Whamcloud - gitweb
LU-8974 osd-ldiskfs: increase supported ldiskfs fs size
[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.gnu.org/licenses/gpl-2.0.html
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Use is subject to license terms.
25  *
26  * Copyright (c) 2011, 2016, Intel Corporation.
27  */
28 /*
29  * This file is part of Lustre, http://www.lustre.org/
30  * Lustre is a trademark of Sun Microsystems, Inc.
31  *
32  * lustre/osd/osd_handler.c
33  *
34  * Top-level entry points into osd module
35  *
36  * Author: Nikita Danilov <nikita@clusterfs.com>
37  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
38  */
39
40 #define DEBUG_SUBSYSTEM S_OSD
41
42 #include <linux/module.h>
43 #include <linux/user_namespace.h>
44 #ifdef HAVE_UIDGID_HEADER
45 # include <linux/uidgid.h>
46 #endif
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 #include <ldiskfs/ldiskfs.h>
58 #include <ldiskfs/xattr.h>
59 #include <ldiskfs/ldiskfs_extents.h>
60 #undef ENTRY
61 /*
62  * struct OBD_{ALLOC,FREE}*()
63  * OBD_FAIL_CHECK
64  */
65 #include <obd_support.h>
66 /* struct ptlrpc_thread */
67 #include <lustre_net.h>
68 #include <lustre_fid.h>
69 /* process_config */
70 #include <lustre_param.h>
71
72 #include "osd_internal.h"
73 #include "osd_dynlocks.h"
74
75 /* llo_* api support */
76 #include <md_object.h>
77 #include <lustre_quota.h>
78
79 #include <lustre_linkea.h>
80
81 int ldiskfs_pdo = 1;
82 module_param(ldiskfs_pdo, int, 0644);
83 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
84
85 int ldiskfs_track_declares_assert;
86 module_param(ldiskfs_track_declares_assert, int, 0644);
87 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
88
89 /* Slab to allocate dynlocks */
90 struct kmem_cache *dynlock_cachep;
91
92 /* Slab to allocate osd_it_ea */
93 struct kmem_cache *osd_itea_cachep;
94
95 static struct lu_kmem_descr ldiskfs_caches[] = {
96         {
97                 .ckd_cache = &dynlock_cachep,
98                 .ckd_name  = "dynlock_cache",
99                 .ckd_size  = sizeof(struct dynlock_handle)
100         },
101         {
102                 .ckd_cache = &osd_itea_cachep,
103                 .ckd_name  = "osd_itea_cache",
104                 .ckd_size  = sizeof(struct osd_it_ea)
105         },
106         {
107                 .ckd_cache = NULL
108         }
109 };
110
111 static const char dot[] = ".";
112 static const char dotdot[] = "..";
113 static const char remote_obj_dir[] = "REM_OBJ_DIR";
114
115 static const struct lu_object_operations      osd_lu_obj_ops;
116 static const struct dt_object_operations      osd_obj_ops;
117 static const struct dt_object_operations      osd_obj_ea_ops;
118 static const struct dt_object_operations      osd_obj_otable_it_ops;
119 static const struct dt_index_operations       osd_index_iam_ops;
120 static const struct dt_index_operations       osd_index_ea_ops;
121
122 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
123                           const struct lu_fid *fid);
124 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
125                                                 struct osd_device *osd);
126
127 int osd_trans_declare_op2rb[] = {
128         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
129         [OSD_OT_PUNCH]          = OSD_OT_MAX,
130         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
131         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
132         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
133         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
134         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
135         [OSD_OT_WRITE]          = OSD_OT_WRITE,
136         [OSD_OT_INSERT]         = OSD_OT_DELETE,
137         [OSD_OT_DELETE]         = OSD_OT_INSERT,
138         [OSD_OT_QUOTA]          = OSD_OT_MAX,
139 };
140
141 static int osd_has_index(const struct osd_object *obj)
142 {
143         return obj->oo_dt.do_index_ops != NULL;
144 }
145
146 static int osd_object_invariant(const struct lu_object *l)
147 {
148         return osd_invariant(osd_obj(l));
149 }
150
151 /*
152  * Concurrency: doesn't matter
153  */
154
155 /*
156  * Concurrency: doesn't matter
157  */
158 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
159 {
160         struct osd_thread_info *oti = osd_oti_get(env);
161         return oti->oti_w_locks > 0 && o->oo_owner == env;
162 }
163
164 /*
165  * Concurrency: doesn't access mutable data
166  */
167 static int osd_root_get(const struct lu_env *env,
168                         struct dt_device *dev, struct lu_fid *f)
169 {
170         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
171         return 0;
172 }
173
174 /*
175  * the following set of functions are used to maintain per-thread
176  * cache of FID->ino mapping. this mechanism is needed to resolve
177  * FID to inode at dt_insert() which in turn stores ino in the
178  * directory entries to keep ldiskfs compatible with ext[34].
179  * due to locking-originated restrictions we can't lookup ino
180  * using LU cache (deadlock is possible). lookup using OI is quite
181  * expensive. so instead we maintain this cache and methods like
182  * dt_create() fill it. so in the majority of cases dt_insert() is
183  * able to find needed mapping in lockless manner.
184  */
185 static struct osd_idmap_cache *
186 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
187              const struct lu_fid *fid)
188 {
189         struct osd_thread_info  *oti   = osd_oti_get(env);
190         struct osd_idmap_cache  *idc    = oti->oti_ins_cache;
191         int i;
192         for (i = 0; i < oti->oti_ins_cache_used; i++) {
193                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
194                         continue;
195                 if (idc[i].oic_dev != osd)
196                         continue;
197
198                 return idc + i;
199         }
200
201         return NULL;
202 }
203
204 static struct osd_idmap_cache *
205 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
206             const struct lu_fid *fid)
207 {
208         struct osd_thread_info  *oti   = osd_oti_get(env);
209         struct osd_idmap_cache  *idc;
210         int i;
211
212         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
213                 i = oti->oti_ins_cache_size * 2;
214                 if (i == 0)
215                         i = OSD_INS_CACHE_SIZE;
216                 OBD_ALLOC(idc, sizeof(*idc) * i);
217                 if (idc == NULL)
218                         return ERR_PTR(-ENOMEM);
219                 if (oti->oti_ins_cache != NULL) {
220                         memcpy(idc, oti->oti_ins_cache,
221                                oti->oti_ins_cache_used * sizeof(*idc));
222                         OBD_FREE(oti->oti_ins_cache,
223                                  oti->oti_ins_cache_used * sizeof(*idc));
224                 }
225                 oti->oti_ins_cache = idc;
226                 oti->oti_ins_cache_size = i;
227         }
228
229         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
230         idc->oic_fid = *fid;
231         idc->oic_dev = osd;
232         idc->oic_lid.oii_ino = 0;
233         idc->oic_lid.oii_gen = 0;
234         idc->oic_remote = 0;
235
236         return idc;
237 }
238
239 /*
240  * lookup mapping for the given fid in the cache, initialize a
241  * new one if not found. the initialization checks whether the
242  * object is local or remote. for local objects, OI is used to
243  * learn ino/generation. the function is used when the caller
244  * has no information about the object, e.g. at dt_insert().
245  */
246 static struct osd_idmap_cache *
247 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
248                      const struct lu_fid *fid)
249 {
250         struct osd_idmap_cache *idc;
251         int rc;
252
253         idc = osd_idc_find(env, osd, fid);
254         LASSERT(!IS_ERR(idc));
255         if (idc != NULL)
256                 return idc;
257
258         /* new mapping is needed */
259         idc = osd_idc_add(env, osd, fid);
260         if (IS_ERR(idc))
261                 return idc;
262
263         /* initialize it */
264         rc = osd_remote_fid(env, osd, fid);
265         if (unlikely(rc < 0))
266                 return ERR_PTR(rc);
267
268         if (rc == 0) {
269                 /* the object is local, lookup in OI */
270                 /* XXX: probably cheaper to lookup in LU first? */
271                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
272                                    &idc->oic_lid, 0);
273                 if (unlikely(rc < 0)) {
274                         CERROR("can't lookup: rc = %d\n", rc);
275                         return ERR_PTR(rc);
276                 }
277         } else {
278                 /* the object is remote */
279                 idc->oic_remote = 1;
280         }
281
282         return idc;
283 }
284
285 /*
286  * lookup mapping for given FID and fill it from the given object.
287  * the object is lolcal by definition.
288  */
289 static int osd_idc_find_and_init(const struct lu_env *env,
290                                  struct osd_device *osd,
291                                  struct osd_object *obj)
292 {
293         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
294         struct osd_idmap_cache  *idc;
295
296         idc = osd_idc_find(env, osd, fid);
297         LASSERT(!IS_ERR(idc));
298         if (idc != NULL) {
299                 if (obj->oo_inode == NULL)
300                         return 0;
301                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
302                         LASSERT(idc->oic_lid.oii_ino == 0);
303                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
304                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
305                 }
306                 return 0;
307         }
308
309         /* new mapping is needed */
310         idc = osd_idc_add(env, osd, fid);
311         if (IS_ERR(idc))
312                 return PTR_ERR(idc);
313
314         if (obj->oo_inode != NULL) {
315                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
316                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
317         }
318         return 0;
319 }
320
321 /*
322  * OSD object methods.
323  */
324
325 /*
326  * Concurrency: no concurrent access is possible that early in object
327  * life-cycle.
328  */
329 static struct lu_object *osd_object_alloc(const struct lu_env *env,
330                                           const struct lu_object_header *hdr,
331                                           struct lu_device *d)
332 {
333         struct osd_object *mo;
334
335         OBD_ALLOC_PTR(mo);
336         if (mo != NULL) {
337                 struct lu_object *l;
338
339                 l = &mo->oo_dt.do_lu;
340                 dt_object_init(&mo->oo_dt, NULL, d);
341                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
342                 l->lo_ops = &osd_lu_obj_ops;
343                 init_rwsem(&mo->oo_sem);
344                 init_rwsem(&mo->oo_ext_idx_sem);
345                 spin_lock_init(&mo->oo_guard);
346                 INIT_LIST_HEAD(&mo->oo_xattr_list);
347                 return l;
348         } else {
349                 return NULL;
350         }
351 }
352
353 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
354                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
355 {
356         int rc;
357
358         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
359         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
360                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
361         if (rc > 0) {
362                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
363                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
364                 rc = 0;
365                 lustre_lma_swab(lma);
366                 /* Check LMA compatibility */
367                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
368                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
369                               "for fid = "DFID", ino = %lu\n",
370                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
371                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
372                               PFID(&lma->lma_self_fid), inode->i_ino);
373                         rc = -EOPNOTSUPP;
374                 }
375         } else if (rc == 0) {
376                 rc = -ENODATA;
377         }
378
379         return rc;
380 }
381
382 /*
383  * retrieve object from backend ext fs.
384  **/
385 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
386                        struct osd_inode_id *id)
387 {
388         int rc;
389         struct inode *inode = NULL;
390
391         /* if we look for an inode withing a running
392          * transaction, then we risk to deadlock */
393         /* osd_dirent_check_repair() breaks this */
394         /*LASSERT(current->journal_info == NULL);*/
395
396         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
397         if (IS_ERR(inode)) {
398                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
399                        id->oii_ino, PTR_ERR(inode));
400         } else if (id->oii_gen != OSD_OII_NOGEN &&
401                    inode->i_generation != id->oii_gen) {
402                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
403                        "i_generation = %u\n",
404                        id->oii_ino, id->oii_gen, inode->i_generation);
405                 iput(inode);
406                 inode = ERR_PTR(-ESTALE);
407         } else if (inode->i_nlink == 0) {
408                 /* due to parallel readdir and unlink,
409                 * we can have dead inode here. */
410                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
411                 iput(inode);
412                 inode = ERR_PTR(-ESTALE);
413         } else if (is_bad_inode(inode)) {
414                 CWARN("%.16s: bad inode: ino = %u\n",
415                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
416                 iput(inode);
417                 inode = ERR_PTR(-ENOENT);
418         } else if ((rc = osd_attach_jinode(inode))) {
419                 iput(inode);
420                 inode = ERR_PTR(rc);
421         } else {
422                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
423                 if (id->oii_gen == OSD_OII_NOGEN)
424                         osd_id_gen(id, inode->i_ino, inode->i_generation);
425
426                 /* Do not update file c/mtime in ldiskfs.
427                  * NB: we don't have any lock to protect this because we don't
428                  * have reference on osd_object now, but contention with
429                  * another lookup + attr_set can't happen in the tiny window
430                  * between if (...) and set S_NOCMTIME. */
431                 if (!(inode->i_flags & S_NOCMTIME))
432                         inode->i_flags |= S_NOCMTIME;
433         }
434         return inode;
435 }
436
437 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
438                           handle_t *handle, struct dentry *child,
439                           struct inode *inode, struct htree_lock *hlock)
440 {
441         int rc, rc2;
442
443         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
444         if (rc == -ENOBUFS || rc == -ENOSPC) {
445                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
446                 struct inode *parent = child->d_parent->d_inode;
447                 struct lu_fid *fid = NULL;
448
449                 rc2 = osd_get_lma(info, parent, child->d_parent, lma);
450                 if (rc2 == 0) {
451                         fid = &lma->lma_self_fid;
452                 } else if (rc2 == -ENODATA) {
453                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
454                                 fid = &info->oti_fid3;
455                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
456                         } else if (!osd->od_is_ost && osd->od_index == 0) {
457                                 fid = &info->oti_fid3;
458                                 lu_igif_build(fid, parent->i_ino,
459                                               parent->i_generation);
460                         }
461                 }
462
463                 if (fid != NULL)
464                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
465                               "maximum entry limit\n",
466                               osd_name(osd), parent->i_ino, PFID(fid),
467                               rc == -ENOSPC ? "has reached" : "is approaching");
468                 else
469                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
470                               "maximum entry limit\n",
471                               osd_name(osd), parent->i_ino,
472                               rc == -ENOSPC ? "has reached" : "is approaching");
473
474                 /* ignore such error now */
475                 if (rc == -ENOBUFS)
476                         rc = 0;
477         }
478
479         return rc;
480 }
481
482
483 static struct inode *
484 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
485              struct osd_inode_id *id, struct lu_fid *fid)
486 {
487         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
488         struct inode            *inode;
489         int                      rc;
490
491         inode = osd_iget(info, dev, id);
492         if (IS_ERR(inode))
493                 return inode;
494
495         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
496         if (rc == 0) {
497                 *fid = lma->lma_self_fid;
498         } else if (rc == -ENODATA) {
499                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
500                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
501                 else
502                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
503         } else {
504                 iput(inode);
505                 inode = ERR_PTR(rc);
506         }
507         return inode;
508 }
509
510 static struct inode *osd_iget_check(struct osd_thread_info *info,
511                                     struct osd_device *dev,
512                                     const struct lu_fid *fid,
513                                     struct osd_inode_id *id,
514                                     bool cached)
515 {
516         struct inode    *inode;
517         int              rc     = 0;
518         ENTRY;
519
520         /* The cached OI mapping is trustable. If we cannot locate the inode
521          * via the cached OI mapping, then return the failure to the caller
522          * directly without further OI checking. */
523
524         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
525         if (IS_ERR(inode)) {
526                 rc = PTR_ERR(inode);
527                 if (cached || (rc != -ENOENT && rc != -ESTALE)) {
528                         CDEBUG(D_INODE, "no inode: ino = %u, rc = %d\n",
529                                id->oii_ino, rc);
530
531                         GOTO(put, rc);
532                 }
533
534                 goto check_oi;
535         }
536
537         if (is_bad_inode(inode)) {
538                 rc = -ENOENT;
539                 if (cached) {
540                         CDEBUG(D_INODE, "bad inode: ino = %u\n", id->oii_ino);
541
542                         GOTO(put, rc);
543                 }
544
545                 goto check_oi;
546         }
547
548         if (id->oii_gen != OSD_OII_NOGEN &&
549             inode->i_generation != id->oii_gen) {
550                 rc = -ESTALE;
551                 if (cached) {
552                         CDEBUG(D_INODE, "unmatched inode: ino = %u, "
553                                "oii_gen = %u, i_generation = %u\n",
554                                id->oii_ino, id->oii_gen, inode->i_generation);
555
556                         GOTO(put, rc);
557                 }
558
559                 goto check_oi;
560         }
561
562         if (inode->i_nlink == 0) {
563                 rc = -ENOENT;
564                 if (cached) {
565                         CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
566
567                         GOTO(put, rc);
568                 }
569
570                 goto check_oi;
571         }
572
573         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
574
575 check_oi:
576         if (rc != 0) {
577                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
578
579                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
580                 /* XXX: There are four possible cases:
581                  *      1. rc = 0.
582                  *         Backup/restore caused the OI invalid.
583                  *      2. rc = 0.
584                  *         Someone unlinked the object but NOT removed
585                  *         the OI mapping, such as mount target device
586                  *         as ldiskfs, and modify something directly.
587                  *      3. rc = -ENOENT.
588                  *         Someone just removed the object between the
589                  *         former oi_lookup and the iget. It is normal.
590                  *      4. Other failure cases.
591                  *
592                  *      Generally, when the device is mounted, it will
593                  *      auto check whether the system is restored from
594                  *      file-level backup or not. We trust such detect
595                  *      to distinguish the 1st case from the 2nd case:
596                  *      if the OI files are consistent but may contain
597                  *      stale OI mappings because of case 2, if iget()
598                  *      returns -ENOENT or -ESTALE, then it should be
599                  *      the case 2. */
600                 if (rc != 0)
601                         /* If the OI mapping was in OI file before the
602                          * osd_iget_check(), but now, it is disappear,
603                          * then it must be removed by race. That is a
604                          * normal race case. */
605                         GOTO(put, rc);
606
607                 if ((!IS_ERR(inode) && inode->i_generation != 0 &&
608                      inode->i_generation == id->oii_gen) ||
609                     (IS_ERR(inode) && !(dev->od_scrub.os_file.sf_flags &
610                                         SF_INCONSISTENT)))
611                         rc = -ENOENT;
612                 else
613                         rc = -EREMCHG;
614         } else {
615                 if (id->oii_gen == OSD_OII_NOGEN)
616                         osd_id_gen(id, inode->i_ino, inode->i_generation);
617
618                 /* Do not update file c/mtime in ldiskfs.
619                  * NB: we don't have any lock to protect this because we don't
620                  * have reference on osd_object now, but contention with
621                  * another lookup + attr_set can't happen in the tiny window
622                  * between if (...) and set S_NOCMTIME. */
623                 if (!(inode->i_flags & S_NOCMTIME))
624                         inode->i_flags |= S_NOCMTIME;
625         }
626
627         GOTO(put, rc);
628
629 put:
630         if (rc != 0) {
631                 if (!IS_ERR(inode))
632                         iput(inode);
633
634                 inode = ERR_PTR(rc);
635         }
636
637         return inode;
638 }
639
640 /**
641  * \retval +v: new filter_fid, does not contain self-fid
642  * \retval 0:  filter_fid_old, contains self-fid
643  * \retval -v: other failure cases
644  */
645 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
646                  struct dentry *dentry, struct lu_fid *fid)
647 {
648         struct filter_fid_old   *ff     = &info->oti_ff;
649         struct ost_id           *ostid  = &info->oti_ostid;
650         int                      rc;
651
652         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
653         if (rc == sizeof(*ff)) {
654                 rc = 0;
655                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
656                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
657                 /* XXX: use 0 as the index for compatibility, the caller will
658                  *      handle index related issues when necessarry. */
659                 ostid_to_fid(fid, ostid, 0);
660         } else if (rc == sizeof(struct filter_fid)) {
661                 rc = 1;
662         } else if (rc >= 0) {
663                 rc = -EINVAL;
664         }
665
666         return rc;
667 }
668
669 static int osd_lma_self_repair(struct osd_thread_info *info,
670                                struct osd_device *osd, struct inode *inode,
671                                const struct lu_fid *fid, __u32 compat)
672 {
673         handle_t *jh;
674         int       rc;
675
676         LASSERT(current->journal_info == NULL);
677
678         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
679                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
680         if (IS_ERR(jh)) {
681                 rc = PTR_ERR(jh);
682                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
683                       osd_name(osd), rc);
684                 return rc;
685         }
686
687         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
688         if (rc != 0)
689                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
690                       osd_name(osd), rc);
691         ldiskfs_journal_stop(jh);
692         return rc;
693 }
694
695 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
696 {
697         struct osd_thread_info  *info   = osd_oti_get(env);
698         struct osd_device       *osd    = osd_obj2dev(obj);
699         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
700         struct inode            *inode  = obj->oo_inode;
701         struct dentry           *dentry = &info->oti_obj_dentry;
702         struct lu_fid           *fid    = NULL;
703         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
704         int                      rc;
705         ENTRY;
706
707         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
708         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
709                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
710         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
711                 fid = &lma->lma_self_fid;
712                 rc = osd_get_idif(info, inode, dentry, fid);
713                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
714                         /* For the given OST-object, if it has neither LMA nor
715                          * FID in XATTR_NAME_FID, then the given FID (which is
716                          * contained in the @obj, from client RPC for locating
717                          * the OST-object) is trusted. We use it to generate
718                          * the LMA. */
719                         osd_lma_self_repair(info, osd, inode, rfid,
720                                             LMAC_FID_ON_OST);
721                         RETURN(0);
722                 }
723         }
724
725         if (rc < 0)
726                 RETURN(rc);
727
728         if (rc > 0) {
729                 rc = 0;
730                 lustre_lma_swab(lma);
731                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
732                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
733                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
734                               "fid = "DFID", ino = %lu\n", osd_name(osd),
735                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
736                               PFID(rfid), inode->i_ino);
737                         rc = -EOPNOTSUPP;
738                 } else {
739                         fid = &lma->lma_self_fid;
740                 }
741         }
742
743         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
744                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
745                         struct ost_id   *oi   = &info->oti_ostid;
746                         struct lu_fid   *fid1 = &info->oti_fid3;
747                         __u32            idx  = fid_idif_ost_idx(rfid);
748
749                         /* For old IDIF, the OST index is not part of the IDIF,
750                          * Means that different OSTs may have the same IDIFs.
751                          * Under such case, we need to make some compatible
752                          * check to make sure to trigger OI scrub properly. */
753                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
754                                 /* Given @rfid is new, LMA is old. */
755                                 fid_to_ostid(fid, oi);
756                                 ostid_to_fid(fid1, oi, idx);
757                                 if (lu_fid_eq(fid1, rfid)) {
758                                         if (osd->od_index_in_idif)
759                                                 osd_lma_self_repair(info, osd,
760                                                         inode, rfid,
761                                                         LMAC_FID_ON_OST);
762                                         RETURN(0);
763                                 }
764                         }
765                 }
766
767                 rc = -EREMCHG;
768         }
769
770         RETURN(rc);
771 }
772
773 struct osd_check_lmv_buf {
774 #ifdef HAVE_DIR_CONTEXT
775         /* please keep it as first member */
776         struct dir_context       ctx;
777 #endif
778         struct osd_thread_info  *oclb_info;
779         struct osd_device       *oclb_dev;
780         struct osd_idmap_cache  *oclb_oic;
781 };
782
783 /**
784  * It is called internally by ->readdir() to filter out the
785  * local slave object's FID of the striped directory.
786  *
787  * \retval      1 found the local slave's FID
788  * \retval      0 continue to check next item
789  * \retval      -ve for failure
790  */
791 #ifdef HAVE_FILLDIR_USE_CTX
792 static int osd_stripe_dir_filldir(struct dir_context *buf,
793 #else
794 static int osd_stripe_dir_filldir(void *buf,
795 #endif
796                                   const char *name, int namelen,
797                                   loff_t offset, __u64 ino, unsigned d_type)
798 {
799         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
800         struct osd_thread_info *oti = oclb->oclb_info;
801         struct lu_fid *fid = &oti->oti_fid3;
802         struct osd_inode_id *id = &oti->oti_id3;
803         struct osd_device *dev = oclb->oclb_dev;
804         struct osd_idmap_cache *oic = oclb->oclb_oic;
805         struct inode *inode;
806         int rc;
807
808         if (name[0] == '.')
809                 return 0;
810
811         fid_zero(fid);
812         sscanf(name + 1, SFID, RFID(fid));
813         if (!fid_is_sane(fid))
814                 return 0;
815
816         if (osd_remote_fid(oti->oti_env, dev, fid))
817                 return 0;
818
819         osd_id_gen(id, ino, OSD_OII_NOGEN);
820         inode = osd_iget(oti, dev, id);
821         if (IS_ERR(inode))
822                 return PTR_ERR(inode);
823
824         iput(inode);
825         osd_add_oi_cache(oti, dev, id, fid);
826         oic->oic_fid = *fid;
827         oic->oic_lid = *id;
828         oic->oic_dev = dev;
829         rc = osd_oii_insert(dev, oic, true);
830
831         return rc == 0 ? 1 : rc;
832 }
833
834 /* When lookup item under striped directory, we need to locate the master
835  * MDT-object of the striped directory firstly, then the client will send
836  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
837  * and the item's name. If the system is restored from MDT file level backup,
838  * then before the OI scrub completely built the OI files, the OI mappings of
839  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
840  * not a problem for the master MDT-object. Because when locate the master
841  * MDT-object, we will do name based lookup (for the striped directory itself)
842  * firstly, during such process we can setup the correct OI mapping for the
843  * master MDT-object. But it will be trouble for the slave MDT-object. Because
844  * the client will not trigger name based lookup on the MDT to locate the slave
845  * MDT-object before locating item under the striped directory, then when
846  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
847  * is invalid and does not know what the right OI mapping is, then the MDT has
848  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
849  * the OI file, related OI mapping is unknown yet, please try again later. And
850  * then client will re-try the RPC again and again until related OI mapping has
851  * been updated. That is quite inefficient.
852  *
853  * To resolve above trouble, we will handle it as the following two cases:
854  *
855  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
856  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
857  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
858  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
859  *    directly. Please check osd_fid_lookup().
860  *
861  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
862  *    Under such case, during lookup the master MDT-object, we will lookup the
863  *    slave MDT-object via readdir against the master MDT-object, because the
864  *    slave MDT-objects information are stored as sub-directories with the name
865  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
866  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
867  *    the correct OI mapping for the slave MDT-object. */
868 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
869                          struct inode *inode, struct osd_idmap_cache *oic)
870 {
871         struct lu_buf *buf = &oti->oti_big_buf;
872         struct dentry *dentry = &oti->oti_obj_dentry;
873         struct file *filp = &oti->oti_file;
874         const struct file_operations *fops;
875         struct lmv_mds_md_v1 *lmv1;
876         struct osd_check_lmv_buf oclb = {
877 #ifdef HAVE_DIR_CONTEXT
878                 .ctx.actor = osd_stripe_dir_filldir,
879 #endif
880                 .oclb_info = oti,
881                 .oclb_dev = dev,
882                 .oclb_oic = oic
883         };
884         int rc = 0;
885         ENTRY;
886
887 again:
888         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
889                              buf->lb_len);
890         if (rc == -ERANGE) {
891                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
892                 if (rc > 0) {
893                         lu_buf_realloc(buf, rc);
894                         if (buf->lb_buf == NULL)
895                                 GOTO(out, rc = -ENOMEM);
896
897                         goto again;
898                 }
899         }
900
901         if (unlikely(rc == 0 || rc == -ENODATA))
902                 GOTO(out, rc = 0);
903
904         if (rc < 0)
905                 GOTO(out, rc);
906
907         if (unlikely(buf->lb_buf == NULL)) {
908                 lu_buf_realloc(buf, rc);
909                 if (buf->lb_buf == NULL)
910                         GOTO(out, rc = -ENOMEM);
911
912                 goto again;
913         }
914
915         lmv1 = buf->lb_buf;
916         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
917                 GOTO(out, rc = 0);
918
919         fops = inode->i_fop;
920         dentry->d_inode = inode;
921         dentry->d_sb = inode->i_sb;
922         filp->f_pos = 0;
923         filp->f_path.dentry = dentry;
924         filp->f_mode = FMODE_64BITHASH;
925         filp->f_mapping = inode->i_mapping;
926         filp->f_op = fops;
927         filp->private_data = NULL;
928         set_file_inode(filp, inode);
929
930 #ifdef HAVE_DIR_CONTEXT
931         oclb.ctx.pos = filp->f_pos;
932         rc = fops->iterate(filp, &oclb.ctx);
933         filp->f_pos = oclb.ctx.pos;
934 #else
935         rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
936 #endif
937         fops->release(inode, filp);
938
939 out:
940         if (rc < 0)
941                 CDEBUG(D_LFSCK, "%.16s: fail to check LMV EA, inode = %lu/%u,"
942                        DFID": rc = %d\n",
943                        LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
944                        inode->i_ino, inode->i_generation,
945                        PFID(&oic->oic_fid), rc);
946         else
947                 rc = 0;
948
949         RETURN(rc);
950 }
951
952 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
953                           const struct lu_fid *fid,
954                           const struct lu_object_conf *conf)
955 {
956         struct osd_thread_info *info;
957         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
958         struct osd_device      *dev;
959         struct osd_idmap_cache *oic;
960         struct osd_inode_id    *id;
961         struct osd_inode_id    *tid;
962         struct inode           *inode = NULL;
963         struct osd_scrub       *scrub;
964         struct scrub_file      *sf;
965         __u32                   flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT |
966                                         SS_AUTO_FULL;
967         __u32                   saved_ino;
968         __u32                   saved_gen;
969         int                     result  = 0;
970         int                     rc1     = 0;
971         bool                    cached  = true;
972         bool                    remote  = false;
973         ENTRY;
974
975         LINVRNT(osd_invariant(obj));
976         LASSERT(obj->oo_inode == NULL);
977         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
978
979         dev = osd_dev(ldev);
980         scrub = &dev->od_scrub;
981         sf = &scrub->os_file;
982         info = osd_oti_get(env);
983         LASSERT(info);
984         oic = &info->oti_cache;
985
986         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
987                 RETURN(-ENOENT);
988
989         /* For the object is created as locking anchor, or for the object to
990          * be created on disk. No need to osd_oi_lookup() at here because FID
991          * shouldn't never be re-used, if it's really a duplicate FID from
992          * unexpected reason, we should be able to detect it later by calling
993          * do_create->osd_oi_insert(). */
994         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
995                 GOTO(out, result = 0);
996
997         /* Search order: 1. per-thread cache. */
998         if (lu_fid_eq(fid, &oic->oic_fid) &&
999             likely(oic->oic_dev == dev)) {
1000                 id = &oic->oic_lid;
1001                 goto iget;
1002         }
1003
1004         id = &info->oti_id;
1005         if (!list_empty(&scrub->os_inconsistent_items)) {
1006                 /* Search order: 2. OI scrub pending list. */
1007                 result = osd_oii_lookup(dev, fid, id);
1008                 if (result == 0)
1009                         goto iget;
1010         }
1011
1012         cached = false;
1013         /* Search order: 3. OI files. */
1014         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1015         if (result == -ENOENT) {
1016                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1017                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1018                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1019                                       sf->sf_oi_bitmap))
1020                         GOTO(out, result = 0);
1021
1022                 goto trigger;
1023         }
1024
1025         if (result != 0)
1026                 GOTO(out, result);
1027
1028 iget:
1029         inode = osd_iget_check(info, dev, fid, id, cached);
1030         if (IS_ERR(inode)) {
1031                 result = PTR_ERR(inode);
1032                 if (result == -ENOENT || result == -ESTALE)
1033                         GOTO(out, result = 0);
1034
1035                 if (result == -EREMCHG) {
1036
1037 trigger:
1038                         /* We still have chance to get the valid inode: for the
1039                          * object which is referenced by remote name entry, the
1040                          * object on the local MDT will be linked under the dir
1041                          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1042                          *
1043                          * We do not know whether the object for the given FID
1044                          * is referenced by some remote name entry or not, and
1045                          * especially for DNE II, a multiple-linked object may
1046                          * have many name entries reside on many MDTs.
1047                          *
1048                          * To simplify the operation, OSD will not distinguish
1049                          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1050                          * only happened for the RPC from other MDT during the
1051                          * OI scrub, or for the client side RPC with FID only,
1052                          * such as FID to path, or from old connected client. */
1053                         if (!remote &&
1054                             !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1055                                 rc1 = osd_lookup_in_remote_parent(info, dev,
1056                                                                   fid, id);
1057                                 if (rc1 == 0) {
1058                                         remote = true;
1059                                         cached = true;
1060                                         flags |= SS_AUTO_PARTIAL;
1061                                         flags &= ~SS_AUTO_FULL;
1062                                         goto iget;
1063                                 }
1064                         }
1065
1066                         if (thread_is_running(&scrub->os_thread)) {
1067                                 if (scrub->os_partial_scan &&
1068                                     !scrub->os_in_join) {
1069                                         goto join;
1070                                 } else {
1071                                         if (inode != NULL && !IS_ERR(inode)) {
1072                                                 LASSERT(remote);
1073
1074                                                 osd_add_oi_cache(info, dev, id,
1075                                                                  fid);
1076                                                 osd_oii_insert(dev, oic, true);
1077                                         } else {
1078                                                 result = -EINPROGRESS;
1079                                         }
1080                                 }
1081                         } else if (!dev->od_noscrub) {
1082
1083 join:
1084                                 rc1 = osd_scrub_start(dev, flags);
1085                                 LCONSOLE_WARN("%.16s: trigger OI scrub by RPC "
1086                                               "for the "DFID" with flags 0x%x,"
1087                                               " rc = %d\n", osd_name(dev),
1088                                               PFID(fid), flags, rc1);
1089                                 if (rc1 == 0 || rc1 == -EALREADY) {
1090                                         if (inode != NULL && !IS_ERR(inode)) {
1091                                                 LASSERT(remote);
1092
1093                                                 osd_add_oi_cache(info, dev, id,
1094                                                                  fid);
1095                                                 osd_oii_insert(dev, oic, true);
1096                                         } else {
1097                                                 result = -EINPROGRESS;
1098                                         }
1099                                 } else {
1100                                         result = -EREMCHG;
1101                                 }
1102                         } else {
1103                                 result = -EREMCHG;
1104                         }
1105                 }
1106
1107                 if (inode == NULL || IS_ERR(inode))
1108                         GOTO(out, result);
1109         } else if (remote) {
1110                 goto trigger;
1111         }
1112
1113         obj->oo_inode = inode;
1114         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
1115
1116         result = osd_check_lma(env, obj);
1117         if (result == 0)
1118                 goto found;
1119
1120         tid = &info->oti_id3;
1121         LASSERT(tid != id);
1122
1123         if (result == -ENODATA) {
1124                 if (!cached)
1125                         /* The current OI mapping is from the OI file,
1126                          * since the inode has been found via
1127                          * osd_iget_check(), no need recheck OI. */
1128                         goto found;
1129
1130                 result = osd_oi_lookup(info, dev, fid, tid, OI_CHECK_FLD);
1131                 if (result == 0) {
1132                         LASSERTF(tid->oii_ino == id->oii_ino &&
1133                                  tid->oii_gen == id->oii_gen,
1134                                  "OI mapping changed(1): %u/%u => %u/%u",
1135                                  tid->oii_ino, tid->oii_gen,
1136                                  id->oii_ino, id->oii_gen);
1137
1138                         LASSERTF(tid->oii_ino == inode->i_ino &&
1139                                  tid->oii_gen == inode->i_generation,
1140                                  "locate wrong inode(1): %u/%u => %ld/%u",
1141                                  tid->oii_ino, tid->oii_gen,
1142                                  inode->i_ino, inode->i_generation);
1143
1144                         /* "result == 0" means the cached OI mapping is still in
1145                          * the OI file, so the target the inode is valid. */
1146                         goto found;
1147                 }
1148
1149                 /* "result == -ENOENT" means that the OI mappinghas been removed
1150                  * by race, the target inode belongs to other object.
1151                  *
1152                  * Others error can be returned  directly. */
1153                 if (result == -ENOENT)
1154                         result = 0;
1155         }
1156
1157         saved_ino = inode->i_ino;
1158         saved_gen = inode->i_generation;
1159         iput(inode);
1160         inode = NULL;
1161         obj->oo_inode = NULL;
1162
1163         if (result != -EREMCHG)
1164                 GOTO(out, result);
1165
1166         if (!cached)
1167                 /* The current OI mapping is from the OI file,
1168                  * since the inode has been found via
1169                  * osd_iget_check(), no need recheck OI. */
1170                 goto trigger;
1171
1172         result = osd_oi_lookup(info, dev, fid, tid, OI_CHECK_FLD);
1173         /* "result == -ENOENT" means the cached OI mapping has been removed from
1174          * the OI file by race, above target inode belongs to other object.
1175          *
1176          * Others error can be returned directly. */
1177         if (result != 0)
1178                 GOTO(out, result = (result == -ENOENT ? 0 : result));
1179
1180         LASSERTF(tid->oii_ino == id->oii_ino && tid->oii_gen == id->oii_gen,
1181                  "OI mapping changed(2): %u/%u => %u/%u",
1182                  tid->oii_ino, tid->oii_gen, id->oii_ino, id->oii_gen);
1183
1184         LASSERTF(tid->oii_ino == saved_ino && tid->oii_gen == saved_gen,
1185                  "locate wrong inode(2): %u/%u => %u/%u",
1186                  tid->oii_ino, tid->oii_gen, saved_ino, saved_gen);
1187
1188         goto trigger;
1189
1190 found:
1191         obj->oo_compat_dot_created = 1;
1192         obj->oo_compat_dotdot_created = 1;
1193
1194         if (S_ISDIR(inode->i_mode) &&
1195             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1196                 osd_check_lmv(info, dev, inode, oic);
1197
1198         result = osd_attach_jinode(inode);
1199         if (result) {
1200                 obj->oo_inode = NULL;
1201                 iput(inode);
1202                 GOTO(out, result);
1203         }
1204
1205         if (!ldiskfs_pdo)
1206                 GOTO(out, result = 0);
1207
1208         LASSERT(obj->oo_hl_head == NULL);
1209         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1210         if (obj->oo_hl_head == NULL) {
1211                 obj->oo_inode = NULL;
1212                 iput(inode);
1213                 GOTO(out, result = -ENOMEM);
1214         }
1215         GOTO(out, result = 0);
1216
1217 out:
1218         if (result != 0 && cached)
1219                 fid_zero(&oic->oic_fid);
1220
1221         LINVRNT(osd_invariant(obj));
1222         return result;
1223 }
1224
1225 /*
1226  * Concurrency: shouldn't matter.
1227  */
1228 static void osd_object_init0(struct osd_object *obj)
1229 {
1230         LASSERT(obj->oo_inode != NULL);
1231         obj->oo_dt.do_body_ops = &osd_body_ops;
1232         obj->oo_dt.do_lu.lo_header->loh_attr |=
1233                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1234 }
1235
1236 /*
1237  * Concurrency: no concurrent access is possible that early in object
1238  * life-cycle.
1239  */
1240 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1241                            const struct lu_object_conf *conf)
1242 {
1243         struct osd_object *obj = osd_obj(l);
1244         int result;
1245
1246         LINVRNT(osd_invariant(obj));
1247
1248         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1249                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1250                 l->lo_header->loh_attr |= LOHA_EXISTS;
1251                 return 0;
1252         }
1253
1254         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1255         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1256         if (result == 0 && obj->oo_inode != NULL) {
1257                 struct osd_thread_info *oti = osd_oti_get(env);
1258                 struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
1259
1260                 osd_object_init0(obj);
1261                 result = osd_get_lma(oti, obj->oo_inode,
1262                                      &oti->oti_obj_dentry, lma);
1263                 if (result == 0) {
1264                         /* Convert LMAI flags to lustre LMA flags
1265                          * and cache it to oo_lma_flags */
1266                         obj->oo_lma_flags =
1267                                 lma_to_lustre_flags(lma->lma_incompat);
1268                 } else if (result == -ENODATA) {
1269                         result = 0;
1270                 }
1271         }
1272
1273         LINVRNT(osd_invariant(obj));
1274         return result;
1275 }
1276
1277 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1278  * The left part is for value, binary mode. */
1279 struct osd_xattr_entry {
1280         struct list_head        oxe_list;
1281         size_t                  oxe_len;
1282         size_t                  oxe_namelen;
1283         bool                    oxe_exist;
1284         struct rcu_head         oxe_rcu;
1285         char                    oxe_buf[0];
1286 };
1287
1288 static int osd_oxc_get(struct osd_object *obj, const char *name,
1289                        struct lu_buf *buf)
1290 {
1291         struct osd_xattr_entry *tmp;
1292         struct osd_xattr_entry *oxe = NULL;
1293         size_t namelen = strlen(name);
1294         int rc;
1295         ENTRY;
1296
1297         rcu_read_lock();
1298         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1299                 if (namelen == tmp->oxe_namelen &&
1300                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1301                         oxe = tmp;
1302                         break;
1303                 }
1304         }
1305
1306         if (oxe == NULL)
1307                 GOTO(out, rc = -ENOENT);
1308
1309         if (!oxe->oxe_exist)
1310                 GOTO(out, rc = -ENODATA);
1311
1312         /* vallen */
1313         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1314         LASSERT(rc > 0);
1315
1316         if (buf->lb_buf == NULL)
1317                 GOTO(out, rc);
1318
1319         if (buf->lb_len < rc)
1320                 GOTO(out, rc = -ERANGE);
1321
1322         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1323         EXIT;
1324 out:
1325         rcu_read_unlock();
1326
1327         return rc;
1328 }
1329
1330 static void osd_oxc_free(struct rcu_head *head)
1331 {
1332         struct osd_xattr_entry *oxe;
1333
1334         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1335         OBD_FREE(oxe, oxe->oxe_len);
1336 }
1337
1338 static void osd_oxc_add(struct osd_object *obj, const char *name,
1339                         const char *buf, int buflen)
1340 {
1341         struct osd_xattr_entry *oxe;
1342         struct osd_xattr_entry *old = NULL;
1343         struct osd_xattr_entry *tmp;
1344         size_t namelen = strlen(name);
1345         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1346
1347         OBD_ALLOC(oxe, len);
1348         if (oxe == NULL)
1349                 return;
1350
1351         INIT_LIST_HEAD(&oxe->oxe_list);
1352         oxe->oxe_len = len;
1353         oxe->oxe_namelen = namelen;
1354         memcpy(oxe->oxe_buf, name, namelen);
1355         if (buflen > 0) {
1356                 LASSERT(buf != NULL);
1357                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1358                 oxe->oxe_exist = true;
1359         } else {
1360                 oxe->oxe_exist = false;
1361         }
1362
1363         /* this should be rarely called, just remove old and add new */
1364         spin_lock(&obj->oo_guard);
1365         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1366                 if (namelen == tmp->oxe_namelen &&
1367                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1368                         old = tmp;
1369                         break;
1370                 }
1371         }
1372         if (old != NULL) {
1373                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1374                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1375         } else {
1376                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1377         }
1378         spin_unlock(&obj->oo_guard);
1379 }
1380
1381 static void osd_oxc_del(struct osd_object *obj, const char *name)
1382 {
1383         struct osd_xattr_entry *oxe;
1384         size_t namelen = strlen(name);
1385
1386         spin_lock(&obj->oo_guard);
1387         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1388                 if (namelen == oxe->oxe_namelen &&
1389                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1390                         list_del_rcu(&oxe->oxe_list);
1391                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1392                         break;
1393                 }
1394         }
1395         spin_unlock(&obj->oo_guard);
1396 }
1397
1398 static void osd_oxc_fini(struct osd_object *obj)
1399 {
1400         struct osd_xattr_entry *oxe, *next;
1401
1402         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1403                 list_del(&oxe->oxe_list);
1404                 OBD_FREE(oxe, oxe->oxe_len);
1405         }
1406 }
1407
1408 /*
1409  * Concurrency: no concurrent access is possible that late in object
1410  * life-cycle.
1411  */
1412 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1413 {
1414         struct osd_object *obj = osd_obj(l);
1415
1416         LINVRNT(osd_invariant(obj));
1417
1418         osd_oxc_fini(obj);
1419         dt_object_fini(&obj->oo_dt);
1420         if (obj->oo_hl_head != NULL)
1421                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1422         OBD_FREE_PTR(obj);
1423 }
1424
1425 /*
1426  * Concurrency: no concurrent access is possible that late in object
1427  * life-cycle.
1428  */
1429 static void osd_index_fini(struct osd_object *o)
1430 {
1431         struct iam_container *bag;
1432
1433         if (o->oo_dir != NULL) {
1434                 bag = &o->oo_dir->od_container;
1435                 if (o->oo_inode != NULL) {
1436                         if (bag->ic_object == o->oo_inode)
1437                                 iam_container_fini(bag);
1438                 }
1439                 OBD_FREE_PTR(o->oo_dir);
1440                 o->oo_dir = NULL;
1441         }
1442 }
1443
1444 /*
1445  * Concurrency: no concurrent access is possible that late in object
1446  * life-cycle (for all existing callers, that is. New callers have to provide
1447  * their own locking.)
1448  */
1449 static int osd_inode_unlinked(const struct inode *inode)
1450 {
1451         return inode->i_nlink == 0;
1452 }
1453
1454 enum {
1455         OSD_TXN_OI_DELETE_CREDITS    = 20,
1456         OSD_TXN_INODE_DELETE_CREDITS = 20
1457 };
1458
1459 /*
1460  * Journal
1461  */
1462
1463 #if OSD_THANDLE_STATS
1464 /**
1465  * Set time when the handle is allocated
1466  */
1467 static void osd_th_alloced(struct osd_thandle *oth)
1468 {
1469         oth->oth_alloced = ktime_get();
1470 }
1471
1472 /**
1473  * Set time when the handle started
1474  */
1475 static void osd_th_started(struct osd_thandle *oth)
1476 {
1477         oth->oth_started = ktime_get();
1478 }
1479
1480 /**
1481  * Check whether the we deal with this handle for too long.
1482  */
1483 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1484                                 ktime_t alloced, ktime_t started,
1485                                 ktime_t closed)
1486 {
1487         ktime_t now = ktime_get();
1488
1489         LASSERT(dev != NULL);
1490
1491         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1492                             ktime_us_delta(started, alloced));
1493         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1494                             ktime_us_delta(closed, started));
1495         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1496                             ktime_us_delta(now, closed));
1497
1498         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1499                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1500                       oth, now, alloced, started, closed);
1501                 libcfs_debug_dumpstack(NULL);
1502         }
1503 }
1504
1505 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1506 {                                                                       \
1507         ktime_t __closed = ktime_get();                                 \
1508         ktime_t __alloced = oth->oth_alloced;                           \
1509         ktime_t __started = oth->oth_started;                           \
1510                                                                         \
1511         expr;                                                           \
1512         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1513 }
1514
1515 #else /* OSD_THANDLE_STATS */
1516
1517 #define osd_th_alloced(h)                  do {} while(0)
1518 #define osd_th_started(h)                  do {} while(0)
1519 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1520
1521 #endif /* OSD_THANDLE_STATS */
1522
1523 /*
1524  * Concurrency: doesn't access mutable data.
1525  */
1526 static int osd_param_is_not_sane(const struct osd_device *dev,
1527                                  const struct thandle *th)
1528 {
1529         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1530
1531         return oh->ot_credits > osd_transaction_size(dev);
1532 }
1533
1534 /*
1535  * Concurrency: shouldn't matter.
1536  */
1537 static void osd_trans_commit_cb(struct super_block *sb,
1538                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1539 {
1540         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1541         struct thandle     *th  = &oh->ot_super;
1542         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
1543         struct dt_txn_commit_cb *dcb, *tmp;
1544
1545         LASSERT(oh->ot_handle == NULL);
1546
1547         if (error)
1548                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1549
1550         dt_txn_hook_commit(th);
1551
1552         /* call per-transaction callbacks if any */
1553         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1554                                  dcb_linkage) {
1555                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1556                          "commit callback entry: magic=%x name='%s'\n",
1557                          dcb->dcb_magic, dcb->dcb_name);
1558                 list_del_init(&dcb->dcb_linkage);
1559                 dcb->dcb_func(NULL, th, dcb, error);
1560         }
1561
1562         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1563         lu_device_put(lud);
1564         th->th_dev = NULL;
1565
1566         lu_context_exit(&th->th_ctx);
1567         lu_context_fini(&th->th_ctx);
1568         OBD_FREE_PTR(oh);
1569 }
1570
1571 #ifndef HAVE_SB_START_WRITE
1572 # define sb_start_write(sb) do {} while (0)
1573 # define sb_end_write(sb) do {} while (0)
1574 #endif
1575
1576 static struct thandle *osd_trans_create(const struct lu_env *env,
1577                                         struct dt_device *d)
1578 {
1579         struct osd_thread_info  *oti = osd_oti_get(env);
1580         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1581         struct osd_thandle      *oh;
1582         struct thandle          *th;
1583         ENTRY;
1584
1585         /* on pending IO in this thread should left from prev. request */
1586         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1587
1588         sb_start_write(osd_sb(osd_dt_dev(d)));
1589
1590         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1591         if (oh != NULL) {
1592                 oh->ot_quota_trans = &oti->oti_quota_trans;
1593                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1594                 th = &oh->ot_super;
1595                 th->th_dev = d;
1596                 th->th_result = 0;
1597                 th->th_tags = LCT_TX_HANDLE;
1598                 oh->ot_credits = 0;
1599                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1600                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1601                 osd_th_alloced(oh);
1602
1603                 memset(oti->oti_declare_ops, 0,
1604                        sizeof(oti->oti_declare_ops));
1605                 memset(oti->oti_declare_ops_cred, 0,
1606                        sizeof(oti->oti_declare_ops_cred));
1607                 memset(oti->oti_declare_ops_used, 0,
1608                        sizeof(oti->oti_declare_ops_used));
1609         } else {
1610                 sb_end_write(osd_sb(osd_dt_dev(d)));
1611                 th = ERR_PTR(-ENOMEM);
1612         }
1613         RETURN(th);
1614 }
1615
1616 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1617 {
1618         struct osd_thread_info  *oti = osd_oti_get(env);
1619         struct osd_thandle      *oh;
1620
1621         oh = container_of0(th, struct osd_thandle, ot_super);
1622         LASSERT(oh != NULL);
1623
1624         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1625               oti->oti_declare_ops[OSD_OT_CREATE],
1626               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1627               oti->oti_declare_ops_used[OSD_OT_CREATE],
1628               oti->oti_declare_ops[OSD_OT_DESTROY],
1629               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1630               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1631         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1632               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1633               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1634               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1635               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1636               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1637               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1638         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1639               oti->oti_declare_ops[OSD_OT_WRITE],
1640               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1641               oti->oti_declare_ops_used[OSD_OT_WRITE],
1642               oti->oti_declare_ops[OSD_OT_PUNCH],
1643               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1644               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1645               oti->oti_declare_ops[OSD_OT_QUOTA],
1646               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1647               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1648         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1649               oti->oti_declare_ops[OSD_OT_INSERT],
1650               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1651               oti->oti_declare_ops_used[OSD_OT_INSERT],
1652               oti->oti_declare_ops[OSD_OT_DELETE],
1653               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1654               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1655         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1656               oti->oti_declare_ops[OSD_OT_REF_ADD],
1657               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1658               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1659               oti->oti_declare_ops[OSD_OT_REF_DEL],
1660               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1661               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1662 }
1663
1664 /*
1665  * Concurrency: shouldn't matter.
1666  */
1667 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1668                            struct thandle *th)
1669 {
1670         struct osd_thread_info *oti = osd_oti_get(env);
1671         struct osd_device  *dev = osd_dt_dev(d);
1672         handle_t           *jh;
1673         struct osd_thandle *oh;
1674         int rc;
1675
1676         ENTRY;
1677
1678         LASSERT(current->journal_info == NULL);
1679
1680         oh = container_of0(th, struct osd_thandle, ot_super);
1681         LASSERT(oh != NULL);
1682         LASSERT(oh->ot_handle == NULL);
1683
1684         rc = dt_txn_hook_start(env, d, th);
1685         if (rc != 0)
1686                 GOTO(out, rc);
1687
1688         if (unlikely(osd_param_is_not_sane(dev, th))) {
1689                 static unsigned long last_printed;
1690                 static int last_credits;
1691
1692                 /* don't make noise on a tiny testing systems
1693                  * actual credits misuse will be caught anyway */
1694                 if (last_credits != oh->ot_credits &&
1695                     time_after(jiffies, last_printed +
1696                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1697                     osd_transaction_size(dev) > 512) {
1698                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1699                               oh->ot_credits, osd_transaction_size(dev));
1700                         osd_trans_dump_creds(env, th);
1701                         libcfs_debug_dumpstack(NULL);
1702                         last_credits = oh->ot_credits;
1703                         last_printed = jiffies;
1704                 }
1705                 /* XXX Limit the credits to 'max_transaction_buffers', and
1706                  *     let the underlying filesystem to catch the error if
1707                  *     we really need so many credits.
1708                  *
1709                  *     This should be removed when we can calculate the
1710                  *     credits precisely. */
1711                 oh->ot_credits = osd_transaction_size(dev);
1712         }
1713
1714         /*
1715          * XXX temporary stuff. Some abstraction layer should
1716          * be used.
1717          */
1718         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1719         osd_th_started(oh);
1720         if (!IS_ERR(jh)) {
1721                 oh->ot_handle = jh;
1722                 LASSERT(oti->oti_txns == 0);
1723                 lu_context_init(&th->th_ctx, th->th_tags);
1724                 lu_context_enter(&th->th_ctx);
1725
1726                 lu_device_get(&d->dd_lu_dev);
1727                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1728                               "osd-tx", th);
1729                 oti->oti_txns++;
1730                 rc = 0;
1731         } else {
1732                 rc = PTR_ERR(jh);
1733         }
1734 out:
1735         RETURN(rc);
1736 }
1737
1738 static int osd_seq_exists(const struct lu_env *env,
1739                           struct osd_device *osd, u64 seq)
1740 {
1741         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1742         struct seq_server_site  *ss = osd_seq_site(osd);
1743         int                     rc;
1744         ENTRY;
1745
1746         LASSERT(ss != NULL);
1747         LASSERT(ss->ss_server_fld != NULL);
1748
1749         rc = osd_fld_lookup(env, osd, seq, range);
1750         if (rc != 0) {
1751                 if (rc != -ENOENT)
1752                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1753                                osd_name(osd), seq, rc);
1754                 RETURN(0);
1755         }
1756
1757         RETURN(ss->ss_node_id == range->lsr_index);
1758 }
1759
1760 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1761 {
1762         struct dt_txn_commit_cb *dcb;
1763         struct dt_txn_commit_cb *tmp;
1764
1765         /* call per-transaction stop callbacks if any */
1766         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1767                                  dcb_linkage) {
1768                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1769                          "commit callback entry: magic=%x name='%s'\n",
1770                          dcb->dcb_magic, dcb->dcb_name);
1771                 list_del_init(&dcb->dcb_linkage);
1772                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1773         }
1774 }
1775
1776 /*
1777  * Concurrency: shouldn't matter.
1778  */
1779 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1780                           struct thandle *th)
1781 {
1782         int                     rc = 0, remove_agents = 0;
1783         struct osd_thandle     *oh;
1784         struct osd_thread_info *oti = osd_oti_get(env);
1785         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1786         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1787         struct qsd_instance    *qsd = osd->od_quota_slave;
1788         struct lquota_trans    *qtrans;
1789         ENTRY;
1790
1791         oh = container_of0(th, struct osd_thandle, ot_super);
1792
1793         /* reset OI cache for safety */
1794         oti->oti_ins_cache_used = 0;
1795
1796         remove_agents = oh->ot_remove_agents;
1797
1798         qtrans = oh->ot_quota_trans;
1799         oh->ot_quota_trans = NULL;
1800
1801         if (oh->ot_handle != NULL) {
1802                 int rc2;
1803                 handle_t *hdl = oh->ot_handle;
1804
1805                 /*
1806                  * add commit callback
1807                  * notice we don't do this in osd_trans_start()
1808                  * as underlying transaction can change during truncate
1809                  */
1810                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1811                                          &oh->ot_jcb);
1812
1813                 LASSERT(oti->oti_txns == 1);
1814                 oti->oti_txns--;
1815
1816                 rc = dt_txn_hook_stop(env, th);
1817                 if (rc != 0)
1818                         CERROR("%s: failed in transaction hook: rc = %d\n",
1819                                osd_name(osd), rc);
1820
1821                 osd_trans_stop_cb(oh, rc);
1822                 /* hook functions might modify th_sync */
1823                 hdl->h_sync = th->th_sync;
1824
1825                 oh->ot_handle = NULL;
1826                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1827                 if (rc2 != 0)
1828                         CERROR("%s: failed to stop transaction: rc = %d\n",
1829                                osd_name(osd), rc2);
1830                 if (!rc)
1831                         rc = rc2;
1832         } else {
1833                 osd_trans_stop_cb(oh, th->th_result);
1834                 OBD_FREE_PTR(oh);
1835         }
1836
1837         /* inform the quota slave device that the transaction is stopping */
1838         qsd_op_end(env, qsd, qtrans);
1839
1840         /* as we want IO to journal and data IO be concurrent, we don't block
1841          * awaiting data IO completion in osd_do_bio(), instead we wait here
1842          * once transaction is submitted to the journal. all reqular requests
1843          * don't do direct IO (except read/write), thus this wait_event becomes
1844          * no-op for them.
1845          *
1846          * IMPORTANT: we have to wait till any IO submited by the thread is
1847          * completed otherwise iobuf may be corrupted by different request
1848          */
1849         wait_event(iobuf->dr_wait,
1850                        atomic_read(&iobuf->dr_numreqs) == 0);
1851         osd_fini_iobuf(osd, iobuf);
1852         if (!rc)
1853                 rc = iobuf->dr_error;
1854
1855         if (unlikely(remove_agents != 0))
1856                 osd_process_scheduled_agent_removals(env, osd);
1857
1858         sb_end_write(osd_sb(osd));
1859
1860         RETURN(rc);
1861 }
1862
1863 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1864 {
1865         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1866                                                ot_super);
1867
1868         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1869         LASSERT(&dcb->dcb_func != NULL);
1870         if (dcb->dcb_flags & DCB_TRANS_STOP)
1871                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1872         else
1873                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1874
1875         return 0;
1876 }
1877
1878 /*
1879  * Called just before object is freed. Releases all resources except for
1880  * object itself (that is released by osd_object_free()).
1881  *
1882  * Concurrency: no concurrent access is possible that late in object
1883  * life-cycle.
1884  */
1885 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1886 {
1887         struct osd_object *obj   = osd_obj(l);
1888         struct inode      *inode = obj->oo_inode;
1889
1890         LINVRNT(osd_invariant(obj));
1891
1892         /*
1893          * If object is unlinked remove fid->ino mapping from object index.
1894          */
1895
1896         osd_index_fini(obj);
1897         if (inode != NULL) {
1898                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1899                 qid_t                    uid = i_uid_read(inode);
1900                 qid_t                    gid = i_gid_read(inode);
1901
1902                 iput(inode);
1903                 obj->oo_inode = NULL;
1904
1905                 if (qsd != NULL) {
1906                         struct osd_thread_info  *info = osd_oti_get(env);
1907                         struct lquota_id_info   *qi = &info->oti_qi;
1908
1909                         /* Release granted quota to master if necessary */
1910                         qi->lqi_id.qid_uid = uid;
1911                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1912
1913                         qi->lqi_id.qid_uid = gid;
1914                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1915                 }
1916         }
1917 }
1918
1919 /*
1920  * Concurrency: ->loo_object_release() is called under site spin-lock.
1921  */
1922 static void osd_object_release(const struct lu_env *env,
1923                                struct lu_object *l)
1924 {
1925         struct osd_object *o = osd_obj(l);
1926         /* nobody should be releasing a non-destroyed object with nlink=0
1927          * the API allows this, but ldiskfs doesn't like and then report
1928          * this inode as deleted */
1929         if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1930                 LBUG();
1931 }
1932
1933 /*
1934  * Concurrency: shouldn't matter.
1935  */
1936 static int osd_object_print(const struct lu_env *env, void *cookie,
1937                             lu_printer_t p, const struct lu_object *l)
1938 {
1939         struct osd_object *o = osd_obj(l);
1940         struct iam_descr  *d;
1941
1942         if (o->oo_dir != NULL)
1943                 d = o->oo_dir->od_container.ic_descr;
1944         else
1945                 d = NULL;
1946         return (*p)(env, cookie,
1947                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1948                     o, o->oo_inode,
1949                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1950                     o->oo_inode ? o->oo_inode->i_generation : 0,
1951                     d ? d->id_ops->id_name : "plain");
1952 }
1953
1954 /*
1955  * Concurrency: shouldn't matter.
1956  */
1957 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1958                struct obd_statfs *sfs)
1959 {
1960         struct osd_device       *osd = osd_dt_dev(d);
1961         struct super_block      *sb = osd_sb(osd);
1962         struct kstatfs          *ksfs;
1963         __u64                    reserved;
1964         int                      result = 0;
1965
1966         if (unlikely(osd->od_mnt == NULL))
1967                 return -EINPROGRESS;
1968
1969         /* osd_lproc.c call this without env, allocate ksfs for that case */
1970         if (unlikely(env == NULL)) {
1971                 OBD_ALLOC_PTR(ksfs);
1972                 if (ksfs == NULL)
1973                         return -ENOMEM;
1974         } else {
1975                 ksfs = &osd_oti_get(env)->oti_ksfs;
1976         }
1977
1978         result = sb->s_op->statfs(sb->s_root, ksfs);
1979         if (result)
1980                 goto out;
1981
1982         statfs_pack(sfs, ksfs);
1983         if (unlikely(sb->s_flags & MS_RDONLY))
1984                 sfs->os_state |= OS_STATE_READONLY;
1985         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
1986                                          LDISKFS_FEATURE_INCOMPAT_EXTENTS))
1987                 sfs->os_maxbytes = sb->s_maxbytes;
1988         else
1989                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
1990
1991         /*
1992          * Reserve some space so to avoid fragmenting the filesystem too much.
1993          * Fragmentation not only impacts performance, but can also increase
1994          * metadata overhead significantly, causing grant calculation to be
1995          * wrong.
1996          *
1997          * Reserve 0.78% of total space, at least 8MB for small filesystems.
1998          */
1999         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2000         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2001         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2002                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2003
2004         sfs->os_blocks -= reserved;
2005         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2006         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2007
2008 out:
2009         if (unlikely(env == NULL))
2010                 OBD_FREE_PTR(ksfs);
2011         return result;
2012 }
2013
2014 /**
2015  * Estimate space needed for file creations. We assume the largest filename
2016  * which is 2^64 - 1, hence a filename of 20 chars.
2017  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2018  */
2019 #ifdef __LDISKFS_DIR_REC_LEN
2020 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2021 #else
2022 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2023 #endif
2024
2025 /*
2026  * Concurrency: doesn't access mutable data.
2027  */
2028 static void osd_conf_get(const struct lu_env *env,
2029                          const struct dt_device *dev,
2030                          struct dt_device_param *param)
2031 {
2032         struct super_block *sb = osd_sb(osd_dt_dev(dev));
2033         int                ea_overhead;
2034
2035         /*
2036          * XXX should be taken from not-yet-existing fs abstraction layer.
2037          */
2038         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2039         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2040         param->ddp_symlink_max  = sb->s_blocksize;
2041         param->ddp_mount_type     = LDD_MT_LDISKFS;
2042         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2043                 param->ddp_maxbytes = sb->s_maxbytes;
2044         else
2045                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2046         /* inode are statically allocated, so per-inode space consumption
2047          * is the space consumed by the directory entry */
2048         param->ddp_inodespace     = PER_OBJ_USAGE;
2049         /* EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2050          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2051          * maximum length to not under count the actual number of extents
2052          * needed for writing a file. */
2053         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2054         /* worst-case extent insertion metadata overhead */
2055         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2056         param->ddp_mntopts      = 0;
2057         if (test_opt(sb, XATTR_USER))
2058                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2059         if (test_opt(sb, POSIX_ACL))
2060                 param->ddp_mntopts |= MNTOPT_ACL;
2061
2062         /* LOD might calculate the max stripe count based on max_ea_size,
2063          * so we need take account in the overhead as well,
2064          * xattr_header + magic + xattr_entry_head */
2065         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2066                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2067
2068 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2069         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2070                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2071                                                                 ea_overhead;
2072         else
2073 #endif
2074                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2075 }
2076
2077 /*
2078  * Concurrency: shouldn't matter.
2079  */
2080 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2081 {
2082         int rc;
2083
2084         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2085
2086         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2087
2088         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
2089                LUSTRE_OSD_LDISKFS_NAME, rc);
2090
2091         return rc;
2092 }
2093
2094 /**
2095  * Start commit for OSD device.
2096  *
2097  * An implementation of dt_commit_async method for OSD device.
2098  * Asychronously starts underlayng fs sync and thereby a transaction
2099  * commit.
2100  *
2101  * \param env environment
2102  * \param d dt device
2103  *
2104  * \see dt_device_operations
2105  */
2106 static int osd_commit_async(const struct lu_env *env,
2107                             struct dt_device *d)
2108 {
2109         struct super_block *s = osd_sb(osd_dt_dev(d));
2110         ENTRY;
2111
2112         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
2113         RETURN(s->s_op->sync_fs(s, 0));
2114 }
2115
2116 /*
2117  * Concurrency: shouldn't matter.
2118  */
2119
2120 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2121 {
2122         struct super_block *sb = osd_sb(osd_dt_dev(d));
2123         struct block_device *dev = sb->s_bdev;
2124 #ifdef HAVE_DEV_SET_RDONLY
2125         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2126         int rc = 0;
2127 #else
2128         int rc = -EOPNOTSUPP;
2129 #endif
2130         ENTRY;
2131
2132 #ifdef HAVE_DEV_SET_RDONLY
2133         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
2134
2135         if (sb->s_op->freeze_fs) {
2136                 rc = sb->s_op->freeze_fs(sb);
2137                 if (rc)
2138                         goto out;
2139         }
2140
2141         if (jdev && (jdev != dev)) {
2142                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2143                        (long)jdev);
2144                 dev_set_rdonly(jdev);
2145         }
2146         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2147         dev_set_rdonly(dev);
2148
2149         if (sb->s_op->unfreeze_fs)
2150                 sb->s_op->unfreeze_fs(sb);
2151
2152 out:
2153 #endif
2154         if (rc)
2155                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2156                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2157
2158         RETURN(rc);
2159 }
2160
2161 /**
2162  * Note: we do not count into QUOTA here.
2163  * If we mount with --data_journal we may need more.
2164  */
2165 const int osd_dto_credits_noquota[DTO_NR] = {
2166         /**
2167          * Insert.
2168          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2169          * SINGLEDATA_TRANS_BLOCKS(8)
2170          * XXX Note: maybe iam need more, since iam have more level than
2171          *           EXT3 htree.
2172          */
2173         [DTO_INDEX_INSERT]  = 16,
2174         /**
2175          * Delete
2176          * just modify a single entry, probably merge few within a block
2177          */
2178         [DTO_INDEX_DELETE]  = 1,
2179         /**
2180          * Used for OI scrub
2181          */
2182         [DTO_INDEX_UPDATE]  = 16,
2183         /**
2184          * 4(inode, inode bits, groups, GDT)
2185          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2186          */
2187         [DTO_OBJECT_CREATE] = 4,
2188         /**
2189          * 4(inode, inode bits, groups, GDT)
2190          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2191          */
2192         [DTO_OBJECT_DELETE] = 4,
2193         /**
2194          * Attr set credits (inode)
2195          */
2196         [DTO_ATTR_SET_BASE] = 1,
2197         /**
2198          * Xattr set. The same as xattr of EXT3.
2199          * DATA_TRANS_BLOCKS(14)
2200          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2201          * are also counted in. Do not know why?
2202          */
2203         [DTO_XATTR_SET]     = 14,
2204         /**
2205          * credits for inode change during write.
2206          */
2207         [DTO_WRITE_BASE]    = 3,
2208         /**
2209          * credits for single block write.
2210          */
2211         [DTO_WRITE_BLOCK]   = 14,
2212         /**
2213          * Attr set credits for chown.
2214          * This is extra credits for setattr, and it is null without quota
2215          */
2216         [DTO_ATTR_SET_CHOWN] = 0
2217 };
2218
2219 static const struct dt_device_operations osd_dt_ops = {
2220         .dt_root_get       = osd_root_get,
2221         .dt_statfs         = osd_statfs,
2222         .dt_trans_create   = osd_trans_create,
2223         .dt_trans_start    = osd_trans_start,
2224         .dt_trans_stop     = osd_trans_stop,
2225         .dt_trans_cb_add   = osd_trans_cb_add,
2226         .dt_conf_get       = osd_conf_get,
2227         .dt_sync           = osd_sync,
2228         .dt_ro             = osd_ro,
2229         .dt_commit_async   = osd_commit_async,
2230 };
2231
2232 static void osd_object_read_lock(const struct lu_env *env,
2233                                  struct dt_object *dt, unsigned role)
2234 {
2235         struct osd_object *obj = osd_dt_obj(dt);
2236         struct osd_thread_info *oti = osd_oti_get(env);
2237
2238         LINVRNT(osd_invariant(obj));
2239
2240         LASSERT(obj->oo_owner != env);
2241         down_read_nested(&obj->oo_sem, role);
2242
2243         LASSERT(obj->oo_owner == NULL);
2244         oti->oti_r_locks++;
2245 }
2246
2247 static void osd_object_write_lock(const struct lu_env *env,
2248                                   struct dt_object *dt, unsigned role)
2249 {
2250         struct osd_object *obj = osd_dt_obj(dt);
2251         struct osd_thread_info *oti = osd_oti_get(env);
2252
2253         LINVRNT(osd_invariant(obj));
2254
2255         LASSERT(obj->oo_owner != env);
2256         down_write_nested(&obj->oo_sem, role);
2257
2258         LASSERT(obj->oo_owner == NULL);
2259         obj->oo_owner = env;
2260         oti->oti_w_locks++;
2261 }
2262
2263 static void osd_object_read_unlock(const struct lu_env *env,
2264                                    struct dt_object *dt)
2265 {
2266         struct osd_object *obj = osd_dt_obj(dt);
2267         struct osd_thread_info *oti = osd_oti_get(env);
2268
2269         LINVRNT(osd_invariant(obj));
2270
2271         LASSERT(oti->oti_r_locks > 0);
2272         oti->oti_r_locks--;
2273         up_read(&obj->oo_sem);
2274 }
2275
2276 static void osd_object_write_unlock(const struct lu_env *env,
2277                                     struct dt_object *dt)
2278 {
2279         struct osd_object *obj = osd_dt_obj(dt);
2280         struct osd_thread_info *oti = osd_oti_get(env);
2281
2282         LINVRNT(osd_invariant(obj));
2283
2284         LASSERT(obj->oo_owner == env);
2285         LASSERT(oti->oti_w_locks > 0);
2286         oti->oti_w_locks--;
2287         obj->oo_owner = NULL;
2288         up_write(&obj->oo_sem);
2289 }
2290
2291 static int osd_object_write_locked(const struct lu_env *env,
2292                                    struct dt_object *dt)
2293 {
2294         struct osd_object *obj = osd_dt_obj(dt);
2295
2296         LINVRNT(osd_invariant(obj));
2297
2298         return obj->oo_owner == env;
2299 }
2300
2301 static struct timespec *osd_inode_time(const struct lu_env *env,
2302                                        struct inode *inode, __u64 seconds)
2303 {
2304         struct osd_thread_info  *oti = osd_oti_get(env);
2305         struct timespec         *t   = &oti->oti_time;
2306
2307         t->tv_sec = seconds;
2308         t->tv_nsec = 0;
2309         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2310         return t;
2311 }
2312
2313 static void osd_inode_getattr(const struct lu_env *env,
2314                               struct inode *inode, struct lu_attr *attr)
2315 {
2316         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2317                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2318                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
2319                            LA_TYPE;
2320
2321         attr->la_atime   = LTIME_S(inode->i_atime);
2322         attr->la_mtime   = LTIME_S(inode->i_mtime);
2323         attr->la_ctime   = LTIME_S(inode->i_ctime);
2324         attr->la_mode    = inode->i_mode;
2325         attr->la_size    = i_size_read(inode);
2326         attr->la_blocks  = inode->i_blocks;
2327         attr->la_uid     = i_uid_read(inode);
2328         attr->la_gid     = i_gid_read(inode);
2329         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2330         attr->la_nlink   = inode->i_nlink;
2331         attr->la_rdev    = inode->i_rdev;
2332         attr->la_blksize = 1 << inode->i_blkbits;
2333         attr->la_blkbits = inode->i_blkbits;
2334 }
2335
2336 static int osd_attr_get(const struct lu_env *env,
2337                         struct dt_object *dt,
2338                         struct lu_attr *attr)
2339 {
2340         struct osd_object *obj = osd_dt_obj(dt);
2341
2342         if (unlikely(!dt_object_exists(dt)))
2343                 return -ENOENT;
2344         if (unlikely(obj->oo_destroyed))
2345                 return -ENOENT;
2346
2347         LASSERT(!dt_object_remote(dt));
2348         LINVRNT(osd_invariant(obj));
2349
2350         spin_lock(&obj->oo_guard);
2351         osd_inode_getattr(env, obj->oo_inode, attr);
2352         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2353                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2354         spin_unlock(&obj->oo_guard);
2355
2356         return 0;
2357 }
2358
2359 static int osd_declare_attr_set(const struct lu_env *env,
2360                                 struct dt_object *dt,
2361                                 const struct lu_attr *attr,
2362                                 struct thandle *handle)
2363 {
2364         struct osd_thandle     *oh;
2365         struct osd_object      *obj;
2366         struct osd_thread_info *info = osd_oti_get(env);
2367         struct lquota_id_info  *qi = &info->oti_qi;
2368         qid_t                   uid;
2369         qid_t                   gid;
2370         long long               bspace;
2371         int                     rc = 0;
2372         bool                    enforce;
2373         ENTRY;
2374
2375         LASSERT(dt != NULL);
2376         LASSERT(handle != NULL);
2377
2378         obj = osd_dt_obj(dt);
2379         LASSERT(osd_invariant(obj));
2380
2381         oh = container_of0(handle, struct osd_thandle, ot_super);
2382         LASSERT(oh->ot_handle == NULL);
2383
2384         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2385                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2386
2387         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2388                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2389
2390         if (attr == NULL || obj->oo_inode == NULL)
2391                 RETURN(rc);
2392
2393         bspace   = obj->oo_inode->i_blocks;
2394         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2395         bspace   = toqb(bspace);
2396
2397         /* Changing ownership is always preformed by super user, it should not
2398          * fail with EDQUOT.
2399          *
2400          * We still need to call the osd_declare_qid() to calculate the journal
2401          * credits for updating quota accounting files and to trigger quota
2402          * space adjustment once the operation is completed.*/
2403         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2404                 /* USERQUOTA */
2405                 uid = i_uid_read(obj->oo_inode);
2406                 qi->lqi_type = USRQUOTA;
2407                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2408                 /* inode accounting */
2409                 qi->lqi_is_blk = false;
2410
2411                 /* one more inode for the new uid ... */
2412                 qi->lqi_id.qid_uid = attr->la_uid;
2413                 qi->lqi_space      = 1;
2414                 /* Reserve credits for the new uid */
2415                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2416                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2417                         rc = 0;
2418                 if (rc)
2419                         RETURN(rc);
2420
2421                 /* and one less inode for the current uid */
2422                 qi->lqi_id.qid_uid = uid;
2423                 qi->lqi_space      = -1;
2424                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2425                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2426                         rc = 0;
2427                 if (rc)
2428                         RETURN(rc);
2429
2430                 /* block accounting */
2431                 qi->lqi_is_blk = true;
2432
2433                 /* more blocks for the new uid ... */
2434                 qi->lqi_id.qid_uid = attr->la_uid;
2435                 qi->lqi_space      = bspace;
2436                 /*
2437                  * Credits for the new uid has been reserved, re-use "obj"
2438                  * to save credit reservation.
2439                  */
2440                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2441                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2442                         rc = 0;
2443                 if (rc)
2444                         RETURN(rc);
2445
2446                 /* and finally less blocks for the current uid */
2447                 qi->lqi_id.qid_uid = uid;
2448                 qi->lqi_space      = -bspace;
2449                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2450                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2451                         rc = 0;
2452                 if (rc)
2453                         RETURN(rc);
2454
2455                 /* GROUP QUOTA */
2456                 gid = i_gid_read(obj->oo_inode);
2457                 qi->lqi_type = GRPQUOTA;
2458                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2459
2460                 /* inode accounting */
2461                 qi->lqi_is_blk = false;
2462
2463                 /* one more inode for the new gid ... */
2464                 qi->lqi_id.qid_gid = attr->la_gid;
2465                 qi->lqi_space      = 1;
2466                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2467                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2468                         rc = 0;
2469                 if (rc)
2470                         RETURN(rc);
2471
2472                 /* and one less inode for the current gid */
2473                 qi->lqi_id.qid_gid = gid;
2474                 qi->lqi_space      = -1;
2475                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2476                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2477                         rc = 0;
2478                 if (rc)
2479                         RETURN(rc);
2480
2481                 /* block accounting */
2482                 qi->lqi_is_blk = true;
2483
2484                 /* more blocks for the new gid ... */
2485                 qi->lqi_id.qid_gid = attr->la_gid;
2486                 qi->lqi_space      = bspace;
2487                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2488                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2489                         rc = 0;
2490                 if (rc)
2491                         RETURN(rc);
2492
2493                 /* and finally less blocks for the current gid */
2494                 qi->lqi_id.qid_gid = gid;
2495                 qi->lqi_space      = -bspace;
2496                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2497                 if (rc == -EDQUOT || rc == -EINPROGRESS)
2498                         rc = 0;
2499                 if (rc)
2500                         RETURN(rc);
2501         }
2502
2503         RETURN(rc);
2504 }
2505
2506 static int osd_inode_setattr(const struct lu_env *env,
2507                              struct inode *inode, const struct lu_attr *attr)
2508 {
2509         __u64 bits = attr->la_valid;
2510
2511         /* Only allow set size for regular file */
2512         if (!S_ISREG(inode->i_mode))
2513                 bits &= ~(LA_SIZE | LA_BLOCKS);
2514
2515         if (bits == 0)
2516                 return 0;
2517
2518         if (bits & LA_ATIME)
2519                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2520         if (bits & LA_CTIME)
2521                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2522         if (bits & LA_MTIME)
2523                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2524         if (bits & LA_SIZE) {
2525                 spin_lock(&inode->i_lock);
2526                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2527                 i_size_write(inode, attr->la_size);
2528                 spin_unlock(&inode->i_lock);
2529         }
2530
2531         /* OSD should not change "i_blocks" which is used by quota.
2532          * "i_blocks" should be changed by ldiskfs only. */
2533         if (bits & LA_MODE)
2534                 inode->i_mode = (inode->i_mode & S_IFMT) |
2535                                 (attr->la_mode & ~S_IFMT);
2536         if (bits & LA_UID)
2537                 i_uid_write(inode, attr->la_uid);
2538         if (bits & LA_GID)
2539                 i_gid_write(inode, attr->la_gid);
2540         if (bits & LA_NLINK)
2541                 set_nlink(inode, attr->la_nlink);
2542         if (bits & LA_RDEV)
2543                 inode->i_rdev = attr->la_rdev;
2544
2545         if (bits & LA_FLAGS) {
2546                 /* always keep S_NOCMTIME */
2547                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2548                                  S_NOCMTIME;
2549         }
2550         return 0;
2551 }
2552
2553 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2554 {
2555         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2556             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2557                 struct iattr    iattr;
2558                 int             rc;
2559
2560                 ll_vfs_dq_init(inode);
2561                 iattr.ia_valid = 0;
2562                 if (attr->la_valid & LA_UID)
2563                         iattr.ia_valid |= ATTR_UID;
2564                 if (attr->la_valid & LA_GID)
2565                         iattr.ia_valid |= ATTR_GID;
2566                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2567                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2568
2569                 rc = ll_vfs_dq_transfer(inode, &iattr);
2570                 if (rc) {
2571                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2572                                "enforcement enabled on the ldiskfs "
2573                                "filesystem?\n", inode->i_sb->s_id, rc);
2574                         return rc;
2575                 }
2576         }
2577         return 0;
2578 }
2579
2580 static int osd_attr_set(const struct lu_env *env,
2581                         struct dt_object *dt,
2582                         const struct lu_attr *attr,
2583                         struct thandle *handle)
2584 {
2585         struct osd_object *obj = osd_dt_obj(dt);
2586         struct inode      *inode;
2587         int rc;
2588
2589         if (!dt_object_exists(dt))
2590                 return -ENOENT;
2591
2592         LASSERT(handle != NULL);
2593         LASSERT(!dt_object_remote(dt));
2594         LASSERT(osd_invariant(obj));
2595
2596         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2597
2598         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2599                 struct osd_thread_info  *oti  = osd_oti_get(env);
2600                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2601                 struct lu_fid           *fid1 = &oti->oti_fid;
2602                 struct osd_inode_id     *id   = &oti->oti_id;
2603                 struct iam_path_descr   *ipd;
2604                 struct iam_container    *bag;
2605                 struct osd_thandle      *oh;
2606                 int                      rc;
2607
2608                 fid_cpu_to_be(fid1, fid0);
2609                 memset(id, 1, sizeof(*id));
2610                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2611                                   fid0)->oi_dir.od_container;
2612                 ipd = osd_idx_ipd_get(env, bag);
2613                 if (unlikely(ipd == NULL))
2614                         RETURN(-ENOMEM);
2615
2616                 oh = container_of0(handle, struct osd_thandle, ot_super);
2617                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2618                                 (const struct iam_rec *)id, ipd);
2619                 osd_ipd_put(env, bag, ipd);
2620                 return(rc > 0 ? 0 : rc);
2621         }
2622
2623         inode = obj->oo_inode;
2624
2625         rc = osd_quota_transfer(inode, attr);
2626         if (rc)
2627                 return rc;
2628
2629         spin_lock(&obj->oo_guard);
2630         rc = osd_inode_setattr(env, inode, attr);
2631         spin_unlock(&obj->oo_guard);
2632         if (rc != 0)
2633                 GOTO(out, rc);
2634
2635         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2636
2637         if (!(attr->la_valid & LA_FLAGS))
2638                 GOTO(out, rc);
2639
2640         /* Let's check if there are extra flags need to be set into LMA */
2641         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2642                 struct osd_thread_info *info = osd_oti_get(env);
2643                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2644
2645                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2646                 if (rc != 0)
2647                         GOTO(out, rc);
2648
2649                 lma->lma_incompat |=
2650                         lustre_to_lma_flags(attr->la_flags);
2651                 lustre_lma_swab(lma);
2652                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2653                                      lma, sizeof(*lma), XATTR_REPLACE);
2654                 if (rc != 0) {
2655                         struct osd_device *osd = osd_obj2dev(obj);
2656
2657                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2658                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2659                               lma->lma_incompat, rc);
2660                 } else {
2661                         obj->oo_lma_flags =
2662                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2663                 }
2664                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2665         }
2666 out:
2667         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2668
2669         return rc;
2670 }
2671
2672 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2673                                            struct osd_object *obj,
2674                                            const char *name, const int namelen)
2675 {
2676         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2677 }
2678
2679 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2680                       umode_t mode, struct dt_allocation_hint *hint,
2681                       struct thandle *th)
2682 {
2683         int result;
2684         struct osd_device  *osd = osd_obj2dev(obj);
2685         struct osd_thandle *oth;
2686         struct dt_object   *parent = NULL;
2687         struct inode       *inode;
2688
2689         LINVRNT(osd_invariant(obj));
2690         LASSERT(obj->oo_inode == NULL);
2691         LASSERT(obj->oo_hl_head == NULL);
2692
2693         if (S_ISDIR(mode) && ldiskfs_pdo) {
2694                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2695                 if (obj->oo_hl_head == NULL)
2696                         return -ENOMEM;
2697         }
2698
2699         oth = container_of(th, struct osd_thandle, ot_super);
2700         LASSERT(oth->ot_handle->h_transaction != NULL);
2701
2702         if (hint != NULL && hint->dah_parent != NULL &&
2703             !dt_object_remote(hint->dah_parent))
2704                 parent = hint->dah_parent;
2705
2706         inode = ldiskfs_create_inode(oth->ot_handle,
2707                                      parent ? osd_dt_obj(parent)->oo_inode :
2708                                               osd_sb(osd)->s_root->d_inode,
2709                                      mode);
2710         if (!IS_ERR(inode)) {
2711                 /* Do not update file c/mtime in ldiskfs. */
2712                 inode->i_flags |= S_NOCMTIME;
2713
2714                 /* For new created object, it must be consistent,
2715                  * and it is unnecessary to scrub against it. */
2716                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2717
2718                 obj->oo_inode = inode;
2719                 result = 0;
2720         } else {
2721                 if (obj->oo_hl_head != NULL) {
2722                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2723                         obj->oo_hl_head = NULL;
2724                 }
2725                 result = PTR_ERR(inode);
2726         }
2727         LINVRNT(osd_invariant(obj));
2728         return result;
2729 }
2730
2731 enum {
2732         OSD_NAME_LEN = 255
2733 };
2734
2735 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2736                      struct lu_attr *attr,
2737                      struct dt_allocation_hint *hint,
2738                      struct dt_object_format *dof,
2739                      struct thandle *th)
2740 {
2741         int result;
2742         struct osd_thandle *oth;
2743         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2744
2745         LASSERT(S_ISDIR(attr->la_mode));
2746
2747         oth = container_of(th, struct osd_thandle, ot_super);
2748         LASSERT(oth->ot_handle->h_transaction != NULL);
2749         result = osd_mkfile(info, obj, mode, hint, th);
2750
2751         return result;
2752 }
2753
2754 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2755                         struct lu_attr *attr,
2756                         struct dt_allocation_hint *hint,
2757                         struct dt_object_format *dof,
2758                         struct thandle *th)
2759 {
2760         int result;
2761         struct osd_thandle *oth;
2762         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2763
2764         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2765
2766         LASSERT(S_ISREG(attr->la_mode));
2767
2768         oth = container_of(th, struct osd_thandle, ot_super);
2769         LASSERT(oth->ot_handle->h_transaction != NULL);
2770
2771         result = osd_mkfile(info, obj, mode, hint, th);
2772         if (result == 0) {
2773                 LASSERT(obj->oo_inode != NULL);
2774                 if (feat->dif_flags & DT_IND_VARKEY)
2775                         result = iam_lvar_create(obj->oo_inode,
2776                                                  feat->dif_keysize_max,
2777                                                  feat->dif_ptrsize,
2778                                                  feat->dif_recsize_max,
2779                                                  oth->ot_handle);
2780                 else
2781                         result = iam_lfix_create(obj->oo_inode,
2782                                                  feat->dif_keysize_max,
2783                                                  feat->dif_ptrsize,
2784                                                  feat->dif_recsize_max,
2785                                                  oth->ot_handle);
2786
2787         }
2788         return result;
2789 }
2790
2791 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2792                      struct lu_attr *attr,
2793                      struct dt_allocation_hint *hint,
2794                      struct dt_object_format *dof,
2795                      struct thandle *th)
2796 {
2797         LASSERT(S_ISREG(attr->la_mode));
2798         return osd_mkfile(info, obj, (attr->la_mode &
2799                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2800 }
2801
2802 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2803                      struct lu_attr *attr,
2804                      struct dt_allocation_hint *hint,
2805                      struct dt_object_format *dof,
2806                      struct thandle *th)
2807 {
2808         LASSERT(S_ISLNK(attr->la_mode));
2809         return osd_mkfile(info, obj, (attr->la_mode &
2810                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2811 }
2812
2813 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2814                      struct lu_attr *attr,
2815                      struct dt_allocation_hint *hint,
2816                      struct dt_object_format *dof,
2817                      struct thandle *th)
2818 {
2819         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2820         int result;
2821
2822         LINVRNT(osd_invariant(obj));
2823         LASSERT(obj->oo_inode == NULL);
2824         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2825                 S_ISFIFO(mode) || S_ISSOCK(mode));
2826
2827         result = osd_mkfile(info, obj, mode, hint, th);
2828         if (result == 0) {
2829                 LASSERT(obj->oo_inode != NULL);
2830                 /*
2831                  * This inode should be marked dirty for i_rdev.  Currently
2832                  * that is done in the osd_attr_init().
2833                  */
2834                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2835                                    attr->la_rdev);
2836         }
2837         LINVRNT(osd_invariant(obj));
2838         return result;
2839 }
2840
2841 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2842                               struct lu_attr *,
2843                               struct dt_allocation_hint *hint,
2844                               struct dt_object_format *dof,
2845                               struct thandle *);
2846
2847 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2848 {
2849         osd_obj_type_f result;
2850
2851         switch (type) {
2852         case DFT_DIR:
2853                 result = osd_mkdir;
2854                 break;
2855         case DFT_REGULAR:
2856                 result = osd_mkreg;
2857                 break;
2858         case DFT_SYM:
2859                 result = osd_mksym;
2860                 break;
2861         case DFT_NODE:
2862                 result = osd_mknod;
2863                 break;
2864         case DFT_INDEX:
2865                 result = osd_mk_index;
2866                 break;
2867
2868         default:
2869                 LBUG();
2870                 break;
2871         }
2872         return result;
2873 }
2874
2875
2876 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2877                         struct dt_object *parent, struct dt_object *child,
2878                         umode_t child_mode)
2879 {
2880         LASSERT(ah);
2881
2882         ah->dah_parent = parent;
2883         ah->dah_mode = child_mode;
2884
2885         if (parent != NULL && !dt_object_remote(parent)) {
2886                 /* will help to find FID->ino at dt_insert("..") */
2887                 struct osd_object *pobj = osd_dt_obj(parent);
2888                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2889         }
2890 }
2891
2892 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2893                           struct lu_attr *attr, struct dt_object_format *dof)
2894 {
2895         struct inode   *inode = obj->oo_inode;
2896         __u64           valid = attr->la_valid;
2897         int             result;
2898
2899         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2900
2901         if (dof->dof_type != DFT_NODE)
2902                 attr->la_valid &= ~LA_RDEV;
2903         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2904                 attr->la_valid &= ~LA_ATIME;
2905         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2906                 attr->la_valid &= ~LA_CTIME;
2907         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2908                 attr->la_valid &= ~LA_MTIME;
2909
2910         result = osd_quota_transfer(inode, attr);
2911         if (result)
2912                 return;
2913
2914         if (attr->la_valid != 0) {
2915                 result = osd_inode_setattr(info->oti_env, inode, attr);
2916                 /*
2917                  * The osd_inode_setattr() should always succeed here.  The
2918                  * only error that could be returned is EDQUOT when we are
2919                  * trying to change the UID or GID of the inode. However, this
2920                  * should not happen since quota enforcement is no longer
2921                  * enabled on ldiskfs (lquota takes care of it).
2922                  */
2923                 LASSERTF(result == 0, "%d\n", result);
2924                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2925         }
2926
2927         attr->la_valid = valid;
2928 }
2929
2930 /**
2931  * Helper function for osd_object_create()
2932  *
2933  * \retval 0, on success
2934  */
2935 static int __osd_object_create(struct osd_thread_info *info,
2936                                struct osd_object *obj, struct lu_attr *attr,
2937                                struct dt_allocation_hint *hint,
2938                                struct dt_object_format *dof,
2939                                struct thandle *th)
2940 {
2941         int     result;
2942         __u32   umask;
2943
2944         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
2945
2946         /* we drop umask so that permissions we pass are not affected */
2947         umask = current->fs->umask;
2948         current->fs->umask = 0;
2949
2950         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2951                                                   th);
2952         if (likely(obj->oo_inode != NULL)) {
2953                 LASSERT(obj->oo_inode->i_state & I_NEW);
2954
2955                 /* Unlock the inode before attr initialization to avoid
2956                  * unnecessary dqget operations. LU-6378 */
2957                 unlock_new_inode(obj->oo_inode);
2958         }
2959
2960         if (likely(result == 0)) {
2961                 osd_attr_init(info, obj, attr, dof);
2962                 osd_object_init0(obj);
2963         }
2964
2965         /* restore previous umask value */
2966         current->fs->umask = umask;
2967
2968         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
2969
2970         return result;
2971 }
2972
2973 /**
2974  * Helper function for osd_object_create()
2975  *
2976  * \retval 0, on success
2977  */
2978 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2979                            const struct lu_fid *fid, struct thandle *th)
2980 {
2981         struct osd_thread_info *info = osd_oti_get(env);
2982         struct osd_inode_id    *id   = &info->oti_id;
2983         struct osd_device      *osd  = osd_obj2dev(obj);
2984         struct osd_thandle     *oh;
2985         int                     rc;
2986
2987         LASSERT(obj->oo_inode != NULL);
2988
2989         oh = container_of0(th, struct osd_thandle, ot_super);
2990         LASSERT(oh->ot_handle);
2991         osd_trans_exec_op(env, th, OSD_OT_INSERT);
2992
2993         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2994         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
2995                            OI_CHECK_FLD, NULL);
2996         osd_trans_exec_check(env, th, OSD_OT_INSERT);
2997
2998         return rc;
2999 }
3000
3001 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3002                    u64 seq, struct lu_seq_range *range)
3003 {
3004         struct seq_server_site  *ss = osd_seq_site(osd);
3005
3006         if (fid_seq_is_idif(seq)) {
3007                 fld_range_set_ost(range);
3008                 range->lsr_index = idif_ost_idx(seq);
3009                 return 0;
3010         }
3011
3012         if (!fid_seq_in_fldb(seq)) {
3013                 fld_range_set_mdt(range);
3014                 if (ss != NULL)
3015                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3016                          * at all */
3017                         range->lsr_index = ss->ss_node_id;
3018                 return 0;
3019         }
3020
3021         LASSERT(ss != NULL);
3022         fld_range_set_any(range);
3023         /* OSD will only do local fld lookup */
3024         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3025 }
3026
3027 /*
3028  * Concurrency: no external locking is necessary.
3029  */
3030 static int osd_declare_object_create(const struct lu_env *env,
3031                                      struct dt_object *dt,
3032                                      struct lu_attr *attr,
3033                                      struct dt_allocation_hint *hint,
3034                                      struct dt_object_format *dof,
3035                                      struct thandle *handle)
3036 {
3037         struct osd_thandle      *oh;
3038         int                      rc;
3039         ENTRY;
3040
3041         LASSERT(handle != NULL);
3042
3043         oh = container_of0(handle, struct osd_thandle, ot_super);
3044         LASSERT(oh->ot_handle == NULL);
3045
3046         /* EA object consumes more credits than regular object: osd_mk_index
3047          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3048          * leaf_node, could involves the block, block bitmap, groups, GDT
3049          * change for each block, so add 4 * 2 credits in that case. */
3050         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3051                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3052                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3053         /* Reuse idle OI block may cause additional one OI block
3054          * to be changed. */
3055         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3056                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3057
3058         if (!attr)
3059                 RETURN(0);
3060
3061         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
3062                                    osd_dt_obj(dt), false, NULL, false);
3063         if (rc != 0)
3064                 RETURN(rc);
3065
3066         /* will help to find FID->ino mapping at dt_insert() */
3067         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3068                                    osd_dt_obj(dt));
3069
3070         RETURN(rc);
3071 }
3072
3073 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
3074                              struct lu_attr *attr,
3075                              struct dt_allocation_hint *hint,
3076                              struct dt_object_format *dof, struct thandle *th)
3077 {
3078         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
3079         struct osd_object       *obj    = osd_dt_obj(dt);
3080         struct osd_thread_info  *info   = osd_oti_get(env);
3081         int result;
3082         ENTRY;
3083
3084         if (dt_object_exists(dt))
3085                 return -EEXIST;
3086
3087         LINVRNT(osd_invariant(obj));
3088         LASSERT(!dt_object_remote(dt));
3089         LASSERT(osd_write_locked(env, obj));
3090         LASSERT(th != NULL);
3091
3092         if (unlikely(fid_is_acct(fid)))
3093                 /* Quota files can't be created from the kernel any more,
3094                  * 'tune2fs -O quota' will take care of creating them */
3095                 RETURN(-EPERM);
3096
3097         result = __osd_object_create(info, obj, attr, hint, dof, th);
3098         if (result == 0) {
3099                 result = __osd_oi_insert(env, obj, fid, th);
3100                 if (obj->oo_dt.do_body_ops == &osd_body_ops_new)
3101                         obj->oo_dt.do_body_ops = &osd_body_ops;
3102         }
3103         LASSERT(ergo(result == 0,
3104                 dt_object_exists(dt) && !dt_object_remote(dt)));
3105
3106         LASSERT(osd_invariant(obj));
3107         RETURN(result);
3108 }
3109
3110 /**
3111  * Called to destroy on-disk representation of the object
3112  *
3113  * Concurrency: must be locked
3114  */
3115 static int osd_declare_object_destroy(const struct lu_env *env,
3116                                       struct dt_object *dt,
3117                                       struct thandle *th)
3118 {
3119         struct osd_object  *obj = osd_dt_obj(dt);
3120         struct inode       *inode = obj->oo_inode;
3121         struct osd_thandle *oh;
3122         int                 rc;
3123         ENTRY;
3124
3125         if (inode == NULL)
3126                 RETURN(-ENOENT);
3127
3128         oh = container_of0(th, struct osd_thandle, ot_super);
3129         LASSERT(oh->ot_handle == NULL);
3130
3131         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3132                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3133         /* Recycle idle OI leaf may cause additional three OI blocks
3134          * to be changed. */
3135         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3136                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3137                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3138         /* one less inode */
3139         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3140                                    -1, oh, obj, false, NULL, false);
3141         if (rc)
3142                 RETURN(rc);
3143         /* data to be truncated */
3144         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3145                                    0, oh, obj, true, NULL, false);
3146         if (rc)
3147                 RETURN(rc);
3148