_INODE_DELETE_CREDITS = 20
803 };
804
805 /*
806  * Journal
807  */
808
809 #if OSD_THANDLE_STATS
810 /**
811  * Set time when the handle is allocated
812  */
813 static void osd_th_alloced(struct osd_thandle *oth)
814 {
815         oth->oth_alloced = cfs_time_current();
816 }
817
818 /**
819  * Set time when the handle started
820  */
821 static void osd_th_started(struct osd_thandle *oth)
822 {
823         oth->oth_started = cfs_time_current();
824 }
825
826 /**
827  * Helper function to convert time interval to microseconds packed in
828  * long int.
829  */
830 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
831 {
832         struct timeval val;
833
834         cfs_duration_usec(cfs_time_sub(end, start), &val);
835         return val.tv_sec * 1000000 + val.tv_usec;
836 }
837
838 /**
839  * Check whether the we deal with this handle for too long.
840  */
841 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
842                                 cfs_time_t alloced, cfs_time_t started,
843                                 cfs_time_t closed)
844 {
845         cfs_time_t now = cfs_time_current();
846
847         LASSERT(dev != NULL);
848
849         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
850                             interval_to_usec(alloced, started));
851         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
852                             interval_to_usec(started, closed));
853         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
854                             interval_to_usec(closed, now));
855
856         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
857                 CWARN("transaction handle %p was open for too long: "
858                       "now "CFS_TIME_T" ,"
859                       "alloced "CFS_TIME_T" ,"
860                       "started "CFS_TIME_T" ,"
861                       "closed "CFS_TIME_T"\n",
862                       oth, now, alloced, started, closed);
863                 libcfs_debug_dumpstack(NULL);
864         }
865 }
866
867 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
868 {                                                                       \
869         cfs_time_t __closed = cfs_time_current();                       \
870         cfs_time_t __alloced = oth->oth_alloced;                        \
871         cfs_time_t __started = oth->oth_started;                        \
872                                                                         \
873         expr;                                                           \
874         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
875 }
876
877 #else /* OSD_THANDLE_STATS */
878
879 #define osd_th_alloced(h)                  do {} while(0)
880 #define osd_th_started(h)                  do {} while(0)
881 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
882
883 #endif /* OSD_THANDLE_STATS */
884
885 /*
886  * Concurrency: doesn't access mutable data.
887  */
888 static int osd_param_is_not_sane(const struct osd_device *dev,
889                                  const struct thandle *th)
890 {
891         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
892
893         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
894 }
895
896 /*
897  * Concurrency: shouldn't matter.
898  */
899 static void osd_trans_commit_cb(struct super_block *sb,
900                                 struct ldiskfs_journal_cb_entry *jcb, int error)
901 {
902         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
903         struct thandle     *th  = &oh->ot_super;
904         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
905         struct dt_txn_commit_cb *dcb, *tmp;
906
907         LASSERT(oh->ot_handle == NULL);
908
909         if (error)
910                 CERROR("transaction @0x%p commit error: %d\n", th, error);
911
912         dt_txn_hook_commit(th);
913
914         /* call per-transaction callbacks if any */
915         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
916                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
917                          "commit callback entry: magic=%x name='%s'\n",
918                          dcb->dcb_magic, dcb->dcb_name);
919                 list_del_init(&dcb->dcb_linkage);
920                 dcb->dcb_func(NULL, th, dcb, error);
921         }
922
923         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
924         lu_device_put(lud);
925         th->th_dev = NULL;
926
927         lu_context_exit(&th->th_ctx);
928         lu_context_fini(&th->th_ctx);
929         thandle_put(th);
930 }
931
932 static struct thandle *osd_trans_create(const struct lu_env *env,
933                                         struct dt_device *d)
934 {
935         struct osd_thread_info  *oti = osd_oti_get(env);
936         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
937         struct osd_thandle      *oh;
938         struct thandle          *th;
939         ENTRY;
940
941         /* on pending IO in this thread should left from prev. request */
942         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
943
944         th = ERR_PTR(-ENOMEM);
945         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
946         if (oh != NULL) {
947                 oh->ot_quota_trans = &oti->oti_quota_trans;
948                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
949                 th = &oh->ot_super;
950                 th->th_dev = d;
951                 th->th_result = 0;
952                 th->th_tags = LCT_TX_HANDLE;
953                 oh->ot_credits = 0;
954                 atomic_set(&th->th_refc, 1);
955                 th->th_alloc_size = sizeof(*oh);
956                 oti->oti_dev = osd_dt_dev(d);
957                 INIT_LIST_HEAD(&oh->ot_dcb_list);
958                 osd_th_alloced(oh);
959
960                 memset(oti->oti_declare_ops, 0,
961                                         sizeof(oti->oti_declare_ops));
962                 memset(oti->oti_declare_ops_rb, 0,
963                                         sizeof(oti->oti_declare_ops_rb));
964                 memset(oti->oti_declare_ops_cred, 0,
965                                         sizeof(oti->oti_declare_ops_cred));
966                 oti->oti_rollback = false;
967         }
968         RETURN(th);
969 }
970
971 /*
972  * Concurrency: shouldn't matter.
973  */
974 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
975                     struct thandle *th)
976 {
977         struct osd_thread_info *oti = osd_oti_get(env);
978         struct osd_device  *dev = osd_dt_dev(d);
979         handle_t           *jh;
980         struct osd_thandle *oh;
981         int rc;
982
983         ENTRY;
984
985         LASSERT(current->journal_info == NULL);
986
987         oh = container_of0(th, struct osd_thandle, ot_super);
988         LASSERT(oh != NULL);
989         LASSERT(oh->ot_handle == NULL);
990
991         rc = dt_txn_hook_start(env, d, th);
992         if (rc != 0)
993                 GOTO(out, rc);
994
995         if (unlikely(osd_param_is_not_sane(dev, th))) {
996                 static unsigned long last_printed;
997                 static int last_credits;
998
999                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1000                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1001                       oh->ot_credits,
1002                       osd_journal(dev)->j_max_transaction_buffers);
1003                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1004                       oti->oti_declare_ops[OSD_OT_CREATE],
1005                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1006                       oti->oti_declare_ops[OSD_OT_DESTROY],
1007                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1008                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1009                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1010                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1011                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1012                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1013                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1014                       oti->oti_declare_ops[OSD_OT_WRITE],
1015                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1016                       oti->oti_declare_ops[OSD_OT_PUNCH],
1017                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1018                       oti->oti_declare_ops[OSD_OT_QUOTA],
1019                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1020                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1021                       oti->oti_declare_ops[OSD_OT_INSERT],
1022                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1023                       oti->oti_declare_ops[OSD_OT_DELETE],
1024                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1025                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1026                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1027                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1028                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1029                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1030
1031                 if (last_credits != oh->ot_credits &&
1032                     time_after(jiffies, last_printed + 60 * HZ)) {
1033                         libcfs_debug_dumpstack(NULL);
1034                         last_credits = oh->ot_credits;
1035                         last_printed = jiffies;
1036                 }
1037                 /* XXX Limit the credits to 'max_transaction_buffers', and
1038                  *     let the underlying filesystem to catch the error if
1039                  *     we really need so many credits.
1040                  *
1041                  *     This should be removed when we can calculate the
1042                  *     credits precisely. */
1043                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
1044         }
1045
1046         /*
1047          * XXX temporary stuff. Some abstraction layer should
1048          * be used.
1049          */
1050         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1051         osd_th_started(oh);
1052         if (!IS_ERR(jh)) {
1053                 oh->ot_handle = jh;
1054                 LASSERT(oti->oti_txns == 0);
1055                 lu_context_init(&th->th_ctx, th->th_tags);
1056                 lu_context_enter(&th->th_ctx);
1057
1058                 lu_device_get(&d->dd_lu_dev);
1059                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1060                               "osd-tx", th);
1061                 oti->oti_txns++;
1062                 rc = 0;
1063         } else {
1064                 rc = PTR_ERR(jh);
1065         }
1066 out:
1067         RETURN(rc);
1068 }
1069
1070 static int osd_seq_exists(const struct lu_env *env,
1071                               struct osd_device *osd, obd_seq seq)
1072 {
1073         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1074         struct seq_server_site  *ss = osd_seq_site(osd);
1075         int                     rc;
1076         ENTRY;
1077
1078         LASSERT(ss != NULL);
1079         LASSERT(ss->ss_server_fld != NULL);
1080
1081         rc = osd_fld_lookup(env, osd, seq, range);
1082         if (rc != 0) {
1083                 if (rc != -ENOENT)
1084                         CERROR("%s: can't lookup FLD sequence "LPX64
1085                                ": rc = %d\n", osd_name(osd), seq, rc);
1086                 RETURN(0);
1087         }
1088
1089         RETURN(ss->ss_node_id == range->lsr_index);
1090 }
1091
1092 /*
1093  * Concurrency: shouldn't matter.
1094  */
1095 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1096                           struct thandle *th)
1097 {
1098         int                     rc = 0;
1099         struct osd_thandle     *oh;
1100         struct osd_thread_info *oti = osd_oti_get(env);
1101         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1102         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1103         struct lquota_trans    *qtrans;
1104         ENTRY;
1105
1106         oh = container_of0(th, struct osd_thandle, ot_super);
1107
1108         qtrans = oh->ot_quota_trans;
1109         oh->ot_quota_trans = NULL;
1110
1111         if (oh->ot_handle != NULL) {
1112                 handle_t *hdl = oh->ot_handle;
1113
1114                 /*
1115                  * add commit callback
1116                  * notice we don't do this in osd_trans_start()
1117                  * as underlying transaction can change during truncate
1118                  */
1119                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1120                                          &oh->ot_jcb);
1121
1122                 LASSERT(oti->oti_txns == 1);
1123                 oti->oti_txns--;
1124                 rc = dt_txn_hook_stop(env, th);
1125                 if (rc != 0)
1126                         CERROR("Failure in transaction hook: %d\n", rc);
1127
1128                 /* hook functions might modify th_sync */
1129                 hdl->h_sync = th->th_sync;
1130
1131                 oh->ot_handle = NULL;
1132                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1133                                   rc = ldiskfs_journal_stop(hdl));
1134                 if (rc != 0)
1135                         CERROR("Failure to stop transaction: %d\n", rc);
1136         } else {
1137                 thandle_put(&oh->ot_super);
1138         }
1139
1140         /* inform the quota slave device that the transaction is stopping */
1141         qsd_op_end(env, qsd, qtrans);
1142
1143         /* as we want IO to journal and data IO be concurrent, we don't block
1144          * awaiting data IO completion in osd_do_bio(), instead we wait here
1145          * once transaction is submitted to the journal. all reqular requests
1146          * don't do direct IO (except read/write), thus this wait_event becomes
1147          * no-op for them.
1148          *
1149          * IMPORTANT: we have to wait till any IO submited by the thread is
1150          * completed otherwise iobuf may be corrupted by different request
1151          */
1152         wait_event(iobuf->dr_wait,
1153                        atomic_read(&iobuf->dr_numreqs) == 0);
1154         osd_fini_iobuf(oti->oti_dev, iobuf);
1155         if (!rc)
1156                 rc = iobuf->dr_error;
1157
1158         RETURN(rc);
1159 }
1160
1161 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1162 {
1163         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1164                                                ot_super);
1165
1166         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1167         LASSERT(&dcb->dcb_func != NULL);
1168         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1169
1170         return 0;
1171 }
1172
1173 /*
1174  * Called just before object is freed. Releases all resources except for
1175  * object itself (that is released by osd_object_free()).
1176  *
1177  * Concurrency: no concurrent access is possible that late in object
1178  * life-cycle.
1179  */
1180 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1181 {
1182         struct osd_object *obj   = osd_obj(l);
1183         struct inode      *inode = obj->oo_inode;
1184
1185         LINVRNT(osd_invariant(obj));
1186
1187         /*
1188          * If object is unlinked remove fid->ino mapping from object index.
1189          */
1190
1191         osd_index_fini(obj);
1192         if (inode != NULL) {
1193                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1194                 qid_t                    uid = i_uid_read(inode);
1195                 qid_t                    gid = i_gid_read(inode);
1196
1197                 iput(inode);
1198                 obj->oo_inode = NULL;
1199
1200                 if (qsd != NULL) {
1201                         struct osd_thread_info  *info = osd_oti_get(env);
1202                         struct lquota_id_info   *qi = &info->oti_qi;
1203
1204                         /* Release granted quota to master if necessary */
1205                         qi->lqi_id.qid_uid = uid;
1206                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1207
1208                         qi->lqi_id.qid_uid = gid;
1209                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1210                 }
1211         }
1212 }
1213
1214 /*
1215  * Concurrency: ->loo_object_release() is called under site spin-lock.
1216  */
1217 static void osd_object_release(const struct lu_env *env,
1218                                struct lu_object *l)
1219 {
1220 }
1221
1222 /*
1223  * Concurrency: shouldn't matter.
1224  */
1225 static int osd_object_print(const struct lu_env *env, void *cookie,
1226                             lu_printer_t p, const struct lu_object *l)
1227 {
1228         struct osd_object *o = osd_obj(l);
1229         struct iam_descr  *d;
1230
1231         if (o->oo_dir != NULL)
1232                 d = o->oo_dir->od_container.ic_descr;
1233         else
1234                 d = NULL;
1235         return (*p)(env, cookie,
1236                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1237                     o, o->oo_inode,
1238                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1239                     o->oo_inode ? o->oo_inode->i_generation : 0,
1240                     d ? d->id_ops->id_name : "plain");
1241 }
1242
1243 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1244
1245 /*
1246  * Concurrency: shouldn't matter.
1247  */
1248 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1249                struct obd_statfs *sfs)
1250 {
1251         struct osd_device  *osd = osd_dt_dev(d);
1252         struct super_block *sb = osd_sb(osd);
1253         struct kstatfs     *ksfs;
1254         int result = 0;
1255
1256         if (unlikely(osd->od_mnt == NULL))
1257                 return -EINPROGRESS;
1258
1259         /* osd_lproc.c call this without env, allocate ksfs for that case */
1260         if (unlikely(env == NULL)) {
1261                 OBD_ALLOC_PTR(ksfs);
1262                 if (ksfs == NULL)
1263                         return -ENOMEM;
1264         } else {
1265                 ksfs = &osd_oti_get(env)->oti_ksfs;
1266         }
1267
1268         spin_lock(&osd->od_osfs_lock);
1269         result = sb->s_op->statfs(sb->s_root, ksfs);
1270         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1271                 statfs_pack(sfs, ksfs);
1272                 if (sb->s_flags & MS_RDONLY)
1273                         sfs->os_state = OS_STATE_READONLY;
1274         }
1275
1276         spin_unlock(&osd->od_osfs_lock);
1277
1278         if (unlikely(env == NULL))
1279                 OBD_FREE_PTR(ksfs);
1280
1281         /* Reserve a small amount of space for local objects like last_rcvd,
1282          * llog, quota files, ... */
1283         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1284                 sfs->os_bavail = 0;
1285         } else {
1286                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1287                 /** Take out metadata overhead for indirect blocks */
1288                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1289         }
1290
1291         return result;
1292 }
1293
1294 /**
1295  * Estimate space needed for file creations. We assume the largest filename
1296  * which is 2^64 - 1, hence a filename of 20 chars.
1297  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1298  */
1299 #ifdef __LDISKFS_DIR_REC_LEN
1300 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1301 #else
1302 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1303 #endif
1304
1305 /*
1306  * Concurrency: doesn't access mutable data.
1307  */
1308 static void osd_conf_get(const struct lu_env *env,
1309                          const struct dt_device *dev,
1310                          struct dt_device_param *param)
1311 {
1312         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1313         int                ea_overhead;
1314
1315         /*
1316          * XXX should be taken from not-yet-existing fs abstraction layer.
1317          */
1318         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1319         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1320         param->ddp_block_shift  = sb->s_blocksize_bits;
1321         param->ddp_mount_type     = LDD_MT_LDISKFS;
1322         param->ddp_maxbytes       = sb->s_maxbytes;
1323         /* Overhead estimate should be fairly accurate, so we really take a tiny
1324          * error margin which also avoids fragmenting the filesystem too much */
1325         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1326         /* inode are statically allocated, so per-inode space consumption
1327          * is the space consumed by the directory entry */
1328         param->ddp_inodespace     = PER_OBJ_USAGE;
1329         /* per-fragment overhead to be used by the client code */
1330         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1331         param->ddp_mntopts      = 0;
1332         if (test_opt(sb, XATTR_USER))
1333                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1334         if (test_opt(sb, POSIX_ACL))
1335                 param->ddp_mntopts |= MNTOPT_ACL;
1336
1337         /* LOD might calculate the max stripe count based on max_ea_size,
1338          * so we need take account in the overhead as well,
1339          * xattr_header + magic + xattr_entry_head */
1340         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1341                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1342
1343 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1344         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1345                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1346                                                                 ea_overhead;
1347         else
1348 #endif
1349                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1350 }
1351
1352 /*
1353  * Concurrency: shouldn't matter.
1354  */
1355 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1356 {
1357         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1358         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1359 }
1360
1361 /**
1362  * Start commit for OSD device.
1363  *
1364  * An implementation of dt_commit_async method for OSD device.
1365  * Asychronously starts underlayng fs sync and thereby a transaction
1366  * commit.
1367  *
1368  * \param env environment
1369  * \param d dt device
1370  *
1371  * \see dt_device_operations
1372  */
1373 static int osd_commit_async(const struct lu_env *env,
1374                             struct dt_device *d)
1375 {
1376         struct super_block *s = osd_sb(osd_dt_dev(d));
1377         ENTRY;
1378
1379         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1380         RETURN(s->s_op->sync_fs(s, 0));
1381 }
1382
1383 /*
1384  * Concurrency: shouldn't matter.
1385  */
1386
1387 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1388 {
1389         struct super_block *sb = osd_sb(osd_dt_dev(d));
1390         struct block_device *dev = sb->s_bdev;
1391 #ifdef HAVE_DEV_SET_RDONLY
1392         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1393         int rc = 0;
1394 #else
1395         int rc = -EOPNOTSUPP;
1396 #endif
1397         ENTRY;
1398
1399 #ifdef HAVE_DEV_SET_RDONLY
1400         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1401
1402         if (jdev && (jdev != dev)) {
1403                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1404                        (long)jdev);
1405                 dev_set_rdonly(jdev);
1406         }
1407         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1408         dev_set_rdonly(dev);
1409 #else
1410         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1411                osd_dt_dev(d)->od_svname, (long)dev, rc);
1412 #endif
1413         RETURN(rc);
1414 }
1415
1416 /*
1417  * Concurrency: serialization provided by callers.
1418  */
1419 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1420                               int mode, unsigned long timeout, __u32 alg,
1421                               struct lustre_capa_key *keys)
1422 {
1423         struct osd_device *dev = osd_dt_dev(d);
1424         ENTRY;
1425
1426         dev->od_fl_capa = mode;
1427         dev->od_capa_timeout = timeout;
1428         dev->od_capa_alg = alg;
1429         dev->od_capa_keys = keys;
1430         RETURN(0);
1431 }
1432
1433 /**
1434  * Note: we do not count into QUOTA here.
1435  * If we mount with --data_journal we may need more.
1436  */
1437 const int osd_dto_credits_noquota[DTO_NR] = {
1438         /**
1439          * Insert.
1440          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1441          * SINGLEDATA_TRANS_BLOCKS(8)
1442          * XXX Note: maybe iam need more, since iam have more level than
1443          *           EXT3 htree.
1444          */
1445         [DTO_INDEX_INSERT]  = 16,
1446         /**
1447          * Delete
1448          * just modify a single entry, probably merge few within a block
1449          */
1450         [DTO_INDEX_DELETE]  = 1,
1451         /**
1452          * Used for OI scrub
1453          */
1454         [DTO_INDEX_UPDATE]  = 16,
1455         /**
1456          * 4(inode, inode bits, groups, GDT)
1457          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1458          */
1459         [DTO_OBJECT_CREATE] = 4,
1460         /**
1461          * 4(inode, inode bits, groups, GDT)
1462          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1463          */
1464         [DTO_OBJECT_DELETE] = 4,
1465         /**
1466          * Attr set credits (inode)
1467          */
1468         [DTO_ATTR_SET_BASE] = 1,
1469         /**
1470          * Xattr set. The same as xattr of EXT3.
1471          * DATA_TRANS_BLOCKS(14)
1472          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1473          * are also counted in. Do not know why?
1474          */
1475         [DTO_XATTR_SET]     = 14,
1476         /**
1477          * credits for inode change during write.
1478          */
1479         [DTO_WRITE_BASE]    = 3,
1480         /**
1481          * credits for single block write.
1482          */
1483         [DTO_WRITE_BLOCK]   = 14,
1484         /**
1485          * Attr set credits for chown.
1486          * This is extra credits for setattr, and it is null without quota
1487          */
1488         [DTO_ATTR_SET_CHOWN] = 0
1489 };
1490
1491 static const struct dt_device_operations osd_dt_ops = {
1492         .dt_root_get       = osd_root_get,
1493         .dt_statfs         = osd_statfs,
1494         .dt_trans_create   = osd_trans_create,
1495         .dt_trans_start    = osd_trans_start,
1496         .dt_trans_stop     = osd_trans_stop,
1497         .dt_trans_cb_add   = osd_trans_cb_add,
1498         .dt_conf_get       = osd_conf_get,
1499         .dt_sync           = osd_sync,
1500         .dt_ro             = osd_ro,
1501         .dt_commit_async   = osd_commit_async,
1502         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1503 };
1504
1505 static void osd_object_read_lock(const struct lu_env *env,
1506                                  struct dt_object *dt, unsigned role)
1507 {
1508         struct osd_object *obj = osd_dt_obj(dt);
1509         struct osd_thread_info *oti = osd_oti_get(env);
1510
1511         LINVRNT(osd_invariant(obj));
1512
1513         LASSERT(obj->oo_owner != env);
1514         down_read_nested(&obj->oo_sem, role);
1515
1516         LASSERT(obj->oo_owner == NULL);
1517         oti->oti_r_locks++;
1518 }
1519
1520 static void osd_object_write_lock(const struct lu_env *env,
1521                                   struct dt_object *dt, unsigned role)
1522 {
1523         struct osd_object *obj = osd_dt_obj(dt);
1524         struct osd_thread_info *oti = osd_oti_get(env);
1525
1526         LINVRNT(osd_invariant(obj));
1527
1528         LASSERT(obj->oo_owner != env);
1529         down_write_nested(&obj->oo_sem, role);
1530
1531         LASSERT(obj->oo_owner == NULL);
1532         obj->oo_owner = env;
1533         oti->oti_w_locks++;
1534 }
1535
1536 static void osd_object_read_unlock(const struct lu_env *env,
1537                                    struct dt_object *dt)
1538 {
1539         struct osd_object *obj = osd_dt_obj(dt);
1540         struct osd_thread_info *oti = osd_oti_get(env);
1541
1542         LINVRNT(osd_invariant(obj));
1543
1544         LASSERT(oti->oti_r_locks > 0);
1545         oti->oti_r_locks--;
1546         up_read(&obj->oo_sem);
1547 }
1548
1549 static void osd_object_write_unlock(const struct lu_env *env,
1550                                     struct dt_object *dt)
1551 {
1552         struct osd_object *obj = osd_dt_obj(dt);
1553         struct osd_thread_info *oti = osd_oti_get(env);
1554
1555         LINVRNT(osd_invariant(obj));
1556
1557         LASSERT(obj->oo_owner == env);
1558         LASSERT(oti->oti_w_locks > 0);
1559         oti->oti_w_locks--;
1560         obj->oo_owner = NULL;
1561         up_write(&obj->oo_sem);
1562 }
1563
1564 static int osd_object_write_locked(const struct lu_env *env,
1565                                    struct dt_object *dt)
1566 {
1567         struct osd_object *obj = osd_dt_obj(dt);
1568
1569         LINVRNT(osd_invariant(obj));
1570
1571         return obj->oo_owner == env;
1572 }
1573
1574 static int capa_is_sane(const struct lu_env *env,
1575                         struct osd_device *dev,
1576                         struct lustre_capa *capa,
1577                         struct lustre_capa_key *keys)
1578 {
1579         struct osd_thread_info *oti = osd_oti_get(env);
1580         struct lustre_capa *tcapa = &oti->oti_capa;
1581         struct obd_capa *oc;
1582         int i, rc = 0;
1583         ENTRY;
1584
1585         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1586         if (oc) {
1587                 if (capa_is_expired(oc)) {
1588                         DEBUG_CAPA(D_ERROR, capa, "expired");
1589                         rc = -ESTALE;
1590                 }
1591                 capa_put(oc);
1592                 RETURN(rc);
1593         }
1594
1595         if (capa_is_expired_sec(capa)) {
1596                 DEBUG_CAPA(D_ERROR, capa, "expired");
1597                 RETURN(-ESTALE);
1598         }
1599
1600         spin_lock(&capa_lock);
1601         for (i = 0; i < 2; i++) {
1602                 if (keys[i].lk_keyid == capa->lc_keyid) {
1603                         oti->oti_capa_key = keys[i];
1604                         break;
1605                 }
1606         }
1607         spin_unlock(&capa_lock);
1608
1609         if (i == 2) {
1610                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1611                 RETURN(-ESTALE);
1612         }
1613
1614         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1615         if (rc)
1616                 RETURN(rc);
1617
1618         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1619                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1620                 RETURN(-EACCES);
1621         }
1622
1623         oc = capa_add(dev->od_capa_hash, capa);
1624         capa_put(oc);
1625
1626         RETURN(0);
1627 }
1628
1629 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1630                     struct lustre_capa *capa, __u64 opc)
1631 {
1632         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1633         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1634         struct lu_capainfo *lci;
1635         int rc;
1636
1637         if (!osd->od_fl_capa)
1638                 return 0;
1639
1640         if (capa == BYPASS_CAPA)
1641                 return 0;
1642
1643         lci = lu_capainfo_get(env);
1644         if (unlikely(lci == NULL))
1645                 return 0;
1646
1647         if (lci->lci_auth == LC_ID_NONE)
1648                 return 0;
1649
1650         if (capa == NULL) {
1651                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1652                        osd_name(osd), PFID(fid), -EACCES);
1653                 return -EACCES;
1654         }
1655
1656         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1657                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1658                            PFID(fid));
1659                 return -EACCES;
1660         }
1661
1662         if (!capa_opc_supported(capa, opc)) {
1663                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1664                 return -EACCES;
1665         }
1666
1667         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1668         if (rc != 0) {
1669                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1670                 return -EACCES;
1671         }
1672
1673         return 0;
1674 }
1675
1676 static struct timespec *osd_inode_time(const struct lu_env *env,
1677                                        struct inode *inode, __u64 seconds)
1678 {
1679         struct osd_thread_info  *oti = osd_oti_get(env);
1680         struct timespec         *t   = &oti->oti_time;
1681
1682         t->tv_sec = seconds;
1683         t->tv_nsec = 0;
1684         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1685         return t;
1686 }
1687
1688
1689 static void osd_inode_getattr(const struct lu_env *env,
1690                               struct inode *inode, struct lu_attr *attr)
1691 {
1692         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1693                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1694                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1695                            LA_TYPE;
1696
1697         attr->la_atime   = LTIME_S(inode->i_atime);
1698         attr->la_mtime   = LTIME_S(inode->i_mtime);
1699         attr->la_ctime   = LTIME_S(inode->i_ctime);
1700         attr->la_mode    = inode->i_mode;
1701         attr->la_size    = i_size_read(inode);
1702         attr->la_blocks  = inode->i_blocks;
1703         attr->la_uid     = i_uid_read(inode);
1704         attr->la_gid     = i_gid_read(inode);
1705         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1706         attr->la_nlink   = inode->i_nlink;
1707         attr->la_rdev    = inode->i_rdev;
1708         attr->la_blksize = 1 << inode->i_blkbits;
1709         attr->la_blkbits = inode->i_blkbits;
1710 }
1711
1712 static int osd_attr_get(const struct lu_env *env,
1713                         struct dt_object *dt,
1714                         struct lu_attr *attr,
1715                         struct lustre_capa *capa)
1716 {
1717         struct osd_object *obj = osd_dt_obj(dt);
1718
1719         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1720         LINVRNT(osd_invariant(obj));
1721
1722         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1723                 return -EACCES;
1724
1725         spin_lock(&obj->oo_guard);
1726         osd_inode_getattr(env, obj->oo_inode, attr);
1727         spin_unlock(&obj->oo_guard);
1728         return 0;
1729 }
1730
1731 static int osd_declare_attr_set(const struct lu_env *env,
1732                                 struct dt_object *dt,
1733                                 const struct lu_attr *attr,
1734                                 struct thandle *handle)
1735 {
1736         struct osd_thandle     *oh;
1737         struct osd_object      *obj;
1738         struct osd_thread_info *info = osd_oti_get(env);
1739         struct lquota_id_info  *qi = &info->oti_qi;
1740         qid_t                   uid;
1741         qid_t                   gid;
1742         long long               bspace;
1743         int                     rc = 0;
1744         bool                    allocated;
1745         ENTRY;
1746
1747         LASSERT(dt != NULL);
1748         LASSERT(handle != NULL);
1749
1750         obj = osd_dt_obj(dt);
1751         LASSERT(osd_invariant(obj));
1752
1753         oh = container_of0(handle, struct osd_thandle, ot_super);
1754         LASSERT(oh->ot_handle == NULL);
1755
1756         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1757                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1758
1759         if (attr == NULL || obj->oo_inode == NULL)
1760                 RETURN(rc);
1761
1762         bspace   = obj->oo_inode->i_blocks;
1763         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1764         bspace   = toqb(bspace);
1765
1766         /* Changing ownership is always preformed by super user, it should not
1767          * fail with EDQUOT.
1768          *
1769          * We still need to call the osd_declare_qid() to calculate the journal
1770          * credits for updating quota accounting files and to trigger quota
1771          * space adjustment once the operation is completed.*/
1772         if ((attr->la_valid & LA_UID) != 0 &&
1773              attr->la_uid != (uid = i_uid_read(obj->oo_inode))) {
1774                 qi->lqi_type = USRQUOTA;
1775
1776                 /* inode accounting */
1777                 qi->lqi_is_blk = false;
1778
1779                 /* one more inode for the new owner ... */
1780                 qi->lqi_id.qid_uid = attr->la_uid;
1781                 qi->lqi_space      = 1;
1782                 allocated = (attr->la_uid == 0) ? true : false;
1783                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1784                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1785                         rc = 0;
1786                 if (rc)
1787                         RETURN(rc);
1788
1789                 /* and one less inode for the current uid */
1790                 qi->lqi_id.qid_uid = uid;
1791                 qi->lqi_space      = -1;
1792                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1793                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1794                         rc = 0;
1795                 if (rc)
1796                         RETURN(rc);
1797
1798                 /* block accounting */
1799                 qi->lqi_is_blk = true;
1800
1801                 /* more blocks for the new owner ... */
1802                 qi->lqi_id.qid_uid = attr->la_uid;
1803                 qi->lqi_space      = bspace;
1804                 allocated = (attr->la_uid == 0) ? true : false;
1805                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1806                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1807                         rc = 0;
1808                 if (rc)
1809                         RETURN(rc);
1810
1811                 /* and finally less blocks for the current owner */
1812                 qi->lqi_id.qid_uid = uid;
1813                 qi->lqi_space      = -bspace;
1814                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1815                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1816                         rc = 0;
1817                 if (rc)
1818                         RETURN(rc);
1819         }
1820
1821         if (attr->la_valid & LA_GID &&
1822             attr->la_gid != (gid = i_gid_read(obj->oo_inode))) {
1823                 qi->lqi_type = GRPQUOTA;
1824
1825                 /* inode accounting */
1826                 qi->lqi_is_blk = false;
1827
1828                 /* one more inode for the new group owner ... */
1829                 qi->lqi_id.qid_gid = attr->la_gid;
1830                 qi->lqi_space      = 1;
1831                 allocated = (attr->la_gid == 0) ? true : false;
1832                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1833                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1834                         rc = 0;
1835                 if (rc)
1836                         RETURN(rc);
1837
1838                 /* and one less inode for the current gid */
1839                 qi->lqi_id.qid_gid = gid;
1840                 qi->lqi_space      = -1;
1841                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1842                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1843                         rc = 0;
1844                 if (rc)
1845                         RETURN(rc);
1846
1847                 /* block accounting */
1848                 qi->lqi_is_blk = true;
1849
1850                 /* more blocks for the new owner ... */
1851                 qi->lqi_id.qid_gid = attr->la_gid;
1852                 qi->lqi_space      = bspace;
1853                 allocated = (attr->la_gid == 0) ? true : false;
1854                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1855                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1856                         rc = 0;
1857                 if (rc)
1858                         RETURN(rc);
1859
1860                 /* and finally less blocks for the current owner */
1861                 qi->lqi_id.qid_gid = gid;
1862                 qi->lqi_space      = -bspace;
1863                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1864                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1865                         rc = 0;
1866                 if (rc)
1867                         RETURN(rc);
1868         }
1869
1870         RETURN(rc);
1871 }
1872
1873 static int osd_inode_setattr(const struct lu_env *env,
1874                              struct inode *inode, const struct lu_attr *attr)
1875 {
1876         __u64 bits = attr->la_valid;
1877
1878         /* Only allow set size for regular file */
1879         if (!S_ISREG(inode->i_mode))
1880                 bits &= ~(LA_SIZE | LA_BLOCKS);
1881
1882         if (bits == 0)
1883                 return 0;
1884
1885         if (bits & LA_ATIME)
1886                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1887         if (bits & LA_CTIME)
1888                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1889         if (bits & LA_MTIME)
1890                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1891         if (bits & LA_SIZE) {
1892                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1893                 i_size_write(inode, attr->la_size);
1894         }
1895
1896 #if 0
1897         /* OSD should not change "i_blocks" which is used by quota.
1898          * "i_blocks" should be changed by ldiskfs only. */
1899         if (bits & LA_BLOCKS)
1900                 inode->i_blocks = attr->la_blocks;
1901 #endif
1902         if (bits & LA_MODE)
1903                 inode->i_mode = (inode->i_mode & S_IFMT) |
1904                                 (attr->la_mode & ~S_IFMT);
1905         if (bits & LA_UID)
1906                 i_uid_write(inode, attr->la_uid);
1907         if (bits & LA_GID)
1908                 i_gid_write(inode, attr->la_gid);
1909         if (bits & LA_NLINK)
1910                 set_nlink(inode, attr->la_nlink);
1911         if (bits & LA_RDEV)
1912                 inode->i_rdev = attr->la_rdev;
1913
1914         if (bits & LA_FLAGS) {
1915                 /* always keep S_NOCMTIME */
1916                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1917                                  S_NOCMTIME;
1918         }
1919         return 0;
1920 }
1921
1922 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1923 {
1924         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1925             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1926                 struct iattr    iattr;
1927                 int             rc;
1928
1929                 iattr.ia_valid = 0;
1930                 if (attr->la_valid & LA_UID)
1931                         iattr.ia_valid |= ATTR_UID;
1932                 if (attr->la_valid & LA_GID)
1933                         iattr.ia_valid |= ATTR_GID;
1934                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1935                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1936
1937                 rc = ll_vfs_dq_transfer(inode, &iattr);
1938                 if (rc) {
1939                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1940                                "enforcement enabled on the ldiskfs filesystem?",
1941                                inode->i_sb->s_id, rc);
1942                         return rc;
1943                 }
1944         }
1945         return 0;
1946 }
1947
1948 static int osd_attr_set(const struct lu_env *env,
1949                         struct dt_object *dt,
1950                         const struct lu_attr *attr,
1951                         struct thandle *handle,
1952                         struct lustre_capa *capa)
1953 {
1954         struct osd_object *obj = osd_dt_obj(dt);
1955         struct inode      *inode;
1956         int rc;
1957
1958         LASSERT(handle != NULL);
1959         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1960         LASSERT(osd_invariant(obj));
1961
1962         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1963                 return -EACCES;
1964
1965         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1966
1967         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1968                 struct osd_thread_info  *oti  = osd_oti_get(env);
1969                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1970                 struct lu_fid           *fid1 = &oti->oti_fid;
1971                 struct osd_inode_id     *id   = &oti->oti_id;
1972                 struct iam_path_descr   *ipd;
1973                 struct iam_container    *bag;
1974                 struct osd_thandle      *oh;
1975                 int                      rc;
1976
1977                 fid_cpu_to_be(fid1, fid0);
1978                 memset(id, 1, sizeof(*id));
1979                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1980                                   fid0)->oi_dir.od_container;
1981                 ipd = osd_idx_ipd_get(env, bag);
1982                 if (unlikely(ipd == NULL))
1983                         RETURN(-ENOMEM);
1984
1985                 oh = container_of0(handle, struct osd_thandle, ot_super);
1986                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1987                                 (const struct iam_rec *)id, ipd);
1988                 osd_ipd_put(env, bag, ipd);
1989                 return(rc > 0 ? 0 : rc);
1990         }
1991
1992         inode = obj->oo_inode;
1993         ll_vfs_dq_init(inode);
1994
1995         rc = osd_quota_transfer(inode, attr);
1996         if (rc)
1997                 return rc;
1998
1999         spin_lock(&obj->oo_guard);
2000         rc = osd_inode_setattr(env, inode, attr);
2001         spin_unlock(&obj->oo_guard);
2002
2003         if (!rc)
2004                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2005         return rc;
2006 }
2007
2008 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2009                                     struct osd_object *obj,
2010                                     const char *name, const int namelen)
2011 {
2012         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2013 }
2014
2015 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2016                       umode_t mode, struct dt_allocation_hint *hint,
2017                       struct thandle *th)
2018 {
2019         int result;
2020         struct osd_device  *osd = osd_obj2dev(obj);
2021         struct osd_thandle *oth;
2022         struct dt_object   *parent = NULL;
2023         struct inode       *inode;
2024
2025         LINVRNT(osd_invariant(obj));
2026         LASSERT(obj->oo_inode == NULL);
2027         LASSERT(obj->oo_hl_head == NULL);
2028
2029         if (S_ISDIR(mode) && ldiskfs_pdo) {
2030                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2031                 if (obj->oo_hl_head == NULL)
2032                         return -ENOMEM;
2033         }
2034
2035         oth = container_of(th, struct osd_thandle, ot_super);
2036         LASSERT(oth->ot_handle->h_transaction != NULL);
2037
2038         if (hint && hint->dah_parent)
2039                 parent = hint->dah_parent;
2040
2041         inode = ldiskfs_create_inode(oth->ot_handle,
2042                                      parent ? osd_dt_obj(parent)->oo_inode :
2043                                               osd_sb(osd)->s_root->d_inode,
2044                                      mode);
2045         if (!IS_ERR(inode)) {
2046                 /* Do not update file c/mtime in ldiskfs.
2047                  * NB: don't need any lock because no contention at this
2048                  * early stage */
2049                 inode->i_flags |= S_NOCMTIME;
2050
2051                 /* For new created object, it must be consistent,
2052                  * and it is unnecessary to scrub against it. */
2053                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2054                 obj->oo_inode = inode;
2055                 result = 0;
2056         } else {
2057                 if (obj->oo_hl_head != NULL) {
2058                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2059                         obj->oo_hl_head = NULL;
2060                 }
2061                 result = PTR_ERR(inode);
2062         }
2063         LINVRNT(osd_invariant(obj));
2064         return result;
2065 }
2066
2067 enum {
2068         OSD_NAME_LEN = 255
2069 };
2070
2071 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2072                      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         struct osd_thandle *oth;
2079         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2080
2081         LASSERT(S_ISDIR(attr->la_mode));
2082
2083         oth = container_of(th, struct osd_thandle, ot_super);
2084         LASSERT(oth->ot_handle->h_transaction != NULL);
2085         result = osd_mkfile(info, obj, mode, hint, th);
2086
2087         return result;
2088 }
2089
2090 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2091                         struct lu_attr *attr,
2092                         struct dt_allocation_hint *hint,
2093                         struct dt_object_format *dof,
2094                         struct thandle *th)
2095 {
2096         int result;
2097         struct osd_thandle *oth;
2098         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2099
2100         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2101
2102         LASSERT(S_ISREG(attr->la_mode));
2103
2104         oth = container_of(th, struct osd_thandle, ot_super);
2105         LASSERT(oth->ot_handle->h_transaction != NULL);
2106
2107         result = osd_mkfile(info, obj, mode, hint, th);
2108         if (result == 0) {
2109                 LASSERT(obj->oo_inode != NULL);
2110                 if (feat->dif_flags & DT_IND_VARKEY)
2111                         result = iam_lvar_create(obj->oo_inode,
2112                                                  feat->dif_keysize_max,
2113                                                  feat->dif_ptrsize,
2114                                                  feat->dif_recsize_max,
2115                                                  oth->ot_handle);
2116                 else
2117                         result = iam_lfix_create(obj->oo_inode,
2118                                                  feat->dif_keysize_max,
2119                                                  feat->dif_ptrsize,
2120                                                  feat->dif_recsize_max,
2121                                                  oth->ot_handle);
2122
2123         }
2124         return result;
2125 }
2126
2127 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2128                      struct lu_attr *attr,
2129                      struct dt_allocation_hint *hint,
2130                      struct dt_object_format *dof,
2131                      struct thandle *th)
2132 {
2133         LASSERT(S_ISREG(attr->la_mode));
2134         return osd_mkfile(info, obj, (attr->la_mode &
2135                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2136 }
2137
2138 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2139                      struct lu_attr *attr,
2140                      struct dt_allocation_hint *hint,
2141                      struct dt_object_format *dof,
2142                      struct thandle *th)
2143 {
2144         LASSERT(S_ISLNK(attr->la_mode));
2145         return osd_mkfile(info, obj, (attr->la_mode &
2146                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2147 }
2148
2149 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2150                      struct lu_attr *attr,
2151                      struct dt_allocation_hint *hint,
2152                      struct dt_object_format *dof,
2153                      struct thandle *th)
2154 {
2155         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2156         int result;
2157
2158         LINVRNT(osd_invariant(obj));
2159         LASSERT(obj->oo_inode == NULL);
2160         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2161                 S_ISFIFO(mode) || S_ISSOCK(mode));
2162
2163         result = osd_mkfile(info, obj, mode, hint, th);
2164         if (result == 0) {
2165                 LASSERT(obj->oo_inode != NULL);
2166                 /*
2167                  * This inode should be marked dirty for i_rdev.  Currently
2168                  * that is done in the osd_attr_init().
2169                  */
2170                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2171                                    attr->la_rdev);
2172         }
2173         LINVRNT(osd_invariant(obj));
2174         return result;
2175 }
2176
2177 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2178                               struct lu_attr *,
2179                               struct dt_allocation_hint *hint,
2180                               struct dt_object_format *dof,
2181                               struct thandle *);
2182
2183 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2184 {
2185         osd_obj_type_f result;
2186
2187         switch (type) {
2188         case DFT_DIR:
2189                 result = osd_mkdir;
2190                 break;
2191         case DFT_REGULAR:
2192                 result = osd_mkreg;
2193                 break;
2194         case DFT_SYM:
2195                 result = osd_mksym;
2196                 break;
2197         case DFT_NODE:
2198                 result = osd_mknod;
2199                 break;
2200         case DFT_INDEX:
2201                 result = osd_mk_index;
2202                 break;
2203
2204         default:
2205                 LBUG();
2206                 break;
2207         }
2208         return result;
2209 }
2210
2211
2212 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2213                         struct dt_object *parent, struct dt_object *child,
2214                         umode_t child_mode)
2215 {
2216         LASSERT(ah);
2217
2218         ah->dah_parent = parent;
2219         ah->dah_mode = child_mode;
2220 }
2221
2222 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2223                           struct lu_attr *attr, struct dt_object_format *dof)
2224 {
2225         struct inode   *inode = obj->oo_inode;
2226         __u64           valid = attr->la_valid;
2227         int             result;
2228
2229         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2230
2231         if (dof->dof_type != DFT_NODE)
2232                 attr->la_valid &= ~LA_RDEV;
2233         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2234                 attr->la_valid &= ~LA_ATIME;
2235         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2236                 attr->la_valid &= ~LA_CTIME;
2237         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2238                 attr->la_valid &= ~LA_MTIME;
2239
2240         result = osd_quota_transfer(inode, attr);
2241         if (result)
2242                 return;
2243
2244         if (attr->la_valid != 0) {
2245                 result = osd_inode_setattr(info->oti_env, inode, attr);
2246                 /*
2247                  * The osd_inode_setattr() should always succeed here.  The
2248                  * only error that could be returned is EDQUOT when we are
2249                  * trying to change the UID or GID of the inode. However, this
2250                  * should not happen since quota enforcement is no longer
2251                  * enabled on ldiskfs (lquota takes care of it).
2252                  */
2253                 LASSERTF(result == 0, "%d", result);
2254                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2255         }
2256
2257         attr->la_valid = valid;
2258 }
2259
2260 /**
2261  * Helper function for osd_object_create()
2262  *
2263  * \retval 0, on success
2264  */
2265 static int __osd_object_create(struct osd_thread_info *info,
2266                                struct osd_object *obj, struct lu_attr *attr,
2267                                struct dt_allocation_hint *hint,
2268                                struct dt_object_format *dof,
2269                                struct thandle *th)
2270 {
2271         int     result;
2272         __u32   umask;
2273
2274         /* we drop umask so that permissions we pass are not affected */
2275         umask = current->fs->umask;
2276         current->fs->umask = 0;
2277
2278         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2279                                                   th);
2280         if (result == 0) {
2281                 osd_attr_init(info, obj, attr, dof);
2282                 osd_object_init0(obj);
2283                 /* bz 24037 */
2284                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2285                         unlock_new_inode(obj->oo_inode);
2286         }
2287
2288         /* restore previous umask value */
2289         current->fs->umask = umask;
2290
2291         return result;
2292 }
2293
2294 /**
2295  * Helper function for osd_object_create()
2296  *
2297  * \retval 0, on success
2298  */
2299 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2300                            const struct lu_fid *fid, struct thandle *th)
2301 {
2302         struct osd_thread_info *info = osd_oti_get(env);
2303         struct osd_inode_id    *id   = &info->oti_id;
2304         struct osd_device      *osd  = osd_obj2dev(obj);
2305         struct osd_thandle     *oh;
2306
2307         LASSERT(obj->oo_inode != NULL);
2308
2309         oh = container_of0(th, struct osd_thandle, ot_super);
2310         LASSERT(oh->ot_handle);
2311
2312         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2313         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2314 }
2315
2316 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2317                    obd_seq seq, struct lu_seq_range *range)
2318 {
2319         struct seq_server_site  *ss = osd_seq_site(osd);
2320
2321         if (fid_seq_is_idif(seq)) {
2322                 fld_range_set_ost(range);
2323                 range->lsr_index = idif_ost_idx(seq);
2324                 return 0;
2325         }
2326
2327         if (!fid_seq_in_fldb(seq)) {
2328                 fld_range_set_mdt(range);
2329                 if (ss != NULL)
2330                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2331                          * at all */
2332                         range->lsr_index = ss->ss_node_id;
2333                 return 0;
2334         }
2335
2336         LASSERT(ss != NULL);
2337         fld_range_set_any(range);
2338         /* OSD will only do local fld lookup */
2339         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2340 }
2341
2342 /*
2343  * Concurrency: no external locking is necessary.
2344  */
2345 static int osd_declare_object_create(const struct lu_env *env,
2346                                      struct dt_object *dt,
2347                                      struct lu_attr *attr,
2348                                      struct dt_allocation_hint *hint,
2349                                      struct dt_object_format *dof,
2350                                      struct thandle *handle)
2351 {
2352         struct osd_thandle      *oh;
2353         int                      rc;
2354         ENTRY;
2355
2356         LASSERT(handle != NULL);
2357
2358         oh = container_of0(handle, struct osd_thandle, ot_super);
2359         LASSERT(oh->ot_handle == NULL);
2360
2361         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2362                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2363         /* Reuse idle OI block may cause additional one OI block
2364          * to be changed. */
2365         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2366                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2367
2368         /* If this is directory, then we expect . and .. to be inserted as
2369          * well. The one directory block always needs to be created for the
2370          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2371          * block), there is no danger of needing a tree for the first block.
2372          */
2373         if (attr && S_ISDIR(attr->la_mode)) {
2374                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2375                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2376                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2377         }
2378
2379         if (!attr)
2380                 RETURN(0);
2381
2382         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2383                                    false, false, NULL, false);
2384         if (rc != 0)
2385                 RETURN(rc);
2386
2387         RETURN(rc);
2388 }
2389
2390 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2391                              struct lu_attr *attr,
2392                              struct dt_allocation_hint *hint,
2393                              struct dt_object_format *dof,
2394                              struct thandle *th)
2395 {
2396         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2397         struct osd_object      *obj    = osd_dt_obj(dt);
2398         struct osd_thread_info *info   = osd_oti_get(env);
2399         int result;
2400
2401         ENTRY;
2402
2403         LINVRNT(osd_invariant(obj));
2404         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2405         LASSERT(osd_write_locked(env, obj));
2406         LASSERT(th != NULL);
2407
2408         if (unlikely(fid_is_acct(fid)))
2409                 /* Quota files can't be created from the kernel any more,
2410                  * 'tune2fs -O quota' will take care of creating them */
2411                 RETURN(-EPERM);
2412
2413         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2414         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2415
2416         result = __osd_object_create(info, obj, attr, hint, dof, th);
2417         if (result == 0)
2418                 result = __osd_oi_insert(env, obj, fid, th);
2419
2420         LASSERT(ergo(result == 0,
2421                      dt_object_exists(dt) && !dt_object_remote(dt)));
2422
2423         LASSERT(osd_invariant(obj));
2424         RETURN(result);
2425 }
2426
2427 /**
2428  * Called to destroy on-disk representation of the object
2429  *
2430  * Concurrency: must be locked
2431  */
2432 static int osd_declare_object_destroy(const struct lu_env *env,
2433                                       struct dt_object *dt,
2434                                       struct thandle *th)
2435 {
2436         struct osd_object  *obj = osd_dt_obj(dt);
2437         struct inode       *inode = obj->oo_inode;
2438         struct osd_thandle *oh;
2439         int                 rc;
2440         ENTRY;
2441
2442         oh = container_of0(th, struct osd_thandle, ot_super);
2443         LASSERT(oh->ot_handle == NULL);
2444         LASSERT(inode);
2445
2446         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2447                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2448         /* Recycle idle OI leaf may cause additional three OI blocks
2449          * to be changed. */
2450         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2451                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2452         /* one less inode */
2453         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2454                                    -1, oh, false, true, NULL, false);
2455         if (rc)
2456                 RETURN(rc);
2457         /* data to be truncated */
2458         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2459                                    0, oh, true, true, NULL, false);
2460         RETURN(rc);
2461 }
2462
2463 static int osd_object_destroy(const struct lu_env *env,
2464                               struct dt_object *dt,
2465                               struct thandle *th)
2466 {
2467         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2468         struct osd_object      *obj = osd_dt_obj(dt);
2469         struct inode           *inode = obj->oo_inode;
2470         struct osd_device      *osd = osd_obj2dev(obj);
2471         struct osd_thandle     *oh;
2472         int                     result;
2473         ENTRY;
2474
2475         oh = container_of0(th, struct osd_thandle, ot_super);
2476         LASSERT(oh->ot_handle);
2477         LASSERT(inode);
2478         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2479
2480         if (unlikely(fid_is_acct(fid)))
2481                 RETURN(-EPERM);
2482
2483         if (S_ISDIR(inode->i_mode)) {
2484                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2485                         inode->i_nlink == 2);
2486                 /* it will check/delete the inode from remote parent,
2487                  * how to optimize it? unlink performance impaction XXX */
2488                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2489                 if (result != 0 && result != -ENOENT) {
2490                         CERROR("%s: delete inode "DFID": rc = %d\n",
2491                                osd_name(osd), PFID(fid), result);
2492                 }
2493                 spin_lock(&obj->oo_guard);
2494                 clear_nlink(inode);
2495                 spin_unlock(&obj->oo_guard);
2496                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2497         }
2498
2499         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2500
2501         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2502                                OI_CHECK_FLD);
2503
2504         /* XXX: add to ext3 orphan list */
2505         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2506
2507         /* not needed in the cache anymore */
2508         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2509
2510         RETURN(0);
2511 }
2512
2513 /**
2514  * Put the fid into lustre_mdt_attrs, and then place the structure
2515  * inode's ea. This fid should not be altered during the life time
2516  * of the inode.
2517  *
2518  * \retval +ve, on success
2519  * \retval -ve, on error
2520  *
2521  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2522  */
2523 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2524                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2525 {
2526         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2527         int                      rc;
2528         ENTRY;
2529
2530         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2531                 RETURN(0);
2532
2533         lustre_lma_init(lma, fid, compat, incompat);
2534         lustre_lma_swab(lma);
2535
2536         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2537                              XATTR_CREATE);
2538         /* LMA may already exist, but we need to check that all the
2539          * desired compat/incompat flags have been added. */
2540         if (unlikely(rc == -EEXIST)) {
2541                 if (compat == 0 && incompat == 0)
2542                         RETURN(0);
2543
2544                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2545                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2546                                      LMA_OLD_SIZE);
2547                 if (rc <= 0)
2548                         RETURN(-EINVAL);
2549
2550                 lustre_lma_swab(lma);
2551                 if (!(~lma->lma_compat & compat) &&
2552                     !(~lma->lma_incompat & incompat))
2553                         RETURN(0);
2554
2555                 lma->lma_compat |= compat;
2556                 lma->lma_incompat |= incompat;
2557                 lustre_lma_swab(lma);
2558                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2559                                      sizeof(*lma), XATTR_REPLACE);
2560         }
2561
2562         RETURN(rc);
2563 }
2564
2565 /**
2566  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2567  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2568  * To have compatilibility with 1.8 ldiskfs driver we need to have
2569  * magic number at start of fid data.
2570  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2571  * its inmemory API.
2572  */
2573 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2574                                   const struct lu_fid *fid)
2575 {
2576         if (!fid_is_namespace_visible(fid) ||
2577             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2578                 param->edp_magic = 0;
2579                 return;
2580         }
2581
2582         param->edp_magic = LDISKFS_LUFID_MAGIC;
2583         param->edp_len =  sizeof(struct lu_fid) + 1;
2584         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2585 }
2586
2587 /**
2588  * Try to read the fid from inode ea into dt_rec.
2589  *
2590  * \param fid object fid.
2591  *
2592  * \retval 0 on success
2593  */
2594 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2595                           __u32 ino, struct lu_fid *fid,
2596                           struct osd_inode_id *id)
2597 {
2598         struct osd_thread_info *info  = osd_oti_get(env);
2599         struct inode           *inode;
2600         ENTRY;
2601
2602         osd_id_gen(id, ino, OSD_OII_NOGEN);
2603         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2604         if (IS_ERR(inode))
2605                 RETURN(PTR_ERR(inode));
2606
2607         iput(inode);
2608         RETURN(0);
2609 }
2610
2611 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2612                                         struct inode *dir,
2613                                         struct inode *parent_dir,
2614                                         const struct lu_fid *dot_fid,
2615                                         const struct lu_fid *dot_dot_fid,
2616                                         struct osd_thandle *oth)
2617 {
2618         struct ldiskfs_dentry_param *dot_ldp;
2619         struct ldiskfs_dentry_param *dot_dot_ldp;
2620
2621         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2622         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2623
2624         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2625         dot_ldp->edp_magic = 0;
2626         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2627                                         dir, dot_ldp, dot_dot_ldp);
2628 }
2629
2630 /**
2631  * Create an local agent inode for remote entry
2632  */
2633 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2634                                                   struct osd_device *osd,
2635                                                   struct osd_object *pobj,
2636                                                   const struct lu_fid *fid,
2637                                                   __u32 type,
2638                                                   struct thandle *th)
2639 {
2640         struct osd_thread_info  *info = osd_oti_get(env);
2641         struct inode            *local;
2642         struct osd_thandle      *oh;
2643         int                     rc;
2644         ENTRY;
2645
2646         LASSERT(th);
2647         oh = container_of(th, struct osd_thandle, ot_super);
2648         LASSERT(oh->ot_handle->h_transaction != NULL);
2649
2650         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2651         if (IS_ERR(local)) {
2652                 CERROR("%s: create local error %d\n", osd_name(osd),
2653                        (int)PTR_ERR(local));
2654                 RETURN(local);
2655         }
2656
2657         /* Set special LMA flag for local agent inode */
2658         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2659         if (rc != 0) {
2660                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2661                        osd_name(osd), PFID(fid), rc);
2662                 RETURN(ERR_PTR(rc));
2663         }
2664
2665         if (!S_ISDIR(type))
2666                 RETURN(local);
2667
2668         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2669                                          lu_object_fid(&pobj->oo_dt.do_lu),
2670                                          fid, oh);
2671         if (rc != 0) {
2672                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2673                         osd_name(osd), PFID(fid), rc);
2674                 RETURN(ERR_PTR(rc));
2675         }
2676
2677         RETURN(local);
2678 }
2679
2680 /**
2681  * Delete local agent inode for remote entry
2682  */
2683 static int osd_delete_local_agent_inode(const struct lu_env *env,
2684                                         struct osd_device *osd,
2685                                         const struct lu_fid *fid,
2686                                         __u32 ino, struct osd_thandle *oh)
2687 {
2688         struct osd_thread_info  *oti = osd_oti_get(env);
2689         struct osd_inode_id     *id = &oti->oti_id;
2690         struct inode            *inode;
2691         ENTRY;
2692
2693         id->oii_ino = le32_to_cpu(ino);
2694         id->oii_gen = OSD_OII_NOGEN;
2695         inode = osd_iget(oti, osd, id);
2696         if (IS_ERR(inode)) {
2697                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2698                        PFID(fid), id->oii_ino, id->oii_gen);
2699                 RETURN(PTR_ERR(inode));
2700         }
2701
2702         clear_nlink(inode);
2703         mark_inode_dirty(inode);
2704         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2705                 osd_name(osd), PFID(fid), inode->i_ino);
2706         iput(inode);
2707         RETURN(0);
2708 }
2709
2710 /**
2711  * OSD layer object create function for interoperability mode (b11826).
2712  * This is mostly similar to osd_object_create(). Only difference being, fid is
2713  * inserted into inode ea here.
2714  *
2715  * \retval   0, on success
2716  * \retval -ve, on error
2717  */
2718 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2719                                 struct lu_attr *attr,
2720                                 struct dt_allocation_hint *hint,
2721                                 struct dt_object_format *dof,
2722                                 struct thandle *th)
2723 {
2724         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2725         struct osd_object      *obj    = osd_dt_obj(dt);
2726         struct osd_thread_info *info   = osd_oti_get(env);
2727         int                     result;
2728
2729         ENTRY;
2730
2731         LASSERT(osd_invariant(obj));
2732         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2733         LASSERT(osd_write_locked(env, obj));
2734         LASSERT(th != NULL);
2735
2736         if (unlikely(fid_is_acct(fid)))
2737                 /* Quota files can't be created from the kernel any more,
2738                  * 'tune2fs -O quota' will take care of creating them */
2739                 RETURN(-EPERM);
2740
2741         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2742         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2743
2744         result = __osd_object_create(info, obj, attr, hint, dof, th);
2745         if (result == 0)
2746                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2747                                 fid_is_on_ost(info, osd_obj2dev(obj),
2748                                               fid, OI_CHECK_FLD) ?
2749                                 LMAC_FID_ON_OST : 0, 0);
2750
2751         if (result == 0)
2752                 result = __osd_oi_insert(env, obj, fid, th);
2753
2754         LASSERT(ergo(result == 0,
2755                      dt_object_exists(dt) && !dt_object_remote(dt)));
2756         LINVRNT(osd_invariant(obj));
2757         RETURN(result);
2758 }
2759
2760 static int osd_declare_object_ref_add(const struct lu_env *env,
2761                                       struct dt_object *dt,
2762                                       struct thandle *handle)
2763 {
2764         struct osd_thandle       *oh;
2765
2766         /* it's possible that object doesn't exist yet */
2767         LASSERT(handle != NULL);
2768
2769         oh = container_of0(handle, struct osd_thandle, ot_super);
2770         LASSERT(oh->ot_handle == NULL);
2771
2772         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2773                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2774
2775         return 0;
2776 }
2777
2778 /*
2779  * Concurrency: @dt is write locked.
2780  */
2781 static int osd_object_ref_add(const struct lu_env *env,
2782                               struct dt_object *dt, struct thandle *th)
2783 {
2784         struct osd_object  *obj = osd_dt_obj(dt);
2785         struct inode       *inode = obj->oo_inode;
2786         struct osd_thandle *oh;
2787         int                 rc = 0;
2788
2789         LINVRNT(osd_invariant(obj));
2790         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2791         LASSERT(osd_write_locked(env, obj));
2792         LASSERT(th != NULL);
2793
2794         oh = container_of0(th, struct osd_thandle, ot_super);
2795         LASSERT(oh->ot_handle != NULL);
2796
2797         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2798
2799         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2800                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2801         /*
2802          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2803          * (65000) subdirectories by storing "1" in i_nlink if the link count
2804          * would otherwise overflow. Directory tranversal tools understand
2805          * that (st_nlink == 1) indicates that the filesystem dose not track
2806          * hard links count on the directory, and will not abort subdirectory
2807          * scanning early once (st_nlink - 2) subdirs have been found.
2808          *
2809          * This also has to properly handle the case of inodes with nlink == 0
2810          * in case they are being linked into the PENDING directory
2811          */
2812 #ifdef I_LINKABLE
2813         /* This is necessary to increment from i_nlink == 0 */
2814         spin_lock(&inode->i_lock);
2815         inode->i_state |= I_LINKABLE;
2816         spin_unlock(&inode->i_lock);
2817 #endif
2818
2819         spin_lock(&obj->oo_guard);
2820         ldiskfs_inc_count(oh->ot_handle, inode);
2821         if (!S_ISDIR(inode->i_mode))
2822                 LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2823         spin_unlock(&obj->oo_guard);
2824
2825         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2826         LINVRNT(osd_invariant(obj));
2827
2828         return rc;
2829 }
2830
2831 static int osd_declare_object_ref_del(const struct lu_env *env,
2832                                       struct dt_object *dt,
2833                                       struct thandle *handle)
2834 {
2835         struct osd_thandle *oh;
2836
2837         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2838         LASSERT(handle != NULL);
2839
2840         oh = container_of0(handle, struct osd_thandle, ot_super);
2841         LASSERT(oh->ot_handle == NULL);
2842
2843         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2844                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2845
2846         return 0;
2847 }
2848
2849 /*
2850  * Concurrency: @dt is write locked.
2851  */
2852 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2853                               struct thandle *th)
2854 {
2855         struct osd_object       *obj = osd_dt_obj(dt);
2856         struct inode            *inode = obj->oo_inode;
2857         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2858         struct osd_thandle      *oh;
2859
2860         LINVRNT(osd_invariant(obj));
2861         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2862         LASSERT(osd_write_locked(env, obj));
2863         LASSERT(th != NULL);
2864
2865         oh = container_of0(th, struct osd_thandle, ot_super);
2866         LASSERT(oh->ot_handle != NULL);
2867
2868         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2869
2870         spin_lock(&obj->oo_guard);
2871         /* That can be result of upgrade from old Lustre version and
2872          * applied only to local files.  Just skip this ref_del call.
2873          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2874         if (inode->i_nlink == 0) {
2875                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2876                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2877                              ", maybe an upgraded file? (LU-3915)\n",
2878                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2879                 spin_unlock(&obj->oo_guard);
2880                 return 0;
2881         }
2882
2883         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2884                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2885
2886         ldiskfs_dec_count(oh->ot_handle, inode);
2887         spin_unlock(&obj->oo_guard);
2888
2889         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2890         LINVRNT(osd_invariant(obj));
2891
2892         return 0;
2893 }
2894
2895 /*
2896  * Get the 64-bit version for an inode.
2897  */
2898 static int osd_object_version_get(const struct lu_env *env,
2899                                   struct dt_object *dt, dt_obj_version_t *ver)
2900 {
2901         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2902
2903         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2904                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2905         *ver = LDISKFS_I(inode)->i_fs_version;
2906         return 0;
2907 }
2908
2909 /*
2910  * Concurrency: @dt is read locked.
2911  */
2912 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2913                          struct lu_buf *buf, const char *name,
2914                          struct lustre_capa *capa)
2915 {
2916         struct osd_object      *obj    = osd_dt_obj(dt);
2917         struct inode           *inode  = obj->oo_inode;
2918         struct osd_thread_info *info   = osd_oti_get(env);
2919         struct dentry          *dentry = &info->oti_obj_dentry;
2920
2921         /* version get is not real XATTR but uses xattr API */
2922         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2923                 /* for version we are just using xattr API but change inode
2924                  * field instead */
2925                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2926                 osd_object_version_get(env, dt, buf->lb_buf);
2927                 return sizeof(dt_obj_version_t);
2928         }
2929
2930         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2931         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2932
2933         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2934                 return -EACCES;
2935
2936         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2937 }
2938
2939
2940 static int osd_declare_xattr_set(const struct lu_env *env,
2941                                  struct dt_object *dt,
2942                                  const struct lu_buf *buf, const char *name,
2943                                  int fl, struct thandle *handle)
2944 {
2945         struct osd_thandle *oh;
2946         int credits;
2947
2948         LASSERT(handle != NULL);
2949
2950         oh = container_of0(handle, struct osd_thandle, ot_super);
2951         LASSERT(oh->ot_handle == NULL);
2952
2953         /* optimistic optimization: LMA is set first and usually fit inode */
2954         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2955                 if (dt_object_exists(dt))
2956                         credits = 0;
2957                 else
2958                         credits = 1;
2959         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2960                 credits = 1;
2961         } else {
2962                 struct osd_device  *osd = osd_dev(dt->do_lu.lo_dev);
2963                 struct super_block *sb = osd_sb(osd);
2964                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2965                 if (buf && buf->lb_len > sb->s_blocksize) {
2966                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2967                                         sb->s_blocksize_bits;
2968                 }
2969         }
2970
2971         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2972
2973         return 0;
2974 }
2975
2976 /*
2977  * Set the 64-bit version for object
2978  */
2979 static void osd_object_version_set(const struct lu_env *env,
2980                                    struct dt_object *dt,
2981                                    dt_obj_version_t *new_version)
2982 {
2983         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2984
2985         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2986                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2987
2988         LDISKFS_I(inode)->i_fs_version = *new_version;
2989         /** Version is set after all inode operations are finished,
2990          *  so we should mark it dirty here */
2991         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2992 }
2993
2994 /*
2995  * Concurrency: @dt is write locked.
2996  */
2997 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2998                          const struct lu_buf *buf, const char *name, int fl,
2999                          struct thandle *handle, struct lustre_capa *capa)
3000 {
3001         struct osd_object      *obj      = osd_dt_obj(dt);
3002         struct inode           *inode    = obj->oo_inode;
3003         struct osd_thread_info *info     = osd_oti_get(env);
3004         int                     fs_flags = 0;
3005         ENTRY;
3006
3007         LASSERT(handle != NULL);
3008
3009         /* version set is not real XATTR */
3010         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3011                 /* for version we are just using xattr API but change inode
3012                  * field instead */
3013                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3014                 osd_object_version_set(env, dt, buf->lb_buf);
3015                 return sizeof(dt_obj_version_t);
3016         }
3017
3018         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3019                 return -EACCES;
3020
3021         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3022                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3023
3024         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3025         if (fl & LU_XATTR_REPLACE)
3026                 fs_flags |= XATTR_REPLACE;
3027
3028         if (fl & LU_XATTR_CREATE)
3029                 fs_flags |= XATTR_CREATE;
3030
3031         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3032                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3033                 int                      rc;
3034
3035                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3036                 if (rc != 0)
3037                         RETURN(rc);
3038
3039                 lma->lma_incompat |= LMAI_STRIPED;
3040                 lustre_lma_swab(lma);
3041                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3042                                      sizeof(*lma), XATTR_REPLACE);
3043                 if (rc != 0)
3044                         RETURN(rc);
3045         }
3046
3047         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3048                                fs_flags);
3049 }
3050
3051 /*
3052  * Concurrency: @dt is read locked.
3053  */
3054 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3055                           struct lu_buf *buf, struct lustre_capa *capa)
3056 {
3057         struct osd_object      *obj    = osd_dt_obj(dt);
3058         struct inode           *inode  = obj->oo_inode;
3059         struct osd_thread_info *info   = osd_oti_get(env);
3060         struct dentry          *dentry = &info->oti_obj_dentry;
3061
3062         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3063         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3064
3065         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3066                 return -EACCES;
3067
3068         dentry->d_inode = inode;
3069         dentry->d_sb = inode->i_sb;
3070         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3071 }
3072
3073 static int osd_declare_xattr_del(const struct lu_env *env,
3074                                  struct dt_object *dt, const char *name,
3075                                  struct thandle *handle)
3076 {
3077         struct osd_thandle *oh;
3078
3079         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3080         LASSERT(handle != NULL);
3081
3082         oh = container_of0(handle, struct osd_thandle, ot_super);
3083         LASSERT(oh->ot_handle == NULL);
3084
3085         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3086                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3087
3088         return 0;
3089 }
3090
3091 /*
3092  * Concurrency: @dt is write locked.
3093  */
3094 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3095                          const char *name, struct thandle *handle,
3096                          struct lustre_capa *capa)
3097 {
3098         struct osd_object      *obj    = osd_dt_obj(dt);
3099         struct inode           *inode  = obj->oo_inode;
3100         struct osd_thread_info *info   = osd_oti_get(env);
3101         struct dentry          *dentry = &info->oti_obj_dentry;
3102         int                     rc;
3103
3104         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3105         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3106         LASSERT(handle != NULL);
3107
3108         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3109                 return -EACCES;
3110
3111         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3112
3113         ll_vfs_dq_init(inode);
3114         dentry->d_inode = inode;
3115         dentry->d_sb = inode->i_sb;
3116         rc = inode->i_op->removexattr(dentry, name);
3117         return rc;
3118 }
3119
3120 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3121                                      struct dt_object *dt,
3122                                      struct lustre_capa *old, __u64 opc)
3123 {
3124         struct osd_thread_info *info = osd_oti_get(env);
3125         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3126         struct osd_object *obj = osd_dt_obj(dt);
3127         struct osd_device *osd = osd_obj2dev(obj);
3128         struct lustre_capa_key *key = &info->oti_capa_key;
3129         struct lustre_capa *capa = &info->oti_capa;
3130         struct obd_capa *oc;
3131         struct lu_capainfo *lci;
3132         int rc;
3133         ENTRY;
3134
3135         if (!osd->od_fl_capa)
3136                 RETURN(ERR_PTR(-ENOENT));
3137
3138         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3139         LINVRNT(osd_invariant(obj));
3140
3141         /* renewal sanity check */
3142         if (old && osd_object_auth(env, dt, old, opc))
3143                 RETURN(ERR_PTR(-EACCES));
3144
3145         lci = lu_capainfo_get(env);
3146         if (unlikely(lci == NULL))
3147                 RETURN(ERR_PTR(-ENOENT));
3148
3149         switch (lci->lci_auth) {
3150         case LC_ID_NONE:
3151                 RETURN(NULL);
3152         case LC_ID_PLAIN:
3153                 capa->lc_uid = i_uid_read(obj->oo_inode);
3154                 capa->lc_gid = i_gid_read(obj->oo_inode);
3155                 capa->lc_flags = LC_ID_PLAIN;
3156                 break;
3157         case LC_ID_CONVERT: {
3158                 __u32 d[4], s[4];
3159
3160                 s[0] = i_uid_read(obj->oo_inode);
3161                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3162                 s[2] = i_uid_read(obj->oo_inode);
3163                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3164                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3165                 if (unlikely(rc))
3166                         RETURN(ERR_PTR(rc));
3167
3168                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3169                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3170                 capa->lc_flags = LC_ID_CONVERT;
3171                 break;
3172         }
3173         default:
3174                 RETURN(ERR_P