Whamcloud - gitweb
LU-9104 obd: Ignore unknown config param while mounting
[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 <uapi/linux/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 #define PFID_STRIPE_IDX_BITS    16
82 #define PFID_STRIPE_COUNT_MASK  ((1 << PFID_STRIPE_IDX_BITS) - 1)
83
84 int ldiskfs_pdo = 1;
85 module_param(ldiskfs_pdo, int, 0644);
86 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
87
88 int ldiskfs_track_declares_assert;
89 module_param(ldiskfs_track_declares_assert, int, 0644);
90 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
91
92 /* Slab to allocate dynlocks */
93 struct kmem_cache *dynlock_cachep;
94
95 /* Slab to allocate osd_it_ea */
96 struct kmem_cache *osd_itea_cachep;
97
98 static struct lu_kmem_descr ldiskfs_caches[] = {
99         {
100                 .ckd_cache = &dynlock_cachep,
101                 .ckd_name  = "dynlock_cache",
102                 .ckd_size  = sizeof(struct dynlock_handle)
103         },
104         {
105                 .ckd_cache = &osd_itea_cachep,
106                 .ckd_name  = "osd_itea_cache",
107                 .ckd_size  = sizeof(struct osd_it_ea)
108         },
109         {
110                 .ckd_cache = NULL
111         }
112 };
113
114 static const char dot[] = ".";
115 static const char dotdot[] = "..";
116
117 static const struct lu_object_operations      osd_lu_obj_ops;
118 static const struct dt_object_operations      osd_obj_ops;
119 static const struct dt_object_operations      osd_obj_otable_it_ops;
120 static const struct dt_index_operations       osd_index_iam_ops;
121 static const struct dt_index_operations       osd_index_ea_ops;
122
123 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
124                           const struct lu_fid *fid);
125 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
126                                                 struct osd_device *osd);
127
128 int osd_trans_declare_op2rb[] = {
129         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
130         [OSD_OT_PUNCH]          = OSD_OT_MAX,
131         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
132         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
133         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
134         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
135         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
136         [OSD_OT_WRITE]          = OSD_OT_WRITE,
137         [OSD_OT_INSERT]         = OSD_OT_DELETE,
138         [OSD_OT_DELETE]         = OSD_OT_INSERT,
139         [OSD_OT_QUOTA]          = OSD_OT_MAX,
140 };
141
142 static int osd_has_index(const struct osd_object *obj)
143 {
144         return obj->oo_dt.do_index_ops != NULL;
145 }
146
147 static int osd_object_invariant(const struct lu_object *l)
148 {
149         return osd_invariant(osd_obj(l));
150 }
151
152 /*
153  * Concurrency: doesn't matter
154  */
155 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
156 {
157         struct osd_thread_info *oti = osd_oti_get(env);
158         return oti->oti_w_locks > 0 && o->oo_owner == env;
159 }
160
161 /*
162  * Concurrency: doesn't access mutable data
163  */
164 static int osd_root_get(const struct lu_env *env,
165                         struct dt_device *dev, struct lu_fid *f)
166 {
167         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
168         return 0;
169 }
170
171 /*
172  * the following set of functions are used to maintain per-thread
173  * cache of FID->ino mapping. this mechanism is needed to resolve
174  * FID to inode at dt_insert() which in turn stores ino in the
175  * directory entries to keep ldiskfs compatible with ext[34].
176  * due to locking-originated restrictions we can't lookup ino
177  * using LU cache (deadlock is possible). lookup using OI is quite
178  * expensive. so instead we maintain this cache and methods like
179  * dt_create() fill it. so in the majority of cases dt_insert() is
180  * able to find needed mapping in lockless manner.
181  */
182 static struct osd_idmap_cache *
183 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
184              const struct lu_fid *fid)
185 {
186         struct osd_thread_info  *oti   = osd_oti_get(env);
187         struct osd_idmap_cache  *idc    = oti->oti_ins_cache;
188         int i;
189         for (i = 0; i < oti->oti_ins_cache_used; i++) {
190                 if (!lu_fid_eq(&idc[i].oic_fid, fid))
191                         continue;
192                 if (idc[i].oic_dev != osd)
193                         continue;
194
195                 return idc + i;
196         }
197
198         return NULL;
199 }
200
201 static struct osd_idmap_cache *
202 osd_idc_add(const struct lu_env *env, struct osd_device *osd,
203             const struct lu_fid *fid)
204 {
205         struct osd_thread_info  *oti   = osd_oti_get(env);
206         struct osd_idmap_cache  *idc;
207         int i;
208
209         if (unlikely(oti->oti_ins_cache_used >= oti->oti_ins_cache_size)) {
210                 i = oti->oti_ins_cache_size * 2;
211                 if (i == 0)
212                         i = OSD_INS_CACHE_SIZE;
213                 OBD_ALLOC(idc, sizeof(*idc) * i);
214                 if (idc == NULL)
215                         return ERR_PTR(-ENOMEM);
216                 if (oti->oti_ins_cache != NULL) {
217                         memcpy(idc, oti->oti_ins_cache,
218                                oti->oti_ins_cache_used * sizeof(*idc));
219                         OBD_FREE(oti->oti_ins_cache,
220                                  oti->oti_ins_cache_used * sizeof(*idc));
221                 }
222                 oti->oti_ins_cache = idc;
223                 oti->oti_ins_cache_size = i;
224         }
225
226         idc = oti->oti_ins_cache + oti->oti_ins_cache_used++;
227         idc->oic_fid = *fid;
228         idc->oic_dev = osd;
229         idc->oic_lid.oii_ino = 0;
230         idc->oic_lid.oii_gen = 0;
231         idc->oic_remote = 0;
232
233         return idc;
234 }
235
236 /*
237  * lookup mapping for the given fid in the cache, initialize a
238  * new one if not found. the initialization checks whether the
239  * object is local or remote. for local objects, OI is used to
240  * learn ino/generation. the function is used when the caller
241  * has no information about the object, e.g. at dt_insert().
242  */
243 static struct osd_idmap_cache *
244 osd_idc_find_or_init(const struct lu_env *env, struct osd_device *osd,
245                      const struct lu_fid *fid)
246 {
247         struct osd_idmap_cache *idc;
248         int rc;
249
250         idc = osd_idc_find(env, osd, fid);
251         LASSERT(!IS_ERR(idc));
252         if (idc != NULL)
253                 return idc;
254
255         /* new mapping is needed */
256         idc = osd_idc_add(env, osd, fid);
257         if (IS_ERR(idc))
258                 return idc;
259
260         /* initialize it */
261         rc = osd_remote_fid(env, osd, fid);
262         if (unlikely(rc < 0))
263                 return ERR_PTR(rc);
264
265         if (rc == 0) {
266                 /* the object is local, lookup in OI */
267                 /* XXX: probably cheaper to lookup in LU first? */
268                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
269                                    &idc->oic_lid, 0);
270                 if (unlikely(rc < 0)) {
271                         CERROR("can't lookup: rc = %d\n", rc);
272                         return ERR_PTR(rc);
273                 }
274         } else {
275                 /* the object is remote */
276                 idc->oic_remote = 1;
277         }
278
279         return idc;
280 }
281
282 /*
283  * lookup mapping for given FID and fill it from the given object.
284  * the object is lolcal by definition.
285  */
286 static int osd_idc_find_and_init(const struct lu_env *env,
287                                  struct osd_device *osd,
288                                  struct osd_object *obj)
289 {
290         const struct lu_fid     *fid = lu_object_fid(&obj->oo_dt.do_lu);
291         struct osd_idmap_cache  *idc;
292
293         idc = osd_idc_find(env, osd, fid);
294         LASSERT(!IS_ERR(idc));
295         if (idc != NULL) {
296                 if (obj->oo_inode == NULL)
297                         return 0;
298                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
299                         LASSERT(idc->oic_lid.oii_ino == 0);
300                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
301                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
302                 }
303                 return 0;
304         }
305
306         /* new mapping is needed */
307         idc = osd_idc_add(env, osd, fid);
308         if (IS_ERR(idc))
309                 return PTR_ERR(idc);
310
311         if (obj->oo_inode != NULL) {
312                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
313                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
314         }
315         return 0;
316 }
317
318 /*
319  * OSD object methods.
320  */
321
322 /*
323  * Concurrency: no concurrent access is possible that early in object
324  * life-cycle.
325  */
326 static struct lu_object *osd_object_alloc(const struct lu_env *env,
327                                           const struct lu_object_header *hdr,
328                                           struct lu_device *d)
329 {
330         struct osd_object *mo;
331
332         OBD_ALLOC_PTR(mo);
333         if (mo != NULL) {
334                 struct lu_object *l;
335
336                 l = &mo->oo_dt.do_lu;
337                 dt_object_init(&mo->oo_dt, NULL, d);
338                 mo->oo_dt.do_ops = &osd_obj_ops;
339                 l->lo_ops = &osd_lu_obj_ops;
340                 init_rwsem(&mo->oo_sem);
341                 init_rwsem(&mo->oo_ext_idx_sem);
342                 spin_lock_init(&mo->oo_guard);
343                 INIT_LIST_HEAD(&mo->oo_xattr_list);
344                 return l;
345         } else {
346                 return NULL;
347         }
348 }
349
350 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
351                 struct dentry *dentry, struct lustre_ost_attrs *loa)
352 {
353         int rc;
354
355         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
356                              (void *)loa, sizeof(*loa));
357         if (rc > 0) {
358                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
359
360                 if (rc < sizeof(*lma))
361                         return -EINVAL;
362
363                 rc = 0;
364                 lustre_loa_swab(loa, true);
365                 /* Check LMA compatibility */
366                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
367                         CWARN("%s: unsupported incompat LMA feature(s) %#x "
368                               "for fid = "DFID", ino = %lu\n",
369                               osd_ino2name(inode),
370                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
371                               PFID(&lma->lma_self_fid), inode->i_ino);
372                         rc = -EOPNOTSUPP;
373                 }
374         } else if (rc == 0) {
375                 rc = -ENODATA;
376         }
377
378         return rc;
379 }
380
381 /*
382  * retrieve object from backend ext fs.
383  **/
384 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
385                        struct osd_inode_id *id)
386 {
387         int rc;
388         struct inode *inode = NULL;
389
390         /* if we look for an inode withing a running
391          * transaction, then we risk to deadlock */
392         /* osd_dirent_check_repair() breaks this */
393         /*LASSERT(current->journal_info == NULL);*/
394
395         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
396         if (IS_ERR(inode)) {
397                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
398                        id->oii_ino, PTR_ERR(inode));
399         } else if (id->oii_gen != OSD_OII_NOGEN &&
400                    inode->i_generation != id->oii_gen) {
401                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
402                        "i_generation = %u\n",
403                        id->oii_ino, id->oii_gen, inode->i_generation);
404                 iput(inode);
405                 inode = ERR_PTR(-ESTALE);
406         } else if (inode->i_nlink == 0) {
407                 /* due to parallel readdir and unlink,
408                 * we can have dead inode here. */
409                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
410                 iput(inode);
411                 inode = ERR_PTR(-ESTALE);
412         } else if (is_bad_inode(inode)) {
413                 CWARN("%s: bad inode: ino = %u\n",
414                 osd_dev2name(dev), id->oii_ino);
415                 iput(inode);
416                 inode = ERR_PTR(-ENOENT);
417         } else if ((rc = osd_attach_jinode(inode))) {
418                 iput(inode);
419                 inode = ERR_PTR(rc);
420         } else {
421                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
422                 if (id->oii_gen == OSD_OII_NOGEN)
423                         osd_id_gen(id, inode->i_ino, inode->i_generation);
424
425                 /* Do not update file c/mtime in ldiskfs.
426                  * NB: we don't have any lock to protect this because we don't
427                  * have reference on osd_object now, but contention with
428                  * another lookup + attr_set can't happen in the tiny window
429                  * between if (...) and set S_NOCMTIME. */
430                 if (!(inode->i_flags & S_NOCMTIME))
431                         inode->i_flags |= S_NOCMTIME;
432         }
433         return inode;
434 }
435
436 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
437                           handle_t *handle, struct dentry *child,
438                           struct inode *inode, struct htree_lock *hlock)
439 {
440         int rc, rc2;
441
442         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
443         if (rc == -ENOBUFS || rc == -ENOSPC) {
444                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
445                 struct inode *parent = child->d_parent->d_inode;
446                 struct lu_fid *fid = NULL;
447
448                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
449                 if (!rc2) {
450                         fid = &loa->loa_lma.lma_self_fid;
451                 } else if (rc2 == -ENODATA) {
452                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
453                                 fid = &info->oti_fid3;
454                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
455                         } else if (!osd->od_is_ost && osd->od_index == 0) {
456                                 fid = &info->oti_fid3;
457                                 lu_igif_build(fid, parent->i_ino,
458                                               parent->i_generation);
459                         }
460                 }
461
462                 if (fid != NULL)
463                         CWARN("%s: directory (inode: %lu, FID: "DFID") %s "
464                               "maximum entry limit\n",
465                               osd_name(osd), parent->i_ino, PFID(fid),
466                               rc == -ENOSPC ? "has reached" : "is approaching");
467                 else
468                         CWARN("%s: directory (inode: %lu, FID: unknown) %s "
469                               "maximum entry limit\n",
470                               osd_name(osd), parent->i_ino,
471                               rc == -ENOSPC ? "has reached" : "is approaching");
472
473                 /* ignore such error now */
474                 if (rc == -ENOBUFS)
475                         rc = 0;
476         }
477
478         return rc;
479 }
480
481
482 static struct inode *
483 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
484              struct osd_inode_id *id, struct lu_fid *fid)
485 {
486         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
487         struct inode *inode;
488         int rc;
489
490         inode = osd_iget(info, dev, id);
491         if (IS_ERR(inode))
492                 return inode;
493
494         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
495         if (!rc) {
496                 *fid = loa->loa_lma.lma_self_fid;
497         } else if (rc == -ENODATA) {
498                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
499                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
500                 else
501                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
502         } else {
503                 iput(inode);
504                 inode = ERR_PTR(rc);
505         }
506         return inode;
507 }
508
509 static struct inode *osd_iget_check(struct osd_thread_info *info,
510                                     struct osd_device *dev,
511                                     const struct lu_fid *fid,
512                                     struct osd_inode_id *id,
513                                     bool trusted)
514 {
515         struct inode *inode;
516         int rc = 0;
517         ENTRY;
518
519         /* The cached OI mapping is trustable. If we cannot locate the inode
520          * via the cached OI mapping, then return the failure to the caller
521          * directly without further OI checking. */
522
523 again:
524         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
525         if (IS_ERR(inode)) {
526                 rc = PTR_ERR(inode);
527                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
528                         goto check_oi;
529
530                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
531                        PFID(fid), id->oii_ino, rc);
532                 GOTO(put, rc);
533         }
534
535         if (is_bad_inode(inode)) {
536                 rc = -ENOENT;
537                 if (!trusted)
538                         goto check_oi;
539
540                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
541                        PFID(fid), id->oii_ino);
542                 GOTO(put, rc);
543         }
544
545         if (id->oii_gen != OSD_OII_NOGEN &&
546             inode->i_generation != id->oii_gen) {
547                 rc = -ESTALE;
548                 if (!trusted)
549                         goto check_oi;
550
551                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
552                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
553                        id->oii_ino, id->oii_gen, inode->i_generation);
554                 GOTO(put, rc);
555         }
556
557         if (inode->i_nlink == 0) {
558                 rc = -ENOENT;
559                 if (!trusted)
560                         goto check_oi;
561
562                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
563                        PFID(fid), id->oii_ino);
564                 GOTO(put, rc);
565         }
566
567         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
568
569 check_oi:
570         if (rc != 0) {
571                 __u32 saved_ino = id->oii_ino;
572                 __u32 saved_gen = id->oii_gen;
573
574                 LASSERT(!trusted);
575                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
576
577                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
578                 /* XXX: There are four possible cases:
579                  *      1. rc = 0.
580                  *         Backup/restore caused the OI invalid.
581                  *      2. rc = 0.
582                  *         Someone unlinked the object but NOT removed
583                  *         the OI mapping, such as mount target device
584                  *         as ldiskfs, and modify something directly.
585                  *      3. rc = -ENOENT.
586                  *         Someone just removed the object between the
587                  *         former oi_lookup and the iget. It is normal.
588                  *      4. Other failure cases.
589                  *
590                  *      Generally, when the device is mounted, it will
591                  *      auto check whether the system is restored from
592                  *      file-level backup or not. We trust such detect
593                  *      to distinguish the 1st case from the 2nd case:
594                  *      if the OI files are consistent but may contain
595                  *      stale OI mappings because of case 2, if iget()
596                  *      returns -ENOENT or -ESTALE, then it should be
597                  *      the case 2. */
598                 if (rc != 0)
599                         /* If the OI mapping was in OI file before the
600                          * osd_iget_check(), but now, it is disappear,
601                          * then it must be removed by race. That is a
602                          * normal race case. */
603                         GOTO(put, rc);
604
605                 /* It is the OI scrub updated the OI mapping by race.
606                  * The new OI mapping must be valid. */
607                 if (saved_ino != id->oii_ino ||
608                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
609                         if (!IS_ERR(inode))
610                                 iput(inode);
611
612                         trusted = true;
613                         goto again;
614                 }
615
616                 if (IS_ERR(inode)) {
617                         if (dev->od_scrub.os_file.sf_flags & SF_INCONSISTENT)
618                                 /* It still can be the case 2, but we cannot
619                                  * distinguish it from the case 1. So return
620                                  * -EREMCHG to block current operation until
621                                  *  OI scrub rebuilt the OI mappings. */
622                                 rc = -EREMCHG;
623                         else
624                                 rc = -ENOENT;
625
626                         GOTO(put, rc);
627                 }
628
629                 if (inode->i_generation == id->oii_gen)
630                         rc = -ENOENT;
631                 else
632                         rc = -EREMCHG;
633         } else {
634                 if (id->oii_gen == OSD_OII_NOGEN)
635                         osd_id_gen(id, inode->i_ino, inode->i_generation);
636
637                 /* Do not update file c/mtime in ldiskfs.
638                  * NB: we don't have any lock to protect this because we don't
639                  * have reference on osd_object now, but contention with
640                  * another lookup + attr_set can't happen in the tiny window
641                  * between if (...) and set S_NOCMTIME. */
642                 if (!(inode->i_flags & S_NOCMTIME))
643                         inode->i_flags |= S_NOCMTIME;
644         }
645
646         GOTO(put, rc);
647
648 put:
649         if (rc != 0) {
650                 if (!IS_ERR(inode))
651                         iput(inode);
652
653                 inode = ERR_PTR(rc);
654         }
655
656         return inode;
657 }
658
659 /**
660  * \retval +v: new filter_fid, does not contain self-fid
661  * \retval 0:  filter_fid_old, contains self-fid
662  * \retval -v: other failure cases
663  */
664 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
665                  struct dentry *dentry, struct lu_fid *fid)
666 {
667         struct filter_fid_old   *ff     = &info->oti_ff;
668         struct ost_id           *ostid  = &info->oti_ostid;
669         int                      rc;
670
671         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
672         if (rc == sizeof(*ff)) {
673                 rc = 0;
674                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
675                 rc = ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
676                 /*
677                  * XXX: use 0 as the index for compatibility, the caller will
678                  *      handle index related issues when necessary.
679                  */
680                 if (!rc)
681                         ostid_to_fid(fid, ostid, 0);
682         } else if (rc == sizeof(struct filter_fid)) {
683                 rc = 1;
684         } else if (rc >= 0) {
685                 rc = -EINVAL;
686         }
687
688         return rc;
689 }
690
691 static int osd_lma_self_repair(struct osd_thread_info *info,
692                                struct osd_device *osd, struct inode *inode,
693                                const struct lu_fid *fid, __u32 compat)
694 {
695         handle_t *jh;
696         int       rc;
697
698         LASSERT(current->journal_info == NULL);
699
700         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
701                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
702         if (IS_ERR(jh)) {
703                 rc = PTR_ERR(jh);
704                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
705                       osd_name(osd), rc);
706                 return rc;
707         }
708
709         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
710         if (rc != 0)
711                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
712                       osd_name(osd), rc);
713         ldiskfs_journal_stop(jh);
714         return rc;
715 }
716
717 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
718 {
719         struct osd_thread_info  *info   = osd_oti_get(env);
720         struct osd_device       *osd    = osd_obj2dev(obj);
721         struct lustre_ost_attrs *loa    = &info->oti_ost_attrs;
722         struct lustre_mdt_attrs *lma    = &loa->loa_lma;
723         struct inode            *inode  = obj->oo_inode;
724         struct dentry           *dentry = &info->oti_obj_dentry;
725         struct lu_fid           *fid    = NULL;
726         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
727         int                      rc;
728         ENTRY;
729
730         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
731                              (void *)loa, sizeof(*loa));
732         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
733                 fid = &lma->lma_self_fid;
734                 rc = osd_get_idif(info, inode, dentry, fid);
735                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
736                         /* For the given OST-object, if it has neither LMA nor
737                          * FID in XATTR_NAME_FID, then the given FID (which is
738                          * contained in the @obj, from client RPC for locating
739                          * the OST-object) is trusted. We use it to generate
740                          * the LMA. */
741                         osd_lma_self_repair(info, osd, inode, rfid,
742                                             LMAC_FID_ON_OST);
743                         RETURN(0);
744                 }
745         }
746
747         if (rc < 0)
748                 RETURN(rc);
749
750         if (rc > 0) {
751                 rc = 0;
752                 lustre_lma_swab(lma);
753                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
754                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
755                               S_ISREG(inode->i_mode)))) {
756                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
757                               "fid = "DFID", ino = %lu\n", osd_name(osd),
758                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
759                               PFID(rfid), inode->i_ino);
760                         rc = -EOPNOTSUPP;
761                 } else {
762                         fid = &lma->lma_self_fid;
763                 }
764         }
765
766         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
767                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
768                         struct ost_id   *oi   = &info->oti_ostid;
769                         struct lu_fid   *fid1 = &info->oti_fid3;
770                         __u32            idx  = fid_idif_ost_idx(rfid);
771
772                         /* For old IDIF, the OST index is not part of the IDIF,
773                          * Means that different OSTs may have the same IDIFs.
774                          * Under such case, we need to make some compatible
775                          * check to make sure to trigger OI scrub properly. */
776                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
777                                 /* Given @rfid is new, LMA is old. */
778                                 fid_to_ostid(fid, oi);
779                                 ostid_to_fid(fid1, oi, idx);
780                                 if (lu_fid_eq(fid1, rfid)) {
781                                         if (osd->od_index_in_idif)
782                                                 osd_lma_self_repair(info, osd,
783                                                         inode, rfid,
784                                                         LMAC_FID_ON_OST);
785                                         RETURN(0);
786                                 }
787                         }
788                 }
789
790                 rc = -EREMCHG;
791         }
792
793         RETURN(rc);
794 }
795
796 struct osd_check_lmv_buf {
797 #ifdef HAVE_DIR_CONTEXT
798         /* please keep it as first member */
799         struct dir_context       ctx;
800 #endif
801         struct osd_thread_info  *oclb_info;
802         struct osd_device       *oclb_dev;
803         struct osd_idmap_cache  *oclb_oic;
804 };
805
806 /**
807  * It is called internally by ->readdir() to filter out the
808  * local slave object's FID of the striped directory.
809  *
810  * \retval      1 found the local slave's FID
811  * \retval      0 continue to check next item
812  * \retval      -ve for failure
813  */
814 #ifdef HAVE_FILLDIR_USE_CTX
815 static int osd_stripe_dir_filldir(struct dir_context *buf,
816 #else
817 static int osd_stripe_dir_filldir(void *buf,
818 #endif
819                                   const char *name, int namelen,
820                                   loff_t offset, __u64 ino, unsigned d_type)
821 {
822         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
823         struct osd_thread_info *oti = oclb->oclb_info;
824         struct lu_fid *fid = &oti->oti_fid3;
825         struct osd_inode_id *id = &oti->oti_id3;
826         struct osd_device *dev = oclb->oclb_dev;
827         struct osd_idmap_cache *oic = oclb->oclb_oic;
828         struct inode *inode;
829         int rc;
830
831         if (name[0] == '.')
832                 return 0;
833
834         fid_zero(fid);
835         sscanf(name + 1, SFID, RFID(fid));
836         if (!fid_is_sane(fid))
837                 return 0;
838
839         if (osd_remote_fid(oti->oti_env, dev, fid))
840                 return 0;
841
842         osd_id_gen(id, ino, OSD_OII_NOGEN);
843         inode = osd_iget(oti, dev, id);
844         if (IS_ERR(inode))
845                 return PTR_ERR(inode);
846
847         iput(inode);
848         osd_add_oi_cache(oti, dev, id, fid);
849         oic->oic_fid = *fid;
850         oic->oic_lid = *id;
851         oic->oic_dev = dev;
852         rc = osd_oii_insert(dev, oic, true);
853
854         return rc == 0 ? 1 : rc;
855 }
856
857 /* When lookup item under striped directory, we need to locate the master
858  * MDT-object of the striped directory firstly, then the client will send
859  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
860  * and the item's name. If the system is restored from MDT file level backup,
861  * then before the OI scrub completely built the OI files, the OI mappings of
862  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
863  * not a problem for the master MDT-object. Because when locate the master
864  * MDT-object, we will do name based lookup (for the striped directory itself)
865  * firstly, during such process we can setup the correct OI mapping for the
866  * master MDT-object. But it will be trouble for the slave MDT-object. Because
867  * the client will not trigger name based lookup on the MDT to locate the slave
868  * MDT-object before locating item under the striped directory, then when
869  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
870  * is invalid and does not know what the right OI mapping is, then the MDT has
871  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
872  * the OI file, related OI mapping is unknown yet, please try again later. And
873  * then client will re-try the RPC again and again until related OI mapping has
874  * been updated. That is quite inefficient.
875  *
876  * To resolve above trouble, we will handle it as the following two cases:
877  *
878  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
879  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
880  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
881  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
882  *    directly. Please check osd_fid_lookup().
883  *
884  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
885  *    Under such case, during lookup the master MDT-object, we will lookup the
886  *    slave MDT-object via readdir against the master MDT-object, because the
887  *    slave MDT-objects information are stored as sub-directories with the name
888  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
889  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
890  *    the correct OI mapping for the slave MDT-object. */
891 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
892                          struct inode *inode, struct osd_idmap_cache *oic)
893 {
894         struct lu_buf *buf = &oti->oti_big_buf;
895         struct dentry *dentry = &oti->oti_obj_dentry;
896         struct file *filp = &oti->oti_file;
897         const struct file_operations *fops;
898         struct lmv_mds_md_v1 *lmv1;
899         struct osd_check_lmv_buf oclb = {
900 #ifdef HAVE_DIR_CONTEXT
901                 .ctx.actor = osd_stripe_dir_filldir,
902 #endif
903                 .oclb_info = oti,
904                 .oclb_dev = dev,
905                 .oclb_oic = oic
906         };
907         int rc = 0;
908         ENTRY;
909
910 again:
911         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
912                              buf->lb_len);
913         if (rc == -ERANGE) {
914                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
915                 if (rc > 0) {
916                         lu_buf_realloc(buf, rc);
917                         if (buf->lb_buf == NULL)
918                                 GOTO(out, rc = -ENOMEM);
919
920                         goto again;
921                 }
922         }
923
924         if (unlikely(rc == 0 || rc == -ENODATA))
925                 GOTO(out, rc = 0);
926
927         if (rc < 0)
928                 GOTO(out, rc);
929
930         if (unlikely(buf->lb_buf == NULL)) {
931                 lu_buf_realloc(buf, rc);
932                 if (buf->lb_buf == NULL)
933                         GOTO(out, rc = -ENOMEM);
934
935                 goto again;
936         }
937
938         lmv1 = buf->lb_buf;
939         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
940                 GOTO(out, rc = 0);
941
942         fops = inode->i_fop;
943         dentry->d_inode = inode;
944         dentry->d_sb = inode->i_sb;
945         filp->f_pos = 0;
946         filp->f_path.dentry = dentry;
947         filp->f_mode = FMODE_64BITHASH;
948         filp->f_mapping = inode->i_mapping;
949         filp->f_op = fops;
950         filp->private_data = NULL;
951         set_file_inode(filp, inode);
952
953 #ifdef HAVE_DIR_CONTEXT
954         oclb.ctx.pos = filp->f_pos;
955         rc = fops->iterate(filp, &oclb.ctx);
956         filp->f_pos = oclb.ctx.pos;
957 #else
958         rc = fops->readdir(filp, &oclb, osd_stripe_dir_filldir);
959 #endif
960         fops->release(inode, filp);
961
962 out:
963         if (rc < 0)
964                 CDEBUG(D_LFSCK, "%s: fail to check LMV EA, inode = %lu/%u,"
965                        DFID": rc = %d\n", osd_ino2name(inode),
966                        inode->i_ino, inode->i_generation,
967                        PFID(&oic->oic_fid), rc);
968         else
969                 rc = 0;
970
971         RETURN(rc);
972 }
973
974 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
975                           const struct lu_fid *fid,
976                           const struct lu_object_conf *conf)
977 {
978         struct osd_thread_info *info;
979         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
980         struct osd_device *dev;
981         struct osd_idmap_cache *oic;
982         struct osd_inode_id *id;
983         struct inode *inode = NULL;
984         struct osd_scrub *scrub;
985         struct scrub_file *sf;
986         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
987         __u32 saved_ino;
988         __u32 saved_gen;
989         int result = 0;
990         int rc1 = 0;
991         bool remote = false;
992         bool trusted = true;
993         bool updated = false;
994         ENTRY;
995
996         LINVRNT(osd_invariant(obj));
997         LASSERT(obj->oo_inode == NULL);
998         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
999
1000         dev = osd_dev(ldev);
1001         scrub = &dev->od_scrub;
1002         sf = &scrub->os_file;
1003         info = osd_oti_get(env);
1004         LASSERT(info);
1005         oic = &info->oti_cache;
1006
1007         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1008                 RETURN(-ENOENT);
1009
1010         /* For the object is created as locking anchor, or for the object to
1011          * be created on disk. No need to osd_oi_lookup() at here because FID
1012          * shouldn't never be re-used, if it's really a duplicate FID from
1013          * unexpected reason, we should be able to detect it later by calling
1014          * do_create->osd_oi_insert(). */
1015         if (conf && conf->loc_flags & LOC_F_NEW)
1016                 GOTO(out, result = 0);
1017
1018         /* Search order: 1. per-thread cache. */
1019         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1020                 id = &oic->oic_lid;
1021                 goto iget;
1022         }
1023
1024         id = &info->oti_id;
1025         if (!list_empty(&scrub->os_inconsistent_items)) {
1026                 /* Search order: 2. OI scrub pending list. */
1027                 result = osd_oii_lookup(dev, fid, id);
1028                 if (!result)
1029                         goto iget;
1030         }
1031
1032         /* The OI mapping in the OI file can be updated by the OI scrub
1033          * when we locate the inode via FID. So it may be not trustable. */
1034         trusted = false;
1035
1036         /* Search order: 3. OI files. */
1037         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1038         if (result == -ENOENT) {
1039                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1040                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1041                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
1042                                       sf->sf_oi_bitmap))
1043                         GOTO(out, result = 0);
1044
1045                 goto trigger;
1046         }
1047
1048         if (result)
1049                 GOTO(out, result);
1050
1051 iget:
1052         obj->oo_inode = NULL;
1053         /* for later passes through checks, not true on first pass */
1054         if (!IS_ERR_OR_NULL(inode))
1055                 iput(inode);
1056
1057         inode = osd_iget_check(info, dev, fid, id, trusted);
1058         if (!IS_ERR(inode)) {
1059                 obj->oo_inode = inode;
1060                 result = 0;
1061                 if (remote)
1062                         goto trigger;
1063
1064                 goto check_lma;
1065         }
1066
1067         result = PTR_ERR(inode);
1068         if (result == -ENOENT || result == -ESTALE)
1069                 GOTO(out, result = 0);
1070
1071         if (result != -EREMCHG)
1072                 GOTO(out, result);
1073
1074 trigger:
1075         /* We still have chance to get the valid inode: for the
1076          * object which is referenced by remote name entry, the
1077          * object on the local MDT will be linked under the dir
1078          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1079          *
1080          * We do not know whether the object for the given FID
1081          * is referenced by some remote name entry or not, and
1082          * especially for DNE II, a multiple-linked object may
1083          * have many name entries reside on many MDTs.
1084          *
1085          * To simplify the operation, OSD will not distinguish
1086          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1087          * only happened for the RPC from other MDT during the
1088          * OI scrub, or for the client side RPC with FID only,
1089          * such as FID to path, or from old connected client. */
1090         if (!remote && !fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1091                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1092                 if (!rc1) {
1093                         remote = true;
1094                         trusted = true;
1095                         flags |= SS_AUTO_PARTIAL;
1096                         flags &= ~SS_AUTO_FULL;
1097                         goto iget;
1098                 }
1099         }
1100
1101         if (thread_is_running(&scrub->os_thread)) {
1102                 if (scrub->os_partial_scan && !scrub->os_in_join)
1103                         goto join;
1104
1105                 if (IS_ERR_OR_NULL(inode) || result)
1106                         GOTO(out, result = -EINPROGRESS);
1107
1108                 LASSERT(remote);
1109                 LASSERT(obj->oo_inode == inode);
1110
1111                 osd_add_oi_cache(info, dev, id, fid);
1112                 osd_oii_insert(dev, oic, true);
1113                 goto found;
1114         }
1115
1116         if (dev->od_noscrub) {
1117                 if (!remote)
1118                         GOTO(out, result = -EREMCHG);
1119
1120                 LASSERT(!result);
1121                 LASSERT(obj->oo_inode == inode);
1122
1123                 osd_add_oi_cache(info, dev, id, fid);
1124                 goto found;
1125         }
1126
1127 join:
1128         rc1 = osd_scrub_start(dev, flags);
1129         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1130                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1131         if (rc1 && rc1 != -EALREADY)
1132                 GOTO(out, result = -EREMCHG);
1133
1134         if (IS_ERR_OR_NULL(inode) || result)
1135                 GOTO(out, result = -EINPROGRESS);
1136
1137         LASSERT(remote);
1138         LASSERT(obj->oo_inode == inode);
1139
1140         osd_add_oi_cache(info, dev, id, fid);
1141         osd_oii_insert(dev, oic, true);
1142         goto found;
1143
1144 check_lma:
1145         result = osd_check_lma(env, obj);
1146         if (!result)
1147                 goto found;
1148
1149         LASSERTF(id->oii_ino == inode->i_ino &&
1150                  id->oii_gen == inode->i_generation,
1151                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1152                  PFID(fid), id->oii_ino, id->oii_gen,
1153                  inode->i_ino, inode->i_generation);
1154
1155         saved_ino = inode->i_ino;
1156         saved_gen = inode->i_generation;
1157
1158         if (unlikely(result == -ENODATA)) {
1159                 /* If the OI scrub updated the OI mapping by race, it
1160                  * must be valid. Trust the inode that has no LMA EA. */
1161                 if (updated)
1162                         goto found;
1163
1164                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1165                 if (!result) {
1166                         /* The OI mapping is still there, the inode is still
1167                          * valid. It is just becaues the inode has no LMA EA. */
1168                         if (saved_ino == id->oii_ino &&
1169                             saved_gen == id->oii_gen)
1170                                 goto found;
1171
1172                         /* It is the OI scrub updated the OI mapping by race.
1173                          * The new OI mapping must be valid. */
1174                         trusted = true;
1175                         updated = true;
1176                         goto iget;
1177                 }
1178
1179                 /* "result == -ENOENT" means that the OI mappinghas been
1180                  * removed by race, so the inode belongs to other object.
1181                  *
1182                  * Others error can be returned  directly. */
1183                 if (result == -ENOENT) {
1184                         LASSERT(trusted);
1185
1186                         obj->oo_inode = NULL;
1187                         result = 0;
1188                 }
1189         }
1190
1191         if (result != -EREMCHG)
1192                 GOTO(out, result);
1193
1194         LASSERT(!updated);
1195
1196         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1197         /* "result == -ENOENT" means the cached OI mapping has been removed
1198          * from the OI file by race, above inode belongs to other object. */
1199         if (result == -ENOENT) {
1200                 LASSERT(trusted);
1201
1202                 obj->oo_inode = NULL;
1203                 GOTO(out, result = 0);
1204         }
1205
1206         if (result)
1207                 GOTO(out, result);
1208
1209         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1210                 result = -EREMCHG;
1211                 goto trigger;
1212         }
1213
1214         /* It is the OI scrub updated the OI mapping by race.
1215          * The new OI mapping must be valid. */
1216         trusted = true;
1217         updated = true;
1218         goto iget;
1219
1220 found:
1221         obj->oo_compat_dot_created = 1;
1222         obj->oo_compat_dotdot_created = 1;
1223
1224         if (S_ISDIR(inode->i_mode) &&
1225             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1226                 osd_check_lmv(info, dev, inode, oic);
1227
1228         result = osd_attach_jinode(inode);
1229         if (result)
1230                 GOTO(out, result);
1231
1232         if (!ldiskfs_pdo)
1233                 GOTO(out, result = 0);
1234
1235         LASSERT(!obj->oo_hl_head);
1236         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1237
1238         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1239
1240 out:
1241         if (result || !obj->oo_inode) {
1242                 if (!IS_ERR_OR_NULL(inode))
1243                         iput(inode);
1244
1245                 obj->oo_inode = NULL;
1246                 if (trusted)
1247                         fid_zero(&oic->oic_fid);
1248         }
1249
1250         LINVRNT(osd_invariant(obj));
1251         return result;
1252 }
1253
1254 /*
1255  * Concurrency: shouldn't matter.
1256  */
1257 static void osd_object_init0(struct osd_object *obj)
1258 {
1259         LASSERT(obj->oo_inode != NULL);
1260         obj->oo_dt.do_body_ops = &osd_body_ops;
1261         obj->oo_dt.do_lu.lo_header->loh_attr |=
1262                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1263 }
1264
1265 /*
1266  * Concurrency: no concurrent access is possible that early in object
1267  * life-cycle.
1268  */
1269 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1270                            const struct lu_object_conf *conf)
1271 {
1272         struct osd_object *obj = osd_obj(l);
1273         int result;
1274
1275         LINVRNT(osd_invariant(obj));
1276
1277         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1278                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1279                 l->lo_header->loh_attr |= LOHA_EXISTS;
1280                 return 0;
1281         }
1282
1283         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1284         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1285         if (result == 0 && obj->oo_inode != NULL) {
1286                 struct osd_thread_info *oti = osd_oti_get(env);
1287                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1288
1289                 osd_object_init0(obj);
1290                 result = osd_get_lma(oti, obj->oo_inode,
1291                                      &oti->oti_obj_dentry, loa);
1292                 if (!result) {
1293                         /* Convert LMAI flags to lustre LMA flags
1294                          * and cache it to oo_lma_flags */
1295                         obj->oo_lma_flags =
1296                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1297                 } else if (result == -ENODATA) {
1298                         result = 0;
1299                 }
1300         }
1301
1302         LINVRNT(osd_invariant(obj));
1303         return result;
1304 }
1305
1306 /* The first part of oxe_buf is xattr name, and is '\0' terminated.
1307  * The left part is for value, binary mode. */
1308 struct osd_xattr_entry {
1309         struct list_head        oxe_list;
1310         size_t                  oxe_len;
1311         size_t                  oxe_namelen;
1312         bool                    oxe_exist;
1313         struct rcu_head         oxe_rcu;
1314         char                    oxe_buf[0];
1315 };
1316
1317 static int osd_oxc_get(struct osd_object *obj, const char *name,
1318                        struct lu_buf *buf)
1319 {
1320         struct osd_xattr_entry *tmp;
1321         struct osd_xattr_entry *oxe = NULL;
1322         size_t namelen = strlen(name);
1323         int rc;
1324         ENTRY;
1325
1326         rcu_read_lock();
1327         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1328                 if (namelen == tmp->oxe_namelen &&
1329                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1330                         oxe = tmp;
1331                         break;
1332                 }
1333         }
1334
1335         if (oxe == NULL)
1336                 GOTO(out, rc = -ENOENT);
1337
1338         if (!oxe->oxe_exist)
1339                 GOTO(out, rc = -ENODATA);
1340
1341         /* vallen */
1342         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1343         LASSERT(rc > 0);
1344
1345         if (buf->lb_buf == NULL)
1346                 GOTO(out, rc);
1347
1348         if (buf->lb_len < rc)
1349                 GOTO(out, rc = -ERANGE);
1350
1351         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1352         EXIT;
1353 out:
1354         rcu_read_unlock();
1355
1356         return rc;
1357 }
1358
1359 static void osd_oxc_free(struct rcu_head *head)
1360 {
1361         struct osd_xattr_entry *oxe;
1362
1363         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1364         OBD_FREE(oxe, oxe->oxe_len);
1365 }
1366
1367 static void osd_oxc_add(struct osd_object *obj, const char *name,
1368                         const char *buf, int buflen)
1369 {
1370         struct osd_xattr_entry *oxe;
1371         struct osd_xattr_entry *old = NULL;
1372         struct osd_xattr_entry *tmp;
1373         size_t namelen = strlen(name);
1374         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1375
1376         OBD_ALLOC(oxe, len);
1377         if (oxe == NULL)
1378                 return;
1379
1380         INIT_LIST_HEAD(&oxe->oxe_list);
1381         oxe->oxe_len = len;
1382         oxe->oxe_namelen = namelen;
1383         memcpy(oxe->oxe_buf, name, namelen);
1384         if (buflen > 0) {
1385                 LASSERT(buf != NULL);
1386                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1387                 oxe->oxe_exist = true;
1388         } else {
1389                 oxe->oxe_exist = false;
1390         }
1391
1392         /* this should be rarely called, just remove old and add new */
1393         spin_lock(&obj->oo_guard);
1394         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1395                 if (namelen == tmp->oxe_namelen &&
1396                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1397                         old = tmp;
1398                         break;
1399                 }
1400         }
1401         if (old != NULL) {
1402                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1403                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1404         } else {
1405                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1406         }
1407         spin_unlock(&obj->oo_guard);
1408 }
1409
1410 static void osd_oxc_del(struct osd_object *obj, const char *name)
1411 {
1412         struct osd_xattr_entry *oxe;
1413         size_t namelen = strlen(name);
1414
1415         spin_lock(&obj->oo_guard);
1416         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1417                 if (namelen == oxe->oxe_namelen &&
1418                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1419                         list_del_rcu(&oxe->oxe_list);
1420                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1421                         break;
1422                 }
1423         }
1424         spin_unlock(&obj->oo_guard);
1425 }
1426
1427 static void osd_oxc_fini(struct osd_object *obj)
1428 {
1429         struct osd_xattr_entry *oxe, *next;
1430
1431         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1432                 list_del(&oxe->oxe_list);
1433                 OBD_FREE(oxe, oxe->oxe_len);
1434         }
1435 }
1436
1437 /*
1438  * Concurrency: no concurrent access is possible that late in object
1439  * life-cycle.
1440  */
1441 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1442 {
1443         struct osd_object *obj = osd_obj(l);
1444
1445         LINVRNT(osd_invariant(obj));
1446
1447         osd_oxc_fini(obj);
1448         dt_object_fini(&obj->oo_dt);
1449         if (obj->oo_hl_head != NULL)
1450                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1451         OBD_FREE_PTR(obj);
1452 }
1453
1454 /*
1455  * Concurrency: no concurrent access is possible that late in object
1456  * life-cycle.
1457  */
1458 static void osd_index_fini(struct osd_object *o)
1459 {
1460         struct iam_container *bag;
1461
1462         if (o->oo_dir != NULL) {
1463                 bag = &o->oo_dir->od_container;
1464                 if (o->oo_inode != NULL) {
1465                         if (bag->ic_object == o->oo_inode)
1466                                 iam_container_fini(bag);
1467                 }
1468                 OBD_FREE_PTR(o->oo_dir);
1469                 o->oo_dir = NULL;
1470         }
1471 }
1472
1473 /*
1474  * Concurrency: no concurrent access is possible that late in object
1475  * life-cycle (for all existing callers, that is. New callers have to provide
1476  * their own locking.)
1477  */
1478 static int osd_inode_unlinked(const struct inode *inode)
1479 {
1480         return inode->i_nlink == 0;
1481 }
1482
1483 enum {
1484         OSD_TXN_OI_DELETE_CREDITS    = 20,
1485         OSD_TXN_INODE_DELETE_CREDITS = 20
1486 };
1487
1488 /*
1489  * Journal
1490  */
1491
1492 #if OSD_THANDLE_STATS
1493 /**
1494  * Set time when the handle is allocated
1495  */
1496 static void osd_th_alloced(struct osd_thandle *oth)
1497 {
1498         oth->oth_alloced = ktime_get();
1499 }
1500
1501 /**
1502  * Set time when the handle started
1503  */
1504 static void osd_th_started(struct osd_thandle *oth)
1505 {
1506         oth->oth_started = ktime_get();
1507 }
1508
1509 /**
1510  * Check whether the we deal with this handle for too long.
1511  */
1512 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1513                                 ktime_t alloced, ktime_t started,
1514                                 ktime_t closed)
1515 {
1516         ktime_t now = ktime_get();
1517
1518         LASSERT(dev != NULL);
1519
1520         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1521                             ktime_us_delta(started, alloced));
1522         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1523                             ktime_us_delta(closed, started));
1524         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1525                             ktime_us_delta(now, closed));
1526
1527         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1528                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1529                       oth, now, alloced, started, closed);
1530                 libcfs_debug_dumpstack(NULL);
1531         }
1532 }
1533
1534 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1535 {                                                                       \
1536         ktime_t __closed = ktime_get();                                 \
1537         ktime_t __alloced = oth->oth_alloced;                           \
1538         ktime_t __started = oth->oth_started;                           \
1539                                                                         \
1540         expr;                                                           \
1541         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1542 }
1543
1544 #else /* OSD_THANDLE_STATS */
1545
1546 #define osd_th_alloced(h)                  do {} while(0)
1547 #define osd_th_started(h)                  do {} while(0)
1548 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1549
1550 #endif /* OSD_THANDLE_STATS */
1551
1552 /*
1553  * Concurrency: doesn't access mutable data.
1554  */
1555 static int osd_param_is_not_sane(const struct osd_device *dev,
1556                                  const struct thandle *th)
1557 {
1558         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1559
1560         return oh->ot_credits > osd_transaction_size(dev);
1561 }
1562
1563 /*
1564  * Concurrency: shouldn't matter.
1565  */
1566 static void osd_trans_commit_cb(struct super_block *sb,
1567                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1568 {
1569         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
1570         struct thandle     *th  = &oh->ot_super;
1571         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
1572         struct dt_txn_commit_cb *dcb, *tmp;
1573
1574         LASSERT(oh->ot_handle == NULL);
1575
1576         if (error)
1577                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1578
1579         dt_txn_hook_commit(th);
1580
1581         /* call per-transaction callbacks if any */
1582         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1583                                  dcb_linkage) {
1584                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1585                          "commit callback entry: magic=%x name='%s'\n",
1586                          dcb->dcb_magic, dcb->dcb_name);
1587                 list_del_init(&dcb->dcb_linkage);
1588                 dcb->dcb_func(NULL, th, dcb, error);
1589         }
1590
1591         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1592         lu_device_put(lud);
1593         th->th_dev = NULL;
1594
1595         lu_context_exit(&th->th_ctx);
1596         lu_context_fini(&th->th_ctx);
1597         OBD_FREE_PTR(oh);
1598 }
1599
1600 #ifndef HAVE_SB_START_WRITE
1601 # define sb_start_write(sb) do {} while (0)
1602 # define sb_end_write(sb) do {} while (0)
1603 #endif
1604
1605 static struct thandle *osd_trans_create(const struct lu_env *env,
1606                                         struct dt_device *d)
1607 {
1608         struct osd_thread_info  *oti = osd_oti_get(env);
1609         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
1610         struct osd_thandle      *oh;
1611         struct thandle          *th;
1612         ENTRY;
1613
1614         if (d->dd_rdonly) {
1615                 CERROR("%s: someone try to start transaction under "
1616                        "readonly mode, should be disabled.\n",
1617                        osd_name(osd_dt_dev(d)));
1618                 dump_stack();
1619                 RETURN(ERR_PTR(-EROFS));
1620         }
1621
1622         /* on pending IO in this thread should left from prev. request */
1623         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1624
1625         sb_start_write(osd_sb(osd_dt_dev(d)));
1626
1627         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
1628         if (oh != NULL) {
1629                 oh->ot_quota_trans = &oti->oti_quota_trans;
1630                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1631                 th = &oh->ot_super;
1632                 th->th_dev = d;
1633                 th->th_result = 0;
1634                 th->th_tags = LCT_TX_HANDLE;
1635                 oh->ot_credits = 0;
1636                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1637                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1638                 osd_th_alloced(oh);
1639
1640                 memset(oti->oti_declare_ops, 0,
1641                        sizeof(oti->oti_declare_ops));
1642                 memset(oti->oti_declare_ops_cred, 0,
1643                        sizeof(oti->oti_declare_ops_cred));
1644                 memset(oti->oti_declare_ops_used, 0,
1645                        sizeof(oti->oti_declare_ops_used));
1646         } else {
1647                 sb_end_write(osd_sb(osd_dt_dev(d)));
1648                 th = ERR_PTR(-ENOMEM);
1649         }
1650         RETURN(th);
1651 }
1652
1653 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1654 {
1655         struct osd_thread_info  *oti = osd_oti_get(env);
1656         struct osd_thandle      *oh;
1657
1658         oh = container_of0(th, struct osd_thandle, ot_super);
1659         LASSERT(oh != NULL);
1660
1661         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1662               oti->oti_declare_ops[OSD_OT_CREATE],
1663               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1664               oti->oti_declare_ops_used[OSD_OT_CREATE],
1665               oti->oti_declare_ops[OSD_OT_DESTROY],
1666               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1667               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1668         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1669               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1670               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1671               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1672               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1673               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1674               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1675         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1676               oti->oti_declare_ops[OSD_OT_WRITE],
1677               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1678               oti->oti_declare_ops_used[OSD_OT_WRITE],
1679               oti->oti_declare_ops[OSD_OT_PUNCH],
1680               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1681               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1682               oti->oti_declare_ops[OSD_OT_QUOTA],
1683               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1684               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1685         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1686               oti->oti_declare_ops[OSD_OT_INSERT],
1687               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1688               oti->oti_declare_ops_used[OSD_OT_INSERT],
1689               oti->oti_declare_ops[OSD_OT_DELETE],
1690               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1691               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1692         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1693               oti->oti_declare_ops[OSD_OT_REF_ADD],
1694               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1695               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1696               oti->oti_declare_ops[OSD_OT_REF_DEL],
1697               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1698               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1699 }
1700
1701 /*
1702  * Concurrency: shouldn't matter.
1703  */
1704 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1705                            struct thandle *th)
1706 {
1707         struct osd_thread_info *oti = osd_oti_get(env);
1708         struct osd_device  *dev = osd_dt_dev(d);
1709         handle_t           *jh;
1710         struct osd_thandle *oh;
1711         int rc;
1712
1713         ENTRY;
1714
1715         LASSERT(current->journal_info == NULL);
1716
1717         oh = container_of0(th, struct osd_thandle, ot_super);
1718         LASSERT(oh != NULL);
1719         LASSERT(oh->ot_handle == NULL);
1720
1721         rc = dt_txn_hook_start(env, d, th);
1722         if (rc != 0)
1723                 GOTO(out, rc);
1724
1725         if (unlikely(osd_param_is_not_sane(dev, th))) {
1726                 static unsigned long last_printed;
1727                 static int last_credits;
1728
1729                 /* don't make noise on a tiny testing systems
1730                  * actual credits misuse will be caught anyway */
1731                 if (last_credits != oh->ot_credits &&
1732                     time_after(jiffies, last_printed +
1733                                msecs_to_jiffies(60 * MSEC_PER_SEC)) &&
1734                     osd_transaction_size(dev) > 512) {
1735                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1736                               oh->ot_credits, osd_transaction_size(dev));
1737                         osd_trans_dump_creds(env, th);
1738                         libcfs_debug_dumpstack(NULL);
1739                         last_credits = oh->ot_credits;
1740                         last_printed = jiffies;
1741                 }
1742                 /* XXX Limit the credits to 'max_transaction_buffers', and
1743                  *     let the underlying filesystem to catch the error if
1744                  *     we really need so many credits.
1745                  *
1746                  *     This should be removed when we can calculate the
1747                  *     credits precisely. */
1748                 oh->ot_credits = osd_transaction_size(dev);
1749         } else if (ldiskfs_track_declares_assert != 0) {
1750                 /* reserve few credits to prevent an assertion in JBD
1751                  * our debugging mechanism will be able to detected
1752                  * overuse. this can help to debug single-update
1753                  * transactions */
1754                 oh->ot_credits += 10;
1755                 if (unlikely(osd_param_is_not_sane(dev, th)))
1756                         oh->ot_credits = osd_transaction_size(dev);
1757         }
1758
1759         /*
1760          * XXX temporary stuff. Some abstraction layer should
1761          * be used.
1762          */
1763         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1764         osd_th_started(oh);
1765         if (!IS_ERR(jh)) {
1766                 oh->ot_handle = jh;
1767                 LASSERT(oti->oti_txns == 0);
1768                 lu_context_init(&th->th_ctx, th->th_tags);
1769                 lu_context_enter(&th->th_ctx);
1770
1771                 lu_device_get(&d->dd_lu_dev);
1772                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1773                               "osd-tx", th);
1774                 oti->oti_txns++;
1775                 rc = 0;
1776         } else {
1777                 rc = PTR_ERR(jh);
1778         }
1779 out:
1780         RETURN(rc);
1781 }
1782
1783 static int osd_seq_exists(const struct lu_env *env,
1784                           struct osd_device *osd, u64 seq)
1785 {
1786         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1787         struct seq_server_site  *ss = osd_seq_site(osd);
1788         int                     rc;
1789         ENTRY;
1790
1791         LASSERT(ss != NULL);
1792         LASSERT(ss->ss_server_fld != NULL);
1793
1794         rc = osd_fld_lookup(env, osd, seq, range);
1795         if (rc != 0) {
1796                 if (rc != -ENOENT)
1797                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
1798                                osd_name(osd), seq, rc);
1799                 RETURN(0);
1800         }
1801
1802         RETURN(ss->ss_node_id == range->lsr_index);
1803 }
1804
1805 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1806 {
1807         struct dt_txn_commit_cb *dcb;
1808         struct dt_txn_commit_cb *tmp;
1809
1810         /* call per-transaction stop callbacks if any */
1811         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1812                                  dcb_linkage) {
1813                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1814                          "commit callback entry: magic=%x name='%s'\n",
1815                          dcb->dcb_magic, dcb->dcb_name);
1816                 list_del_init(&dcb->dcb_linkage);
1817                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1818         }
1819 }
1820
1821 /*
1822  * Concurrency: shouldn't matter.
1823  */
1824 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1825                           struct thandle *th)
1826 {
1827         int                     rc = 0, remove_agents = 0;
1828         struct osd_thandle     *oh;
1829         struct osd_thread_info *oti = osd_oti_get(env);
1830         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1831         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1832         struct qsd_instance    *qsd = osd->od_quota_slave;
1833         struct lquota_trans    *qtrans;
1834         ENTRY;
1835
1836         oh = container_of0(th, struct osd_thandle, ot_super);
1837
1838         /* reset OI cache for safety */
1839         oti->oti_ins_cache_used = 0;
1840
1841         remove_agents = oh->ot_remove_agents;
1842
1843         qtrans = oh->ot_quota_trans;
1844         oh->ot_quota_trans = NULL;
1845
1846         if (oh->ot_handle != NULL) {
1847                 int rc2;
1848                 handle_t *hdl = oh->ot_handle;
1849
1850                 /*
1851                  * add commit callback
1852                  * notice we don't do this in osd_trans_start()
1853                  * as underlying transaction can change during truncate
1854                  */
1855                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1856                                          &oh->ot_jcb);
1857
1858                 LASSERT(oti->oti_txns == 1);
1859                 oti->oti_txns--;
1860
1861                 rc = dt_txn_hook_stop(env, th);
1862                 if (rc != 0)
1863                         CERROR("%s: failed in transaction hook: rc = %d\n",
1864                                osd_name(osd), rc);
1865
1866                 osd_trans_stop_cb(oh, rc);
1867                 /* hook functions might modify th_sync */
1868                 hdl->h_sync = th->th_sync;
1869
1870                 oh->ot_handle = NULL;
1871                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
1872                 if (rc2 != 0)
1873                         CERROR("%s: failed to stop transaction: rc = %d\n",
1874                                osd_name(osd), rc2);
1875                 if (!rc)
1876                         rc = rc2;
1877         } else {
1878                 osd_trans_stop_cb(oh, th->th_result);
1879                 OBD_FREE_PTR(oh);
1880         }
1881
1882         /* inform the quota slave device that the transaction is stopping */
1883         qsd_op_end(env, qsd, qtrans);
1884
1885         /* as we want IO to journal and data IO be concurrent, we don't block
1886          * awaiting data IO completion in osd_do_bio(), instead we wait here
1887          * once transaction is submitted to the journal. all reqular requests
1888          * don't do direct IO (except read/write), thus this wait_event becomes
1889          * no-op for them.
1890          *
1891          * IMPORTANT: we have to wait till any IO submited by the thread is
1892          * completed otherwise iobuf may be corrupted by different request
1893          */
1894         wait_event(iobuf->dr_wait,
1895                        atomic_read(&iobuf->dr_numreqs) == 0);
1896         osd_fini_iobuf(osd, iobuf);
1897         if (!rc)
1898                 rc = iobuf->dr_error;
1899
1900         if (unlikely(remove_agents != 0))
1901                 osd_process_scheduled_agent_removals(env, osd);
1902
1903         sb_end_write(osd_sb(osd));
1904
1905         RETURN(rc);
1906 }
1907
1908 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1909 {
1910         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1911                                                ot_super);
1912
1913         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1914         LASSERT(&dcb->dcb_func != NULL);
1915         if (dcb->dcb_flags & DCB_TRANS_STOP)
1916                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1917         else
1918                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1919
1920         return 0;
1921 }
1922
1923 /*
1924  * Called just before object is freed. Releases all resources except for
1925  * object itself (that is released by osd_object_free()).
1926  *
1927  * Concurrency: no concurrent access is possible that late in object
1928  * life-cycle.
1929  */
1930 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1931 {
1932         struct osd_object *obj   = osd_obj(l);
1933         struct inode      *inode = obj->oo_inode;
1934
1935         LINVRNT(osd_invariant(obj));
1936
1937         /*
1938          * If object is unlinked remove fid->ino mapping from object index.
1939          */
1940
1941         osd_index_fini(obj);
1942         if (inode != NULL) {
1943                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1944                 qid_t                    uid = i_uid_read(inode);
1945                 qid_t                    gid = i_gid_read(inode);
1946
1947                 iput(inode);
1948                 obj->oo_inode = NULL;
1949
1950                 if (qsd != NULL) {
1951                         struct osd_thread_info  *info = osd_oti_get(env);
1952                         struct lquota_id_info   *qi = &info->oti_qi;
1953
1954                         /* Release granted quota to master if necessary */
1955                         qi->lqi_id.qid_uid = uid;
1956                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1957
1958                         qi->lqi_id.qid_uid = gid;
1959                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1960
1961                         qi->lqi_id.qid_uid = i_projid_read(inode);
1962                         qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
1963                 }
1964         }
1965 }
1966
1967 /*
1968  * Concurrency: ->loo_object_release() is called under site spin-lock.
1969  */
1970 static void osd_object_release(const struct lu_env *env,
1971                                struct lu_object *l)
1972 {
1973         struct osd_object *o = osd_obj(l);
1974         /* nobody should be releasing a non-destroyed object with nlink=0
1975          * the API allows this, but ldiskfs doesn't like and then report
1976          * this inode as deleted */
1977         if (unlikely(!o->oo_destroyed && o->oo_inode && o->oo_inode->i_nlink == 0))
1978                 LBUG();
1979 }
1980
1981 /*
1982  * Concurrency: shouldn't matter.
1983  */
1984 static int osd_object_print(const struct lu_env *env, void *cookie,
1985                             lu_printer_t p, const struct lu_object *l)
1986 {
1987         struct osd_object *o = osd_obj(l);
1988         struct iam_descr  *d;
1989
1990         if (o->oo_dir != NULL)
1991                 d = o->oo_dir->od_container.ic_descr;
1992         else
1993                 d = NULL;
1994         return (*p)(env, cookie,
1995                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1996                     o, o->oo_inode,
1997                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1998                     o->oo_inode ? o->oo_inode->i_generation : 0,
1999                     d ? d->id_ops->id_name : "plain");
2000 }
2001
2002 /*
2003  * Concurrency: shouldn't matter.
2004  */
2005 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2006                struct obd_statfs *sfs)
2007 {
2008         struct osd_device       *osd = osd_dt_dev(d);
2009         struct super_block      *sb = osd_sb(osd);
2010         struct kstatfs          *ksfs;
2011         __u64                    reserved;
2012         int                      result = 0;
2013
2014         if (unlikely(osd->od_mnt == NULL))
2015                 return -EINPROGRESS;
2016
2017         /* osd_lproc.c call this without env, allocate ksfs for that case */
2018         if (unlikely(env == NULL)) {
2019                 OBD_ALLOC_PTR(ksfs);
2020                 if (ksfs == NULL)
2021                         return -ENOMEM;
2022         } else {
2023                 ksfs = &osd_oti_get(env)->oti_ksfs;
2024         }
2025
2026         result = sb->s_op->statfs(sb->s_root, ksfs);
2027         if (result)
2028                 goto out;
2029
2030         statfs_pack(sfs, ksfs);
2031         if (unlikely(sb->s_flags & MS_RDONLY))
2032                 sfs->os_state |= OS_STATE_READONLY;
2033         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb,
2034                                          LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2035                 sfs->os_maxbytes = sb->s_maxbytes;
2036         else
2037                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2038
2039         /*
2040          * Reserve some space so to avoid fragmenting the filesystem too much.
2041          * Fragmentation not only impacts performance, but can also increase
2042          * metadata overhead significantly, causing grant calculation to be
2043          * wrong.
2044          *
2045          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2046          */
2047         CLASSERT(OSD_STATFS_RESERVED > LDISKFS_MAX_BLOCK_SIZE);
2048         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2049         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2050                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2051
2052         sfs->os_blocks -= reserved;
2053         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2054         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2055
2056 out:
2057         if (unlikely(env == NULL))
2058                 OBD_FREE_PTR(ksfs);
2059         return result;
2060 }
2061
2062 /**
2063  * Estimate space needed for file creations. We assume the largest filename
2064  * which is 2^64 - 1, hence a filename of 20 chars.
2065  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2066  */
2067 #ifdef __LDISKFS_DIR_REC_LEN
2068 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2069 #else
2070 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2071 #endif
2072
2073 /*
2074  * Concurrency: doesn't access mutable data.
2075  */
2076 static void osd_conf_get(const struct lu_env *env,
2077                          const struct dt_device *dev,
2078                          struct dt_device_param *param)
2079 {
2080         struct super_block *sb = osd_sb(osd_dt_dev(dev));
2081         int                ea_overhead;
2082
2083         /*
2084          * XXX should be taken from not-yet-existing fs abstraction layer.
2085          */
2086         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2087         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2088         param->ddp_symlink_max  = sb->s_blocksize;
2089         param->ddp_mount_type     = LDD_MT_LDISKFS;
2090         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EXTENTS))
2091                 param->ddp_maxbytes = sb->s_maxbytes;
2092         else
2093                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2094         /* inode are statically allocated, so per-inode space consumption
2095          * is the space consumed by the directory entry */
2096         param->ddp_inodespace     = PER_OBJ_USAGE;
2097         /* EXT_INIT_MAX_LEN is the theoretical maximum extent size  (32k blocks
2098          * = 128MB) which is unlikely to be hit in real life. Report a smaller
2099          * maximum length to not under count the actual number of extents
2100          * needed for writing a file. */
2101         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 2;
2102         /* worst-case extent insertion metadata overhead */
2103         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2104         param->ddp_mntopts      = 0;
2105         if (test_opt(sb, XATTR_USER))
2106                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2107         if (test_opt(sb, POSIX_ACL))
2108                 param->ddp_mntopts |= MNTOPT_ACL;
2109
2110         /* LOD might calculate the max stripe count based on max_ea_size,
2111          * so we need take account in the overhead as well,
2112          * xattr_header + magic + xattr_entry_head */
2113         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2114                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2115
2116 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2117         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
2118                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2119                                                                 ea_overhead;
2120         else
2121 #endif
2122                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2123 }
2124
2125 /*
2126  * Concurrency: shouldn't matter.
2127  */
2128 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2129 {
2130         int rc;
2131
2132         CDEBUG(D_CACHE, "%s: syncing OSD\n", osd_dt_dev(d)->od_svname);
2133
2134         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
2135
2136         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2137                rc);
2138
2139         return rc;
2140 }
2141
2142 /**
2143  * Start commit for OSD device.
2144  *
2145  * An implementation of dt_commit_async method for OSD device.
2146  * Asychronously starts underlayng fs sync and thereby a transaction
2147  * commit.
2148  *
2149  * \param env environment
2150  * \param d dt device
2151  *
2152  * \see dt_device_operations
2153  */
2154 static int osd_commit_async(const struct lu_env *env,
2155                             struct dt_device *d)
2156 {
2157         struct super_block *s = osd_sb(osd_dt_dev(d));
2158         ENTRY;
2159
2160         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2161         RETURN(s->s_op->sync_fs(s, 0));
2162 }
2163
2164 /* Our own copy of the set readonly functions if present, or NU if not. */
2165 static int (*priv_dev_set_rdonly)(struct block_device *bdev);
2166 static int (*priv_dev_check_rdonly)(struct block_device *bdev);
2167 /* static int (*priv_dev_clear_rdonly)(struct block_device *bdev); */
2168
2169 /*
2170  * Concurrency: shouldn't matter.
2171  */
2172 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2173 {
2174         struct super_block *sb = osd_sb(osd_dt_dev(d));
2175         struct block_device *dev = sb->s_bdev;
2176         int rc = -EOPNOTSUPP;
2177         ENTRY;
2178
2179         if (priv_dev_set_rdonly) {
2180                 struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
2181
2182                 rc = 0;
2183                 CERROR("*** setting %s read-only ***\n",
2184                        osd_dt_dev(d)->od_svname);
2185
2186                 if (sb->s_op->freeze_fs) {
2187                         rc = sb->s_op->freeze_fs(sb);
2188                         if (rc)
2189                                 goto out;
2190                 }
2191
2192                 if (jdev && (jdev != dev)) {
2193                         CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
2194                                (long)jdev);
2195                         priv_dev_set_rdonly(jdev);
2196                 }
2197                 CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
2198                 priv_dev_set_rdonly(dev);
2199
2200                 if (sb->s_op->unfreeze_fs)
2201                         sb->s_op->unfreeze_fs(sb);
2202         }
2203
2204 out:
2205         if (rc)
2206                 CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2207                        osd_dt_dev(d)->od_svname, (long)dev, rc);
2208
2209         RETURN(rc);
2210 }
2211
2212 /**
2213  * Note: we do not count into QUOTA here.
2214  * If we mount with --data_journal we may need more.
2215  */
2216 const int osd_dto_credits_noquota[DTO_NR] = {
2217         /**
2218          * Insert.
2219          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2220          * SINGLEDATA_TRANS_BLOCKS(8)
2221          * XXX Note: maybe iam need more, since iam have more level than
2222          *           EXT3 htree.
2223          */
2224         [DTO_INDEX_INSERT]  = 16,
2225         /**
2226          * Delete
2227          * just modify a single entry, probably merge few within a block
2228          */
2229         [DTO_INDEX_DELETE]  = 1,
2230         /**
2231          * Used for OI scrub
2232          */
2233         [DTO_INDEX_UPDATE]  = 16,
2234         /**
2235          * 4(inode, inode bits, groups, GDT)
2236          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2237          */
2238         [DTO_OBJECT_CREATE] = 4,
2239         /**
2240          * 4(inode, inode bits, groups, GDT)
2241          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2242          */
2243         [DTO_OBJECT_DELETE] = 4,
2244         /**
2245          * Attr set credits (inode)
2246          */
2247         [DTO_ATTR_SET_BASE] = 1,
2248         /**
2249          * Xattr set. The same as xattr of EXT3.
2250          * DATA_TRANS_BLOCKS(14)
2251          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2252          * are also counted in. Do not know why?
2253          */
2254         [DTO_XATTR_SET]     = 14,
2255         /**
2256          * credits for inode change during write.
2257          */
2258         [DTO_WRITE_BASE]    = 3,
2259         /**
2260          * credits for single block write.
2261          */
2262         [DTO_WRITE_BLOCK]   = 14,
2263         /**
2264          * Attr set credits for chown.
2265          * This is extra credits for setattr, and it is null without quota
2266          */
2267         [DTO_ATTR_SET_CHOWN] = 0
2268 };
2269
2270 static const struct dt_device_operations osd_dt_ops = {
2271         .dt_root_get       = osd_root_get,
2272         .dt_statfs         = osd_statfs,
2273         .dt_trans_create   = osd_trans_create,
2274         .dt_trans_start    = osd_trans_start,
2275         .dt_trans_stop     = osd_trans_stop,
2276         .dt_trans_cb_add   = osd_trans_cb_add,
2277         .dt_conf_get       = osd_conf_get,
2278         .dt_sync           = osd_sync,
2279         .dt_ro             = osd_ro,
2280         .dt_commit_async   = osd_commit_async,
2281 };
2282
2283 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2284                           unsigned role)
2285 {
2286         struct osd_object *obj = osd_dt_obj(dt);
2287         struct osd_thread_info *oti = osd_oti_get(env);
2288
2289         LINVRNT(osd_invariant(obj));
2290
2291         LASSERT(obj->oo_owner != env);
2292         down_read_nested(&obj->oo_sem, role);
2293
2294         LASSERT(obj->oo_owner == NULL);
2295         oti->oti_r_locks++;
2296 }
2297
2298 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2299                            unsigned role)
2300 {
2301         struct osd_object *obj = osd_dt_obj(dt);
2302         struct osd_thread_info *oti = osd_oti_get(env);
2303
2304         LINVRNT(osd_invariant(obj));
2305
2306         LASSERT(obj->oo_owner != env);
2307         down_write_nested(&obj->oo_sem, role);
2308
2309         LASSERT(obj->oo_owner == NULL);
2310         obj->oo_owner = env;
2311         oti->oti_w_locks++;
2312 }
2313
2314 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2315 {
2316         struct osd_object *obj = osd_dt_obj(dt);
2317         struct osd_thread_info *oti = osd_oti_get(env);
2318
2319         LINVRNT(osd_invariant(obj));
2320
2321         LASSERT(oti->oti_r_locks > 0);
2322         oti->oti_r_locks--;
2323         up_read(&obj->oo_sem);
2324 }
2325
2326 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2327 {
2328         struct osd_object *obj = osd_dt_obj(dt);
2329         struct osd_thread_info *oti = osd_oti_get(env);
2330
2331         LINVRNT(osd_invariant(obj));
2332
2333         LASSERT(obj->oo_owner == env);
2334         LASSERT(oti->oti_w_locks > 0);
2335         oti->oti_w_locks--;
2336         obj->oo_owner = NULL;
2337         up_write(&obj->oo_sem);
2338 }
2339
2340 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2341 {
2342         struct osd_object *obj = osd_dt_obj(dt);
2343
2344         LINVRNT(osd_invariant(obj));
2345
2346         return obj->oo_owner == env;
2347 }
2348
2349 static struct timespec *osd_inode_time(const struct lu_env *env,
2350                                        struct inode *inode, __u64 seconds)
2351 {
2352         struct osd_thread_info  *oti = osd_oti_get(env);
2353         struct timespec         *t   = &oti->oti_time;
2354
2355         t->tv_sec = seconds;
2356         t->tv_nsec = 0;
2357         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
2358         return t;
2359 }
2360
2361 static void osd_inode_getattr(const struct lu_env *env,
2362                               struct inode *inode, struct lu_attr *attr)
2363 {
2364         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2365                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2366                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2367                            LA_BLKSIZE | LA_TYPE;
2368
2369         attr->la_atime   = LTIME_S(inode->i_atime);
2370         attr->la_mtime   = LTIME_S(inode->i_mtime);
2371         attr->la_ctime   = LTIME_S(inode->i_ctime);
2372         attr->la_mode    = inode->i_mode;
2373         attr->la_size    = i_size_read(inode);
2374         attr->la_blocks  = inode->i_blocks;
2375         attr->la_uid     = i_uid_read(inode);
2376         attr->la_gid     = i_gid_read(inode);
2377         attr->la_projid  = i_projid_read(inode);
2378         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2379         attr->la_nlink   = inode->i_nlink;
2380         attr->la_rdev    = inode->i_rdev;
2381         attr->la_blksize = 1 << inode->i_blkbits;
2382         attr->la_blkbits = inode->i_blkbits;
2383         /*
2384          * Ext4 did not transfer inherit flags from raw inode
2385          * to inode flags, and ext4 internally test raw inode
2386          * @i_flags directly. Instead of patching ext4, we do it here.
2387          */
2388         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2389                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2390 }
2391
2392 static int osd_attr_get(const struct lu_env *env,
2393                         struct dt_object *dt,
2394                         struct lu_attr *attr)
2395 {
2396         struct osd_object *obj = osd_dt_obj(dt);
2397
2398         if (unlikely(!dt_object_exists(dt)))
2399                 return -ENOENT;
2400         if (unlikely(obj->oo_destroyed))
2401                 return -ENOENT;
2402
2403         LASSERT(!dt_object_remote(dt));
2404         LINVRNT(osd_invariant(obj));
2405
2406         spin_lock(&obj->oo_guard);
2407         osd_inode_getattr(env, obj->oo_inode, attr);
2408         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL)
2409                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2410         spin_unlock(&obj->oo_guard);
2411
2412         return 0;
2413 }
2414
2415 static int osd_declare_attr_qid(const struct lu_env *env,
2416                                 struct osd_object *obj,
2417                                 struct osd_thandle *oh, long long bspace,
2418                                 qid_t old_id, qid_t new_id, bool enforce,
2419                                 unsigned type)
2420 {
2421         int rc;
2422         struct osd_thread_info *info = osd_oti_get(env);
2423         struct lquota_id_info  *qi = &info->oti_qi;
2424
2425         qi->lqi_type = type;
2426         /* inode accounting */
2427         qi->lqi_is_blk = false;
2428
2429         /* one more inode for the new id ... */
2430         qi->lqi_id.qid_uid = new_id;
2431         qi->lqi_space      = 1;
2432         /* Reserve credits for the new id */
2433         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2434         if (rc == -EDQUOT || rc == -EINPROGRESS)
2435                 rc = 0;
2436         if (rc)
2437                 RETURN(rc);
2438
2439         /* and one less inode for the current id */
2440         qi->lqi_id.qid_uid = old_id;
2441         qi->lqi_space      = -1;
2442         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2443         if (rc == -EDQUOT || rc == -EINPROGRESS)
2444                 rc = 0;
2445         if (rc)
2446                 RETURN(rc);
2447
2448         /* block accounting */
2449         qi->lqi_is_blk = true;
2450
2451         /* more blocks for the new id ... */
2452         qi->lqi_id.qid_uid = new_id;
2453         qi->lqi_space      = bspace;
2454         /*
2455          * Credits for the new uid has been reserved, re-use "obj"
2456          * to save credit reservation.
2457          */
2458         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2459         if (rc == -EDQUOT || rc == -EINPROGRESS)
2460                 rc = 0;
2461         if (rc)
2462                 RETURN(rc);
2463
2464         /* and finally less blocks for the current uid */
2465         qi->lqi_id.qid_uid = old_id;
2466         qi->lqi_space      = -bspace;
2467         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2468         if (rc == -EDQUOT || rc == -EINPROGRESS)
2469                 rc = 0;
2470
2471         RETURN(rc);
2472 }
2473
2474 static int osd_declare_attr_set(const struct lu_env *env,
2475                                 struct dt_object *dt,
2476                                 const struct lu_attr *attr,
2477                                 struct thandle *handle)
2478 {
2479         struct osd_thandle     *oh;
2480         struct osd_object      *obj;
2481         qid_t                   uid;
2482         qid_t                   gid;
2483         long long               bspace;
2484         int                     rc = 0;
2485         bool                    enforce;
2486         ENTRY;
2487
2488         LASSERT(dt != NULL);
2489         LASSERT(handle != NULL);
2490
2491         obj = osd_dt_obj(dt);
2492         LASSERT(osd_invariant(obj));
2493
2494         oh = container_of0(handle, struct osd_thandle, ot_super);
2495         LASSERT(oh->ot_handle == NULL);
2496
2497         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2498                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2499
2500         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2501                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2502
2503         if (attr == NULL || obj->oo_inode == NULL)
2504                 RETURN(rc);
2505
2506         bspace   = obj->oo_inode->i_blocks;
2507         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
2508         bspace   = toqb(bspace);
2509
2510         /* Changing ownership is always preformed by super user, it should not
2511          * fail with EDQUOT.
2512          *
2513          * We still need to call the osd_declare_qid() to calculate the journal
2514          * credits for updating quota accounting files and to trigger quota
2515          * space adjustment once the operation is completed.*/
2516         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2517                 /* USERQUOTA */
2518                 uid = i_uid_read(obj->oo_inode);
2519                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2520                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2521                                           attr->la_uid, enforce, USRQUOTA);
2522                 if (rc)
2523                         RETURN(rc);
2524
2525                 gid = i_gid_read(obj->oo_inode);
2526                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2527                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2528                                           i_gid_read(obj->oo_inode),
2529                                           attr->la_gid, enforce, GRPQUOTA);
2530                 if (rc)
2531                         RETURN(rc);
2532
2533         }
2534 #ifdef HAVE_PROJECT_QUOTA
2535         if (attr->la_valid & LA_PROJID) {
2536                 __u32 projid = i_projid_read(obj->oo_inode);
2537                 enforce = (attr->la_valid & LA_PROJID) &&
2538                                         (attr->la_projid != projid);
2539                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2540                                           (qid_t)projid, (qid_t)attr->la_projid,
2541                                           enforce, PRJQUOTA);
2542                 if (rc)
2543                         RETURN(rc);
2544         }
2545 #endif
2546         RETURN(rc);
2547 }
2548
2549 static int osd_inode_setattr(const struct lu_env *env,
2550                              struct inode *inode, const struct lu_attr *attr)
2551 {
2552         __u64 bits = attr->la_valid;
2553
2554         /* Only allow set size for regular file */
2555         if (!S_ISREG(inode->i_mode))
2556                 bits &= ~(LA_SIZE | LA_BLOCKS);
2557
2558         if (bits == 0)
2559                 return 0;
2560
2561         if (bits & LA_ATIME)
2562                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
2563         if (bits & LA_CTIME)
2564                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
2565         if (bits & LA_MTIME)
2566                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
2567         if (bits & LA_SIZE) {
2568                 spin_lock(&inode->i_lock);
2569                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2570                 i_size_write(inode, attr->la_size);
2571                 spin_unlock(&inode->i_lock);
2572         }
2573
2574         /* OSD should not change "i_blocks" which is used by quota.
2575          * "i_blocks" should be changed by ldiskfs only. */
2576         if (bits & LA_MODE)
2577                 inode->i_mode = (inode->i_mode & S_IFMT) |
2578                                 (attr->la_mode & ~S_IFMT);
2579         if (bits & LA_UID)
2580                 i_uid_write(inode, attr->la_uid);
2581         if (bits & LA_GID)
2582                 i_gid_write(inode, attr->la_gid);
2583         if (bits & LA_PROJID)
2584                 i_projid_write(inode, attr->la_projid);
2585         if (bits & LA_NLINK)
2586                 set_nlink(inode, attr->la_nlink);
2587         if (bits & LA_RDEV)
2588                 inode->i_rdev = attr->la_rdev;
2589
2590         if (bits & LA_FLAGS) {
2591                 /* always keep S_NOCMTIME */
2592                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2593                                  S_NOCMTIME;
2594                 /*
2595                  * Ext4 did not transfer inherit flags from
2596                  * @inode->i_flags to raw inode i_flags when writing
2597                  * flags, we do it explictly here.
2598                  */
2599                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2600                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2601         }
2602         return 0;
2603 }
2604
2605 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
2606 {
2607         int rc;
2608
2609         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
2610             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
2611                 struct iattr    iattr;
2612
2613                 ll_vfs_dq_init(inode);
2614                 iattr.ia_valid = 0;
2615                 if (attr->la_valid & LA_UID)
2616                         iattr.ia_valid |= ATTR_UID;
2617                 if (attr->la_valid & LA_GID)
2618                         iattr.ia_valid |= ATTR_GID;
2619                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
2620                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
2621
2622                 rc = ll_vfs_dq_transfer(inode, &iattr);
2623                 if (rc) {
2624                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2625                                "enforcement enabled on the ldiskfs "
2626                                "filesystem?\n", inode->i_sb->s_id, rc);
2627                         return rc;
2628                 }
2629         }
2630
2631 #ifdef HAVE_PROJECT_QUOTA
2632         /* Handle project id transfer here properly */
2633         if (attr->la_valid & LA_PROJID &&
2634             attr->la_projid != i_projid_read(inode)) {
2635                 rc = __ldiskfs_ioctl_setproject(inode, attr->la_projid);
2636                 if (rc) {
2637                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
2638                                "enforcement enabled on the ldiskfs "
2639                                "filesystem?\n", inode->i_sb->s_id, rc);
2640                         return rc;
2641                 }
2642         }
2643 #endif
2644         return 0;
2645 }
2646
2647 static int osd_attr_set(const struct lu_env *env,
2648                         struct dt_object *dt,
2649                         const struct lu_attr *attr,
2650                         struct thandle *handle)
2651 {
2652         struct osd_object *obj = osd_dt_obj(dt);
2653         struct inode      *inode;
2654         int rc;
2655
2656         if (!dt_object_exists(dt))
2657                 return -ENOENT;
2658
2659         LASSERT(handle != NULL);
2660         LASSERT(!dt_object_remote(dt));
2661         LASSERT(osd_invariant(obj));
2662
2663         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
2664
2665         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
2666                 struct osd_thread_info  *oti  = osd_oti_get(env);
2667                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
2668                 struct lu_fid           *fid1 = &oti->oti_fid;
2669                 struct osd_inode_id     *id   = &oti->oti_id;
2670                 struct iam_path_descr   *ipd;
2671                 struct iam_container    *bag;
2672                 struct osd_thandle      *oh;
2673                 int                      rc;
2674
2675                 fid_cpu_to_be(fid1, fid0);
2676                 memset(id, 1, sizeof(*id));
2677                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
2678                                   fid0)->oi_dir.od_container;
2679                 ipd = osd_idx_ipd_get(env, bag);
2680                 if (unlikely(ipd == NULL))
2681                         RETURN(-ENOMEM);
2682
2683                 oh = container_of0(handle, struct osd_thandle, ot_super);
2684                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
2685                                 (const struct iam_rec *)id, ipd);
2686                 osd_ipd_put(env, bag, ipd);
2687                 return(rc > 0 ? 0 : rc);
2688         }
2689
2690         inode = obj->oo_inode;
2691
2692         rc = osd_quota_transfer(inode, attr);
2693         if (rc)
2694                 return rc;
2695
2696         spin_lock(&obj->oo_guard);
2697         rc = osd_inode_setattr(env, inode, attr);
2698         spin_unlock(&obj->oo_guard);
2699         if (rc != 0)
2700                 GOTO(out, rc);
2701
2702         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2703
2704         if (!(attr->la_valid & LA_FLAGS))
2705                 GOTO(out, rc);
2706
2707         /* Let's check if there are extra flags need to be set into LMA */
2708         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
2709                 struct osd_thread_info *info = osd_oti_get(env);
2710                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
2711
2712                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
2713                                  &info->oti_ost_attrs);
2714                 if (rc)
2715                         GOTO(out, rc);
2716
2717                 lma->lma_incompat |=
2718                         lustre_to_lma_flags(attr->la_flags);
2719                 lustre_lma_swab(lma);
2720                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA,
2721                                      lma, sizeof(*lma), XATTR_REPLACE);
2722                 if (rc != 0) {
2723                         struct osd_device *osd = osd_obj2dev(obj);
2724
2725                         CWARN("%s: set "DFID" lma flags %u failed: rc = %d\n",
2726                               osd_name(osd), PFID(lu_object_fid(&dt->do_lu)),
2727                               lma->lma_incompat, rc);
2728                 } else {
2729                         obj->oo_lma_flags =
2730                                 attr->la_flags & LUSTRE_LMA_FL_MASKS;
2731                 }
2732                 osd_trans_exec_check(env, handle, OSD_OT_XATTR_SET);
2733         }
2734 out:
2735         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
2736
2737         return rc;
2738 }
2739
2740 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
2741                                            struct osd_object *obj,
2742                                            const char *name, const int namelen)
2743 {
2744         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
2745 }
2746
2747 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
2748                       umode_t mode, struct dt_allocation_hint *hint,
2749                       struct thandle *th)
2750 {
2751         int result;
2752         struct osd_device  *osd = osd_obj2dev(obj);
2753         struct osd_thandle *oth;
2754         struct dt_object   *parent = NULL;
2755         struct inode       *inode;
2756
2757         LINVRNT(osd_invariant(obj));
2758         LASSERT(obj->oo_inode == NULL);
2759         LASSERT(obj->oo_hl_head == NULL);
2760
2761         if (S_ISDIR(mode) && ldiskfs_pdo) {
2762                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
2763                 if (obj->oo_hl_head == NULL)
2764                         return -ENOMEM;
2765         }
2766
2767         oth = container_of(th, struct osd_thandle, ot_super);
2768         LASSERT(oth->ot_handle->h_transaction != NULL);
2769
2770         if (hint != NULL && hint->dah_parent != NULL &&
2771             !dt_object_remote(hint->dah_parent))
2772                 parent = hint->dah_parent;
2773
2774         inode = ldiskfs_create_inode(oth->ot_handle,
2775                                      parent ? osd_dt_obj(parent)->oo_inode :
2776                                               osd_sb(osd)->s_root->d_inode,
2777                                      mode);
2778         if (!IS_ERR(inode)) {
2779                 /* Do not update file c/mtime in ldiskfs. */
2780                 inode->i_flags |= S_NOCMTIME;
2781
2782                 /* For new created object, it must be consistent,
2783                  * and it is unnecessary to scrub against it. */
2784                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
2785
2786                 obj->oo_inode = inode;
2787                 result = 0;
2788         } else {
2789                 if (obj->oo_hl_head != NULL) {
2790                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
2791                         obj->oo_hl_head = NULL;
2792                 }
2793                 result = PTR_ERR(inode);
2794         }
2795         LINVRNT(osd_invariant(obj));
2796         return result;
2797 }
2798
2799 enum {
2800         OSD_NAME_LEN = 255
2801 };
2802
2803 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
2804                      struct lu_attr *attr,
2805                      struct dt_allocation_hint *hint,
2806                      struct dt_object_format *dof,
2807                      struct thandle *th)
2808 {
2809         int result;
2810         struct osd_thandle *oth;
2811         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX | S_ISGID));
2812
2813         LASSERT(S_ISDIR(attr->la_mode));
2814
2815         oth = container_of(th, struct osd_thandle, ot_super);
2816         LASSERT(oth->ot_handle->h_transaction != NULL);
2817         result = osd_mkfile(info, obj, mode, hint, th);
2818
2819         return result;
2820 }
2821
2822 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
2823                         struct lu_attr *attr,
2824                         struct dt_allocation_hint *hint,
2825                         struct dt_object_format *dof,
2826                         struct thandle *th)
2827 {
2828         int result;
2829         struct osd_thandle *oth;
2830         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
2831
2832         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
2833
2834         LASSERT(S_ISREG(attr->la_mode));
2835
2836         oth = container_of(th, struct osd_thandle, ot_super);
2837         LASSERT(oth->ot_handle->h_transaction != NULL);
2838
2839         result = osd_mkfile(info, obj, mode, hint, th);
2840         if (result == 0) {
2841                 LASSERT(obj->oo_inode != NULL);
2842                 if (feat->dif_flags & DT_IND_VARKEY)
2843                         result = iam_lvar_create(obj->oo_inode,
2844                                                  feat->dif_keysize_max,
2845                                                  feat->dif_ptrsize,
2846                                                  feat->dif_recsize_max,
2847                                                  oth->ot_handle);
2848                 else
2849                         result = iam_lfix_create(obj->oo_inode,
2850                                                  feat->dif_keysize_max,
2851                                                  feat->dif_ptrsize,
2852                                                  feat->dif_recsize_max,
2853                                                  oth->ot_handle);
2854
2855         }
2856         return result;
2857 }
2858
2859 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
2860                      struct lu_attr *attr,
2861                      struct dt_allocation_hint *hint,
2862                      struct dt_object_format *dof,
2863                      struct thandle *th)
2864 {
2865         LASSERT(S_ISREG(attr->la_mode));
2866         return osd_mkfile(info, obj, (attr->la_mode &
2867                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2868 }
2869
2870 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
2871                      struct lu_attr *attr,
2872                      struct dt_allocation_hint *hint,
2873                      struct dt_object_format *dof,
2874                      struct thandle *th)
2875 {
2876         LASSERT(S_ISLNK(attr->la_mode));
2877         return osd_mkfile(info, obj, (attr->la_mode &
2878                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
2879 }
2880
2881 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
2882                      struct lu_attr *attr,
2883                      struct dt_allocation_hint *hint,
2884                      struct dt_object_format *dof,
2885                      struct thandle *th)
2886 {
2887         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
2888         int result;
2889
2890         LINVRNT(osd_invariant(obj));
2891         LASSERT(obj->oo_inode == NULL);
2892         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
2893                 S_ISFIFO(mode) || S_ISSOCK(mode));
2894
2895         result = osd_mkfile(info, obj, mode, hint, th);
2896         if (result == 0) {
2897                 LASSERT(obj->oo_inode != NULL);
2898                 /*
2899                  * This inode should be marked dirty for i_rdev.  Currently
2900                  * that is done in the osd_attr_init().
2901                  */
2902                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2903                                    attr->la_rdev);
2904         }
2905         LINVRNT(osd_invariant(obj));
2906         return result;
2907 }
2908
2909 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2910                               struct lu_attr *,
2911                               struct dt_allocation_hint *hint,
2912                               struct dt_object_format *dof,
2913                               struct thandle *);
2914
2915 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2916 {
2917         osd_obj_type_f result;
2918
2919         switch (type) {
2920         case DFT_DIR:
2921                 result = osd_mkdir;
2922                 break;
2923         case DFT_REGULAR:
2924                 result = osd_mkreg;
2925                 break;
2926         case DFT_SYM:
2927                 result = osd_mksym;
2928                 break;
2929         case DFT_NODE:
2930                 result = osd_mknod;
2931                 break;
2932         case DFT_INDEX:
2933                 result = osd_mk_index;
2934                 break;
2935
2936         default:
2937                 LBUG();
2938                 break;
2939         }
2940         return result;
2941 }
2942
2943
2944 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2945                         struct dt_object *parent, struct dt_object *child,
2946                         umode_t child_mode)
2947 {
2948         LASSERT(ah);
2949
2950         ah->dah_parent = parent;
2951         ah->dah_mode = child_mode;
2952
2953         if (parent != NULL && !dt_object_remote(parent)) {
2954                 /* will help to find FID->ino at dt_insert("..") */
2955                 struct osd_object *pobj = osd_dt_obj(parent);
2956                 osd_idc_find_and_init(env, osd_obj2dev(pobj), pobj);
2957         }
2958 }
2959
2960 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2961                           struct lu_attr *attr, struct dt_object_format *dof)
2962 {
2963         struct inode   *inode = obj->oo_inode;
2964         __u64           valid = attr->la_valid;
2965         int             result;
2966
2967         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2968
2969         if (dof->dof_type != DFT_NODE)
2970                 attr->la_valid &= ~LA_RDEV;
2971         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2972                 attr->la_valid &= ~LA_ATIME;
2973         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2974                 attr->la_valid &= ~LA_CTIME;
2975         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2976                 attr->la_valid &= ~LA_MTIME;
2977
2978         result = osd_quota_transfer(inode, attr);
2979         if (result)
2980                 return;
2981
2982         if (attr->la_valid != 0) {
2983                 result = osd_inode_setattr(info->oti_env, inode, attr);
2984                 /*
2985                  * The osd_inode_setattr() should always succeed here.  The
2986                  * only error that could be returned is EDQUOT when we are
2987                  * trying to change the UID or GID of the inode. However, this
2988                  * should not happen since quota enforcement is no longer
2989                  * enabled on ldiskfs (lquota takes care of it).
2990                  */
2991                 LASSERTF(result == 0, "%d\n", result);
2992                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2993         }
2994
2995         attr->la_valid = valid;
2996 }
2997
2998 /**
2999  * Helper function for osd_create()
3000  *
3001  * \retval 0, on success
3002  */
3003 static int __osd_create(struct osd_thread_info *info, struct osd_object *obj,
3004                         struct lu_attr *attr, struct dt_allocation_hint *hint,
3005                         struct dt_object_format *dof, struct thandle *th)
3006 {
3007         int     result;
3008         __u32   umask;
3009
3010         osd_trans_exec_op(info->oti_env, th, OSD_OT_CREATE);
3011
3012         /* we drop umask so that permissions we pass are not affected */
3013         umask = current->fs->umask;
3014         current->fs->umask = 0;
3015
3016         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
3017                                                   th);
3018         if (likely(obj->oo_inode != NULL)) {
3019                 LASSERT(obj->oo_inode->i_state & I_NEW);
3020
3021                 /* Unlock the inode before attr initialization to avoid
3022                  * unnecessary dqget operations. LU-6378 */
3023                 unlock_new_inode(obj->oo_inode);
3024         }
3025
3026         if (likely(result == 0)) {
3027                 osd_attr_init(info, obj, attr, dof);
3028                 osd_object_init0(obj);
3029         }
3030
3031         /* restore previous umask value */
3032         current->fs->umask = umask;
3033
3034         osd_trans_exec_check(info->oti_env, th, OSD_OT_CREATE);
3035
3036         return result;
3037 }
3038
3039 /**
3040  * Helper function for osd_create()
3041  *
3042  * \retval 0, on success
3043  */
3044 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
3045                            const struct lu_fid *fid, struct thandle *th)
3046 {
3047         struct osd_thread_info *info = osd_oti_get(env);
3048         struct osd_inode_id    *id   = &info->oti_id;
3049         struct osd_device      *osd  = osd_obj2dev(obj);
3050         struct osd_thandle     *oh;
3051         int                     rc;
3052
3053         LASSERT(obj->oo_inode != NULL);
3054
3055         oh = container_of0(th, struct osd_thandle, ot_super);
3056         LASSERT(oh->ot_handle);
3057         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3058
3059         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
3060         rc = osd_oi_insert(info, osd, fid, id, oh->ot_handle,
3061                            OI_CHECK_FLD, NULL);
3062         osd_trans_exec_check(env, th, OSD_OT_INSERT);
3063
3064         return rc;
3065 }
3066
3067 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
3068                    u64 seq, struct lu_seq_range *range)
3069 {
3070         struct seq_server_site  *ss = osd_seq_site(osd);
3071
3072         if (fid_seq_is_idif(seq)) {
3073                 fld_range_set_ost(range);
3074                 range->lsr_index = idif_ost_idx(seq);
3075                 return 0;
3076         }
3077
3078         if (!fid_seq_in_fldb(seq)) {
3079                 fld_range_set_mdt(range);
3080                 if (ss != NULL)
3081                         /* FIXME: If ss is NULL, it suppose not get lsr_index
3082                          * at all */
3083                         range->lsr_index = ss->ss_node_id;
3084                 return 0;
3085         }
3086
3087         LASSERT(ss != NULL);
3088         fld_range_set_any(range);
3089         /* OSD will only do local fld lookup */
3090         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
3091 }
3092
3093 static int osd_declare_create(const struct lu_env *env, struct dt_object *dt,
3094                               struct lu_attr *attr,
3095                               struct dt_allocation_hint *hint,
3096                               struct dt_object_format *dof,
3097                               struct thandle *handle)
3098 {
3099         struct osd_thandle *oh;
3100         int rc;
3101         ENTRY;
3102
3103         LASSERT(handle != NULL);
3104
3105         oh = container_of0(handle, struct osd_thandle, ot_super);
3106         LASSERT(oh->ot_handle == NULL);
3107
3108         /* EA object consumes more credits than regular object: osd_mk_index
3109          * vs. osd_mkreg: osd_mk_index will create 2 blocks for root_node and
3110          * leaf_node, could involves the block, block bitmap, groups, GDT
3111          * change for each block, so add 4 * 2 credits in that case. */
3112         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
3113                              osd_dto_credits_noquota[DTO_OBJECT_CREATE] +
3114                              (dof->dof_type == DFT_INDEX) ? 4 * 2 : 0);
3115         /* Reuse idle OI block may cause additional one OI block
3116          * to be changed. */
3117         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3118                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
3119
3120         if (!attr)
3121                 RETURN(0);
3122
3123         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid,
3124                                    attr->la_projid, 1, oh, osd_dt_obj(dt),
3125                                    NULL, OSD_QID_INODE);
3126         if (rc != 0)
3127                 RETURN(rc);
3128
3129         /* will help to find FID->ino mapping at dt_insert() */
3130         rc = osd_idc_find_and_init(env, osd_obj2dev(osd_dt_obj(dt)),
3131                                    osd_dt_obj(dt));
3132
3133         RETURN(rc);
3134 }
3135
3136 /**
3137  * Called to destroy on-disk representation of the object
3138  *
3139  * Concurrency: must be locked
3140  */
3141 static int osd_declare_destroy(const struct lu_env *env, struct dt_object *dt,
3142                                struct thandle *th)
3143 {
3144         struct osd_object  *obj = osd_dt_obj(dt);
3145         struct inode       *inode = obj->oo_inode;
3146         struct osd_thandle *oh;
3147         int                 rc;
3148         ENTRY;
3149
3150         if (inode == NULL)
3151                 RETURN(-ENOENT);
3152
3153         oh = container_of0(th, struct osd_thandle, ot_super);
3154         LASSERT(oh->ot_handle == NULL);
3155
3156         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
3157                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
3158         /* Recycle idle OI leaf may cause additional three OI blocks
3159          * to be changed. */
3160         if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LOST_MDTOBJ2))
3161                 osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3162                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
3163         /* one less inode */
3164         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3165                                    i_projid_read(inode), -1, oh, obj, NULL,
3166                                    OSD_QID_INODE);
3167         if (rc)
3168                 RETURN(rc);
3169         /* data to be truncated */
3170         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3171                                    i_projid_read(inode), 0, oh, obj, NULL,
3172                                    OSD_QID_BLK);
3173         if (rc)
3174                 RETURN(rc);
3175
3176         /* will help to find FID->ino when this object is being
3177          * added to PENDING/ */
3178         rc = osd_idc_find_and_init(env, osd_obj2dev(obj), obj);
3179