Whamcloud - gitweb
LU-3536 osp: move update packing into out_lib.c
[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 #include <lustre_linkea.h>
77
78 int ldiskfs_pdo = 1;
79 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
80                 "ldiskfs with parallel directory operations");
81
82 int ldiskfs_track_declares_assert;
83 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
84                 "LBUG during tracking of declares");
85
86 /* Slab to allocate dynlocks */
87 struct kmem_cache *dynlock_cachep;
88
89 static struct lu_kmem_descr ldiskfs_caches[] = {
90         {
91                 .ckd_cache = &dynlock_cachep,
92                 .ckd_name  = "dynlock_cache",
93                 .ckd_size  = sizeof(struct dynlock_handle)
94         },
95         {
96                 .ckd_cache = NULL
97         }
98 };
99
100 static const char dot[] = ".";
101 static const char dotdot[] = "..";
102 static const char remote_obj_dir[] = "REM_OBJ_DIR";
103
104 static const struct lu_object_operations      osd_lu_obj_ops;
105 static const struct dt_object_operations      osd_obj_ops;
106 static const struct dt_object_operations      osd_obj_ea_ops;
107 static const struct dt_object_operations      osd_obj_otable_it_ops;
108 static const struct dt_index_operations       osd_index_iam_ops;
109 static const struct dt_index_operations       osd_index_ea_ops;
110
111 int osd_trans_declare_op2rb[] = {
112         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
113         [OSD_OT_PUNCH]          = OSD_OT_MAX,
114         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
115         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
116         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
117         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
118         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
119         [OSD_OT_WRITE]          = OSD_OT_WRITE,
120         [OSD_OT_INSERT]         = OSD_OT_DELETE,
121         [OSD_OT_DELETE]         = OSD_OT_INSERT,
122         [OSD_OT_UPDATE]         = OSD_OT_MAX,
123         [OSD_OT_QUOTA]          = OSD_OT_MAX,
124 };
125
126 static int osd_has_index(const struct osd_object *obj)
127 {
128         return obj->oo_dt.do_index_ops != NULL;
129 }
130
131 static int osd_object_invariant(const struct lu_object *l)
132 {
133         return osd_invariant(osd_obj(l));
134 }
135
136 /*
137  * Concurrency: doesn't matter
138  */
139
140 /*
141  * Concurrency: doesn't matter
142  */
143 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
144 {
145         struct osd_thread_info *oti = osd_oti_get(env);
146         return oti->oti_w_locks > 0 && o->oo_owner == env;
147 }
148
149 /*
150  * Concurrency: doesn't access mutable data
151  */
152 static int osd_root_get(const struct lu_env *env,
153                         struct dt_device *dev, struct lu_fid *f)
154 {
155         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
156         return 0;
157 }
158
159 /*
160  * OSD object methods.
161  */
162
163 /*
164  * Concurrency: no concurrent access is possible that early in object
165  * life-cycle.
166  */
167 static struct lu_object *osd_object_alloc(const struct lu_env *env,
168                                           const struct lu_object_header *hdr,
169                                           struct lu_device *d)
170 {
171         struct osd_object *mo;
172
173         OBD_ALLOC_PTR(mo);
174         if (mo != NULL) {
175                 struct lu_object *l;
176
177                 l = &mo->oo_dt.do_lu;
178                 dt_object_init(&mo->oo_dt, NULL, d);
179                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
180                 l->lo_ops = &osd_lu_obj_ops;
181                 init_rwsem(&mo->oo_sem);
182                 init_rwsem(&mo->oo_ext_idx_sem);
183                 spin_lock_init(&mo->oo_guard);
184                 return l;
185         } else {
186                 return NULL;
187         }
188 }
189
190 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
191                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
192 {
193         int rc;
194
195         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
196         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
197                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
198         if (rc > 0) {
199                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
200                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
201                 rc = 0;
202                 lustre_lma_swab(lma);
203                 /* Check LMA compatibility */
204                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
205                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
206                               "for fid = "DFID", ino = %lu\n",
207                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
208                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
209                               PFID(&lma->lma_self_fid), inode->i_ino);
210                         rc = -EOPNOTSUPP;
211                 }
212         } else if (rc == 0) {
213                 rc = -ENODATA;
214         }
215
216         return rc;
217 }
218
219 /*
220  * retrieve object from backend ext fs.
221  **/
222 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
223                        struct osd_inode_id *id)
224 {
225         struct inode *inode = NULL;
226
227         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
228         if (IS_ERR(inode)) {
229                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
230                        id->oii_ino, PTR_ERR(inode));
231         } else if (id->oii_gen != OSD_OII_NOGEN &&
232                    inode->i_generation != id->oii_gen) {
233                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
234                        "i_generation = %u\n",
235                        id->oii_ino, id->oii_gen, inode->i_generation);
236                 iput(inode);
237                 inode = ERR_PTR(-ESTALE);
238         } else if (inode->i_nlink == 0) {
239                 /* due to parallel readdir and unlink,
240                 * we can have dead inode here. */
241                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
242                 iput(inode);
243                 inode = ERR_PTR(-ESTALE);
244         } else if (is_bad_inode(inode)) {
245                 CWARN("%.16s: bad inode: ino = %u\n",
246                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
247                 iput(inode);
248                 inode = ERR_PTR(-ENOENT);
249         } else {
250                 if (id->oii_gen == OSD_OII_NOGEN)
251                         osd_id_gen(id, inode->i_ino, inode->i_generation);
252
253                 /* Do not update file c/mtime in ldiskfs.
254                  * NB: we don't have any lock to protect this because we don't
255                  * have reference on osd_object now, but contention with
256                  * another lookup + attr_set can't happen in the tiny window
257                  * between if (...) and set S_NOCMTIME. */
258                 if (!(inode->i_flags & S_NOCMTIME))
259                         inode->i_flags |= S_NOCMTIME;
260         }
261         return inode;
262 }
263
264 static struct inode *
265 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
266              struct osd_inode_id *id, struct lu_fid *fid)
267 {
268         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
269         struct inode            *inode;
270         int                      rc;
271
272         inode = osd_iget(info, dev, id);
273         if (IS_ERR(inode))
274                 return inode;
275
276         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
277         if (rc == 0) {
278                 *fid = lma->lma_self_fid;
279         } else if (rc == -ENODATA) {
280                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
281                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
282                 else
283                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
284         } else {
285                 iput(inode);
286                 inode = ERR_PTR(rc);
287         }
288         return inode;
289 }
290
291 static struct inode *osd_iget_check(struct osd_thread_info *info,
292                                     struct osd_device *dev,
293                                     const struct lu_fid *fid,
294                                     struct osd_inode_id *id,
295                                     bool in_oi)
296 {
297         struct inode    *inode;
298         int              rc     = 0;
299         ENTRY;
300
301         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
302         if (IS_ERR(inode)) {
303                 rc = PTR_ERR(inode);
304                 if (!in_oi || (rc != -ENOENT && rc != -ESTALE)) {
305                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
306                                id->oii_ino, rc);
307
308                         GOTO(put, rc);
309                 }
310
311                 goto check_oi;
312         }
313
314         if (is_bad_inode(inode)) {
315                 rc = -ENOENT;
316                 if (!in_oi) {
317                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
318
319                         GOTO(put, rc);
320                 }
321
322                 goto check_oi;
323         }
324
325         if (id->oii_gen != OSD_OII_NOGEN &&
326             inode->i_generation != id->oii_gen) {
327                 rc = -ESTALE;
328                 if (!in_oi) {
329                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
330                                "oii_gen = %u, i_generation = %u\n",
331                                id->oii_ino, id->oii_gen, inode->i_generation);
332
333                         GOTO(put, rc);
334                 }
335
336                 goto check_oi;
337         }
338
339         if (inode->i_nlink == 0) {
340                 rc = -ENOENT;
341                 if (!in_oi) {
342                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
343
344                         GOTO(put, rc);
345                 }
346
347                 goto check_oi;
348         }
349
350 check_oi:
351         if (rc != 0) {
352                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
353
354                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
355                 /* XXX: There are some possible cases:
356                  *      1. rc = 0.
357                  *         Backup/restore caused the OI invalid.
358                  *      2. rc = 0.
359                  *         Someone unlinked the object but NOT removed
360                  *         the OI mapping, such as mount target device
361                  *         as ldiskfs, and modify something directly.
362                  *      3. rc = -ENOENT.
363                  *         Someone just removed the object between the
364                  *         former oi_lookup and the iget. It is normal.
365                  *      4. Other failure cases.
366                  *
367                  *      Generally, when the device is mounted, it will
368                  *      auto check whether the system is restored from
369                  *      file-level backup or not. We trust such detect
370                  *      to distinguish the 1st case from the 2nd case. */
371                 if (rc == 0) {
372                         if (!IS_ERR(inode) && inode->i_generation != 0 &&
373                             inode->i_generation == id->oii_gen)
374                                 rc = -ENOENT;
375                         else
376                                 rc = -EREMCHG;
377                 }
378         } else {
379                 if (id->oii_gen == OSD_OII_NOGEN)
380                         osd_id_gen(id, inode->i_ino, inode->i_generation);
381
382                 /* Do not update file c/mtime in ldiskfs.
383                  * NB: we don't have any lock to protect this because we don't
384                  * have reference on osd_object now, but contention with
385                  * another lookup + attr_set can't happen in the tiny window
386                  * between if (...) and set S_NOCMTIME. */
387                 if (!(inode->i_flags & S_NOCMTIME))
388                         inode->i_flags |= S_NOCMTIME;
389         }
390
391         GOTO(put, rc);
392
393 put:
394         if (rc != 0) {
395                 if (!IS_ERR(inode))
396                         iput(inode);
397
398                 inode = ERR_PTR(rc);
399         }
400
401         return inode;
402 }
403
404 /**
405  * \retval +v: new filter_fid, does not contain self-fid
406  * \retval 0:  filter_fid_old, contains self-fid
407  * \retval -v: other failure cases
408  */
409 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
410                  struct dentry *dentry, struct lu_fid *fid)
411 {
412         struct filter_fid_old   *ff     = &info->oti_ff;
413         struct ost_id           *ostid  = &info->oti_ostid;
414         int                      rc;
415
416         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
417         if (rc == sizeof(*ff)) {
418                 rc = 0;
419                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
420                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
421                 /* XXX: should use real OST index in the future. LU-3569 */
422                 ostid_to_fid(fid, ostid, 0);
423         } else if (rc == sizeof(struct filter_fid)) {
424                 rc = 1;
425         } else if (rc >= 0) {
426                 rc = -EINVAL;
427         }
428
429         return rc;
430 }
431
432 static int osd_lma_self_repair(struct osd_thread_info *info,
433                                struct osd_device *osd, struct inode *inode,
434                                const struct lu_fid *fid, __u32 compat)
435 {
436         handle_t *jh;
437         int       rc;
438
439         LASSERT(current->journal_info == NULL);
440
441         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
442                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
443         if (IS_ERR(jh)) {
444                 rc = PTR_ERR(jh);
445                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
446                       osd_name(osd), rc);
447                 return rc;
448         }
449
450         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
451         if (rc != 0)
452                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
453                       osd_name(osd), rc);
454         ldiskfs_journal_stop(jh);
455         return rc;
456 }
457
458 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
459 {
460         struct osd_thread_info  *info   = osd_oti_get(env);
461         struct osd_device       *osd    = osd_obj2dev(obj);
462         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
463         struct inode            *inode  = obj->oo_inode;
464         struct dentry           *dentry = &info->oti_obj_dentry;
465         struct lu_fid           *fid    = NULL;
466         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
467         int                      rc;
468         ENTRY;
469
470         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
471         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
472                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
473         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
474                 fid = &lma->lma_self_fid;
475                 rc = osd_get_idif(info, inode, dentry, fid);
476                 if ((rc > 0) || (rc == -ENODATA && osd->od_lma_self_repair)) {
477                         /* For the given OST-object, if it has neither LMA nor
478                          * FID in XATTR_NAME_FID, then the given FID (which is
479                          * contained in the @obj, from client RPC for locating
480                          * the OST-object) is trusted. We use it to generate
481                          * the LMA. */
482                         osd_lma_self_repair(info, osd, inode, rfid,
483                                 fid_is_on_ost(info, osd, fid, OI_CHECK_FLD) ?
484                                 LMAC_FID_ON_OST : 0);
485                         RETURN(0);
486                 }
487         }
488
489         if (unlikely(rc == -ENODATA))
490                 RETURN(0);
491
492         if (rc < 0)
493                 RETURN(rc);
494
495         if (rc > 0) {
496                 rc = 0;
497                 lustre_lma_swab(lma);
498                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
499                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
500                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
501                               "fid = "DFID", ino = %lu\n", osd_name(osd),
502                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
503                               PFID(rfid), inode->i_ino);
504                         rc = -EOPNOTSUPP;
505                 } else if (!(lma->lma_compat & LMAC_NOT_IN_OI)) {
506                         fid = &lma->lma_self_fid;
507                 }
508         }
509
510         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
511                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
512                         struct ost_id   *oi   = &info->oti_ostid;
513                         struct lu_fid   *fid1 = &info->oti_fid3;
514                         __u32            idx  = fid_idif_ost_idx(rfid);
515
516                         /* For old IDIF, the OST index is not part of the IDIF,
517                          * Means that different OSTs may have the same IDIFs.
518                          * Under such case, we need to make some compatible
519                          * check to make sure to trigger OI scrub properly. */
520                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
521                                 /* Given @rfid is new, LMA is old. */
522                                 fid_to_ostid(fid, oi);
523                                 ostid_to_fid(fid1, oi, idx);
524                                 if (lu_fid_eq(fid1, rfid)) {
525                                         if (osd->od_lma_self_repair)
526                                                 osd_lma_self_repair(info, osd,
527                                                         inode, rfid,
528                                                         LMAC_FID_ON_OST);
529                                         RETURN(0);
530                                 }
531                         }
532                 }
533
534                 CDEBUG(D_INODE, "%s: FID "DFID" != self_fid "DFID"\n",
535                        osd_name(osd), PFID(rfid), PFID(fid));
536                 rc = -EREMCHG;
537         }
538
539         RETURN(rc);
540 }
541
542 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
543                           const struct lu_fid *fid,
544                           const struct lu_object_conf *conf)
545 {
546         struct osd_thread_info *info;
547         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
548         struct osd_device      *dev;
549         struct osd_idmap_cache *oic;
550         struct osd_inode_id    *id;
551         struct inode           *inode;
552         struct osd_scrub       *scrub;
553         struct scrub_file      *sf;
554         int                     result;
555         int                     saved  = 0;
556         bool                    in_oi  = false;
557         bool                    triggered = false;
558         ENTRY;
559
560         LINVRNT(osd_invariant(obj));
561         LASSERT(obj->oo_inode == NULL);
562         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
563
564         dev = osd_dev(ldev);
565         scrub = &dev->od_scrub;
566         sf = &scrub->os_file;
567         info = osd_oti_get(env);
568         LASSERT(info);
569         oic = &info->oti_cache;
570
571         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
572                 RETURN(-ENOENT);
573
574         /* For the object is created as locking anchor, or for the object to
575          * be created on disk. No need to osd_oi_lookup() at here because FID
576          * shouldn't never be re-used, if it's really a duplicate FID from
577          * unexpected reason, we should be able to detect it later by calling
578          * do_create->osd_oi_insert(). */
579         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
580                 GOTO(out, result = 0);
581
582         /* Search order: 1. per-thread cache. */
583         if (lu_fid_eq(fid, &oic->oic_fid) &&
584             likely(oic->oic_dev == dev)) {
585                 id = &oic->oic_lid;
586                 goto iget;
587         }
588
589         id = &info->oti_id;
590         if (!list_empty(&scrub->os_inconsistent_items)) {
591                 /* Search order: 2. OI scrub pending list. */
592                 result = osd_oii_lookup(dev, fid, id);
593                 if (result == 0)
594                         goto iget;
595         }
596
597         /* Search order: 3. OI files. */
598         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
599         if (result == -ENOENT) {
600                 if (!fid_is_norm(fid) ||
601                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
602                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
603                                       sf->sf_oi_bitmap))
604                         GOTO(out, result = 0);
605
606                 goto trigger;
607         }
608
609         if (result != 0)
610                 GOTO(out, result);
611
612         in_oi = true;
613
614 iget:
615         inode = osd_iget_check(info, dev, fid, id, in_oi);
616         if (IS_ERR(inode)) {
617                 result = PTR_ERR(inode);
618                 if (result == -ENOENT || result == -ESTALE) {
619                         if (!in_oi)
620                                 fid_zero(&oic->oic_fid);
621
622                         GOTO(out, result = -ENOENT);
623                 } else if (result == -EREMCHG) {
624
625 trigger:
626                         if (!in_oi)
627                                 fid_zero(&oic->oic_fid);
628
629                         if (unlikely(triggered))
630                                 GOTO(out, result = saved);
631
632                         triggered = true;
633                         if (thread_is_running(&scrub->os_thread)) {
634                                 result = -EINPROGRESS;
635                         } else if (!dev->od_noscrub) {
636                                 result = osd_scrub_start(dev);
637                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
638                                               "for "DFID", rc = %d [1]\n",
639                                               osd_name(dev), PFID(fid),result);
640                                 if (result == 0 || result == -EALREADY)
641                                         result = -EINPROGRESS;
642                                 else
643                                         result = -EREMCHG;
644                         }
645
646                         /* We still have chance to get the valid inode: for the
647                          * object which is referenced by remote name entry, the
648                          * object on the local MDT will be linked under the dir
649                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
650                          *
651                          * We do not know whether the object for the given FID
652                          * is referenced by some remote name entry or not, and
653                          * especially for DNE II, a multiple-linked object may
654                          * have many name entries reside on many MDTs.
655                          *
656                          * To simplify the operation, OSD will not distinguish
657                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
658                          * only happened for the RPC from other MDT during the
659                          * OI scrub, or for the client side RPC with FID only,
660                          * such as FID to path, or from old connected client. */
661                         saved = result;
662                         result = osd_lookup_in_remote_parent(info, dev,
663                                                              fid, id);
664                         if (result == 0) {
665                                 in_oi = false;
666                                 goto iget;
667                         }
668
669                         result = saved;
670                 }
671
672                 GOTO(out, result);
673         }
674
675         obj->oo_inode = inode;
676         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
677
678         result = osd_check_lma(env, obj);
679         if (result != 0) {
680                 iput(inode);
681                 obj->oo_inode = NULL;
682                 if (result == -EREMCHG) {
683                         if (!in_oi) {
684                                 result = osd_oi_lookup(info, dev, fid, id,
685                                                        OI_CHECK_FLD);
686                                 if (result != 0) {
687                                         fid_zero(&oic->oic_fid);
688                                         GOTO(out, result);
689                                 }
690                         }
691
692                         goto trigger;
693                 }
694
695                 GOTO(out, result);
696         }
697
698         obj->oo_compat_dot_created = 1;
699         obj->oo_compat_dotdot_created = 1;
700
701         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
702                 GOTO(out, result = 0);
703
704         LASSERT(obj->oo_hl_head == NULL);
705         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
706         if (obj->oo_hl_head == NULL) {
707                 obj->oo_inode = NULL;
708                 iput(inode);
709                 GOTO(out, result = -ENOMEM);
710         }
711         GOTO(out, result = 0);
712
713 out:
714         LINVRNT(osd_invariant(obj));
715         return result;
716 }
717
718 /*
719  * Concurrency: shouldn't matter.
720  */
721 static void osd_object_init0(struct osd_object *obj)
722 {
723         LASSERT(obj->oo_inode != NULL);
724         obj->oo_dt.do_body_ops = &osd_body_ops;
725         obj->oo_dt.do_lu.lo_header->loh_attr |=
726                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
727 }
728
729 /*
730  * Concurrency: no concurrent access is possible that early in object
731  * life-cycle.
732  */
733 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
734                            const struct lu_object_conf *conf)
735 {
736         struct osd_object *obj = osd_obj(l);
737         int result;
738
739         LINVRNT(osd_invariant(obj));
740
741         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
742                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
743                 l->lo_header->loh_attr |= LOHA_EXISTS;
744                 return 0;
745         }
746
747         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
748         obj->oo_dt.do_body_ops = &osd_body_ops_new;
749         if (result == 0 && obj->oo_inode != NULL)
750                 osd_object_init0(obj);
751
752         LINVRNT(osd_invariant(obj));
753         return result;
754 }
755
756 /*
757  * Concurrency: no concurrent access is possible that late in object
758  * life-cycle.
759  */
760 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
761 {
762         struct osd_object *obj = osd_obj(l);
763
764         LINVRNT(osd_invariant(obj));
765
766         dt_object_fini(&obj->oo_dt);
767         if (obj->oo_hl_head != NULL)
768                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
769         OBD_FREE_PTR(obj);
770 }
771
772 /*
773  * Concurrency: no concurrent access is possible that late in object
774  * life-cycle.
775  */
776 static void osd_index_fini(struct osd_object *o)
777 {
778         struct iam_container *bag;
779
780         if (o->oo_dir != NULL) {
781                 bag = &o->oo_dir->od_container;
782                 if (o->oo_inode != NULL) {
783                         if (bag->ic_object == o->oo_inode)
784                                 iam_container_fini(bag);
785                 }
786                 OBD_FREE_PTR(o->oo_dir);
787                 o->oo_dir = NULL;
788         }
789 }
790
791 /*
792  * Concurrency: no concurrent access is possible that late in object
793  * life-cycle (for all existing callers, that is. New callers have to provide
794  * their own locking.)
795  */
796 static int osd_inode_unlinked(const struct inode *inode)
797 {
798         return inode->i_nlink == 0;
799 }
800
801 enum {
802         OSD_TXN_OI_DELETE_CREDITS    = 20,
803         OSD_TXN_INODE_DELETE_CREDITS = 20
804 };
805
806 /*
807  * Journal
808  */
809
810 #if OSD_THANDLE_STATS
811 /**
812  * Set time when the handle is allocated
813  */
814 static void osd_th_alloced(struct osd_thandle *oth)
815 {
816         oth->oth_alloced = cfs_time_current();
817 }
818
819 /**
820  * Set time when the handle started
821  */
822 static void osd_th_started(struct osd_thandle *oth)
823 {
824         oth->oth_started = cfs_time_current();
825 }
826
827 /**
828  * Helper function to convert time interval to microseconds packed in
829  * long int.
830  */
831 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
832 {
833         struct timeval val;
834
835         cfs_duration_usec(cfs_time_sub(end, start), &val);
836         return val.tv_sec * 1000000 + val.tv_usec;
837 }
838
839 /**
840  * Check whether the we deal with this handle for too long.
841  */
842 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
843                                 cfs_time_t alloced, cfs_time_t started,
844                                 cfs_time_t closed)
845 {
846         cfs_time_t now = cfs_time_current();
847
848         LASSERT(dev != NULL);
849
850         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
851                             interval_to_usec(alloced, started));
852         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
853                             interval_to_usec(started, closed));
854         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
855                             interval_to_usec(closed, now));
856
857         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
858                 CWARN("transaction handle %p was open for too long: "
859                       "now "CFS_TIME_T" ,"
860                       "alloced "CFS_TIME_T" ,"
861                       "started "CFS_TIME_T" ,"
862                       "closed "CFS_TIME_T"\n",
863                       oth, now, alloced, started, closed);
864                 libcfs_debug_dumpstack(NULL);
865         }
866 }
867
868 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
869 {                                                                       \
870         cfs_time_t __closed = cfs_time_current();                       \
871         cfs_time_t __alloced = oth->oth_alloced;                        \
872         cfs_time_t __started = oth->oth_started;                        \
873                                                                         \
874         expr;                                                           \
875         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
876 }
877
878 #else /* OSD_THANDLE_STATS */
879
880 #define osd_th_alloced(h)                  do {} while(0)
881 #define osd_th_started(h)                  do {} while(0)
882 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
883
884 #endif /* OSD_THANDLE_STATS */
885
886 /*
887  * Concurrency: doesn't access mutable data.
888  */
889 static int osd_param_is_not_sane(const struct osd_device *dev,
890                                  const struct thandle *th)
891 {
892         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
893
894         return oh->ot_credits > osd_transaction_size(dev);
895 }
896
897 /*
898  * Concurrency: shouldn't matter.
899  */
900 static void osd_trans_commit_cb(struct super_block *sb,
901                                 struct ldiskfs_journal_cb_entry *jcb, int error)
902 {
903         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
904         struct thandle     *th  = &oh->ot_super;
905         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
906         struct dt_txn_commit_cb *dcb, *tmp;
907
908         LASSERT(oh->ot_handle == NULL);
909
910         if (error)
911                 CERROR("transaction @0x%p commit error: %d\n", th, error);
912
913         dt_txn_hook_commit(th);
914
915         /* call per-transaction callbacks if any */
916         list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
917                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
918                          "commit callback entry: magic=%x name='%s'\n",
919                          dcb->dcb_magic, dcb->dcb_name);
920                 list_del_init(&dcb->dcb_linkage);
921                 dcb->dcb_func(NULL, th, dcb, error);
922         }
923
924         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
925         lu_device_put(lud);
926         th->th_dev = NULL;
927
928         lu_context_exit(&th->th_ctx);
929         lu_context_fini(&th->th_ctx);
930         thandle_put(th);
931 }
932
933 static struct thandle *osd_trans_create(const struct lu_env *env,
934                                         struct dt_device *d)
935 {
936         struct osd_thread_info  *oti = osd_oti_get(env);
937         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
938         struct osd_thandle      *oh;
939         struct thandle          *th;
940         ENTRY;
941
942         /* on pending IO in this thread should left from prev. request */
943         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
944
945         th = ERR_PTR(-ENOMEM);
946         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
947         if (oh != NULL) {
948                 oh->ot_quota_trans = &oti->oti_quota_trans;
949                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
950                 th = &oh->ot_super;
951                 th->th_dev = d;
952                 th->th_result = 0;
953                 th->th_tags = LCT_TX_HANDLE;
954                 oh->ot_credits = 0;
955                 atomic_set(&th->th_refc, 1);
956                 th->th_alloc_size = sizeof(*oh);
957                 oti->oti_dev = osd_dt_dev(d);
958                 INIT_LIST_HEAD(&oh->ot_dcb_list);
959                 osd_th_alloced(oh);
960
961                 memset(oti->oti_declare_ops, 0,
962                                         sizeof(oti->oti_declare_ops));
963                 memset(oti->oti_declare_ops_rb, 0,
964                                         sizeof(oti->oti_declare_ops_rb));
965                 memset(oti->oti_declare_ops_cred, 0,
966                                         sizeof(oti->oti_declare_ops_cred));
967                 oti->oti_rollback = false;
968         }
969         RETURN(th);
970 }
971
972 /*
973  * Concurrency: shouldn't matter.
974  */
975 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
976                     struct thandle *th)
977 {
978         struct osd_thread_info *oti = osd_oti_get(env);
979         struct osd_device  *dev = osd_dt_dev(d);
980         handle_t           *jh;
981         struct osd_thandle *oh;
982         int rc;
983
984         ENTRY;
985
986         LASSERT(current->journal_info == NULL);
987
988         oh = container_of0(th, struct osd_thandle, ot_super);
989         LASSERT(oh != NULL);
990         LASSERT(oh->ot_handle == NULL);
991
992         rc = dt_txn_hook_start(env, d, th);
993         if (rc != 0)
994                 GOTO(out, rc);
995
996         if (unlikely(osd_param_is_not_sane(dev, th))) {
997                 static unsigned long last_printed;
998                 static int last_credits;
999
1000                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
1001                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
1002                       oh->ot_credits,
1003                       osd_journal(dev)->j_max_transaction_buffers);
1004                 CWARN("  create: %u/%u, destroy: %u/%u\n",
1005                       oti->oti_declare_ops[OSD_OT_CREATE],
1006                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
1007                       oti->oti_declare_ops[OSD_OT_DESTROY],
1008                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
1009                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
1010                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
1011                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1012                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
1013                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
1014                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
1015                       oti->oti_declare_ops[OSD_OT_WRITE],
1016                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
1017                       oti->oti_declare_ops[OSD_OT_PUNCH],
1018                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1019                       oti->oti_declare_ops[OSD_OT_QUOTA],
1020                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
1021                 CWARN("  insert: %u/%u, delete: %u/%u\n",
1022                       oti->oti_declare_ops[OSD_OT_INSERT],
1023                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
1024                       oti->oti_declare_ops[OSD_OT_DELETE],
1025                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
1026                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
1027                       oti->oti_declare_ops[OSD_OT_REF_ADD],
1028                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1029                       oti->oti_declare_ops[OSD_OT_REF_DEL],
1030                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
1031
1032                 if (last_credits != oh->ot_credits &&
1033                     time_after(jiffies, last_printed + 60 * HZ)) {
1034                         libcfs_debug_dumpstack(NULL);
1035                         last_credits = oh->ot_credits;
1036                         last_printed = jiffies;
1037                 }
1038                 /* XXX Limit the credits to 'max_transaction_buffers', and
1039                  *     let the underlying filesystem to catch the error if
1040                  *     we really need so many credits.
1041                  *
1042                  *     This should be removed when we can calculate the
1043                  *     credits precisely. */
1044                 oh->ot_credits = osd_transaction_size(dev);
1045         }
1046
1047         /*
1048          * XXX temporary stuff. Some abstraction layer should
1049          * be used.
1050          */
1051         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1052         osd_th_started(oh);
1053         if (!IS_ERR(jh)) {
1054                 oh->ot_handle = jh;
1055                 LASSERT(oti->oti_txns == 0);
1056                 lu_context_init(&th->th_ctx, th->th_tags);
1057                 lu_context_enter(&th->th_ctx);
1058
1059                 lu_device_get(&d->dd_lu_dev);
1060                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1061                               "osd-tx", th);
1062                 oti->oti_txns++;
1063                 rc = 0;
1064         } else {
1065                 rc = PTR_ERR(jh);
1066         }
1067 out:
1068         RETURN(rc);
1069 }
1070
1071 static int osd_seq_exists(const struct lu_env *env,
1072                               struct osd_device *osd, obd_seq seq)
1073 {
1074         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1075         struct seq_server_site  *ss = osd_seq_site(osd);
1076         int                     rc;
1077         ENTRY;
1078
1079         LASSERT(ss != NULL);
1080         LASSERT(ss->ss_server_fld != NULL);
1081
1082         rc = osd_fld_lookup(env, osd, seq, range);
1083         if (rc != 0) {
1084                 if (rc != -ENOENT)
1085                         CERROR("%s: can't lookup FLD sequence "LPX64
1086                                ": rc = %d\n", osd_name(osd), seq, rc);
1087                 RETURN(0);
1088         }
1089
1090         RETURN(ss->ss_node_id == range->lsr_index);
1091 }
1092
1093 /*
1094  * Concurrency: shouldn't matter.
1095  */
1096 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1097                           struct thandle *th)
1098 {
1099         int                     rc = 0;
1100         struct osd_thandle     *oh;
1101         struct osd_thread_info *oti = osd_oti_get(env);
1102         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1103         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
1104         struct lquota_trans    *qtrans;
1105         ENTRY;
1106
1107         oh = container_of0(th, struct osd_thandle, ot_super);
1108
1109         qtrans = oh->ot_quota_trans;
1110         oh->ot_quota_trans = NULL;
1111
1112         if (oh->ot_handle != NULL) {
1113                 handle_t *hdl = oh->ot_handle;
1114
1115                 /*
1116                  * add commit callback
1117                  * notice we don't do this in osd_trans_start()
1118                  * as underlying transaction can change during truncate
1119                  */
1120                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1121                                          &oh->ot_jcb);
1122
1123                 LASSERT(oti->oti_txns == 1);
1124                 oti->oti_txns--;
1125                 rc = dt_txn_hook_stop(env, th);
1126                 if (rc != 0)
1127                         CERROR("Failure in transaction hook: %d\n", rc);
1128
1129                 /* hook functions might modify th_sync */
1130                 hdl->h_sync = th->th_sync;
1131
1132                 oh->ot_handle = NULL;
1133                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
1134                                   rc = ldiskfs_journal_stop(hdl));
1135                 if (rc != 0)
1136                         CERROR("Failure to stop transaction: %d\n", rc);
1137         } else {
1138                 thandle_put(&oh->ot_super);
1139         }
1140
1141         /* inform the quota slave device that the transaction is stopping */
1142         qsd_op_end(env, qsd, qtrans);
1143
1144         /* as we want IO to journal and data IO be concurrent, we don't block
1145          * awaiting data IO completion in osd_do_bio(), instead we wait here
1146          * once transaction is submitted to the journal. all reqular requests
1147          * don't do direct IO (except read/write), thus this wait_event becomes
1148          * no-op for them.
1149          *
1150          * IMPORTANT: we have to wait till any IO submited by the thread is
1151          * completed otherwise iobuf may be corrupted by different request
1152          */
1153         wait_event(iobuf->dr_wait,
1154                        atomic_read(&iobuf->dr_numreqs) == 0);
1155         osd_fini_iobuf(oti->oti_dev, iobuf);
1156         if (!rc)
1157                 rc = iobuf->dr_error;
1158
1159         RETURN(rc);
1160 }
1161
1162 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1163 {
1164         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1165                                                ot_super);
1166
1167         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1168         LASSERT(&dcb->dcb_func != NULL);
1169         list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
1170
1171         return 0;
1172 }
1173
1174 /*
1175  * Called just before object is freed. Releases all resources except for
1176  * object itself (that is released by osd_object_free()).
1177  *
1178  * Concurrency: no concurrent access is possible that late in object
1179  * life-cycle.
1180  */
1181 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1182 {
1183         struct osd_object *obj   = osd_obj(l);
1184         struct inode      *inode = obj->oo_inode;
1185
1186         LINVRNT(osd_invariant(obj));
1187
1188         /*
1189          * If object is unlinked remove fid->ino mapping from object index.
1190          */
1191
1192         osd_index_fini(obj);
1193         if (inode != NULL) {
1194                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1195                 qid_t                    uid = i_uid_read(inode);
1196                 qid_t                    gid = i_gid_read(inode);
1197
1198                 iput(inode);
1199                 obj->oo_inode = NULL;
1200
1201                 if (qsd != NULL) {
1202                         struct osd_thread_info  *info = osd_oti_get(env);
1203                         struct lquota_id_info   *qi = &info->oti_qi;
1204
1205                         /* Release granted quota to master if necessary */
1206                         qi->lqi_id.qid_uid = uid;
1207                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1208
1209                         qi->lqi_id.qid_uid = gid;
1210                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1211                 }
1212         }
1213 }
1214
1215 /*
1216  * Concurrency: ->loo_object_release() is called under site spin-lock.
1217  */
1218 static void osd_object_release(const struct lu_env *env,
1219                                struct lu_object *l)
1220 {
1221 }
1222
1223 /*
1224  * Concurrency: shouldn't matter.
1225  */
1226 static int osd_object_print(const struct lu_env *env, void *cookie,
1227                             lu_printer_t p, const struct lu_object *l)
1228 {
1229         struct osd_object *o = osd_obj(l);
1230         struct iam_descr  *d;
1231
1232         if (o->oo_dir != NULL)
1233                 d = o->oo_dir->od_container.ic_descr;
1234         else
1235                 d = NULL;
1236         return (*p)(env, cookie,
1237                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1238                     o, o->oo_inode,
1239                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1240                     o->oo_inode ? o->oo_inode->i_generation : 0,
1241                     d ? d->id_ops->id_name : "plain");
1242 }
1243
1244 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1245
1246 /*
1247  * Concurrency: shouldn't matter.
1248  */
1249 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1250                struct obd_statfs *sfs)
1251 {
1252         struct osd_device  *osd = osd_dt_dev(d);
1253         struct super_block *sb = osd_sb(osd);
1254         struct kstatfs     *ksfs;
1255         int result = 0;
1256
1257         if (unlikely(osd->od_mnt == NULL))
1258                 return -EINPROGRESS;
1259
1260         /* osd_lproc.c call this without env, allocate ksfs for that case */
1261         if (unlikely(env == NULL)) {
1262                 OBD_ALLOC_PTR(ksfs);
1263                 if (ksfs == NULL)
1264                         return -ENOMEM;
1265         } else {
1266                 ksfs = &osd_oti_get(env)->oti_ksfs;
1267         }
1268
1269         spin_lock(&osd->od_osfs_lock);
1270         result = sb->s_op->statfs(sb->s_root, ksfs);
1271         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1272                 statfs_pack(sfs, ksfs);
1273                 if (sb->s_flags & MS_RDONLY)
1274                         sfs->os_state = OS_STATE_READONLY;
1275         }
1276
1277         spin_unlock(&osd->od_osfs_lock);
1278
1279         if (unlikely(env == NULL))
1280                 OBD_FREE_PTR(ksfs);
1281
1282         /* Reserve a small amount of space for local objects like last_rcvd,
1283          * llog, quota files, ... */
1284         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1285                 sfs->os_bavail = 0;
1286         } else {
1287                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1288                 /** Take out metadata overhead for indirect blocks */
1289                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1290         }
1291
1292         return result;
1293 }
1294
1295 /**
1296  * Estimate space needed for file creations. We assume the largest filename
1297  * which is 2^64 - 1, hence a filename of 20 chars.
1298  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1299  */
1300 #ifdef __LDISKFS_DIR_REC_LEN
1301 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1302 #else
1303 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1304 #endif
1305
1306 /*
1307  * Concurrency: doesn't access mutable data.
1308  */
1309 static void osd_conf_get(const struct lu_env *env,
1310                          const struct dt_device *dev,
1311                          struct dt_device_param *param)
1312 {
1313         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1314         int                ea_overhead;
1315
1316         /*
1317          * XXX should be taken from not-yet-existing fs abstraction layer.
1318          */
1319         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1320         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1321         param->ddp_block_shift  = sb->s_blocksize_bits;
1322         param->ddp_mount_type     = LDD_MT_LDISKFS;
1323         param->ddp_maxbytes       = sb->s_maxbytes;
1324         /* Overhead estimate should be fairly accurate, so we really take a tiny
1325          * error margin which also avoids fragmenting the filesystem too much */
1326         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1327         /* inode are statically allocated, so per-inode space consumption
1328          * is the space consumed by the directory entry */
1329         param->ddp_inodespace     = PER_OBJ_USAGE;
1330         /* per-fragment overhead to be used by the client code */
1331         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1332         param->ddp_mntopts      = 0;
1333         if (test_opt(sb, XATTR_USER))
1334                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1335         if (test_opt(sb, POSIX_ACL))
1336                 param->ddp_mntopts |= MNTOPT_ACL;
1337
1338         /* LOD might calculate the max stripe count based on max_ea_size,
1339          * so we need take account in the overhead as well,
1340          * xattr_header + magic + xattr_entry_head */
1341         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1342                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1343
1344 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1345         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1346                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1347                                                                 ea_overhead;
1348         else
1349 #endif
1350                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1351 }
1352
1353 /*
1354  * Concurrency: shouldn't matter.
1355  */
1356 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1357 {
1358         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1359         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1360 }
1361
1362 /**
1363  * Start commit for OSD device.
1364  *
1365  * An implementation of dt_commit_async method for OSD device.
1366  * Asychronously starts underlayng fs sync and thereby a transaction
1367  * commit.
1368  *
1369  * \param env environment
1370  * \param d dt device
1371  *
1372  * \see dt_device_operations
1373  */
1374 static int osd_commit_async(const struct lu_env *env,
1375                             struct dt_device *d)
1376 {
1377         struct super_block *s = osd_sb(osd_dt_dev(d));
1378         ENTRY;
1379
1380         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1381         RETURN(s->s_op->sync_fs(s, 0));
1382 }
1383
1384 /*
1385  * Concurrency: shouldn't matter.
1386  */
1387
1388 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1389 {
1390         struct super_block *sb = osd_sb(osd_dt_dev(d));
1391         struct block_device *dev = sb->s_bdev;
1392 #ifdef HAVE_DEV_SET_RDONLY
1393         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1394         int rc = 0;
1395 #else
1396         int rc = -EOPNOTSUPP;
1397 #endif
1398         ENTRY;
1399
1400 #ifdef HAVE_DEV_SET_RDONLY
1401         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1402
1403         if (jdev && (jdev != dev)) {
1404                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1405                        (long)jdev);
1406                 dev_set_rdonly(jdev);
1407         }
1408         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1409         dev_set_rdonly(dev);
1410 #else
1411         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1412                osd_dt_dev(d)->od_svname, (long)dev, rc);
1413 #endif
1414         RETURN(rc);
1415 }
1416
1417 /*
1418  * Concurrency: serialization provided by callers.
1419  */
1420 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1421                               int mode, unsigned long timeout, __u32 alg,
1422                               struct lustre_capa_key *keys)
1423 {
1424         struct osd_device *dev = osd_dt_dev(d);
1425         ENTRY;
1426
1427         dev->od_fl_capa = mode;
1428         dev->od_capa_timeout = timeout;
1429         dev->od_capa_alg = alg;
1430         dev->od_capa_keys = keys;
1431         RETURN(0);
1432 }
1433
1434 /**
1435  * Note: we do not count into QUOTA here.
1436  * If we mount with --data_journal we may need more.
1437  */
1438 const int osd_dto_credits_noquota[DTO_NR] = {
1439         /**
1440          * Insert.
1441          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1442          * SINGLEDATA_TRANS_BLOCKS(8)
1443          * XXX Note: maybe iam need more, since iam have more level than
1444          *           EXT3 htree.
1445          */
1446         [DTO_INDEX_INSERT]  = 16,
1447         /**
1448          * Delete
1449          * just modify a single entry, probably merge few within a block
1450          */
1451         [DTO_INDEX_DELETE]  = 1,
1452         /**
1453          * Used for OI scrub
1454          */
1455         [DTO_INDEX_UPDATE]  = 16,
1456         /**
1457          * 4(inode, inode bits, groups, GDT)
1458          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1459          */
1460         [DTO_OBJECT_CREATE] = 4,
1461         /**
1462          * 4(inode, inode bits, groups, GDT)
1463          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1464          */
1465         [DTO_OBJECT_DELETE] = 4,
1466         /**
1467          * Attr set credits (inode)
1468          */
1469         [DTO_ATTR_SET_BASE] = 1,
1470         /**
1471          * Xattr set. The same as xattr of EXT3.
1472          * DATA_TRANS_BLOCKS(14)
1473          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1474          * are also counted in. Do not know why?
1475          */
1476         [DTO_XATTR_SET]     = 14,
1477         /**
1478          * credits for inode change during write.
1479          */
1480         [DTO_WRITE_BASE]    = 3,
1481         /**
1482          * credits for single block write.
1483          */
1484         [DTO_WRITE_BLOCK]   = 14,
1485         /**
1486          * Attr set credits for chown.
1487          * This is extra credits for setattr, and it is null without quota
1488          */
1489         [DTO_ATTR_SET_CHOWN] = 0
1490 };
1491
1492 static const struct dt_device_operations osd_dt_ops = {
1493         .dt_root_get       = osd_root_get,
1494         .dt_statfs         = osd_statfs,
1495         .dt_trans_create   = osd_trans_create,
1496         .dt_trans_start    = osd_trans_start,
1497         .dt_trans_stop     = osd_trans_stop,
1498         .dt_trans_cb_add   = osd_trans_cb_add,
1499         .dt_conf_get       = osd_conf_get,
1500         .dt_sync           = osd_sync,
1501         .dt_ro             = osd_ro,
1502         .dt_commit_async   = osd_commit_async,
1503         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1504 };
1505
1506 static void osd_object_read_lock(const struct lu_env *env,
1507                                  struct dt_object *dt, unsigned role)
1508 {
1509         struct osd_object *obj = osd_dt_obj(dt);
1510         struct osd_thread_info *oti = osd_oti_get(env);
1511
1512         LINVRNT(osd_invariant(obj));
1513
1514         LASSERT(obj->oo_owner != env);
1515         down_read_nested(&obj->oo_sem, role);
1516
1517         LASSERT(obj->oo_owner == NULL);
1518         oti->oti_r_locks++;
1519 }
1520
1521 static void osd_object_write_lock(const struct lu_env *env,
1522                                   struct dt_object *dt, unsigned role)
1523 {
1524         struct osd_object *obj = osd_dt_obj(dt);
1525         struct osd_thread_info *oti = osd_oti_get(env);
1526
1527         LINVRNT(osd_invariant(obj));
1528
1529         LASSERT(obj->oo_owner != env);
1530         down_write_nested(&obj->oo_sem, role);
1531
1532         LASSERT(obj->oo_owner == NULL);
1533         obj->oo_owner = env;
1534         oti->oti_w_locks++;
1535 }
1536
1537 static void osd_object_read_unlock(const struct lu_env *env,
1538                                    struct dt_object *dt)
1539 {
1540         struct osd_object *obj = osd_dt_obj(dt);
1541         struct osd_thread_info *oti = osd_oti_get(env);
1542
1543         LINVRNT(osd_invariant(obj));
1544
1545         LASSERT(oti->oti_r_locks > 0);
1546         oti->oti_r_locks--;
1547         up_read(&obj->oo_sem);
1548 }
1549
1550 static void osd_object_write_unlock(const struct lu_env *env,
1551                                     struct dt_object *dt)
1552 {
1553         struct osd_object *obj = osd_dt_obj(dt);
1554         struct osd_thread_info *oti = osd_oti_get(env);
1555
1556         LINVRNT(osd_invariant(obj));
1557
1558         LASSERT(obj->oo_owner == env);
1559         LASSERT(oti->oti_w_locks > 0);
1560         oti->oti_w_locks--;
1561         obj->oo_owner = NULL;
1562         up_write(&obj->oo_sem);
1563 }
1564
1565 static int osd_object_write_locked(const struct lu_env *env,
1566                                    struct dt_object *dt)
1567 {
1568         struct osd_object *obj = osd_dt_obj(dt);
1569
1570         LINVRNT(osd_invariant(obj));
1571
1572         return obj->oo_owner == env;
1573 }
1574
1575 static int capa_is_sane(const struct lu_env *env,
1576                         struct osd_device *dev,
1577                         struct lustre_capa *capa,
1578                         struct lustre_capa_key *keys)
1579 {
1580         struct osd_thread_info *oti = osd_oti_get(env);
1581         struct lustre_capa *tcapa = &oti->oti_capa;
1582         struct obd_capa *oc;
1583         int i, rc = 0;
1584         ENTRY;
1585
1586         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1587         if (oc) {
1588                 if (capa_is_expired(oc)) {
1589                         DEBUG_CAPA(D_ERROR, capa, "expired");
1590                         rc = -ESTALE;
1591                 }
1592                 capa_put(oc);
1593                 RETURN(rc);
1594         }
1595
1596         if (capa_is_expired_sec(capa)) {
1597                 DEBUG_CAPA(D_ERROR, capa, "expired");
1598                 RETURN(-ESTALE);
1599         }
1600
1601         spin_lock(&capa_lock);
1602         for (i = 0; i < 2; i++) {
1603                 if (keys[i].lk_keyid == capa->lc_keyid) {
1604                         oti->oti_capa_key = keys[i];
1605                         break;
1606                 }
1607         }
1608         spin_unlock(&capa_lock);
1609
1610         if (i == 2) {
1611                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1612                 RETURN(-ESTALE);
1613         }
1614
1615         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1616         if (rc)
1617                 RETURN(rc);
1618
1619         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1620                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1621                 RETURN(-EACCES);
1622         }
1623
1624         oc = capa_add(dev->od_capa_hash, capa);
1625         capa_put(oc);
1626
1627         RETURN(0);
1628 }
1629
1630 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1631                     struct lustre_capa *capa, __u64 opc)
1632 {
1633         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1634         struct osd_device *osd = osd_dev(dt->do_lu.lo_dev);
1635         struct lu_capainfo *lci;
1636         int rc;
1637
1638         if (!osd->od_fl_capa)
1639                 return 0;
1640
1641         if (capa == BYPASS_CAPA)
1642                 return 0;
1643
1644         lci = lu_capainfo_get(env);
1645         if (unlikely(lci == NULL))
1646                 return 0;
1647
1648         if (lci->lci_auth == LC_ID_NONE)
1649                 return 0;
1650
1651         if (capa == NULL) {
1652                 CERROR("%s: no capability provided for FID "DFID": rc = %d\n",
1653                        osd_name(osd), PFID(fid), -EACCES);
1654                 return -EACCES;
1655         }
1656
1657         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1658                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1659                            PFID(fid));
1660                 return -EACCES;
1661         }
1662
1663         if (!capa_opc_supported(capa, opc)) {
1664                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1665                 return -EACCES;
1666         }
1667
1668         rc = capa_is_sane(env, osd, capa, osd->od_capa_keys);
1669         if (rc != 0) {
1670                 DEBUG_CAPA(D_ERROR, capa, "insane: rc = %d", rc);
1671                 return -EACCES;
1672         }
1673
1674         return 0;
1675 }
1676
1677 static struct timespec *osd_inode_time(const struct lu_env *env,
1678                                        struct inode *inode, __u64 seconds)
1679 {
1680         struct osd_thread_info  *oti = osd_oti_get(env);
1681         struct timespec         *t   = &oti->oti_time;
1682
1683         t->tv_sec = seconds;
1684         t->tv_nsec = 0;
1685         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1686         return t;
1687 }
1688
1689
1690 static void osd_inode_getattr(const struct lu_env *env,
1691                               struct inode *inode, struct lu_attr *attr)
1692 {
1693         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1694                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1695                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1696                            LA_TYPE;
1697
1698         attr->la_atime   = LTIME_S(inode->i_atime);
1699         attr->la_mtime   = LTIME_S(inode->i_mtime);
1700         attr->la_ctime   = LTIME_S(inode->i_ctime);
1701         attr->la_mode    = inode->i_mode;
1702         attr->la_size    = i_size_read(inode);
1703         attr->la_blocks  = inode->i_blocks;
1704         attr->la_uid     = i_uid_read(inode);
1705         attr->la_gid     = i_gid_read(inode);
1706         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1707         attr->la_nlink   = inode->i_nlink;
1708         attr->la_rdev    = inode->i_rdev;
1709         attr->la_blksize = 1 << inode->i_blkbits;
1710         attr->la_blkbits = inode->i_blkbits;
1711 }
1712
1713 static int osd_attr_get(const struct lu_env *env,
1714                         struct dt_object *dt,
1715                         struct lu_attr *attr,
1716                         struct lustre_capa *capa)
1717 {
1718         struct osd_object *obj = osd_dt_obj(dt);
1719
1720         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1721         LINVRNT(osd_invariant(obj));
1722
1723         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1724                 return -EACCES;
1725
1726         spin_lock(&obj->oo_guard);
1727         osd_inode_getattr(env, obj->oo_inode, attr);
1728         spin_unlock(&obj->oo_guard);
1729         return 0;
1730 }
1731
1732 static int osd_declare_attr_set(const struct lu_env *env,
1733                                 struct dt_object *dt,
1734                                 const struct lu_attr *attr,
1735                                 struct thandle *handle)
1736 {
1737         struct osd_thandle     *oh;
1738         struct osd_object      *obj;
1739         struct osd_thread_info *info = osd_oti_get(env);
1740         struct lquota_id_info  *qi = &info->oti_qi;
1741         qid_t                   uid;
1742         qid_t                   gid;
1743         long long               bspace;
1744         int                     rc = 0;
1745         bool                    allocated;
1746         ENTRY;
1747
1748         LASSERT(dt != NULL);
1749         LASSERT(handle != NULL);
1750
1751         obj = osd_dt_obj(dt);
1752         LASSERT(osd_invariant(obj));
1753
1754         oh = container_of0(handle, struct osd_thandle, ot_super);
1755         LASSERT(oh->ot_handle == NULL);
1756
1757         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1758                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1759
1760         if (attr == NULL || obj->oo_inode == NULL)
1761                 RETURN(rc);
1762
1763         bspace   = obj->oo_inode->i_blocks;
1764         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1765         bspace   = toqb(bspace);
1766
1767         /* Changing ownership is always preformed by super user, it should not
1768          * fail with EDQUOT.
1769          *
1770          * We still need to call the osd_declare_qid() to calculate the journal
1771          * credits for updating quota accounting files and to trigger quota
1772          * space adjustment once the operation is completed.*/
1773         if ((attr->la_valid & LA_UID) != 0 &&
1774              attr->la_uid != (uid = i_uid_read(obj->oo_inode))) {
1775                 qi->lqi_type = USRQUOTA;
1776
1777                 /* inode accounting */
1778                 qi->lqi_is_blk = false;
1779
1780                 /* one more inode for the new owner ... */
1781                 qi->lqi_id.qid_uid = attr->la_uid;
1782                 qi->lqi_space      = 1;
1783                 allocated = (attr->la_uid == 0) ? true : false;
1784                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1785                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1786                         rc = 0;
1787                 if (rc)
1788                         RETURN(rc);
1789
1790                 /* and one less inode for the current uid */
1791                 qi->lqi_id.qid_uid = uid;
1792                 qi->lqi_space      = -1;
1793                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1794                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1795                         rc = 0;
1796                 if (rc)
1797                         RETURN(rc);
1798
1799                 /* block accounting */
1800                 qi->lqi_is_blk = true;
1801
1802                 /* more blocks for the new owner ... */
1803                 qi->lqi_id.qid_uid = attr->la_uid;
1804                 qi->lqi_space      = bspace;
1805                 allocated = (attr->la_uid == 0) ? true : false;
1806                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1807                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1808                         rc = 0;
1809                 if (rc)
1810                         RETURN(rc);
1811
1812                 /* and finally less blocks for the current owner */
1813                 qi->lqi_id.qid_uid = uid;
1814                 qi->lqi_space      = -bspace;
1815                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1816                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1817                         rc = 0;
1818                 if (rc)
1819                         RETURN(rc);
1820         }
1821
1822         if (attr->la_valid & LA_GID &&
1823             attr->la_gid != (gid = i_gid_read(obj->oo_inode))) {
1824                 qi->lqi_type = GRPQUOTA;
1825
1826                 /* inode accounting */
1827                 qi->lqi_is_blk = false;
1828
1829                 /* one more inode for the new group owner ... */
1830                 qi->lqi_id.qid_gid = attr->la_gid;
1831                 qi->lqi_space      = 1;
1832                 allocated = (attr->la_gid == 0) ? true : false;
1833                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1834                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1835                         rc = 0;
1836                 if (rc)
1837                         RETURN(rc);
1838
1839                 /* and one less inode for the current gid */
1840                 qi->lqi_id.qid_gid = gid;
1841                 qi->lqi_space      = -1;
1842                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1843                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1844                         rc = 0;
1845                 if (rc)
1846                         RETURN(rc);
1847
1848                 /* block accounting */
1849                 qi->lqi_is_blk = true;
1850
1851                 /* more blocks for the new owner ... */
1852                 qi->lqi_id.qid_gid = attr->la_gid;
1853                 qi->lqi_space      = bspace;
1854                 allocated = (attr->la_gid == 0) ? true : false;
1855                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1856                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1857                         rc = 0;
1858                 if (rc)
1859                         RETURN(rc);
1860
1861                 /* and finally less blocks for the current owner */
1862                 qi->lqi_id.qid_gid = gid;
1863                 qi->lqi_space      = -bspace;
1864                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1865                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1866                         rc = 0;
1867                 if (rc)
1868                         RETURN(rc);
1869         }
1870
1871         RETURN(rc);
1872 }
1873
1874 static int osd_inode_setattr(const struct lu_env *env,
1875                              struct inode *inode, const struct lu_attr *attr)
1876 {
1877         __u64 bits = attr->la_valid;
1878
1879         /* Only allow set size for regular file */
1880         if (!S_ISREG(inode->i_mode))
1881                 bits &= ~(LA_SIZE | LA_BLOCKS);
1882
1883         if (bits == 0)
1884                 return 0;
1885
1886         if (bits & LA_ATIME)
1887                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1888         if (bits & LA_CTIME)
1889                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1890         if (bits & LA_MTIME)
1891                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1892         if (bits & LA_SIZE) {
1893                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1894                 i_size_write(inode, attr->la_size);
1895         }
1896
1897 #if 0
1898         /* OSD should not change "i_blocks" which is used by quota.
1899          * "i_blocks" should be changed by ldiskfs only. */
1900         if (bits & LA_BLOCKS)
1901                 inode->i_blocks = attr->la_blocks;
1902 #endif
1903         if (bits & LA_MODE)
1904                 inode->i_mode = (inode->i_mode & S_IFMT) |
1905                                 (attr->la_mode & ~S_IFMT);
1906         if (bits & LA_UID)
1907                 i_uid_write(inode, attr->la_uid);
1908         if (bits & LA_GID)
1909                 i_gid_write(inode, attr->la_gid);
1910         if (bits & LA_NLINK)
1911                 set_nlink(inode, attr->la_nlink);
1912         if (bits & LA_RDEV)
1913                 inode->i_rdev = attr->la_rdev;
1914
1915         if (bits & LA_FLAGS) {
1916                 /* always keep S_NOCMTIME */
1917                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1918                                  S_NOCMTIME;
1919         }
1920         return 0;
1921 }
1922
1923 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1924 {
1925         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1926             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1927                 struct iattr    iattr;
1928                 int             rc;
1929
1930                 iattr.ia_valid = 0;
1931                 if (attr->la_valid & LA_UID)
1932                         iattr.ia_valid |= ATTR_UID;
1933                 if (attr->la_valid & LA_GID)
1934                         iattr.ia_valid |= ATTR_GID;
1935                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1936                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1937
1938                 rc = ll_vfs_dq_transfer(inode, &iattr);
1939                 if (rc) {
1940                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1941                                "enforcement enabled on the ldiskfs "
1942                                "filesystem?\n", inode->i_sb->s_id, rc);
1943                         return rc;
1944                 }
1945         }
1946         return 0;
1947 }
1948
1949 static int osd_attr_set(const struct lu_env *env,
1950                         struct dt_object *dt,
1951                         const struct lu_attr *attr,
1952                         struct thandle *handle,
1953                         struct lustre_capa *capa)
1954 {
1955         struct osd_object *obj = osd_dt_obj(dt);
1956         struct inode      *inode;
1957         int rc;
1958
1959         LASSERT(handle != NULL);
1960         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1961         LASSERT(osd_invariant(obj));
1962
1963         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1964                 return -EACCES;
1965
1966         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1967
1968         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1969                 struct osd_thread_info  *oti  = osd_oti_get(env);
1970                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1971                 struct lu_fid           *fid1 = &oti->oti_fid;
1972                 struct osd_inode_id     *id   = &oti->oti_id;
1973                 struct iam_path_descr   *ipd;
1974                 struct iam_container    *bag;
1975                 struct osd_thandle      *oh;
1976                 int                      rc;
1977
1978                 fid_cpu_to_be(fid1, fid0);
1979                 memset(id, 1, sizeof(*id));
1980                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1981                                   fid0)->oi_dir.od_container;
1982                 ipd = osd_idx_ipd_get(env, bag);
1983                 if (unlikely(ipd == NULL))
1984                         RETURN(-ENOMEM);
1985
1986                 oh = container_of0(handle, struct osd_thandle, ot_super);
1987                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1988                                 (const struct iam_rec *)id, ipd);
1989                 osd_ipd_put(env, bag, ipd);
1990                 return(rc > 0 ? 0 : rc);
1991         }
1992
1993         inode = obj->oo_inode;
1994         ll_vfs_dq_init(inode);
1995
1996         rc = osd_quota_transfer(inode, attr);
1997         if (rc)
1998                 return rc;
1999
2000         spin_lock(&obj->oo_guard);
2001         rc = osd_inode_setattr(env, inode, attr);
2002         spin_unlock(&obj->oo_guard);
2003
2004         if (!rc)
2005                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2006         return rc;
2007 }
2008
2009 struct dentry *osd_child_dentry_get(const struct lu_env *env,
2010                                     struct osd_object *obj,
2011                                     const char *name, const int namelen)
2012 {
2013         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2014 }
2015
2016 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2017                       umode_t mode, struct dt_allocation_hint *hint,
2018                       struct thandle *th)
2019 {
2020         int result;
2021         struct osd_device  *osd = osd_obj2dev(obj);
2022         struct osd_thandle *oth;
2023         struct dt_object   *parent = NULL;
2024         struct inode       *inode;
2025
2026         LINVRNT(osd_invariant(obj));
2027         LASSERT(obj->oo_inode == NULL);
2028         LASSERT(obj->oo_hl_head == NULL);
2029
2030         if (S_ISDIR(mode) && ldiskfs_pdo) {
2031                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2032                 if (obj->oo_hl_head == NULL)
2033                         return -ENOMEM;
2034         }
2035
2036         oth = container_of(th, struct osd_thandle, ot_super);
2037         LASSERT(oth->ot_handle->h_transaction != NULL);
2038
2039         if (hint && hint->dah_parent)
2040                 parent = hint->dah_parent;
2041
2042         inode = ldiskfs_create_inode(oth->ot_handle,
2043                                      parent ? osd_dt_obj(parent)->oo_inode :
2044                                               osd_sb(osd)->s_root->d_inode,
2045                                      mode);
2046         if (!IS_ERR(inode)) {
2047                 /* Do not update file c/mtime in ldiskfs.
2048                  * NB: don't need any lock because no contention at this
2049                  * early stage */
2050                 inode->i_flags |= S_NOCMTIME;
2051
2052                 /* For new created object, it must be consistent,
2053                  * and it is unnecessary to scrub against it. */
2054                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2055                 obj->oo_inode = inode;
2056                 result = 0;
2057         } else {
2058                 if (obj->oo_hl_head != NULL) {
2059                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2060                         obj->oo_hl_head = NULL;
2061                 }
2062                 result = PTR_ERR(inode);
2063         }
2064         LINVRNT(osd_invariant(obj));
2065         return result;
2066 }
2067
2068 enum {
2069         OSD_NAME_LEN = 255
2070 };
2071
2072 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2073                      struct lu_attr *attr,
2074                      struct dt_allocation_hint *hint,
2075                      struct dt_object_format *dof,
2076                      struct thandle *th)
2077 {
2078         int result;
2079         struct osd_thandle *oth;
2080         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
2081
2082         LASSERT(S_ISDIR(attr->la_mode));
2083
2084         oth = container_of(th, struct osd_thandle, ot_super);
2085         LASSERT(oth->ot_handle->h_transaction != NULL);
2086         result = osd_mkfile(info, obj, mode, hint, th);
2087
2088         return result;
2089 }
2090
2091 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2092                         struct lu_attr *attr,
2093                         struct dt_allocation_hint *hint,
2094                         struct dt_object_format *dof,
2095                         struct thandle *th)
2096 {
2097         int result;
2098         struct osd_thandle *oth;
2099         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2100
2101         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2102
2103         LASSERT(S_ISREG(attr->la_mode));
2104
2105         oth = container_of(th, struct osd_thandle, ot_super);
2106         LASSERT(oth->ot_handle->h_transaction != NULL);
2107
2108         result = osd_mkfile(info, obj, mode, hint, th);
2109         if (result == 0) {
2110                 LASSERT(obj->oo_inode != NULL);
2111                 if (feat->dif_flags & DT_IND_VARKEY)
2112                         result = iam_lvar_create(obj->oo_inode,
2113                                                  feat->dif_keysize_max,
2114                                                  feat->dif_ptrsize,
2115                                                  feat->dif_recsize_max,
2116                                                  oth->ot_handle);
2117                 else
2118                         result = iam_lfix_create(obj->oo_inode,
2119                                                  feat->dif_keysize_max,
2120                                                  feat->dif_ptrsize,
2121                                                  feat->dif_recsize_max,
2122                                                  oth->ot_handle);
2123
2124         }
2125         return result;
2126 }
2127
2128 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2129                      struct lu_attr *attr,
2130                      struct dt_allocation_hint *hint,
2131                      struct dt_object_format *dof,
2132                      struct thandle *th)
2133 {
2134         LASSERT(S_ISREG(attr->la_mode));
2135         return osd_mkfile(info, obj, (attr->la_mode &
2136                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2137 }
2138
2139 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2140                      struct lu_attr *attr,
2141                      struct dt_allocation_hint *hint,
2142                      struct dt_object_format *dof,
2143                      struct thandle *th)
2144 {
2145         LASSERT(S_ISLNK(attr->la_mode));
2146         return osd_mkfile(info, obj, (attr->la_mode &
2147                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2148 }
2149
2150 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2151                      struct lu_attr *attr,
2152                      struct dt_allocation_hint *hint,
2153                      struct dt_object_format *dof,
2154                      struct thandle *th)
2155 {
2156         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2157         int result;
2158
2159         LINVRNT(osd_invariant(obj));
2160         LASSERT(obj->oo_inode == NULL);
2161         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2162                 S_ISFIFO(mode) || S_ISSOCK(mode));
2163
2164         result = osd_mkfile(info, obj, mode, hint, th);
2165         if (result == 0) {
2166                 LASSERT(obj->oo_inode != NULL);
2167                 /*
2168                  * This inode should be marked dirty for i_rdev.  Currently
2169                  * that is done in the osd_attr_init().
2170                  */
2171                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2172                                    attr->la_rdev);
2173         }
2174         LINVRNT(osd_invariant(obj));
2175         return result;
2176 }
2177
2178 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2179                               struct lu_attr *,
2180                               struct dt_allocation_hint *hint,
2181                               struct dt_object_format *dof,
2182                               struct thandle *);
2183
2184 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2185 {
2186         osd_obj_type_f result;
2187
2188         switch (type) {
2189         case DFT_DIR:
2190                 result = osd_mkdir;
2191                 break;
2192         case DFT_REGULAR:
2193                 result = osd_mkreg;
2194                 break;
2195         case DFT_SYM:
2196                 result = osd_mksym;
2197                 break;
2198         case DFT_NODE:
2199                 result = osd_mknod;
2200                 break;
2201         case DFT_INDEX:
2202                 result = osd_mk_index;
2203                 break;
2204
2205         default:
2206                 LBUG();
2207                 break;
2208         }
2209         return result;
2210 }
2211
2212
2213 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2214                         struct dt_object *parent, struct dt_object *child,
2215                         umode_t child_mode)
2216 {
2217         LASSERT(ah);
2218
2219         ah->dah_parent = parent;
2220         ah->dah_mode = child_mode;
2221 }
2222
2223 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2224                           struct lu_attr *attr, struct dt_object_format *dof)
2225 {
2226         struct inode   *inode = obj->oo_inode;
2227         __u64           valid = attr->la_valid;
2228         int             result;
2229
2230         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2231
2232         if (dof->dof_type != DFT_NODE)
2233                 attr->la_valid &= ~LA_RDEV;
2234         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2235                 attr->la_valid &= ~LA_ATIME;
2236         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2237                 attr->la_valid &= ~LA_CTIME;
2238         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2239                 attr->la_valid &= ~LA_MTIME;
2240
2241         result = osd_quota_transfer(inode, attr);
2242         if (result)
2243                 return;
2244
2245         if (attr->la_valid != 0) {
2246                 result = osd_inode_setattr(info->oti_env, inode, attr);
2247                 /*
2248                  * The osd_inode_setattr() should always succeed here.  The
2249                  * only error that could be returned is EDQUOT when we are
2250                  * trying to change the UID or GID of the inode. However, this
2251                  * should not happen since quota enforcement is no longer
2252                  * enabled on ldiskfs (lquota takes care of it).
2253                  */
2254                 LASSERTF(result == 0, "%d\n", result);
2255                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2256         }
2257
2258         attr->la_valid = valid;
2259 }
2260
2261 /**
2262  * Helper function for osd_object_create()
2263  *
2264  * \retval 0, on success
2265  */
2266 static int __osd_object_create(struct osd_thread_info *info,
2267                                struct osd_object *obj, struct lu_attr *attr,
2268                                struct dt_allocation_hint *hint,
2269                                struct dt_object_format *dof,
2270                                struct thandle *th)
2271 {
2272         int     result;
2273         __u32   umask;
2274
2275         /* we drop umask so that permissions we pass are not affected */
2276         umask = current->fs->umask;
2277         current->fs->umask = 0;
2278
2279         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2280                                                   th);
2281         if (result == 0) {
2282                 osd_attr_init(info, obj, attr, dof);
2283                 osd_object_init0(obj);
2284                 /* bz 24037 */
2285                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
2286                         unlock_new_inode(obj->oo_inode);
2287         }
2288
2289         /* restore previous umask value */
2290         current->fs->umask = umask;
2291
2292         return result;
2293 }
2294
2295 /**
2296  * Helper function for osd_object_create()
2297  *
2298  * \retval 0, on success
2299  */
2300 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2301                            const struct lu_fid *fid, struct thandle *th)
2302 {
2303         struct osd_thread_info *info = osd_oti_get(env);
2304         struct osd_inode_id    *id   = &info->oti_id;
2305         struct osd_device      *osd  = osd_obj2dev(obj);
2306         struct osd_thandle     *oh;
2307
2308         LASSERT(obj->oo_inode != NULL);
2309
2310         oh = container_of0(th, struct osd_thandle, ot_super);
2311         LASSERT(oh->ot_handle);
2312
2313         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2314         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2315 }
2316
2317 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2318                    obd_seq seq, struct lu_seq_range *range)
2319 {
2320         struct seq_server_site  *ss = osd_seq_site(osd);
2321
2322         if (fid_seq_is_idif(seq)) {
2323                 fld_range_set_ost(range);
2324                 range->lsr_index = idif_ost_idx(seq);
2325                 return 0;
2326         }
2327
2328         if (!fid_seq_in_fldb(seq)) {
2329                 fld_range_set_mdt(range);
2330                 if (ss != NULL)
2331                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2332                          * at all */
2333                         range->lsr_index = ss->ss_node_id;
2334                 return 0;
2335         }
2336
2337         LASSERT(ss != NULL);
2338         fld_range_set_any(range);
2339         /* OSD will only do local fld lookup */
2340         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2341 }
2342
2343 /*
2344  * Concurrency: no external locking is necessary.
2345  */
2346 static int osd_declare_object_create(const struct lu_env *env,
2347                                      struct dt_object *dt,
2348                                      struct lu_attr *attr,
2349                                      struct dt_allocation_hint *hint,
2350                                      struct dt_object_format *dof,
2351                                      struct thandle *handle)
2352 {
2353         struct osd_thandle      *oh;
2354         int                      rc;
2355         ENTRY;
2356
2357         LASSERT(handle != NULL);
2358
2359         oh = container_of0(handle, struct osd_thandle, ot_super);
2360         LASSERT(oh->ot_handle == NULL);
2361
2362         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2363                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2364         /* Reuse idle OI block may cause additional one OI block
2365          * to be changed. */
2366         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2367                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2368
2369         /* If this is directory, then we expect . and .. to be inserted as
2370          * well. The one directory block always needs to be created for the
2371          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2372          * block), there is no danger of needing a tree for the first block.
2373          */
2374         if (attr && S_ISDIR(attr->la_mode)) {
2375                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2376                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2377                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2378         }
2379
2380         if (!attr)
2381                 RETURN(0);
2382
2383         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2384                                    false, false, NULL, false);
2385         if (rc != 0)
2386                 RETURN(rc);
2387
2388         RETURN(rc);
2389 }
2390
2391 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2392                              struct lu_attr *attr,
2393                              struct dt_allocation_hint *hint,
2394                              struct dt_object_format *dof,
2395                              struct thandle *th)
2396 {
2397         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2398         struct osd_object      *obj    = osd_dt_obj(dt);
2399         struct osd_thread_info *info   = osd_oti_get(env);
2400         int result;
2401
2402         ENTRY;
2403
2404         LINVRNT(osd_invariant(obj));
2405         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2406         LASSERT(osd_write_locked(env, obj));
2407         LASSERT(th != NULL);
2408
2409         if (unlikely(fid_is_acct(fid)))
2410                 /* Quota files can't be created from the kernel any more,
2411                  * 'tune2fs -O quota' will take care of creating them */
2412                 RETURN(-EPERM);
2413
2414         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2415         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2416
2417         result = __osd_object_create(info, obj, attr, hint, dof, th);
2418         if (result == 0)
2419                 result = __osd_oi_insert(env, obj, fid, th);
2420
2421         LASSERT(ergo(result == 0,
2422                      dt_object_exists(dt) && !dt_object_remote(dt)));
2423
2424         LASSERT(osd_invariant(obj));
2425         RETURN(result);
2426 }
2427
2428 /**
2429  * Called to destroy on-disk representation of the object
2430  *
2431  * Concurrency: must be locked
2432  */
2433 static int osd_declare_object_destroy(const struct lu_env *env,
2434                                       struct dt_object *dt,
2435                                       struct thandle *th)
2436 {
2437         struct osd_object  *obj = osd_dt_obj(dt);
2438         struct inode       *inode = obj->oo_inode;
2439         struct osd_thandle *oh;
2440         int                 rc;
2441         ENTRY;
2442
2443         oh = container_of0(th, struct osd_thandle, ot_super);
2444         LASSERT(oh->ot_handle == NULL);
2445         LASSERT(inode);
2446
2447         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2448                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2449         /* Recycle idle OI leaf may cause additional three OI blocks
2450          * to be changed. */
2451         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2452                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2453         /* one less inode */
2454         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2455                                    -1, oh, false, true, NULL, false);
2456         if (rc)
2457                 RETURN(rc);
2458         /* data to be truncated */
2459         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2460                                    0, oh, true, true, NULL, false);
2461         RETURN(rc);
2462 }
2463
2464 static int osd_object_destroy(const struct lu_env *env,
2465                               struct dt_object *dt,
2466                               struct thandle *th)
2467 {
2468         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2469         struct osd_object      *obj = osd_dt_obj(dt);
2470         struct inode           *inode = obj->oo_inode;
2471         struct osd_device      *osd = osd_obj2dev(obj);
2472         struct osd_thandle     *oh;
2473         int                     result;
2474         ENTRY;
2475
2476         oh = container_of0(th, struct osd_thandle, ot_super);
2477         LASSERT(oh->ot_handle);
2478         LASSERT(inode);
2479         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2480
2481         if (unlikely(fid_is_acct(fid)))
2482                 RETURN(-EPERM);
2483
2484         if (S_ISDIR(inode->i_mode)) {
2485                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2486                         inode->i_nlink == 2);
2487                 /* it will check/delete the inode from remote parent,
2488                  * how to optimize it? unlink performance impaction XXX */
2489                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2490                 if (result != 0 && result != -ENOENT) {
2491                         CERROR("%s: delete inode "DFID": rc = %d\n",
2492                                osd_name(osd), PFID(fid), result);
2493                 }
2494                 spin_lock(&obj->oo_guard);
2495                 clear_nlink(inode);
2496                 spin_unlock(&obj->oo_guard);
2497                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2498         }
2499
2500         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2501
2502         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2503                                OI_CHECK_FLD);
2504
2505         /* XXX: add to ext3 orphan list */
2506         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2507
2508         /* not needed in the cache anymore */
2509         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2510
2511         RETURN(0);
2512 }
2513
2514 /**
2515  * Put the fid into lustre_mdt_attrs, and then place the structure
2516  * inode's ea. This fid should not be altered during the life time
2517  * of the inode.
2518  *
2519  * \retval +ve, on success
2520  * \retval -ve, on error
2521  *
2522  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2523  */
2524 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2525                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2526 {
2527         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2528         int                      rc;
2529         ENTRY;
2530
2531         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2532                 RETURN(0);
2533
2534         lustre_lma_init(lma, fid, compat, incompat);
2535         lustre_lma_swab(lma);
2536
2537         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2538                              XATTR_CREATE);
2539         /* LMA may already exist, but we need to check that all the
2540          * desired compat/incompat flags have been added. */
2541         if (unlikely(rc == -EEXIST)) {
2542                 if (compat == 0 && incompat == 0)
2543                         RETURN(0);
2544
2545                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2546                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2547                                      LMA_OLD_SIZE);
2548                 if (rc <= 0)
2549                         RETURN(-EINVAL);
2550
2551                 lustre_lma_swab(lma);
2552                 if (!(~lma->lma_compat & compat) &&
2553                     !(~lma->lma_incompat & incompat))
2554                         RETURN(0);
2555
2556                 lma->lma_compat |= compat;
2557                 lma->lma_incompat |= incompat;
2558                 lustre_lma_swab(lma);
2559                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2560                                      sizeof(*lma), XATTR_REPLACE);
2561         }
2562
2563         RETURN(rc);
2564 }
2565
2566 /**
2567  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2568  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2569  * To have compatilibility with 1.8 ldiskfs driver we need to have
2570  * magic number at start of fid data.
2571  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2572  * its inmemory API.
2573  */
2574 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2575                                   const struct lu_fid *fid)
2576 {
2577         if (!fid_is_namespace_visible(fid) ||
2578             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2579                 param->edp_magic = 0;
2580                 return;
2581         }
2582
2583         param->edp_magic = LDISKFS_LUFID_MAGIC;
2584         param->edp_len =  sizeof(struct lu_fid) + 1;
2585         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2586 }
2587
2588 /**
2589  * Try to read the fid from inode ea into dt_rec.
2590  *
2591  * \param fid object fid.
2592  *
2593  * \retval 0 on success
2594  */
2595 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2596                           __u32 ino, struct lu_fid *fid,
2597                           struct osd_inode_id *id)
2598 {
2599         struct osd_thread_info *info  = osd_oti_get(env);
2600         struct inode           *inode;
2601         ENTRY;
2602
2603         osd_id_gen(id, ino, OSD_OII_NOGEN);
2604         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2605         if (IS_ERR(inode))
2606                 RETURN(PTR_ERR(inode));
2607
2608         iput(inode);
2609         RETURN(0);
2610 }
2611
2612 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2613                                         struct inode *dir,
2614                                         struct inode *parent_dir,
2615                                         const struct lu_fid *dot_fid,
2616                                         const struct lu_fid *dot_dot_fid,
2617                                         struct osd_thandle *oth)
2618 {
2619         struct ldiskfs_dentry_param *dot_ldp;
2620         struct ldiskfs_dentry_param *dot_dot_ldp;
2621
2622         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2623         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2624
2625         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2626         dot_ldp->edp_magic = 0;
2627         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2628                                         dir, dot_ldp, dot_dot_ldp);
2629 }
2630
2631 /**
2632  * Create an local agent inode for remote entry
2633  */
2634 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2635                                                   struct osd_device *osd,
2636                                                   struct osd_object *pobj,
2637                                                   const struct lu_fid *fid,
2638                                                   __u32 type,
2639                                                   struct thandle *th)
2640 {
2641         struct osd_thread_info  *info = osd_oti_get(env);
2642         struct inode            *local;
2643         struct osd_thandle      *oh;
2644         int                     rc;
2645         ENTRY;
2646
2647         LASSERT(th);
2648         oh = container_of(th, struct osd_thandle, ot_super);
2649         LASSERT(oh->ot_handle->h_transaction != NULL);
2650
2651         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2652         if (IS_ERR(local)) {
2653                 CERROR("%s: create local error %d\n", osd_name(osd),
2654                        (int)PTR_ERR(local));
2655                 RETURN(local);
2656         }
2657
2658         /* Set special LMA flag for local agent inode */
2659         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2660         if (rc != 0) {
2661                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2662                        osd_name(osd), PFID(fid), rc);
2663                 RETURN(ERR_PTR(rc));
2664         }
2665
2666         if (!S_ISDIR(type))
2667                 RETURN(local);
2668
2669         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2670                                          lu_object_fid(&pobj->oo_dt.do_lu),
2671                                          fid, oh);
2672         if (rc != 0) {
2673                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2674                         osd_name(osd), PFID(fid), rc);
2675                 RETURN(ERR_PTR(rc));
2676         }
2677
2678         RETURN(local);
2679 }
2680
2681 /**
2682  * Delete local agent inode for remote entry
2683  */
2684 static int osd_delete_local_agent_inode(const struct lu_env *env,
2685                                         struct osd_device *osd,
2686                                         const struct lu_fid *fid,
2687                                         __u32 ino, struct osd_thandle *oh)
2688 {
2689         struct osd_thread_info  *oti = osd_oti_get(env);
2690         struct osd_inode_id     *id = &oti->oti_id;
2691         struct inode            *inode;
2692         ENTRY;
2693
2694         id->oii_ino = le32_to_cpu(ino);
2695         id->oii_gen = OSD_OII_NOGEN;
2696         inode = osd_iget(oti, osd, id);
2697         if (IS_ERR(inode)) {
2698                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2699                        PFID(fid), id->oii_ino, id->oii_gen);
2700                 RETURN(PTR_ERR(inode));
2701         }
2702
2703         clear_nlink(inode);
2704         mark_inode_dirty(inode);
2705         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2706                 osd_name(osd), PFID(fid), inode->i_ino);
2707         iput(inode);
2708         RETURN(0);
2709 }
2710
2711 /**
2712  * OSD layer object create function for interoperability mode (b11826).
2713  * This is mostly similar to osd_object_create(). Only difference being, fid is
2714  * inserted into inode ea here.
2715  *
2716  * \retval   0, on success
2717  * \retval -ve, on error
2718  */
2719 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2720                                 struct lu_attr *attr,
2721                                 struct dt_allocation_hint *hint,
2722                                 struct dt_object_format *dof,
2723                                 struct thandle *th)
2724 {
2725         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2726         struct osd_object      *obj    = osd_dt_obj(dt);
2727         struct osd_thread_info *info   = osd_oti_get(env);
2728         int                     result;
2729
2730         ENTRY;
2731
2732         LASSERT(osd_invariant(obj));
2733         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2734         LASSERT(osd_write_locked(env, obj));
2735         LASSERT(th != NULL);
2736
2737         if (unlikely(fid_is_acct(fid)))
2738                 /* Quota files can't be created from the kernel any more,
2739                  * 'tune2fs -O quota' will take care of creating them */
2740                 RETURN(-EPERM);
2741
2742         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2743         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2744
2745         result = __osd_object_create(info, obj, attr, hint, dof, th);
2746         if (result == 0)
2747                 result = osd_ea_fid_set(info, obj->oo_inode, fid,
2748                                 fid_is_on_ost(info, osd_obj2dev(obj),
2749                                               fid, OI_CHECK_FLD) ?
2750                                 LMAC_FID_ON_OST : 0, 0);
2751
2752         if (result == 0)
2753                 result = __osd_oi_insert(env, obj, fid, th);
2754
2755         LASSERT(ergo(result == 0,
2756                      dt_object_exists(dt) && !dt_object_remote(dt)));
2757         LINVRNT(osd_invariant(obj));
2758         RETURN(result);
2759 }
2760
2761 static int osd_declare_object_ref_add(const struct lu_env *env,
2762                                       struct dt_object *dt,
2763                                       struct thandle *handle)
2764 {
2765         struct osd_thandle       *oh;
2766
2767         /* it's possible that object doesn't exist yet */
2768         LASSERT(handle != NULL);
2769
2770         oh = container_of0(handle, struct osd_thandle, ot_super);
2771         LASSERT(oh->ot_handle == NULL);
2772
2773         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2774                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2775
2776         return 0;
2777 }
2778
2779 /*
2780  * Concurrency: @dt is write locked.
2781  */
2782 static int osd_object_ref_add(const struct lu_env *env,
2783                               struct dt_object *dt, struct thandle *th)
2784 {
2785         struct osd_object  *obj = osd_dt_obj(dt);
2786         struct inode       *inode = obj->oo_inode;
2787         struct osd_thandle *oh;
2788         int                 rc = 0;
2789
2790         LINVRNT(osd_invariant(obj));
2791         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2792         LASSERT(osd_write_locked(env, obj));
2793         LASSERT(th != NULL);
2794
2795         oh = container_of0(th, struct osd_thandle, ot_super);
2796         LASSERT(oh->ot_handle != NULL);
2797
2798         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2799
2800         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2801                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2802         /*
2803          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2804          * (65000) subdirectories by storing "1" in i_nlink if the link count
2805          * would otherwise overflow. Directory tranversal tools understand
2806          * that (st_nlink == 1) indicates that the filesystem dose not track
2807          * hard links count on the directory, and will not abort subdirectory
2808          * scanning early once (st_nlink - 2) subdirs have been found.
2809          *
2810          * This also has to properly handle the case of inodes with nlink == 0
2811          * in case they are being linked into the PENDING directory
2812          */
2813         spin_lock(&obj->oo_guard);
2814         if (unlikely(inode->i_nlink == 0))
2815                 /* inc_nlink from 0 may cause WARN_ON */
2816                 set_nlink(inode, 1);
2817         else {
2818                 ldiskfs_inc_count(oh->ot_handle, inode);
2819                 if (!S_ISDIR(inode->i_mode))
2820                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2821         }
2822         spin_unlock(&obj->oo_guard);
2823
2824         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2825         LINVRNT(osd_invariant(obj));
2826
2827         return rc;
2828 }
2829
2830 static int osd_declare_object_ref_del(const struct lu_env *env,
2831                                       struct dt_object *dt,
2832                                       struct thandle *handle)
2833 {
2834         struct osd_thandle *oh;
2835
2836         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2837         LASSERT(handle != NULL);
2838
2839         oh = container_of0(handle, struct osd_thandle, ot_super);
2840         LASSERT(oh->ot_handle == NULL);
2841
2842         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2843                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2844
2845         return 0;
2846 }
2847
2848 /*
2849  * Concurrency: @dt is write locked.
2850  */
2851 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2852                               struct thandle *th)
2853 {
2854         struct osd_object       *obj = osd_dt_obj(dt);
2855         struct inode            *inode = obj->oo_inode;
2856         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2857         struct osd_thandle      *oh;
2858
2859         LINVRNT(osd_invariant(obj));
2860         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2861         LASSERT(osd_write_locked(env, obj));
2862         LASSERT(th != NULL);
2863
2864         oh = container_of0(th, struct osd_thandle, ot_super);
2865         LASSERT(oh->ot_handle != NULL);
2866
2867         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2868
2869         spin_lock(&obj->oo_guard);
2870         /* That can be result of upgrade from old Lustre version and
2871          * applied only to local files.  Just skip this ref_del call.
2872          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2873         if (inode->i_nlink == 0) {
2874                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2875                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2876                              ", maybe an upgraded file? (LU-3915)\n",
2877                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2878                 spin_unlock(&obj->oo_guard);
2879                 return 0;
2880         }
2881
2882         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2883                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2884
2885         ldiskfs_dec_count(oh->ot_handle, inode);
2886         spin_unlock(&obj->oo_guard);
2887
2888         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2889         LINVRNT(osd_invariant(obj));
2890
2891         return 0;
2892 }
2893
2894 /*
2895  * Get the 64-bit version for an inode.
2896  */
2897 static int osd_object_version_get(const struct lu_env *env,
2898                                   struct dt_object *dt, dt_obj_version_t *ver)
2899 {
2900         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2901
2902         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2903                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2904         *ver = LDISKFS_I(inode)->i_fs_version;
2905         return 0;
2906 }
2907
2908 /*
2909  * Concurrency: @dt is read locked.
2910  */
2911 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2912                          struct lu_buf *buf, const char *name,
2913                          struct lustre_capa *capa)
2914 {
2915         struct osd_object      *obj    = osd_dt_obj(dt);
2916         struct inode           *inode  = obj->oo_inode;
2917         struct osd_thread_info *info   = osd_oti_get(env);
2918         struct dentry          *dentry = &info->oti_obj_dentry;
2919
2920         /* version get is not real XATTR but uses xattr API */
2921         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2922                 /* for version we are just using xattr API but change inode
2923                  * field instead */
2924                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2925                 osd_object_version_get(env, dt, buf->lb_buf);
2926                 return sizeof(dt_obj_version_t);
2927         }
2928
2929         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2930         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2931
2932         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2933                 return -EACCES;
2934
2935         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2936 }
2937
2938
2939 static int osd_declare_xattr_set(const struct lu_env *env,
2940                                  struct dt_object *dt,
2941                                  const struct lu_buf *buf, const char *name,
2942                                  int fl, struct thandle *handle)
2943 {
2944         struct osd_thandle *oh;
2945         int credits;
2946
2947         LASSERT(handle != NULL);
2948
2949         oh = container_of0(handle, struct osd_thandle, ot_super);
2950         LASSERT(oh->ot_handle == NULL);
2951
2952         /* optimistic optimization: LMA is set first and usually fit inode */
2953         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2954                 if (dt_object_exists(dt))
2955                         credits = 0;
2956                 else
2957                         credits = 1;
2958         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2959                 credits = 1;
2960         } else {
2961                 struct osd_device  *osd = osd_dev(dt->do_lu.lo_dev);
2962                 struct super_block *sb = osd_sb(osd);
2963                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2964                 if (buf && buf->lb_len > sb->s_blocksize) {
2965                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2966                                         sb->s_blocksize_bits;
2967                 }
2968         }
2969
2970         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2971
2972         return 0;
2973 }
2974
2975 /*
2976  * Set the 64-bit version for object
2977  */
2978 static void osd_object_version_set(const struct lu_env *env,
2979                                    struct dt_object *dt,
2980                                    dt_obj_version_t *new_version)
2981 {
2982         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2983
2984         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2985                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2986
2987         LDISKFS_I(inode)->i_fs_version = *new_version;
2988         /** Version is set after all inode operations are finished,
2989          *  so we should mark it dirty here */
2990         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2991 }
2992
2993 /*
2994  * Concurrency: @dt is write locked.
2995  */
2996 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2997                          const struct lu_buf *buf, const char *name, int fl,
2998                          struct thandle *handle, struct lustre_capa *capa)
2999 {
3000         struct osd_object      *obj      = osd_dt_obj(dt);
3001         struct inode           *inode    = obj->oo_inode;
3002         struct osd_thread_info *info     = osd_oti_get(env);
3003         int                     fs_flags = 0;
3004         ENTRY;
3005
3006         LASSERT(handle != NULL);
3007
3008         /* version set is not real XATTR */
3009         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
3010                 /* for version we are just using xattr API but change inode
3011                  * field instead */
3012                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
3013                 osd_object_version_set(env, dt, buf->lb_buf);
3014                 return sizeof(dt_obj_version_t);
3015         }
3016
3017         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3018                 return -EACCES;
3019
3020         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zd\n",
3021                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
3022
3023         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3024         if (fl & LU_XATTR_REPLACE)
3025                 fs_flags |= XATTR_REPLACE;
3026
3027         if (fl & LU_XATTR_CREATE)
3028                 fs_flags |= XATTR_CREATE;
3029
3030         if (strcmp(name, XATTR_NAME_LMV) == 0) {
3031                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
3032                 int                      rc;
3033
3034                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
3035                 if (rc != 0)
3036                         RETURN(rc);
3037
3038                 lma->lma_incompat |= LMAI_STRIPED;
3039                 lustre_lma_swab(lma);
3040                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
3041                                      sizeof(*lma), XATTR_REPLACE);
3042                 if (rc != 0)
3043                         RETURN(rc);
3044         }
3045
3046         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
3047                                fs_flags);
3048 }
3049
3050 /*
3051  * Concurrency: @dt is read locked.
3052  */
3053 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
3054                           struct lu_buf *buf, struct lustre_capa *capa)
3055 {
3056         struct osd_object      *obj    = osd_dt_obj(dt);
3057         struct inode           *inode  = obj->oo_inode;
3058         struct osd_thread_info *info   = osd_oti_get(env);
3059         struct dentry          *dentry = &info->oti_obj_dentry;
3060
3061         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3062         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
3063
3064         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
3065                 return -EACCES;
3066
3067         dentry->d_inode = inode;
3068         dentry->d_sb = inode->i_sb;
3069         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
3070 }
3071
3072 static int osd_declare_xattr_del(const struct lu_env *env,
3073                                  struct dt_object *dt, const char *name,
3074                                  struct thandle *handle)
3075 {
3076         struct osd_thandle *oh;
3077
3078         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3079         LASSERT(handle != NULL);
3080
3081         oh = container_of0(handle, struct osd_thandle, ot_super);
3082         LASSERT(oh->ot_handle == NULL);
3083
3084         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
3085                              osd_dto_credits_noquota[DTO_XATTR_SET]);
3086
3087         return 0;
3088 }
3089
3090 /*
3091  * Concurrency: @dt is write locked.
3092  */
3093 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
3094                          const char *name, struct thandle *handle,
3095                          struct lustre_capa *capa)
3096 {
3097         struct osd_object      *obj    = osd_dt_obj(dt);
3098         struct inode           *inode  = obj->oo_inode;
3099         struct osd_thread_info *info   = osd_oti_get(env);
3100         struct dentry          *dentry = &info->oti_obj_dentry;
3101         int                     rc;
3102
3103         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3104         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
3105         LASSERT(handle != NULL);
3106
3107         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
3108                 return -EACCES;
3109
3110         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
3111
3112         ll_vfs_dq_init(inode);
3113         dentry->d_inode = inode;
3114         dentry->d_sb = inode->i_sb;
3115         rc = inode->i_op->removexattr(dentry, name);
3116         return rc;
3117 }
3118
3119 static struct obd_capa *osd_capa_get(const struct lu_env *env,
3120                                      struct dt_object *dt,
3121                                      struct lustre_capa *old, __u64 opc)
3122 {
3123         struct osd_thread_info *info = osd_oti_get(env);
3124         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
3125         struct osd_object *obj = osd_dt_obj(dt);
3126         struct osd_device *osd = osd_obj2dev(obj);
3127         struct lustre_capa_key *key = &info->oti_capa_key;
3128         struct lustre_capa *capa = &info->oti_capa;
3129         struct obd_capa *oc;
3130         struct lu_capainfo *lci;
3131         int rc;
3132         ENTRY;
3133
3134         if (!osd->od_fl_capa)
3135                 RETURN(ERR_PTR(-ENOENT));
3136
3137         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3138         LINVRNT(osd_invariant(obj));
3139
3140         /* renewal sanity check */
3141         if (old && osd_object_auth(env, dt, old, opc))
3142                 RETURN(ERR_PTR(-EACCES));
3143
3144         lci = lu_capainfo_get(env);
3145         if (unlikely(lci == NULL))
3146                 RETURN(ERR_PTR(-ENOENT));
3147
3148         switch (lci->lci_auth) {
3149         case LC_ID_NONE:
3150                 RETURN(NULL);
3151         case LC_ID_PLAIN:
3152                 capa->lc_uid = i_uid_read(obj->oo_inode);
3153                 capa->lc_gid = i_gid_read(obj->oo_inode);
3154                 capa->lc_flags = LC_ID_PLAIN;
3155                 break;
3156         case LC_ID_CONVERT: {
3157                 __u32 d[4], s[4];
3158
3159                 s[0] = i_uid_read(obj->oo_inode);
3160                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
3161                 s[2] = i_uid_read(obj->oo_inode);
3162                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
3163                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
3164                 if (unlikely(rc))
3165                         RETURN(ERR_PTR(rc));
3166
3167                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
3168                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
3169                 capa->lc_flags = LC_ID_CONVERT;
3170                 break;
3171         }
3172         default:
3173                 RETUR