Whamcloud - gitweb
LU-13974 llog: check stale osp object
[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, 2017, 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/fs_struct.h>
43 #include <linux/kallsyms.h>
44 #include <linux/module.h>
45 #include <linux/user_namespace.h>
46 #include <linux/uidgid.h>
47
48 /* prerequisite for linux/xattr.h */
49 #include <linux/types.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/fs.h>
52 /* XATTR_{REPLACE,CREATE} */
53 #include <linux/xattr.h>
54
55 #include <ldiskfs/ldiskfs.h>
56 #include <ldiskfs/xattr.h>
57 #include <ldiskfs/ldiskfs_extents.h>
58 #undef ENTRY
59 /*
60  * struct OBD_{ALLOC,FREE}*()
61  * OBD_FAIL_CHECK
62  */
63 #include <obd_support.h>
64 /* struct ptlrpc_thread */
65 #include <lustre_net.h>
66 #include <lustre_fid.h>
67 /* process_config */
68 #include <uapi/linux/lustre/lustre_param.h>
69
70 #include "osd_internal.h"
71 #include "osd_dynlocks.h"
72
73 /* llo_* api support */
74 #include <md_object.h>
75 #include <lustre_quota.h>
76
77 #include <lustre_linkea.h>
78
79 /* Maximum EA size is limited by LNET_MTU for remote objects */
80 #define OSD_MAX_EA_SIZE 1048364
81
82 int ldiskfs_pdo = 1;
83 module_param(ldiskfs_pdo, int, 0644);
84 MODULE_PARM_DESC(ldiskfs_pdo, "ldiskfs with parallel directory operations");
85
86 int ldiskfs_track_declares_assert;
87 module_param(ldiskfs_track_declares_assert, int, 0644);
88 MODULE_PARM_DESC(ldiskfs_track_declares_assert, "LBUG during tracking of declares");
89
90 /* Slab to allocate dynlocks */
91 struct kmem_cache *dynlock_cachep;
92
93 /* Slab to allocate osd_it_ea */
94 struct kmem_cache *osd_itea_cachep;
95
96 static struct lu_kmem_descr ldiskfs_caches[] = {
97         {
98                 .ckd_cache = &dynlock_cachep,
99                 .ckd_name  = "dynlock_cache",
100                 .ckd_size  = sizeof(struct dynlock_handle)
101         },
102         {
103                 .ckd_cache = &osd_itea_cachep,
104                 .ckd_name  = "osd_itea_cache",
105                 .ckd_size  = sizeof(struct osd_it_ea)
106         },
107         {
108                 .ckd_cache = NULL
109         }
110 };
111
112 static const char dot[] = ".";
113 static const char dotdot[] = "..";
114
115 static const struct lu_object_operations      osd_lu_obj_ops;
116 static const struct dt_object_operations      osd_obj_ops;
117 static const struct dt_object_operations      osd_obj_otable_it_ops;
118 static const struct dt_index_operations       osd_index_iam_ops;
119 static const struct dt_index_operations       osd_index_ea_ops;
120
121 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
122                           const struct lu_fid *fid);
123 static int osd_process_scheduled_agent_removals(const struct lu_env *env,
124                                                 struct osd_device *osd);
125
126 int osd_trans_declare_op2rb[] = {
127         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
128         [OSD_OT_PUNCH]          = OSD_OT_MAX,
129         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
130         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
131         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
132         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
133         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
134         [OSD_OT_WRITE]          = OSD_OT_WRITE,
135         [OSD_OT_INSERT]         = OSD_OT_DELETE,
136         [OSD_OT_DELETE]         = OSD_OT_INSERT,
137         [OSD_OT_QUOTA]          = OSD_OT_MAX,
138 };
139
140 static int osd_has_index(const struct osd_object *obj)
141 {
142         return obj->oo_dt.do_index_ops != NULL;
143 }
144
145 static int osd_object_invariant(const struct lu_object *l)
146 {
147         return osd_invariant(osd_obj(l));
148 }
149
150 /*
151  * Concurrency: doesn't matter
152  */
153 static int osd_is_write_locked(const struct lu_env *env, struct osd_object *o)
154 {
155         struct osd_thread_info *oti = osd_oti_get(env);
156
157         return oti->oti_w_locks > 0 && o->oo_owner == env;
158 }
159
160 /*
161  * Concurrency: doesn't access mutable data
162  */
163 static int osd_root_get(const struct lu_env *env,
164                         struct dt_device *dev, struct lu_fid *f)
165 {
166         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
167         return 0;
168 }
169
170 /*
171  * the following set of functions are used to maintain per-thread
172  * cache of FID->ino mapping. this mechanism is needed to resolve
173  * FID to inode at dt_insert() which in turn stores ino in the
174  * directory entries to keep ldiskfs compatible with ext[34].
175  * due to locking-originated restrictions we can't lookup ino
176  * using LU cache (deadlock is possible). lookup using OI is quite
177  * expensive. so instead we maintain this cache and methods like
178  * dt_create() fill it. so in the majority of cases dt_insert() is
179  * able to find needed mapping in lockless manner.
180  */
181 static struct osd_idmap_cache *
182 osd_idc_find(const struct lu_env *env, struct osd_device *osd,
183              const struct lu_fid *fid)
184 {
185         struct osd_thread_info *oti = osd_oti_get(env);
186         struct osd_idmap_cache *idc = oti->oti_ins_cache;
187         int i;
188
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_PTR_ARRAY(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_PTR_ARRAY(oti->oti_ins_cache,
220                                            oti->oti_ins_cache_used);
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         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
256                osd->od_svname, PFID(fid));
257
258         /* new mapping is needed */
259         idc = osd_idc_add(env, osd, fid);
260         if (IS_ERR(idc)) {
261                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
262                        osd->od_svname, PFID(fid), PTR_ERR(idc));
263                 return idc;
264         }
265
266         /* initialize it */
267         rc = osd_remote_fid(env, osd, fid);
268         if (unlikely(rc < 0))
269                 return ERR_PTR(rc);
270
271         if (rc == 0) {
272                 /* the object is local, lookup in OI */
273                 /* XXX: probably cheaper to lookup in LU first? */
274                 rc = osd_oi_lookup(osd_oti_get(env), osd, fid,
275                                    &idc->oic_lid, 0);
276                 if (unlikely(rc < 0)) {
277                         CERROR("can't lookup: rc = %d\n", rc);
278                         return ERR_PTR(rc);
279                 }
280         } else {
281                 /* the object is remote */
282                 idc->oic_remote = 1;
283         }
284
285         return idc;
286 }
287
288 static void osd_idc_dump_lma(const struct lu_env *env,
289                                 struct osd_device *osd,
290                                 unsigned long ino,
291                                 bool check_in_oi)
292 {
293         struct osd_thread_info *info = osd_oti_get(env);
294         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
295         const struct lu_fid *fid;
296         struct osd_inode_id lid;
297         struct inode *inode;
298         int rc;
299
300         inode = osd_ldiskfs_iget(osd_sb(osd), ino);
301         if (IS_ERR(inode)) {
302                 CERROR("%s: can't get inode %lu: rc = %d\n",
303                        osd->od_svname, ino, (int)PTR_ERR(inode));
304                 return;
305         }
306         if (is_bad_inode(inode)) {
307                 CERROR("%s: bad inode %lu\n", osd->od_svname, ino);
308                 goto put;
309         }
310         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
311         if (rc) {
312                 CERROR("%s: can't get LMA for %lu: rc = %d\n",
313                        osd->od_svname, ino, rc);
314                 goto put;
315         }
316         fid = &loa->loa_lma.lma_self_fid;
317         LCONSOLE(D_INFO, "%s: "DFID" in inode %lu/%u\n", osd->od_svname,
318                       PFID(fid), ino, (unsigned)inode->i_generation);
319         if (!check_in_oi)
320                 goto put;
321         rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
322         if (rc) {
323                 CERROR("%s: can't lookup "DFID": rc = %d\n",
324                        osd->od_svname, PFID(fid), rc);
325                 goto put;
326         }
327         LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n", osd->od_svname,
328                       PFID(fid), lid.oii_ino, lid.oii_gen);
329 put:
330         iput(inode);
331 }
332
333 static void osd_idc_dump_debug(const struct lu_env *env,
334                                 struct osd_device *osd,
335                                 const struct lu_fid *fid,
336                                 unsigned long ino1,
337                                 unsigned long ino2)
338 {
339         struct osd_inode_id lid;
340
341         int rc;
342
343         rc = osd_oi_lookup(osd_oti_get(env), osd, fid, &lid, 0);
344         if (!rc) {
345                 LCONSOLE(D_INFO, "%s: "DFID" maps to %u/%u\n",
346                         osd->od_svname, PFID(fid), lid.oii_ino, lid.oii_gen);
347                 osd_idc_dump_lma(env, osd, lid.oii_ino, false);
348         } else {
349                 CERROR("%s: can't lookup "DFID": rc = %d\n",
350                        osd->od_svname, PFID(fid), rc);
351         }
352         if (ino1)
353                 osd_idc_dump_lma(env, osd, ino1, true);
354         if (ino2)
355                 osd_idc_dump_lma(env, osd, ino2, true);
356 }
357
358 /*
359  * lookup mapping for given FID and fill it from the given object.
360  * the object is lolcal by definition.
361  */
362 static int osd_idc_find_and_init(const struct lu_env *env,
363                                  struct osd_device *osd,
364                                  struct osd_object *obj)
365 {
366         const struct lu_fid *fid = lu_object_fid(&obj->oo_dt.do_lu);
367         struct osd_idmap_cache *idc;
368
369         idc = osd_idc_find(env, osd, fid);
370         LASSERT(!IS_ERR(idc));
371         if (idc != NULL) {
372                 if (obj->oo_inode == NULL)
373                         return 0;
374                 if (idc->oic_lid.oii_ino != obj->oo_inode->i_ino) {
375                         if (idc->oic_lid.oii_ino) {
376                                 osd_idc_dump_debug(env, osd, fid,
377                                                    idc->oic_lid.oii_ino,
378                                                    obj->oo_inode->i_ino);
379                                 return -EINVAL;
380                         }
381                         idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
382                         idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
383                 }
384                 return 0;
385         }
386
387         CDEBUG(D_INODE, "%s: FID "DFID" not in the id map cache\n",
388                osd->od_svname, PFID(fid));
389
390         /* new mapping is needed */
391         idc = osd_idc_add(env, osd, fid);
392         if (IS_ERR(idc)) {
393                 CERROR("%s: FID "DFID" add id map cache failed: %ld\n",
394                        osd->od_svname, PFID(fid), PTR_ERR(idc));
395                 return PTR_ERR(idc);
396         }
397
398         if (obj->oo_inode != NULL) {
399                 idc->oic_lid.oii_ino = obj->oo_inode->i_ino;
400                 idc->oic_lid.oii_gen = obj->oo_inode->i_generation;
401         }
402         return 0;
403 }
404
405 /*
406  * OSD object methods.
407  */
408
409 /*
410  * Concurrency: no concurrent access is possible that early in object
411  * life-cycle.
412  */
413 static struct lu_object *osd_object_alloc(const struct lu_env *env,
414                                           const struct lu_object_header *hdr,
415                                           struct lu_device *d)
416 {
417         struct osd_object *mo;
418
419         OBD_ALLOC_PTR(mo);
420         if (mo != NULL) {
421                 struct lu_object *l;
422                 struct lu_object_header *h;
423                 struct osd_device *o = osd_dev(d);
424
425                 l = &mo->oo_dt.do_lu;
426                 if (unlikely(o->od_in_init)) {
427                         OBD_ALLOC_PTR(h);
428                         if (!h) {
429                                 OBD_FREE_PTR(mo);
430                                 return NULL;
431                         }
432
433                         lu_object_header_init(h);
434                         lu_object_init(l, h, d);
435                         lu_object_add_top(h, l);
436                         mo->oo_header = h;
437                 } else {
438                         dt_object_init(&mo->oo_dt, NULL, d);
439                         mo->oo_header = NULL;
440                 }
441
442                 mo->oo_dt.do_ops = &osd_obj_ops;
443                 l->lo_ops = &osd_lu_obj_ops;
444                 init_rwsem(&mo->oo_sem);
445                 init_rwsem(&mo->oo_ext_idx_sem);
446                 spin_lock_init(&mo->oo_guard);
447                 INIT_LIST_HEAD(&mo->oo_xattr_list);
448                 return l;
449         }
450         return NULL;
451 }
452
453 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
454                 struct dentry *dentry, struct lustre_ost_attrs *loa)
455 {
456         int rc;
457
458         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
459                              (void *)loa, sizeof(*loa));
460         if (rc > 0) {
461                 struct lustre_mdt_attrs *lma = &loa->loa_lma;
462
463                 if (rc < sizeof(*lma))
464                         return -EINVAL;
465
466                 rc = 0;
467                 lustre_loa_swab(loa, true);
468                 /* Check LMA compatibility */
469                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
470                         rc = -EOPNOTSUPP;
471                         CWARN("%s: unsupported incompat LMA feature(s) %#x for fid = "DFID", ino = %lu: rc = %d\n",
472                               osd_ino2name(inode),
473                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
474                               PFID(&lma->lma_self_fid), inode->i_ino, rc);
475                 }
476         } else if (rc == 0) {
477                 rc = -ENODATA;
478         }
479
480         return rc;
481 }
482
483 /*
484  * retrieve object from backend ext fs.
485  **/
486 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
487                        struct osd_inode_id *id)
488 {
489         int rc;
490         struct inode *inode = NULL;
491
492         /*
493          * if we look for an inode withing a running
494          * transaction, then we risk to deadlock
495          * osd_dirent_check_repair() breaks this
496          */
497          /* LASSERT(current->journal_info == NULL); */
498
499         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
500         if (IS_ERR(inode)) {
501                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
502                        id->oii_ino, PTR_ERR(inode));
503         } else if (id->oii_gen != OSD_OII_NOGEN &&
504                    inode->i_generation != id->oii_gen) {
505                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
506                        "i_generation = %u\n",
507                        id->oii_ino, id->oii_gen, inode->i_generation);
508                 iput(inode);
509                 inode = ERR_PTR(-ESTALE);
510         } else if (inode->i_nlink == 0) {
511                 /*
512                  * due to parallel readdir and unlink,
513                  * we can have dead inode here.
514                  */
515                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
516                 iput(inode);
517                 inode = ERR_PTR(-ESTALE);
518         } else if (is_bad_inode(inode)) {
519                 rc = -ENOENT;
520                 CWARN("%s: bad inode: ino = %u: rc = %d\n",
521                       osd_dev2name(dev), id->oii_ino, rc);
522                 iput(inode);
523                 inode = ERR_PTR(rc);
524         } else if ((rc = osd_attach_jinode(inode))) {
525                 iput(inode);
526                 inode = ERR_PTR(rc);
527         } else {
528                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
529                 if (id->oii_gen == OSD_OII_NOGEN)
530                         osd_id_gen(id, inode->i_ino, inode->i_generation);
531
532                 /*
533                  * Do not update file c/mtime in ldiskfs.
534                  * NB: we don't have any lock to protect this because we don't
535                  * have reference on osd_object now, but contention with
536                  * another lookup + attr_set can't happen in the tiny window
537                  * between if (...) and set S_NOCMTIME.
538                  */
539                 if (!(inode->i_flags & S_NOCMTIME))
540                         inode->i_flags |= S_NOCMTIME;
541         }
542         return inode;
543 }
544
545 int osd_ldiskfs_add_entry(struct osd_thread_info *info, struct osd_device *osd,
546                           handle_t *handle, struct dentry *child,
547                           struct inode *inode, struct htree_lock *hlock)
548 {
549         int rc, rc2;
550
551         rc = __ldiskfs_add_entry(handle, child, inode, hlock);
552         if (rc == -ENOBUFS || rc == -ENOSPC) {
553                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
554                 struct inode *parent = child->d_parent->d_inode;
555                 struct lu_fid *fid = NULL;
556                 char fidstr[FID_LEN + 1] = "unknown";
557
558                 rc2 = osd_get_lma(info, parent, child->d_parent, loa);
559                 if (!rc2) {
560                         fid = &loa->loa_lma.lma_self_fid;
561                 } else if (rc2 == -ENODATA) {
562                         if (unlikely(parent == inode->i_sb->s_root->d_inode)) {
563                                 fid = &info->oti_fid3;
564                                 lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
565                         } else if (!osd->od_is_ost && osd->od_index == 0) {
566                                 fid = &info->oti_fid3;
567                                 lu_igif_build(fid, parent->i_ino,
568                                               parent->i_generation);
569                         }
570                 }
571
572                 if (fid != NULL)
573                         snprintf(fidstr, sizeof(fidstr), DFID, PFID(fid));
574
575                 /* below message is checked in sanity.sh test_129 */
576                 if (rc == -ENOSPC) {
577                         CWARN("%s: directory (inode: %lu, FID: %s) has reached max size limit\n",
578                               osd_name(osd), parent->i_ino, fidstr);
579                 } else {
580                         rc = 0; /* ignore such error now */
581                         CWARN("%s: directory (inode: %lu, FID: %s) is approaching max size limit\n",
582                               osd_name(osd), parent->i_ino, fidstr);
583                 }
584
585         }
586
587         return rc;
588 }
589
590
591 struct inode *
592 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
593              struct osd_inode_id *id, struct lu_fid *fid)
594 {
595         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
596         struct inode *inode;
597         int rc;
598
599         inode = osd_iget(info, dev, id);
600         if (IS_ERR(inode))
601                 return inode;
602
603         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
604         if (!rc) {
605                 *fid = loa->loa_lma.lma_self_fid;
606         } else if (rc == -ENODATA) {
607                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
608                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
609                 else
610                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
611         } else {
612                 iput(inode);
613                 inode = ERR_PTR(rc);
614         }
615         return inode;
616 }
617
618 static struct inode *osd_iget_check(struct osd_thread_info *info,
619                                     struct osd_device *dev,
620                                     const struct lu_fid *fid,
621                                     struct osd_inode_id *id,
622                                     bool trusted)
623 {
624         struct inode *inode;
625         int rc = 0;
626
627         ENTRY;
628
629         /*
630          * The cached OI mapping is trustable. If we cannot locate the inode
631          * via the cached OI mapping, then return the failure to the caller
632          * directly without further OI checking.
633          */
634
635 again:
636         inode = osd_ldiskfs_iget(osd_sb(dev), id->oii_ino);
637         if (IS_ERR(inode)) {
638                 rc = PTR_ERR(inode);
639                 if (!trusted && (rc == -ENOENT || rc == -ESTALE))
640                         goto check_oi;
641
642                 CDEBUG(D_INODE, "no inode for FID: "DFID", ino = %u, rc = %d\n",
643                        PFID(fid), id->oii_ino, rc);
644                 GOTO(put, rc);
645         }
646
647         if (is_bad_inode(inode)) {
648                 rc = -ENOENT;
649                 if (!trusted)
650                         goto check_oi;
651
652                 CDEBUG(D_INODE, "bad inode for FID: "DFID", ino = %u\n",
653                        PFID(fid), id->oii_ino);
654                 GOTO(put, rc);
655         }
656
657         if (id->oii_gen != OSD_OII_NOGEN &&
658             inode->i_generation != id->oii_gen) {
659                 rc = -ESTALE;
660                 if (!trusted)
661                         goto check_oi;
662
663                 CDEBUG(D_INODE, "unmatched inode for FID: "DFID", ino = %u, "
664                        "oii_gen = %u, i_generation = %u\n", PFID(fid),
665                        id->oii_ino, id->oii_gen, inode->i_generation);
666                 GOTO(put, rc);
667         }
668
669         if (inode->i_nlink == 0) {
670                 rc = -ENOENT;
671                 if (!trusted)
672                         goto check_oi;
673
674                 CDEBUG(D_INODE, "stale inode for FID: "DFID", ino = %u\n",
675                        PFID(fid), id->oii_ino);
676                 GOTO(put, rc);
677         }
678
679         ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_DESTROY);
680
681 check_oi:
682         if (rc != 0) {
683                 __u32 saved_ino = id->oii_ino;
684                 __u32 saved_gen = id->oii_gen;
685
686                 LASSERT(!trusted);
687                 LASSERTF(rc == -ESTALE || rc == -ENOENT, "rc = %d\n", rc);
688
689                 rc = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
690                 /*
691                  * XXX: There are four possible cases:
692                  *      1. rc = 0.
693                  *         Backup/restore caused the OI invalid.
694                  *      2. rc = 0.
695                  *         Someone unlinked the object but NOT removed
696                  *         the OI mapping, such as mount target device
697                  *         as ldiskfs, and modify something directly.
698                  *      3. rc = -ENOENT.
699                  *         Someone just removed the object between the
700                  *         former oi_lookup and the iget. It is normal.
701                  *      4. Other failure cases.
702                  *
703                  *      Generally, when the device is mounted, it will
704                  *      auto check whether the system is restored from
705                  *      file-level backup or not. We trust such detect
706                  *      to distinguish the 1st case from the 2nd case:
707                  *      if the OI files are consistent but may contain
708                  *      stale OI mappings because of case 2, if iget()
709                  *      returns -ENOENT or -ESTALE, then it should be
710                  *      the case 2.
711                  */
712                 if (rc != 0)
713                         /*
714                          * If the OI mapping was in OI file before the
715                          * osd_iget_check(), but now, it is disappear,
716                          * then it must be removed by race. That is a
717                          * normal race case.
718                          */
719                         GOTO(put, rc);
720
721                 /*
722                  * It is the OI scrub updated the OI mapping by race.
723                  * The new OI mapping must be valid.
724                  */
725                 if (saved_ino != id->oii_ino ||
726                     (saved_gen != id->oii_gen && saved_gen != OSD_OII_NOGEN)) {
727                         if (!IS_ERR(inode))
728                                 iput(inode);
729
730                         trusted = true;
731                         goto again;
732                 }
733
734                 if (IS_ERR(inode)) {
735                         if (dev->od_scrub.os_scrub.os_file.sf_flags &
736                             SF_INCONSISTENT)
737                                 /*
738                                  * It still can be the case 2, but we cannot
739                                  * distinguish it from the case 1. So return
740                                  * -EREMCHG to block current operation until
741                                  *  OI scrub rebuilt the OI mappings.
742                                  */
743                                 rc = -EREMCHG;
744                         else
745                                 rc = -ENOENT;
746
747                         GOTO(put, rc);
748                 }
749
750                 if (inode->i_generation == id->oii_gen)
751                         rc = -ENOENT;
752                 else
753                         rc = -EREMCHG;
754         } else {
755                 if (id->oii_gen == OSD_OII_NOGEN)
756                         osd_id_gen(id, inode->i_ino, inode->i_generation);
757
758                 /*
759                  * Do not update file c/mtime in ldiskfs.
760                  * NB: we don't have any lock to protect this because we don't
761                  * have reference on osd_object now, but contention with
762                  * another lookup + attr_set can't happen in the tiny window
763                  * between if (...) and set S_NOCMTIME.
764                  */
765                 if (!(inode->i_flags & S_NOCMTIME))
766                         inode->i_flags |= S_NOCMTIME;
767         }
768
769         GOTO(put, rc);
770
771 put:
772         if (rc != 0) {
773                 if (!IS_ERR(inode))
774                         iput(inode);
775
776                 inode = ERR_PTR(rc);
777         }
778
779         return inode;
780 }
781
782 /**
783  * \retval +v: new filter_fid does not contain self-fid
784  * \retval 0:  filter_fid_18_23, contains self-fid
785  * \retval -v: other failure cases
786  */
787 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
788                  struct dentry *dentry, struct lu_fid *fid)
789 {
790         struct filter_fid *ff = &info->oti_ff;
791         struct ost_id *ostid = &info->oti_ostid;
792         int rc;
793
794         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
795         if (rc == sizeof(struct filter_fid_18_23)) {
796                 struct filter_fid_18_23 *ff_old = (void *)ff;
797
798                 ostid_set_seq(ostid, le64_to_cpu(ff_old->ff_seq));
799                 rc = ostid_set_id(ostid, le64_to_cpu(ff_old->ff_objid));
800                 /*
801                  * XXX: use 0 as the index for compatibility, the caller will
802                  * handle index related issues when necessary.
803                  */
804                 if (!rc)
805                         ostid_to_fid(fid, ostid, 0);
806         } else if (rc >= (int)sizeof(struct filter_fid_24_29)) {
807                 rc = 1;
808         } else if (rc >= 0) {
809                 rc = -EINVAL;
810         }
811
812         return rc;
813 }
814
815 static int osd_lma_self_repair(struct osd_thread_info *info,
816                                struct osd_device *osd, struct inode *inode,
817                                const struct lu_fid *fid, __u32 compat)
818 {
819         handle_t *jh;
820         int rc;
821
822         LASSERT(current->journal_info == NULL);
823
824         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
825                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
826         if (IS_ERR(jh)) {
827                 rc = PTR_ERR(jh);
828                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
829                       osd_name(osd), rc);
830                 return rc;
831         }
832
833         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
834         if (rc != 0)
835                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
836                       osd_name(osd), rc);
837         ldiskfs_journal_stop(jh);
838         return rc;
839 }
840
841 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
842 {
843         struct osd_thread_info *info = osd_oti_get(env);
844         struct osd_device *osd = osd_obj2dev(obj);
845         struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
846         struct lustre_mdt_attrs *lma = &loa->loa_lma;
847         struct inode *inode = obj->oo_inode;
848         struct dentry *dentry = &info->oti_obj_dentry;
849         struct lu_fid *fid = NULL;
850         const struct lu_fid *rfid = lu_object_fid(&obj->oo_dt.do_lu);
851         int rc;
852
853         ENTRY;
854
855         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
856                              (void *)loa, sizeof(*loa));
857         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
858                 fid = &lma->lma_self_fid;
859                 rc = osd_get_idif(info, inode, dentry, fid);
860                 if (rc > 0 || (rc == -ENODATA && osd->od_index_in_idif)) {
861                         /*
862                          * For the given OST-object, if it has neither LMA nor
863                          * FID in XATTR_NAME_FID, then the given FID (which is
864                          * contained in the @obj, from client RPC for locating
865                          * the OST-object) is trusted. We use it to generate
866                          * the LMA.
867                          */
868                         osd_lma_self_repair(info, osd, inode, rfid,
869                                             LMAC_FID_ON_OST);
870                         RETURN(0);
871                 }
872         }
873
874         if (rc < 0)
875                 RETURN(rc);
876
877         if (rc > 0) {
878                 rc = 0;
879                 lustre_lma_swab(lma);
880                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
881                              (CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT) &&
882                               S_ISREG(inode->i_mode)))) {
883                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
884                               "fid = "DFID", ino = %lu\n", osd_name(osd),
885                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
886                               PFID(rfid), inode->i_ino);
887                         rc = -EOPNOTSUPP;
888                 } else {
889                         fid = &lma->lma_self_fid;
890                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
891                             osd->od_is_ost)
892                                 obj->oo_pfid_in_lma = 1;
893                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
894                             !osd->od_is_ost)
895                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
896                 }
897         }
898
899         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
900                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
901                         struct ost_id   *oi   = &info->oti_ostid;
902                         struct lu_fid   *fid1 = &info->oti_fid3;
903                         __u32            idx  = fid_idif_ost_idx(rfid);
904
905                         /*
906                          * For old IDIF, the OST index is not part of the IDIF,
907                          * Means that different OSTs may have the same IDIFs.
908                          * Under such case, we need to make some compatible
909                          * check to make sure to trigger OI scrub properly.
910                          */
911                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
912                                 /* Given @rfid is new, LMA is old. */
913                                 fid_to_ostid(fid, oi);
914                                 ostid_to_fid(fid1, oi, idx);
915                                 if (lu_fid_eq(fid1, rfid)) {
916                                         if (osd->od_index_in_idif)
917                                                 osd_lma_self_repair(info, osd,
918                                                         inode, rfid,
919                                                         LMAC_FID_ON_OST);
920                                         RETURN(0);
921                                 }
922                         }
923                 }
924
925                 rc = -EREMCHG;
926         }
927
928         RETURN(rc);
929 }
930
931 struct osd_check_lmv_buf {
932         /* please keep it as first member */
933         struct dir_context ctx;
934         struct osd_thread_info *oclb_info;
935         struct osd_device *oclb_dev;
936         struct osd_idmap_cache *oclb_oic;
937         int oclb_items;
938         bool oclb_found;
939 };
940
941 /**
942  * It is called internally by ->iterate*() to filter out the
943  * local slave object's FID of the striped directory.
944  *
945  * \retval      1 found the local slave's FID
946  * \retval      0 continue to check next item
947  * \retval      -ve for failure
948  */
949 #ifdef HAVE_FILLDIR_USE_CTX
950 static int osd_stripe_dir_filldir(struct dir_context *buf,
951 #else
952 static int osd_stripe_dir_filldir(void *buf,
953 #endif
954                                   const char *name, int namelen,
955                                   loff_t offset, __u64 ino, unsigned int d_type)
956 {
957         struct osd_check_lmv_buf *oclb = (struct osd_check_lmv_buf *)buf;
958         struct osd_thread_info *oti = oclb->oclb_info;
959         struct lu_fid *fid = &oti->oti_fid3;
960         struct osd_inode_id *id = &oti->oti_id3;
961         struct osd_device *dev = oclb->oclb_dev;
962         struct osd_idmap_cache *oic = oclb->oclb_oic;
963         struct inode *inode;
964
965         oclb->oclb_items++;
966
967         if (name[0] == '.')
968                 return 0;
969
970         fid_zero(fid);
971         sscanf(name + 1, SFID, RFID(fid));
972         if (!fid_is_sane(fid))
973                 return 0;
974
975         if (osd_remote_fid(oti->oti_env, dev, fid))
976                 return 0;
977
978         osd_id_gen(id, ino, OSD_OII_NOGEN);
979         inode = osd_iget(oti, dev, id);
980         if (IS_ERR(inode))
981                 return PTR_ERR(inode);
982
983         iput(inode);
984         osd_add_oi_cache(oti, dev, id, fid);
985         oic->oic_fid = *fid;
986         oic->oic_lid = *id;
987         oic->oic_dev = dev;
988         osd_oii_insert(dev, oic, true);
989         oclb->oclb_found = true;
990
991         return 1;
992 }
993
994 /*
995  * When lookup item under striped directory, we need to locate the master
996  * MDT-object of the striped directory firstly, then the client will send
997  * lookup (getattr_by_name) RPC to the MDT with some slave MDT-object's FID
998  * and the item's name. If the system is restored from MDT file level backup,
999  * then before the OI scrub completely built the OI files, the OI mappings of
1000  * the master MDT-object and slave MDT-object may be invalid. Usually, it is
1001  * not a problem for the master MDT-object. Because when locate the master
1002  * MDT-object, we will do name based lookup (for the striped directory itself)
1003  * firstly, during such process we can setup the correct OI mapping for the
1004  * master MDT-object. But it will be trouble for the slave MDT-object. Because
1005  * the client will not trigger name based lookup on the MDT to locate the slave
1006  * MDT-object before locating item under the striped directory, then when
1007  * osd_fid_lookup(), it will find that the OI mapping for the slave MDT-object
1008  * is invalid and does not know what the right OI mapping is, then the MDT has
1009  * to return -EINPROGRESS to the client to notify that the OI scrub is rebuiding
1010  * the OI file, related OI mapping is unknown yet, please try again later. And
1011  * then client will re-try the RPC again and again until related OI mapping has
1012  * been updated. That is quite inefficient.
1013  *
1014  * To resolve above trouble, we will handle it as the following two cases:
1015  *
1016  * 1) The slave MDT-object and the master MDT-object are on different MDTs.
1017  *    It is relative easy. Be as one of remote MDT-objects, the slave MDT-object
1018  *    is linked under /REMOTE_PARENT_DIR with the name of its FID string.
1019  *    We can locate the slave MDT-object via lookup the /REMOTE_PARENT_DIR
1020  *    directly. Please check osd_fid_lookup().
1021  *
1022  * 2) The slave MDT-object and the master MDT-object reside on the same MDT.
1023  *    Under such case, during lookup the master MDT-object, we will lookup the
1024  *    slave MDT-object via readdir against the master MDT-object, because the
1025  *    slave MDT-objects information are stored as sub-directories with the name
1026  *    "${FID}:${index}". Then when find the local slave MDT-object, its OI
1027  *    mapping will be recorded. Then subsequent osd_fid_lookup() will know
1028  *    the correct OI mapping for the slave MDT-object.
1029  */
1030 static int osd_check_lmv(struct osd_thread_info *oti, struct osd_device *dev,
1031                          struct inode *inode, struct osd_idmap_cache *oic)
1032 {
1033         struct lu_buf *buf = &oti->oti_big_buf;
1034         struct dentry *dentry = &oti->oti_obj_dentry;
1035         struct file *filp;
1036         struct lmv_mds_md_v1 *lmv1;
1037         struct osd_check_lmv_buf oclb = {
1038                 .ctx.actor = osd_stripe_dir_filldir,
1039                 .oclb_info = oti,
1040                 .oclb_dev = dev,
1041                 .oclb_oic = oic,
1042                 .oclb_found = false,
1043         };
1044         int rc = 0;
1045
1046         ENTRY;
1047
1048 again:
1049         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, buf->lb_buf,
1050                              buf->lb_len);
1051         if (rc == -ERANGE) {
1052                 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMV, NULL, 0);
1053                 if (rc > 0) {
1054                         lu_buf_realloc(buf, rc);
1055                         if (buf->lb_buf == NULL)
1056                                 GOTO(out, rc = -ENOMEM);
1057
1058                         goto again;
1059                 }
1060         }
1061
1062         if (unlikely(rc == 0 || rc == -ENODATA))
1063                 GOTO(out, rc = 0);
1064
1065         if (rc < 0)
1066                 GOTO(out, rc);
1067
1068         if (unlikely(buf->lb_buf == NULL)) {
1069                 lu_buf_realloc(buf, rc);
1070                 if (buf->lb_buf == NULL)
1071                         GOTO(out, rc = -ENOMEM);
1072
1073                 goto again;
1074         }
1075
1076         lmv1 = buf->lb_buf;
1077         if (le32_to_cpu(lmv1->lmv_magic) != LMV_MAGIC_V1)
1078                 GOTO(out, rc = 0);
1079
1080         filp = osd_quasi_file(oti->oti_env, inode);
1081         rc = osd_security_file_alloc(filp);
1082         if (rc)
1083                 goto out;
1084
1085         do {
1086                 oclb.oclb_items = 0;
1087                 rc = iterate_dir(filp, &oclb.ctx);
1088         } while (rc >= 0 && oclb.oclb_items > 0 && !oclb.oclb_found &&
1089                  filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
1090         inode->i_fop->release(inode, filp);
1091
1092 out:
1093         if (rc < 0)
1094                 CDEBUG(D_LFSCK,
1095                        "%s: cannot check LMV, ino = %lu/%u "DFID": rc = %d\n",
1096                        osd_ino2name(inode), inode->i_ino, inode->i_generation,
1097                        PFID(&oic->oic_fid), rc);
1098         else
1099                 rc = 0;
1100
1101         RETURN(rc);
1102 }
1103
1104 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
1105                           const struct lu_fid *fid,
1106                           const struct lu_object_conf *conf)
1107 {
1108         struct osd_thread_info *info;
1109         struct lu_device *ldev = obj->oo_dt.do_lu.lo_dev;
1110         struct osd_device *dev;
1111         struct osd_idmap_cache *oic;
1112         struct osd_inode_id *id;
1113         struct inode *inode = NULL;
1114         struct lustre_scrub *scrub;
1115         struct scrub_file *sf;
1116         __u32 flags = SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT | SS_AUTO_FULL;
1117         __u32 saved_ino;
1118         __u32 saved_gen;
1119         int result = 0;
1120         int rc1 = 0;
1121         bool remote = false;
1122         bool trusted = true;
1123         bool updated = false;
1124         bool checked = false;
1125
1126         ENTRY;
1127
1128         LINVRNT(osd_invariant(obj));
1129         LASSERT(obj->oo_inode == NULL);
1130
1131         if (fid_is_sane(fid) == 0) {
1132                 CERROR("%s: invalid FID "DFID"\n", ldev->ld_obd->obd_name,
1133                        PFID(fid));
1134                 dump_stack();
1135                 RETURN(-EINVAL);
1136         }
1137
1138         dev = osd_dev(ldev);
1139         scrub = &dev->od_scrub.os_scrub;
1140         sf = &scrub->os_file;
1141         info = osd_oti_get(env);
1142         LASSERT(info);
1143         oic = &info->oti_cache;
1144
1145         if (OBD_FAIL_CHECK(OBD_FAIL_SRV_ENOENT))
1146                 RETURN(-ENOENT);
1147
1148         /*
1149          * For the object is created as locking anchor, or for the object to
1150          * be created on disk. No need to osd_oi_lookup() at here because FID
1151          * shouldn't never be re-used, if it's really a duplicate FID from
1152          * unexpected reason, we should be able to detect it later by calling
1153          * do_create->osd_oi_insert().
1154          */
1155         if (conf && conf->loc_flags & LOC_F_NEW)
1156                 GOTO(out, result = 0);
1157
1158         /* Search order: 1. per-thread cache. */
1159         if (lu_fid_eq(fid, &oic->oic_fid) && likely(oic->oic_dev == dev)) {
1160                 id = &oic->oic_lid;
1161                 goto iget;
1162         }
1163
1164         id = &info->oti_id;
1165         if (!list_empty(&scrub->os_inconsistent_items)) {
1166                 /* Search order: 2. OI scrub pending list. */
1167                 result = osd_oii_lookup(dev, fid, id);
1168                 if (!result)
1169                         goto iget;
1170         }
1171
1172         /*
1173          * The OI mapping in the OI file can be updated by the OI scrub
1174          * when we locate the inode via FID. So it may be not trustable.
1175          */
1176         trusted = false;
1177
1178         /* Search order: 3. OI files. */
1179         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1180         if (result == -ENOENT) {
1181                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
1182                     fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) ||
1183                     !ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
1184                                       sf->sf_oi_bitmap))
1185                         GOTO(out, result = 0);
1186
1187                 goto trigger;
1188         }
1189
1190         /* -ESTALE is returned if inode of OST object doesn't exist */
1191         if (result == -ESTALE &&
1192             fid_is_on_ost(info, dev, fid, OI_CHECK_FLD)) {
1193                 GOTO(out, result = 0);
1194         }
1195
1196         if (result)
1197                 GOTO(out, result);
1198
1199 iget:
1200         obj->oo_inode = NULL;
1201         /* for later passes through checks, not true on first pass */
1202         if (!IS_ERR_OR_NULL(inode))
1203                 iput(inode);
1204
1205         inode = osd_iget_check(info, dev, fid, id, trusted);
1206         if (!IS_ERR(inode)) {
1207                 obj->oo_inode = inode;
1208                 result = 0;
1209                 if (remote)
1210                         goto trigger;
1211
1212                 goto check_lma;
1213         }
1214
1215         result = PTR_ERR(inode);
1216         if (result == -ENOENT || result == -ESTALE)
1217                 GOTO(out, result = 0);
1218
1219         if (result != -EREMCHG)
1220                 GOTO(out, result);
1221
1222 trigger:
1223         /*
1224          * We still have chance to get the valid inode: for the
1225          * object which is referenced by remote name entry, the
1226          * object on the local MDT will be linked under the dir
1227          * of "/REMOTE_PARENT_DIR" with its FID string as name.
1228          *
1229          * We do not know whether the object for the given FID
1230          * is referenced by some remote name entry or not, and
1231          * especially for DNE II, a multiple-linked object may
1232          * have many name entries reside on many MDTs.
1233          *
1234          * To simplify the operation, OSD will not distinguish
1235          * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
1236          * only happened for the RPC from other MDT during the
1237          * OI scrub, or for the client side RPC with FID only,
1238          * such as FID to path, or from old connected client.
1239          */
1240         if (!remote) {
1241                 rc1 = osd_lookup_in_remote_parent(info, dev, fid, id);
1242                 if (!rc1) {
1243                         remote = true;
1244                         trusted = true;
1245                         flags |= SS_AUTO_PARTIAL;
1246                         flags &= ~SS_AUTO_FULL;
1247                         goto iget;
1248                 }
1249         }
1250
1251         if (scrub->os_running) {
1252                 if (scrub->os_partial_scan && !scrub->os_in_join)
1253                         goto join;
1254
1255                 osd_add_oi_cache(info, dev, id, fid);
1256                 if (IS_ERR_OR_NULL(inode) || result) {
1257                         osd_oii_insert(dev, oic, result == -ENOENT);
1258                         GOTO(out, result = -EINPROGRESS);
1259                 }
1260
1261                 LASSERT(remote);
1262                 LASSERT(obj->oo_inode == inode);
1263
1264                 osd_oii_insert(dev, oic, true);
1265                 goto found;
1266         }
1267
1268         if (dev->od_auto_scrub_interval == AS_NEVER) {
1269                 if (!remote)
1270                         GOTO(out, result = -EREMCHG);
1271
1272                 LASSERT(!result);
1273                 LASSERT(obj->oo_inode == inode);
1274
1275                 osd_add_oi_cache(info, dev, id, fid);
1276                 goto found;
1277         }
1278
1279 join:
1280         rc1 = osd_scrub_start(env, dev, flags);
1281         LCONSOLE_WARN("%s: trigger OI scrub by RPC for the " DFID" with flags "
1282                       "0x%x, rc = %d\n", osd_name(dev), PFID(fid), flags, rc1);
1283         if (rc1 && rc1 != -EALREADY)
1284                 GOTO(out, result = -EREMCHG);
1285
1286         osd_add_oi_cache(info, dev, id, fid);
1287         if (IS_ERR_OR_NULL(inode) || result) {
1288                 osd_oii_insert(dev, oic, result == -ENOENT);
1289                 GOTO(out, result = -EINPROGRESS);
1290         }
1291
1292         LASSERT(remote);
1293         LASSERT(obj->oo_inode == inode);
1294
1295         osd_oii_insert(dev, oic, true);
1296         goto found;
1297
1298 check_lma:
1299         checked = true;
1300         if (unlikely(obj->oo_header))
1301                 goto found;
1302
1303         result = osd_check_lma(env, obj);
1304         if (!result)
1305                 goto found;
1306
1307         LASSERTF(id->oii_ino == inode->i_ino &&
1308                  id->oii_gen == inode->i_generation,
1309                  "locate wrong inode for FID: "DFID", %u/%u => %ld/%u\n",
1310                  PFID(fid), id->oii_ino, id->oii_gen,
1311                  inode->i_ino, inode->i_generation);
1312
1313         saved_ino = inode->i_ino;
1314         saved_gen = inode->i_generation;
1315
1316         if (unlikely(result == -ENODATA)) {
1317                 /*
1318                  * If the OI scrub updated the OI mapping by race, it
1319                  * must be valid. Trust the inode that has no LMA EA.
1320                  */
1321                 if (updated)
1322                         goto found;
1323
1324                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1325                 if (!result) {
1326                         /*
1327                          * The OI mapping is still there, the inode is still
1328                          * valid. It is just becaues the inode has no LMA EA.
1329                          */
1330                         if (saved_ino == id->oii_ino &&
1331                             saved_gen == id->oii_gen)
1332                                 goto found;
1333
1334                         /*
1335                          * It is the OI scrub updated the OI mapping by race.
1336                          * The new OI mapping must be valid.
1337                          */
1338                         trusted = true;
1339                         updated = true;
1340                         goto iget;
1341                 }
1342
1343                 /*
1344                  * "result == -ENOENT" means that the OI mappinghas been
1345                  * removed by race, so the inode belongs to other object.
1346                  *
1347                  * Others error can be returned  directly.
1348                  */
1349                 if (result == -ENOENT) {
1350                         LASSERT(trusted);
1351
1352                         obj->oo_inode = NULL;
1353                         result = 0;
1354                 }
1355         }
1356
1357         if (result != -EREMCHG)
1358                 GOTO(out, result);
1359
1360         LASSERT(!updated);
1361
1362         /*
1363          * if two OST objects map to the same inode, and inode mode is
1364          * (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666), which means it's
1365          * reserved by precreate, and not written yet, in this case, don't
1366          * set inode for the object whose FID mismatch, so that it can create
1367          * inode and not block precreate.
1368          */
1369         if (fid_is_on_ost(info, dev, fid, OI_CHECK_FLD) &&
1370             inode->i_mode == (S_IFREG | S_ISUID | S_ISGID | S_ISVTX | 0666)) {
1371                 obj->oo_inode = NULL;
1372                 GOTO(out, result = 0);
1373         }
1374
1375         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
1376         /*
1377          * "result == -ENOENT" means the cached OI mapping has been removed
1378          * from the OI file by race, above inode belongs to other object.
1379          */
1380         if (result == -ENOENT) {
1381                 LASSERT(trusted);
1382
1383                 obj->oo_inode = NULL;
1384                 GOTO(out, result = 0);
1385         }
1386
1387         if (result)
1388                 GOTO(out, result);
1389
1390         if (saved_ino == id->oii_ino && saved_gen == id->oii_gen) {
1391                 result = -EREMCHG;
1392                 goto trigger;
1393         }
1394
1395         /*
1396          * It is the OI scrub updated the OI mapping by race.
1397          * The new OI mapping must be valid.
1398          */
1399         trusted = true;
1400         updated = true;
1401         goto iget;
1402
1403 found:
1404         if (!checked) {
1405                 struct lustre_ost_attrs *loa = &info->oti_ost_attrs;
1406                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1407
1408                 result = osd_get_lma(info, inode, &info->oti_obj_dentry, loa);
1409                 if (!result) {
1410                         if (lma->lma_compat & LMAC_STRIPE_INFO &&
1411                             dev->od_is_ost)
1412                                 obj->oo_pfid_in_lma = 1;
1413                         if (unlikely(lma->lma_incompat & LMAI_REMOTE_PARENT) &&
1414                             !dev->od_is_ost)
1415                                 lu_object_set_agent_entry(&obj->oo_dt.do_lu);
1416                 } else if (result != -ENODATA) {
1417                         GOTO(out, result);
1418                 }
1419         }
1420
1421         obj->oo_compat_dot_created = 1;
1422         obj->oo_compat_dotdot_created = 1;
1423
1424         if (S_ISDIR(inode->i_mode) &&
1425             (flags & SS_AUTO_PARTIAL || sf->sf_status == SS_SCANNING))
1426                 osd_check_lmv(info, dev, inode, oic);
1427
1428         result = osd_attach_jinode(inode);
1429         if (result)
1430                 GOTO(out, result);
1431
1432         if (!ldiskfs_pdo)
1433                 GOTO(out, result = 0);
1434
1435         LASSERT(!obj->oo_hl_head);
1436         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1437
1438         GOTO(out, result = (!obj->oo_hl_head ? -ENOMEM : 0));
1439
1440 out:
1441         if (result || !obj->oo_inode) {
1442                 if (!IS_ERR_OR_NULL(inode))
1443                         iput(inode);
1444
1445                 obj->oo_inode = NULL;
1446                 if (trusted)
1447                         fid_zero(&oic->oic_fid);
1448         }
1449
1450         LINVRNT(osd_invariant(obj));
1451         return result;
1452 }
1453
1454 /*
1455  * Concurrency: shouldn't matter.
1456  */
1457 static void osd_object_init0(struct osd_object *obj)
1458 {
1459         LASSERT(obj->oo_inode != NULL);
1460         obj->oo_dt.do_body_ops = &osd_body_ops;
1461         obj->oo_dt.do_lu.lo_header->loh_attr |=
1462                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
1463 }
1464
1465 /*
1466  * Concurrency: no concurrent access is possible that early in object
1467  * life-cycle.
1468  */
1469 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
1470                            const struct lu_object_conf *conf)
1471 {
1472         struct osd_object *obj = osd_obj(l);
1473         int result;
1474
1475         LINVRNT(osd_invariant(obj));
1476
1477         if (OBD_FAIL_PRECHECK(OBD_FAIL_MDS_LLOG_UMOUNT_RACE) &&
1478             cfs_fail_val == 2) {
1479                 struct osd_thread_info *info = osd_oti_get(env);
1480                 struct osd_idmap_cache *oic = &info->oti_cache;
1481                 /* invalidate thread cache */
1482                 memset(&oic->oic_fid, 0, sizeof(oic->oic_fid));
1483         }
1484         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
1485                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
1486                 l->lo_header->loh_attr |= LOHA_EXISTS;
1487                 return 0;
1488         }
1489
1490         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
1491         obj->oo_dt.do_body_ops = &osd_body_ops_new;
1492         if (result == 0 && obj->oo_inode != NULL) {
1493                 struct osd_thread_info *oti = osd_oti_get(env);
1494                 struct lustre_ost_attrs *loa = &oti->oti_ost_attrs;
1495
1496                 osd_object_init0(obj);
1497                 if (unlikely(obj->oo_header))
1498                         return 0;
1499
1500                 result = osd_get_lma(oti, obj->oo_inode,
1501                                      &oti->oti_obj_dentry, loa);
1502                 if (!result) {
1503                         /*
1504                          * Convert LMAI flags to lustre LMA flags
1505                          * and cache it to oo_lma_flags
1506                          */
1507                         obj->oo_lma_flags =
1508                                 lma_to_lustre_flags(loa->loa_lma.lma_incompat);
1509                 } else if (result == -ENODATA) {
1510                         result = 0;
1511                 }
1512         }
1513         obj->oo_dirent_count = LU_DIRENT_COUNT_UNSET;
1514
1515         LINVRNT(osd_invariant(obj));
1516         return result;
1517 }
1518
1519 /*
1520  * The first part of oxe_buf is xattr name, and is '\0' terminated.
1521  * The left part is for value, binary mode.
1522  */
1523 struct osd_xattr_entry {
1524         struct list_head        oxe_list;
1525         size_t                  oxe_len;
1526         size_t                  oxe_namelen;
1527         bool                    oxe_exist;
1528         struct rcu_head         oxe_rcu;
1529         char                    oxe_buf[0];
1530 };
1531
1532 static int osd_oxc_get(struct osd_object *obj, const char *name,
1533                        struct lu_buf *buf)
1534 {
1535         struct osd_xattr_entry *tmp;
1536         struct osd_xattr_entry *oxe = NULL;
1537         size_t namelen = strlen(name);
1538         int rc;
1539
1540         rcu_read_lock();
1541         list_for_each_entry_rcu(tmp, &obj->oo_xattr_list, oxe_list) {
1542                 if (namelen == tmp->oxe_namelen &&
1543                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1544                         oxe = tmp;
1545                         break;
1546                 }
1547         }
1548
1549         if (oxe == NULL)
1550                 GOTO(out, rc = -ENOENT);
1551
1552         if (!oxe->oxe_exist)
1553                 GOTO(out, rc = -ENODATA);
1554
1555         /* vallen */
1556         rc = oxe->oxe_len - sizeof(*oxe) - oxe->oxe_namelen - 1;
1557         LASSERT(rc > 0);
1558
1559         if (buf->lb_buf == NULL)
1560                 GOTO(out, rc);
1561
1562         if (buf->lb_len < rc)
1563                 GOTO(out, rc = -ERANGE);
1564
1565         memcpy(buf->lb_buf, &oxe->oxe_buf[namelen + 1], rc);
1566 out:
1567         rcu_read_unlock();
1568
1569         return rc;
1570 }
1571
1572 static void osd_oxc_free(struct rcu_head *head)
1573 {
1574         struct osd_xattr_entry *oxe;
1575
1576         oxe = container_of(head, struct osd_xattr_entry, oxe_rcu);
1577         OBD_FREE(oxe, oxe->oxe_len);
1578 }
1579
1580 static void osd_oxc_add(struct osd_object *obj, const char *name,
1581                         const char *buf, int buflen)
1582 {
1583         struct osd_xattr_entry *oxe;
1584         struct osd_xattr_entry *old = NULL;
1585         struct osd_xattr_entry *tmp;
1586         size_t namelen = strlen(name);
1587         size_t len = sizeof(*oxe) + namelen + 1 + buflen;
1588
1589         OBD_ALLOC(oxe, len);
1590         if (oxe == NULL)
1591                 return;
1592
1593         INIT_LIST_HEAD(&oxe->oxe_list);
1594         oxe->oxe_len = len;
1595         oxe->oxe_namelen = namelen;
1596         memcpy(oxe->oxe_buf, name, namelen);
1597         if (buflen > 0) {
1598                 LASSERT(buf != NULL);
1599                 memcpy(oxe->oxe_buf + namelen + 1, buf, buflen);
1600                 oxe->oxe_exist = true;
1601         } else {
1602                 oxe->oxe_exist = false;
1603         }
1604
1605         /* this should be rarely called, just remove old and add new */
1606         spin_lock(&obj->oo_guard);
1607         list_for_each_entry(tmp, &obj->oo_xattr_list, oxe_list) {
1608                 if (namelen == tmp->oxe_namelen &&
1609                     strncmp(name, tmp->oxe_buf, namelen) == 0) {
1610                         old = tmp;
1611                         break;
1612                 }
1613         }
1614         if (old != NULL) {
1615                 list_replace_rcu(&old->oxe_list, &oxe->oxe_list);
1616                 call_rcu(&old->oxe_rcu, osd_oxc_free);
1617         } else {
1618                 list_add_tail_rcu(&oxe->oxe_list, &obj->oo_xattr_list);
1619         }
1620         spin_unlock(&obj->oo_guard);
1621 }
1622
1623 static void osd_oxc_del(struct osd_object *obj, const char *name)
1624 {
1625         struct osd_xattr_entry *oxe;
1626         size_t namelen = strlen(name);
1627
1628         spin_lock(&obj->oo_guard);
1629         list_for_each_entry(oxe, &obj->oo_xattr_list, oxe_list) {
1630                 if (namelen == oxe->oxe_namelen &&
1631                     strncmp(name, oxe->oxe_buf, namelen) == 0) {
1632                         list_del_rcu(&oxe->oxe_list);
1633                         call_rcu(&oxe->oxe_rcu, osd_oxc_free);
1634                         break;
1635                 }
1636         }
1637         spin_unlock(&obj->oo_guard);
1638 }
1639
1640 static void osd_oxc_fini(struct osd_object *obj)
1641 {
1642         struct osd_xattr_entry *oxe, *next;
1643
1644         list_for_each_entry_safe(oxe, next, &obj->oo_xattr_list, oxe_list) {
1645                 list_del(&oxe->oxe_list);
1646                 OBD_FREE(oxe, oxe->oxe_len);
1647         }
1648 }
1649
1650 /*
1651  * Concurrency: no concurrent access is possible that late in object
1652  * life-cycle.
1653  */
1654 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
1655 {
1656         struct osd_object *obj = osd_obj(l);
1657         struct lu_object_header *h = obj->oo_header;
1658
1659         LINVRNT(osd_invariant(obj));
1660
1661         osd_oxc_fini(obj);
1662         dt_object_fini(&obj->oo_dt);
1663         if (obj->oo_hl_head != NULL)
1664                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1665         /* obj doesn't contain an lu_object_header, so we don't need call_rcu */
1666         OBD_FREE_PTR(obj);
1667         if (unlikely(h))
1668                 lu_object_header_free(h);
1669 }
1670
1671 /*
1672  * Concurrency: no concurrent access is possible that late in object
1673  * life-cycle.
1674  */
1675 static void osd_index_fini(struct osd_object *o)
1676 {
1677         struct iam_container *bag;
1678
1679         if (o->oo_dir != NULL) {
1680                 bag = &o->oo_dir->od_container;
1681                 if (o->oo_inode != NULL) {
1682                         if (bag->ic_object == o->oo_inode)
1683                                 iam_container_fini(bag);
1684                 }
1685                 OBD_FREE_PTR(o->oo_dir);
1686                 o->oo_dir = NULL;
1687         }
1688 }
1689
1690 enum {
1691         OSD_TXN_OI_DELETE_CREDITS    = 20,
1692         OSD_TXN_INODE_DELETE_CREDITS = 20
1693 };
1694
1695 /*
1696  * Journal
1697  */
1698
1699 #if OSD_THANDLE_STATS
1700 /**
1701  * Set time when the handle is allocated
1702  */
1703 static void osd_th_alloced(struct osd_thandle *oth)
1704 {
1705         oth->oth_alloced = ktime_get();
1706 }
1707
1708 /**
1709  * Set time when the handle started
1710  */
1711 static void osd_th_started(struct osd_thandle *oth)
1712 {
1713         oth->oth_started = ktime_get();
1714 }
1715
1716 /**
1717  * Check whether the we deal with this handle for too long.
1718  */
1719 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
1720                                 ktime_t alloced, ktime_t started,
1721                                 ktime_t closed)
1722 {
1723         ktime_t now = ktime_get();
1724
1725         LASSERT(dev != NULL);
1726
1727         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
1728                             ktime_us_delta(started, alloced));
1729         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
1730                             ktime_us_delta(closed, started));
1731         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
1732                             ktime_us_delta(now, closed));
1733
1734         if (ktime_before(ktime_add_ns(alloced, 30 * NSEC_PER_SEC), now)) {
1735                 CWARN("transaction handle %p was open for too long: now %lld, alloced %lld, started %lld, closed %lld\n",
1736                                 oth, now, alloced, started, closed);
1737                 libcfs_debug_dumpstack(NULL);
1738         }
1739 }
1740
1741 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
1742 {                                                                       \
1743         ktime_t __closed = ktime_get();                                 \
1744         ktime_t __alloced = oth->oth_alloced;                           \
1745         ktime_t __started = oth->oth_started;                           \
1746                                                                         \
1747         expr;                                                           \
1748         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
1749 }
1750
1751 #else /* OSD_THANDLE_STATS */
1752
1753 #define osd_th_alloced(h)                  do {} while(0)
1754 #define osd_th_started(h)                  do {} while(0)
1755 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
1756
1757 #endif /* OSD_THANDLE_STATS */
1758
1759 /*
1760  * Concurrency: doesn't access mutable data.
1761  */
1762 static int osd_param_is_not_sane(const struct osd_device *dev,
1763                                  const struct thandle *th)
1764 {
1765         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
1766
1767         return oh->ot_credits > osd_transaction_size(dev);
1768 }
1769
1770 /*
1771  * Concurrency: shouldn't matter.
1772  */
1773 static void osd_trans_commit_cb(struct super_block *sb,
1774                                 struct ldiskfs_journal_cb_entry *jcb, int error)
1775 {
1776         struct osd_thandle *oh = container_of(jcb, struct osd_thandle, ot_jcb);
1777         struct thandle *th = &oh->ot_super;
1778         struct lu_device *lud = &th->th_dev->dd_lu_dev;
1779         struct osd_device *osd = osd_dev(lud);
1780         struct dt_txn_commit_cb *dcb, *tmp;
1781
1782         LASSERT(oh->ot_handle == NULL);
1783
1784         if (error)
1785                 CERROR("transaction @0x%p commit error: %d\n", th, error);
1786
1787         OBD_FAIL_TIMEOUT(OBD_FAIL_OST_DELAY_TRANS, 40);
1788         /* call per-transaction callbacks if any */
1789         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
1790                                  dcb_linkage) {
1791                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1792                          "commit callback entry: magic=%x name='%s'\n",
1793                          dcb->dcb_magic, dcb->dcb_name);
1794                 list_del_init(&dcb->dcb_linkage);
1795                 dcb->dcb_func(NULL, th, dcb, error);
1796         }
1797
1798         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
1799         if (atomic_dec_and_test(&osd->od_commit_cb_in_flight))
1800                 wake_up(&osd->od_commit_cb_done);
1801         th->th_dev = NULL;
1802
1803         OBD_FREE_PTR(oh);
1804 }
1805
1806 static struct thandle *osd_trans_create(const struct lu_env *env,
1807                                         struct dt_device *d)
1808 {
1809         struct osd_thread_info *oti = osd_oti_get(env);
1810         struct osd_iobuf *iobuf = &oti->oti_iobuf;
1811         struct osd_thandle *oh;
1812         struct thandle *th;
1813
1814         ENTRY;
1815
1816         if (d->dd_rdonly) {
1817                 CERROR("%s: someone try to start transaction under "
1818                        "readonly mode, should be disabled.\n",
1819                        osd_name(osd_dt_dev(d)));
1820                 dump_stack();
1821                 RETURN(ERR_PTR(-EROFS));
1822         }
1823
1824         /* on pending IO in this thread should left from prev. request */
1825         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
1826
1827         sb_start_write(osd_sb(osd_dt_dev(d)));
1828
1829         OBD_ALLOC_GFP(oh, sizeof(*oh), GFP_NOFS);
1830         if (!oh) {
1831                 sb_end_write(osd_sb(osd_dt_dev(d)));
1832                 RETURN(ERR_PTR(-ENOMEM));
1833         }
1834
1835         oh->ot_quota_trans = &oti->oti_quota_trans;
1836         memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
1837         th = &oh->ot_super;
1838         th->th_dev = d;
1839         th->th_result = 0;
1840         oh->ot_credits = 0;
1841         oh->oh_declared_ext = 0;
1842         INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
1843         INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
1844         INIT_LIST_HEAD(&oh->ot_trunc_locks);
1845         osd_th_alloced(oh);
1846
1847         memset(oti->oti_declare_ops, 0,
1848                sizeof(oti->oti_declare_ops));
1849         memset(oti->oti_declare_ops_cred, 0,
1850                sizeof(oti->oti_declare_ops_cred));
1851         memset(oti->oti_declare_ops_used, 0,
1852                sizeof(oti->oti_declare_ops_used));
1853
1854         oti->oti_ins_cache_depth++;
1855
1856         RETURN(th);
1857 }
1858
1859 void osd_trans_dump_creds(const struct lu_env *env, struct thandle *th)
1860 {
1861         struct osd_thread_info *oti = osd_oti_get(env);
1862         struct osd_thandle *oh;
1863
1864         oh = container_of(th, struct osd_thandle, ot_super);
1865         LASSERT(oh != NULL);
1866
1867         CWARN("  create: %u/%u/%u, destroy: %u/%u/%u\n",
1868               oti->oti_declare_ops[OSD_OT_CREATE],
1869               oti->oti_declare_ops_cred[OSD_OT_CREATE],
1870               oti->oti_declare_ops_used[OSD_OT_CREATE],
1871               oti->oti_declare_ops[OSD_OT_DESTROY],
1872               oti->oti_declare_ops_cred[OSD_OT_DESTROY],
1873               oti->oti_declare_ops_used[OSD_OT_DESTROY]);
1874         CWARN("  attr_set: %u/%u/%u, xattr_set: %u/%u/%u\n",
1875               oti->oti_declare_ops[OSD_OT_ATTR_SET],
1876               oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
1877               oti->oti_declare_ops_used[OSD_OT_ATTR_SET],
1878               oti->oti_declare_ops[OSD_OT_XATTR_SET],
1879               oti->oti_declare_ops_cred[OSD_OT_XATTR_SET],
1880               oti->oti_declare_ops_used[OSD_OT_XATTR_SET]);
1881         CWARN("  write: %u/%u/%u, punch: %u/%u/%u, quota %u/%u/%u\n",
1882               oti->oti_declare_ops[OSD_OT_WRITE],
1883               oti->oti_declare_ops_cred[OSD_OT_WRITE],
1884               oti->oti_declare_ops_used[OSD_OT_WRITE],
1885               oti->oti_declare_ops[OSD_OT_PUNCH],
1886               oti->oti_declare_ops_cred[OSD_OT_PUNCH],
1887               oti->oti_declare_ops_used[OSD_OT_PUNCH],
1888               oti->oti_declare_ops[OSD_OT_QUOTA],
1889               oti->oti_declare_ops_cred[OSD_OT_QUOTA],
1890               oti->oti_declare_ops_used[OSD_OT_QUOTA]);
1891         CWARN("  insert: %u/%u/%u, delete: %u/%u/%u\n",
1892               oti->oti_declare_ops[OSD_OT_INSERT],
1893               oti->oti_declare_ops_cred[OSD_OT_INSERT],
1894               oti->oti_declare_ops_used[OSD_OT_INSERT],
1895               oti->oti_declare_ops[OSD_OT_DELETE],
1896               oti->oti_declare_ops_cred[OSD_OT_DELETE],
1897               oti->oti_declare_ops_used[OSD_OT_DELETE]);
1898         CWARN("  ref_add: %u/%u/%u, ref_del: %u/%u/%u\n",
1899               oti->oti_declare_ops[OSD_OT_REF_ADD],
1900               oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
1901               oti->oti_declare_ops_used[OSD_OT_REF_ADD],
1902               oti->oti_declare_ops[OSD_OT_REF_DEL],
1903               oti->oti_declare_ops_cred[OSD_OT_REF_DEL],
1904               oti->oti_declare_ops_used[OSD_OT_REF_DEL]);
1905 }
1906
1907 /*
1908  * Concurrency: shouldn't matter.
1909  */
1910 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
1911                            struct thandle *th)
1912 {
1913         struct osd_thread_info *oti = osd_oti_get(env);
1914         struct osd_device *dev = osd_dt_dev(d);
1915         handle_t *jh;
1916         struct osd_thandle *oh;
1917         int rc;
1918
1919         ENTRY;
1920
1921         LASSERT(current->journal_info == NULL);
1922
1923         oh = container_of(th, struct osd_thandle, ot_super);
1924         LASSERT(oh != NULL);
1925         LASSERT(oh->ot_handle == NULL);
1926
1927         rc = dt_txn_hook_start(env, d, th);
1928         if (rc != 0)
1929                 GOTO(out, rc);
1930
1931         if (unlikely(osd_param_is_not_sane(dev, th))) {
1932                 static unsigned long last_printed;
1933                 static int last_credits;
1934
1935                 /*
1936                  * don't make noise on a tiny testing systems
1937                  * actual credits misuse will be caught anyway
1938                  */
1939                 if (last_credits != oh->ot_credits &&
1940                     time_after(jiffies, last_printed +
1941                                cfs_time_seconds(60)) &&
1942                     osd_transaction_size(dev) > 512) {
1943                         CWARN("%s: credits %u > trans_max %u\n", osd_name(dev),
1944                               oh->ot_credits, osd_transaction_size(dev));
1945                         osd_trans_dump_creds(env, th);
1946                         libcfs_debug_dumpstack(NULL);
1947                         last_credits = oh->ot_credits;
1948                         last_printed = jiffies;
1949                 }
1950                 /*
1951                  * XXX Limit the credits to 'max_transaction_buffers', and
1952                  *     let the underlying filesystem to catch the error if
1953                  *     we really need so many credits.
1954                  *
1955                  *     This should be removed when we can calculate the
1956                  *     credits precisely.
1957                  */
1958                 oh->ot_credits = osd_transaction_size(dev);
1959         } else if (ldiskfs_track_declares_assert != 0) {
1960                 /*
1961                  * reserve few credits to prevent an assertion in JBD
1962                  * our debugging mechanism will be able to detected
1963                  * overuse. this can help to debug single-update
1964                  * transactions
1965                  */
1966                 oh->ot_credits += 10;
1967                 if (unlikely(osd_param_is_not_sane(dev, th)))
1968                         oh->ot_credits = osd_transaction_size(dev);
1969         }
1970
1971         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_TXN_START))
1972                 GOTO(out, rc = -EIO);
1973
1974         /*
1975          * XXX temporary stuff. Some abstraction layer should
1976          * be used.
1977          */
1978         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
1979         osd_th_started(oh);
1980         if (!IS_ERR(jh)) {
1981                 oh->ot_handle = jh;
1982                 LASSERT(oti->oti_txns == 0);
1983
1984                 atomic_inc(&dev->od_commit_cb_in_flight);
1985                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
1986                               "osd-tx", th);
1987                 oti->oti_txns++;
1988                 rc = 0;
1989         } else {
1990                 rc = PTR_ERR(jh);
1991         }
1992 out:
1993         RETURN(rc);
1994 }
1995
1996 static int osd_seq_exists(const struct lu_env *env,
1997                           struct osd_device *osd, u64 seq)
1998 {
1999         struct lu_seq_range *range = &osd_oti_get(env)->oti_seq_range;
2000         struct seq_server_site *ss = osd_seq_site(osd);
2001         int rc;
2002
2003         ENTRY;
2004
2005         LASSERT(ss != NULL);
2006         LASSERT(ss->ss_server_fld != NULL);
2007
2008         rc = osd_fld_lookup(env, osd, seq, range);
2009         if (rc != 0) {
2010                 if (rc != -ENOENT)
2011                         CERROR("%s: can't lookup FLD sequence %#llx: rc = %d\n",
2012                                osd_name(osd), seq, rc);
2013                 RETURN(0);
2014         }
2015
2016         RETURN(ss->ss_node_id == range->lsr_index);
2017 }
2018
2019 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
2020 {
2021         struct dt_txn_commit_cb *dcb;
2022         struct dt_txn_commit_cb *tmp;
2023
2024         /* call per-transaction stop callbacks if any */
2025         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
2026                                  dcb_linkage) {
2027                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
2028                          "commit callback entry: magic=%x name='%s'\n",
2029                          dcb->dcb_magic, dcb->dcb_name);
2030                 list_del_init(&dcb->dcb_linkage);
2031                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
2032         }
2033 }
2034
2035 /*
2036  * Concurrency: shouldn't matter.
2037  */
2038 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
2039                           struct thandle *th)
2040 {
2041         struct osd_thread_info *oti = osd_oti_get(env);
2042         struct osd_thandle *oh;
2043         struct osd_iobuf *iobuf = &oti->oti_iobuf;
2044         struct osd_device *osd = osd_dt_dev(th->th_dev);
2045         struct qsd_instance *qsd = osd_def_qsd(osd);
2046         struct lquota_trans *qtrans;
2047         LIST_HEAD(truncates);
2048         int rc = 0, remove_agents = 0;
2049
2050         ENTRY;
2051
2052         oh = container_of(th, struct osd_thandle, ot_super);
2053
2054         remove_agents = oh->ot_remove_agents;
2055
2056         qtrans = oh->ot_quota_trans;
2057         oh->ot_quota_trans = NULL;
2058
2059         /* move locks to local list, stop tx, execute truncates */
2060         list_splice(&oh->ot_trunc_locks, &truncates);
2061
2062         if (oh->ot_handle != NULL) {
2063                 int rc2;
2064
2065                 handle_t *hdl = oh->ot_handle;
2066
2067                 /*
2068                  * add commit callback
2069                  * notice we don't do this in osd_trans_start()
2070                  * as underlying transaction can change during truncate
2071                  */
2072                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
2073                                              &oh->ot_jcb);
2074
2075                 LASSERT(oti->oti_txns == 1);
2076                 oti->oti_txns--;
2077
2078                 rc = dt_txn_hook_stop(env, th);
2079                 if (rc != 0)
2080                         CERROR("%s: failed in transaction hook: rc = %d\n",
2081                                osd_name(osd), rc);
2082
2083                 osd_trans_stop_cb(oh, rc);
2084                 /* hook functions might modify th_sync */
2085                 hdl->h_sync = th->th_sync;
2086
2087                 oh->ot_handle = NULL;
2088                 OSD_CHECK_SLOW_TH(oh, osd, rc2 = ldiskfs_journal_stop(hdl));
2089                 if (rc2 != 0)
2090                         CERROR("%s: failed to stop transaction: rc = %d\n",
2091                                osd_name(osd), rc2);
2092                 if (!rc)
2093                         rc = rc2;
2094
2095                 osd_process_truncates(&truncates);
2096         } else {
2097                 osd_trans_stop_cb(oh, th->th_result);
2098                 OBD_FREE_PTR(oh);
2099         }
2100
2101         osd_trunc_unlock_all(env, &truncates);
2102
2103         /* inform the quota slave device that the transaction is stopping */
2104         qsd_op_end(env, qsd, qtrans);
2105
2106         /*
2107          * as we want IO to journal and data IO be concurrent, we don't block
2108          * awaiting data IO completion in osd_do_bio(), instead we wait here
2109          * once transaction is submitted to the journal. all reqular requests
2110          * don't do direct IO (except read/write), thus this wait_event becomes
2111          * no-op for them.
2112          *
2113          * IMPORTANT: we have to wait till any IO submited by the thread is
2114          * completed otherwise iobuf may be corrupted by different request
2115          */
2116         wait_event(iobuf->dr_wait,
2117                        atomic_read(&iobuf->dr_numreqs) == 0);
2118         osd_fini_iobuf(osd, iobuf);
2119         if (!rc)
2120                 rc = iobuf->dr_error;
2121
2122         if (unlikely(remove_agents != 0))
2123                 osd_process_scheduled_agent_removals(env, osd);
2124
2125         LASSERT(oti->oti_ins_cache_depth > 0);
2126         oti->oti_ins_cache_depth--;
2127         /* reset OI cache for safety */
2128         if (oti->oti_ins_cache_depth == 0)
2129                 oti->oti_ins_cache_used = 0;
2130
2131         sb_end_write(osd_sb(osd));
2132
2133         RETURN(rc);
2134 }
2135
2136 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
2137 {
2138         struct osd_thandle *oh = container_of(th, struct osd_thandle,
2139                                               ot_super);
2140
2141         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
2142         LASSERT(&dcb->dcb_func != NULL);
2143         if (dcb->dcb_flags & DCB_TRANS_STOP)
2144                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
2145         else
2146                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
2147
2148         return 0;
2149 }
2150
2151 /*
2152  * Called just before object is freed. Releases all resources except for
2153  * object itself (that is released by osd_object_free()).
2154  *
2155  * Concurrency: no concurrent access is possible that late in object
2156  * life-cycle.
2157  */
2158 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
2159 {
2160         struct osd_object *obj = osd_obj(l);
2161         struct qsd_instance *qsd = osd_def_qsd(osd_obj2dev(obj));
2162         struct inode *inode = obj->oo_inode;
2163         __u64 projid;
2164         qid_t uid;
2165         qid_t gid;
2166
2167         LINVRNT(osd_invariant(obj));
2168
2169         /*
2170          * If object is unlinked remove fid->ino mapping from object index.
2171          */
2172
2173         osd_index_fini(obj);
2174
2175         if (!inode)
2176                 return;
2177
2178         if (osd_has_index(obj) &&  obj->oo_dt.do_index_ops == &osd_index_iam_ops)
2179                 ldiskfs_set_inode_flag(inode, LDISKFS_INODE_JOURNAL_DATA);
2180
2181         uid = i_uid_read(inode);
2182         gid = i_gid_read(inode);
2183         projid = i_projid_read(inode);
2184
2185         obj->oo_inode = NULL;
2186         iput(inode);
2187
2188         /* do not rebalance quota if the caller needs to release memory
2189          * otherwise qsd_refresh_usage() may went into a new ldiskfs
2190          * transaction and risk to deadlock - LU-12178 */
2191         if (current->flags & (PF_MEMALLOC | PF_KSWAPD))
2192                 return;
2193
2194         if (!obj->oo_header && qsd) {
2195                 struct osd_thread_info *info = osd_oti_get(env);
2196                 struct lquota_id_info *qi = &info->oti_qi;
2197
2198                 /* Release granted quota to master if necessary */
2199                 qi->lqi_id.qid_uid = uid;
2200                 qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
2201
2202                 qi->lqi_id.qid_uid = gid;
2203                 qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
2204
2205                 qi->lqi_id.qid_uid = projid;
2206                 qsd_op_adjust(env, qsd, &qi->lqi_id, PRJQUOTA);
2207         }
2208 }
2209
2210 /*
2211  * Concurrency: ->loo_object_release() is called under site spin-lock.
2212  */
2213 static void osd_object_release(const struct lu_env *env,
2214                                struct lu_object *l)
2215 {
2216         struct osd_object *o = osd_obj(l);
2217
2218         /*
2219          * nobody should be releasing a non-destroyed object with nlink=0
2220          * the API allows this, but ldiskfs doesn't like and then report
2221          * this inode as deleted
2222          */
2223         LASSERT(!(o->oo_destroyed == 0 && o->oo_inode &&
2224                   o->oo_inode->i_nlink == 0));
2225 }
2226
2227 /*
2228  * Concurrency: shouldn't matter.
2229  */
2230 static int osd_object_print(const struct lu_env *env, void *cookie,
2231                             lu_printer_t p, const struct lu_object *l)
2232 {
2233         struct osd_object *o = osd_obj(l);
2234         struct iam_descr *d;
2235
2236         if (o->oo_dir != NULL)
2237                 d = o->oo_dir->od_container.ic_descr;
2238         else
2239                 d = NULL;
2240         return (*p)(env, cookie,
2241                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
2242                     o, o->oo_inode,
2243                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
2244                     o->oo_inode ? o->oo_inode->i_generation : 0,
2245                     d ? d->id_ops->id_name : "plain");
2246 }
2247
2248 /*
2249  * Concurrency: shouldn't matter.
2250  */
2251 int osd_statfs(const struct lu_env *env, struct dt_device *d,
2252                 struct obd_statfs *sfs, struct obd_statfs_info *info)
2253 {
2254         struct osd_device *osd = osd_dt_dev(d);
2255         struct super_block *sb = osd_sb(osd);
2256         struct kstatfs *ksfs;
2257         __u64 reserved;
2258         int result = 0;
2259
2260         if (unlikely(osd->od_mnt == NULL))
2261                 return -EINPROGRESS;
2262
2263         /* osd_lproc.c call this without env, allocate ksfs for that case */
2264         if (unlikely(env == NULL)) {
2265                 OBD_ALLOC_PTR(ksfs);
2266                 if (ksfs == NULL)
2267                         return -ENOMEM;
2268         } else {
2269                 ksfs = &osd_oti_get(env)->oti_ksfs;
2270         }
2271
2272         result = sb->s_op->statfs(sb->s_root, ksfs);
2273         if (result)
2274                 goto out;
2275
2276         statfs_pack(sfs, ksfs);
2277         if (unlikely(sb->s_flags & SB_RDONLY))
2278                 sfs->os_state |= OS_STATFS_READONLY;
2279
2280         sfs->os_state |= osd->od_nonrotational ? OS_STATFS_NONROT : 0;
2281
2282         if (ldiskfs_has_feature_extents(sb))
2283                 sfs->os_maxbytes = sb->s_maxbytes;
2284         else
2285                 sfs->os_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2286
2287         /*
2288          * Reserve some space so to avoid fragmenting the filesystem too much.
2289          * Fragmentation not only impacts performance, but can also increase
2290          * metadata overhead significantly, causing grant calculation to be
2291          * wrong.
2292          *
2293          * Reserve 0.78% of total space, at least 8MB for small filesystems.
2294          */
2295         BUILD_BUG_ON(OSD_STATFS_RESERVED <= LDISKFS_MAX_BLOCK_SIZE);
2296         reserved = OSD_STATFS_RESERVED >> sb->s_blocksize_bits;
2297         if (likely(sfs->os_blocks >= reserved << OSD_STATFS_RESERVED_SHIFT))
2298                 reserved = sfs->os_blocks >> OSD_STATFS_RESERVED_SHIFT;
2299
2300         sfs->os_blocks -= reserved;
2301         sfs->os_bfree  -= min(reserved, sfs->os_bfree);
2302         sfs->os_bavail -= min(reserved, sfs->os_bavail);
2303
2304 out:
2305         if (unlikely(env == NULL))
2306                 OBD_FREE_PTR(ksfs);
2307         return result;
2308 }
2309
2310 /**
2311  * Estimate space needed for file creations. We assume the largest filename
2312  * which is 2^64 - 1, hence a filename of 20 chars.
2313  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
2314  */
2315 #ifdef __LDISKFS_DIR_REC_LEN
2316 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
2317 #else
2318 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
2319 #endif
2320
2321 /*
2322  * Concurrency: doesn't access mutable data.
2323  */
2324 static void osd_conf_get(const struct lu_env *env,
2325                          const struct dt_device *dev,
2326                          struct dt_device_param *param)
2327 {
2328         struct osd_device *d = osd_dt_dev(dev);
2329         struct super_block *sb = osd_sb(d);
2330         struct blk_integrity *bi = bdev_get_integrity(sb->s_bdev);
2331         const char *name;
2332         int ea_overhead;
2333
2334         /*
2335          * XXX should be taken from not-yet-existing fs abstraction layer.
2336          */
2337         param->ddp_max_name_len = LDISKFS_NAME_LEN;
2338         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
2339         param->ddp_symlink_max  = sb->s_blocksize;
2340         param->ddp_mount_type   = LDD_MT_LDISKFS;
2341         if (ldiskfs_has_feature_extents(sb))
2342                 param->ddp_maxbytes = sb->s_maxbytes;
2343         else
2344                 param->ddp_maxbytes = LDISKFS_SB(sb)->s_bitmap_maxbytes;
2345         /*
2346          * inode are statically allocated, so per-inode space consumption
2347          * is the space consumed by the directory entry
2348          */
2349         param->ddp_inodespace     = PER_OBJ_USAGE;
2350         /*
2351          * EXT_INIT_MAX_LEN is the theoretical maximum extent size (32k blocks
2352          * is 128MB) which is unlikely to be hit in real life. Report a smaller
2353          * maximum length to not under-count the actual number of extents
2354          * needed for writing a file if there are sub-optimal block allocations.
2355          */
2356         param->ddp_max_extent_blks = EXT_INIT_MAX_LEN >> 1;
2357         /* worst-case extent insertion metadata overhead */
2358         param->ddp_extent_tax = 6 * LDISKFS_BLOCK_SIZE(sb);
2359         param->ddp_mntopts = 0;
2360         if (test_opt(sb, XATTR_USER))
2361                 param->ddp_mntopts |= MNTOPT_USERXATTR;
2362         if (test_opt(sb, POSIX_ACL))
2363                 param->ddp_mntopts |= MNTOPT_ACL;
2364
2365         /*
2366          * LOD might calculate the max stripe count based on max_ea_size,
2367          * so we need take account in the overhead as well,
2368          * xattr_header + magic + xattr_entry_head
2369          */
2370         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
2371                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
2372
2373 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
2374         if (ldiskfs_has_feature_ea_inode(sb))
2375                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
2376                                                                 ea_overhead;
2377         else
2378 #endif
2379                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
2380
2381         if (param->ddp_max_ea_size > OBD_MAX_EA_SIZE)
2382                 param->ddp_max_ea_size = OBD_MAX_EA_SIZE;
2383
2384         /*
2385          * Preferred RPC size for efficient disk IO.  4MB shows good
2386          * all-around performance for ldiskfs, but use bigalloc chunk size
2387          * by default if larger.
2388          */
2389 #if defined(LDISKFS_CLUSTER_SIZE)
2390         if (LDISKFS_CLUSTER_SIZE(sb) > DT_DEF_BRW_SIZE)
2391                 param->ddp_brw_size = LDISKFS_CLUSTER_SIZE(sb);
2392         else
2393 #endif
2394                 param->ddp_brw_size = DT_DEF_BRW_SIZE;
2395
2396         param->ddp_t10_cksum_type = 0;
2397         if (bi) {
2398                 unsigned short interval = blk_integrity_interval(bi);
2399                 name = blk_integrity_name(bi);
2400                 /*
2401                  * Expected values:
2402                  * T10-DIF-TYPE1-CRC
2403                  * T10-DIF-TYPE3-CRC
2404                  * T10-DIF-TYPE1-IP
2405                  * T10-DIF-TYPE3-IP
2406                  */
2407                 if (strncmp(name, "T10-DIF-TYPE",
2408                             sizeof("T10-DIF-TYPE") - 1) == 0) {
2409                         /* also skip "1/3-" at end */
2410                         const int type_off = sizeof("T10-DIF-TYPE.");
2411                         char type_number = name[type_off - 2];
2412
2413                         if (interval != 512 && interval != 4096) {
2414                                 CERROR("%s: unsupported T10PI sector size %u\n",
2415                                        d->od_svname, interval);
2416                         } else if (type_number != '1' && type_number != '3') {
2417                                 CERROR("%s: unsupported T10PI type %s\n",
2418                                        d->od_svname, name);
2419                         } else if (strcmp(name + type_off, "CRC") == 0) {
2420                                 d->od_t10_type = type_number == '1' ?
2421                                         OSD_T10_TYPE1_CRC : OSD_T10_TYPE3_CRC;
2422                                 param->ddp_t10_cksum_type = interval == 512 ?
2423                                         OBD_CKSUM_T10CRC512 :
2424                                         OBD_CKSUM_T10CRC4K;
2425                         } else if (strcmp(name + type_off, "IP") == 0) {
2426                                 d->od_t10_type = type_number == '1' ?
2427                                         OSD_T10_TYPE1_IP : OSD_T10_TYPE3_IP;
2428                                 param->ddp_t10_cksum_type = interval == 512 ?
2429                                         OBD_CKSUM_T10IP512 :
2430                                         OBD_CKSUM_T10IP4K;
2431                         } else {
2432                                 CERROR("%s: unsupported checksum type of "
2433                                        "T10PI type '%s'",
2434                                        d->od_svname, name);
2435                         }
2436
2437                 } else {
2438                         CERROR("%s: unsupported T10PI type '%s'",
2439                                d->od_svname, name);
2440                 }
2441         }
2442
2443         param->ddp_has_lseek_data_hole = true;
2444 }
2445
2446 static struct super_block *osd_mnt_sb_get(const struct dt_device *d)
2447 {
2448         return osd_sb(osd_dt_dev(d));
2449 }
2450
2451 /*
2452  * Concurrency: shouldn't matter.
2453  */
2454 static int osd_sync(const struct lu_env *env, struct dt_device *d)
2455 {
2456         int rc;
2457         struct super_block *s = osd_sb(osd_dt_dev(d));
2458         ENTRY;
2459
2460         down_read(&s->s_umount);
2461         rc = s->s_op->sync_fs(s, 1);
2462         up_read(&s->s_umount);
2463
2464         CDEBUG(D_CACHE, "%s: synced OSD: rc = %d\n", osd_dt_dev(d)->od_svname,
2465                rc);
2466
2467         return rc;
2468 }
2469
2470 /**
2471  * Start commit for OSD device.
2472  *
2473  * An implementation of dt_commit_async method for OSD device.
2474  * Asychronously starts underlayng fs sync and thereby a transaction
2475  * commit.
2476  *
2477  * \param env environment
2478  * \param d dt device
2479  *
2480  * \see dt_device_operations
2481  */
2482 static int osd_commit_async(const struct lu_env *env,
2483                             struct dt_device *d)
2484 {
2485         struct super_block *s = osd_sb(osd_dt_dev(d));
2486         int rc;
2487
2488         ENTRY;
2489
2490         CDEBUG(D_HA, "%s: async commit OSD\n", osd_dt_dev(d)->od_svname);
2491         down_read(&s->s_umount);
2492         rc = s->s_op->sync_fs(s, 0);
2493         up_read(&s->s_umount);
2494
2495         RETURN(rc);
2496 }
2497
2498 static int (*priv_security_file_alloc)(struct file *file);
2499
2500 int osd_security_file_alloc(struct file *file)
2501 {
2502         if (priv_security_file_alloc)
2503                 return priv_security_file_alloc(file);
2504         return 0;
2505 }
2506
2507 /*
2508  * Concurrency: shouldn't matter.
2509  */
2510 static int osd_ro(const struct lu_env *env, struct dt_device *d)
2511 {
2512         struct super_block *sb = osd_sb(osd_dt_dev(d));
2513         struct block_device *dev = sb->s_bdev;
2514         int rc = -EOPNOTSUPP;
2515
2516         ENTRY;
2517
2518         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
2519                osd_dt_dev(d)->od_svname, (long)dev, rc);
2520
2521         RETURN(rc);
2522 }
2523
2524 /**
2525  * Note: we do not count into QUOTA here.
2526  * If we mount with --data_journal we may need more.
2527  */
2528 const int osd_dto_credits_noquota[DTO_NR] = {
2529         /**
2530          * Insert.
2531          * INDEX_EXTRA_TRANS_BLOCKS(8) +
2532          * SINGLEDATA_TRANS_BLOCKS(8)
2533          * XXX Note: maybe iam need more, since iam have more level than
2534          *           EXT3 htree.
2535          */
2536         [DTO_INDEX_INSERT]  = 16,
2537         /**
2538          * Delete
2539          * just modify a single entry, probably merge few within a block
2540          */
2541         [DTO_INDEX_DELETE]  = 1,
2542         /**
2543          * Used for OI scrub
2544          */
2545         [DTO_INDEX_UPDATE]  = 16,
2546         /**
2547          * 4(inode, inode bits, groups, GDT)
2548          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
2549          */
2550         [DTO_OBJECT_CREATE] = 4,
2551         /**
2552          * 4(inode, inode bits, groups, GDT)
2553          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
2554          */
2555         [DTO_OBJECT_DELETE] = 4,
2556         /**
2557          * Attr set credits (inode)
2558          */
2559         [DTO_ATTR_SET_BASE] = 1,
2560         /**
2561          * Xattr set. The same as xattr of EXT3.
2562          * DATA_TRANS_BLOCKS(14)
2563          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
2564          * are also counted in. Do not know why?
2565          */
2566         [DTO_XATTR_SET]     = 14,
2567         /**
2568          * credits for inode change during write.
2569          */
2570         [DTO_WRITE_BASE]    = 3,
2571         /**
2572          * credits for single block write.
2573          */
2574         [DTO_WRITE_BLOCK]   = 14,
2575         /**
2576          * Attr set credits for chown.
2577          * This is extra credits for setattr, and it is null without quota
2578          */
2579         [DTO_ATTR_SET_CHOWN] = 0
2580 };
2581
2582 static const struct dt_device_operations osd_dt_ops = {
2583         .dt_root_get       = osd_root_get,
2584         .dt_statfs         = osd_statfs,
2585         .dt_trans_create   = osd_trans_create,
2586         .dt_trans_start    = osd_trans_start,
2587         .dt_trans_stop     = osd_trans_stop,
2588         .dt_trans_cb_add   = osd_trans_cb_add,
2589         .dt_conf_get       = osd_conf_get,
2590         .dt_mnt_sb_get     = osd_mnt_sb_get,
2591         .dt_sync           = osd_sync,
2592         .dt_ro             = osd_ro,
2593         .dt_commit_async   = osd_commit_async,
2594 };
2595
2596 static void osd_read_lock(const struct lu_env *env, struct dt_object *dt,
2597                           unsigned int role)
2598 {
2599         struct osd_object *obj = osd_dt_obj(dt);
2600         struct osd_thread_info *oti = osd_oti_get(env);
2601
2602         LINVRNT(osd_invariant(obj));
2603
2604         LASSERT(obj->oo_owner != env);
2605         down_read_nested(&obj->oo_sem, role);
2606
2607         LASSERT(obj->oo_owner == NULL);
2608         oti->oti_r_locks++;
2609 }
2610
2611 static void osd_write_lock(const struct lu_env *env, struct dt_object *dt,
2612                            unsigned int role)
2613 {
2614         struct osd_object *obj = osd_dt_obj(dt);
2615         struct osd_thread_info *oti = osd_oti_get(env);
2616
2617         LINVRNT(osd_invariant(obj));
2618
2619         LASSERT(obj->oo_owner != env);
2620         down_write_nested(&obj->oo_sem, role);
2621
2622         LASSERT(obj->oo_owner == NULL);
2623         obj->oo_owner = env;
2624         oti->oti_w_locks++;
2625 }
2626
2627 static void osd_read_unlock(const struct lu_env *env, struct dt_object *dt)
2628 {
2629         struct osd_object *obj = osd_dt_obj(dt);
2630         struct osd_thread_info *oti = osd_oti_get(env);
2631
2632         LINVRNT(osd_invariant(obj));
2633
2634         LASSERT(oti->oti_r_locks > 0);
2635         oti->oti_r_locks--;
2636         up_read(&obj->oo_sem);
2637 }
2638
2639 static void osd_write_unlock(const struct lu_env *env, struct dt_object *dt)
2640 {
2641         struct osd_object *obj = osd_dt_obj(dt);
2642         struct osd_thread_info *oti = osd_oti_get(env);
2643
2644         LINVRNT(osd_invariant(obj));
2645
2646         LASSERT(obj->oo_owner == env);
2647         LASSERT(oti->oti_w_locks > 0);
2648         oti->oti_w_locks--;
2649         obj->oo_owner = NULL;
2650         up_write(&obj->oo_sem);
2651 }
2652
2653 static int osd_write_locked(const struct lu_env *env, struct dt_object *dt)
2654 {
2655         struct osd_object *obj = osd_dt_obj(dt);
2656
2657         LINVRNT(osd_invariant(obj));
2658
2659         return obj->oo_owner == env;
2660 }
2661
2662 static void osd_inode_getattr(const struct lu_env *env,
2663                               struct inode *inode, struct lu_attr *attr)
2664 {
2665         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
2666                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
2667                            LA_PROJID | LA_FLAGS | LA_NLINK | LA_RDEV |
2668                            LA_BLKSIZE | LA_TYPE | LA_BTIME;
2669
2670         attr->la_atime = inode->i_atime.tv_sec;
2671         attr->la_mtime = inode->i_mtime.tv_sec;
2672         attr->la_ctime = inode->i_ctime.tv_sec;
2673         attr->la_btime = LDISKFS_I(inode)->i_crtime.tv_sec;
2674         attr->la_mode    = inode->i_mode;
2675         attr->la_size    = i_size_read(inode);
2676         attr->la_blocks  = inode->i_blocks;
2677         attr->la_uid     = i_uid_read(inode);
2678         attr->la_gid     = i_gid_read(inode);
2679         attr->la_projid  = i_projid_read(inode);
2680         attr->la_flags   = ll_inode_to_ext_flags(inode->i_flags);
2681         attr->la_nlink   = inode->i_nlink;
2682         attr->la_rdev    = inode->i_rdev;
2683         attr->la_blksize = 1 << inode->i_blkbits;
2684         attr->la_blkbits = inode->i_blkbits;
2685         /*
2686          * Ext4 did not transfer inherit flags from raw inode
2687          * to inode flags, and ext4 internally test raw inode
2688          * @i_flags directly. Instead of patching ext4, we do it here.
2689          */
2690         if (LDISKFS_I(inode)->i_flags & LUSTRE_PROJINHERIT_FL)
2691                 attr->la_flags |= LUSTRE_PROJINHERIT_FL;
2692 }
2693
2694 static int osd_dirent_count(const struct lu_env *env, struct dt_object *dt,
2695                             u64 *count)
2696 {
2697         struct osd_object *obj = osd_dt_obj(dt);
2698         const struct dt_it_ops *iops;
2699         struct dt_it *it;
2700         int rc;
2701
2702         ENTRY;
2703
2704         LASSERT(S_ISDIR(obj->oo_inode->i_mode));
2705         LASSERT(fid_is_namespace_visible(lu_object_fid(&obj->oo_dt.do_lu)));
2706
2707         if (obj->oo_dirent_count != LU_DIRENT_COUNT_UNSET) {
2708                 *count = obj->oo_dirent_count;
2709                 RETURN(0);
2710         }
2711
2712         /* directory not initialized yet */
2713         if (!dt->do_index_ops) {
2714                 *count = 0;
2715                 RETURN(0);
2716         }
2717
2718         iops = &dt->do_index_ops->dio_it;
2719         it = iops->init(env, dt, LUDA_64BITHASH);
2720         if (IS_ERR(it))
2721                 RETURN(PTR_ERR(it));
2722
2723         rc = iops->load(env, it, 0);
2724         if (rc < 0) {
2725                 if (rc == -ENODATA) {
2726                         rc = 0;
2727                         *count = 0;
2728                 }
2729                 GOTO(out, rc);
2730         }
2731         if (rc > 0)
2732                 rc = iops->next(env, it);
2733
2734         for (*count = 0; rc == 0 || rc == -ESTALE; rc = iops->next(env, it)) {
2735                 if (rc == -ESTALE)
2736                         continue;
2737
2738                 if (iops->key_size(env, it) == 0)
2739                         continue;
2740
2741                 (*count)++;
2742         }
2743         if (rc == 1) {
2744                 obj->oo_dirent_count = *count;
2745                 rc = 0;
2746         }
2747 out:
2748         iops->put(env, it);
2749         iops->fini(env, it);
2750
2751         RETURN(rc);
2752 }
2753
2754 static int osd_attr_get(const struct lu_env *env, struct dt_object *dt,
2755                         struct lu_attr *attr)
2756 {
2757         struct osd_object *obj = osd_dt_obj(dt);
2758         int rc = 0;
2759
2760         if (unlikely(!dt_object_exists(dt)))
2761                 return -ENOENT;
2762         if (unlikely(obj->oo_destroyed))
2763                 return -ENOENT;
2764
2765         LASSERT(!dt_object_remote(dt));
2766         LINVRNT(osd_invariant(obj));
2767
2768         spin_lock(&obj->oo_guard);
2769         osd_inode_getattr(env, obj->oo_inode, attr);
2770         if (obj->oo_lma_flags & LUSTRE_ORPHAN_FL) {
2771                 attr->la_valid |= LA_FLAGS;
2772                 attr->la_flags |= LUSTRE_ORPHAN_FL;
2773         }
2774         if (obj->oo_lma_flags & LUSTRE_ENCRYPT_FL) {
2775                 attr->la_valid |= LA_FLAGS;
2776                 attr->la_flags |= LUSTRE_ENCRYPT_FL;
2777         }
2778         spin_unlock(&obj->oo_guard);
2779
2780         if (S_ISDIR(obj->oo_inode->i_mode) &&
2781             fid_is_namespace_visible(lu_object_fid(&dt->do_lu)))
2782                 rc = osd_dirent_count(env, dt, &attr->la_dirent_count);
2783
2784         return rc;
2785 }
2786
2787 static int osd_declare_attr_qid(const struct lu_env *env,
2788                                 struct osd_object *obj,
2789                                 struct osd_thandle *oh, long long bspace,
2790                                 qid_t old_id, qid_t new_id, bool enforce,
2791                                 unsigned int type, bool ignore_edquot)
2792 {
2793         int rc;
2794         struct osd_thread_info *info = osd_oti_get(env);
2795         struct lquota_id_info  *qi = &info->oti_qi;
2796
2797         qi->lqi_type = type;
2798         /* inode accounting */
2799         qi->lqi_is_blk = false;
2800
2801         /* one more inode for the new id ... */
2802         qi->lqi_id.qid_uid = new_id;
2803         qi->lqi_space      = 1;
2804         /* Reserve credits for the new id */
2805         rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
2806         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2807                 rc = 0;
2808         if (rc)
2809                 RETURN(rc);
2810
2811         /* and one less inode for the current id */
2812         qi->lqi_id.qid_uid = old_id;
2813         qi->lqi_space = -1;
2814         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2815         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2816                 rc = 0;
2817         if (rc)
2818                 RETURN(rc);
2819
2820         /* block accounting */
2821         qi->lqi_is_blk = true;
2822
2823         /* more blocks for the new id ... */
2824         qi->lqi_id.qid_uid = new_id;
2825         qi->lqi_space      = bspace;
2826         /*
2827          * Credits for the new uid has been reserved, re-use "obj"
2828          * to save credit reservation.
2829          */
2830         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2831         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2832                 rc = 0;
2833         if (rc)
2834                 RETURN(rc);
2835
2836         /* and finally less blocks for the current uid */
2837         qi->lqi_id.qid_uid = old_id;
2838         qi->lqi_space      = -bspace;
2839         rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
2840         if (ignore_edquot && (rc == -EDQUOT || rc == -EINPROGRESS))
2841                 rc = 0;
2842
2843         RETURN(rc);
2844 }
2845
2846 static int osd_declare_attr_set(const struct lu_env *env,
2847                                 struct dt_object *dt,
2848                                 const struct lu_attr *attr,
2849                                 struct thandle *handle)
2850 {
2851         struct osd_thandle *oh;
2852         struct osd_object *obj;
2853         qid_t uid;
2854         qid_t gid;
2855         long long bspace;
2856         int rc = 0;
2857         bool enforce;
2858
2859         ENTRY;
2860
2861         LASSERT(dt != NULL);
2862         LASSERT(handle != NULL);
2863
2864         obj = osd_dt_obj(dt);
2865         LASSERT(osd_invariant(obj));
2866
2867         oh = container_of(handle, struct osd_thandle, ot_super);
2868         LASSERT(oh->ot_handle == NULL);
2869
2870         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
2871                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2872
2873         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2874                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2875
2876         if (attr == NULL || obj->oo_inode == NULL)
2877                 RETURN(rc);
2878
2879         bspace   = obj->oo_inode->i_blocks << 9;
2880         bspace   = toqb(bspace);
2881
2882         /*
2883          * Changing ownership is always preformed by super user, it should not
2884          * fail with EDQUOT unless required explicitly.
2885          *
2886          * We still need to call the osd_declare_qid() to calculate the journal
2887          * credits for updating quota accounting files and to trigger quota
2888          * space adjustment once the operation is completed.
2889          */
2890         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
2891                 bool ignore_edquot = !(attr->la_flags & LUSTRE_SET_SYNC_FL);
2892
2893                 if (!ignore_edquot)
2894                         CDEBUG(D_QUOTA,
2895                                "%s: enforce quota on UID %u, GID %u (quota space is %lld)\n",
2896                                osd_ino2name(obj->oo_inode), attr->la_uid,
2897                                attr->la_gid, bspace);
2898
2899                 /* USERQUOTA */
2900                 uid = i_uid_read(obj->oo_inode);
2901                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
2902                 rc = osd_declare_attr_qid(env, obj, oh, bspace, uid,
2903                                           attr->la_uid, enforce, USRQUOTA,
2904                                           true);
2905                 if (rc)
2906                         RETURN(rc);
2907
2908                 gid = i_gid_read(obj->oo_inode);
2909                 CDEBUG(D_QUOTA, "declare uid %d -> %d gid %d -> %d\n", uid,
2910                        attr->la_uid, gid, attr->la_gid);
2911                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
2912                 rc = osd_declare_attr_qid(env, obj, oh, bspace, gid,
2913                                           attr->la_gid, enforce, GRPQUOTA,
2914                                           ignore_edquot);
2915                 if (rc)
2916                         RETURN(rc);
2917
2918         }
2919 #ifdef HAVE_PROJECT_QUOTA
2920         if (attr->la_valid & LA_PROJID) {
2921                 __u32 projid = i_projid_read(obj->oo_inode);
2922
2923                 enforce = (attr->la_valid & LA_PROJID) &&
2924                                         (attr->la_projid != projid);
2925                 rc = osd_declare_attr_qid(env, obj, oh, bspace,
2926                                           (qid_t)projid, (qid_t)attr->la_projid,
2927                                           enforce, PRJQUOTA, true);
2928                 if (rc)
2929                         RETURN(rc);
2930         }
2931 #endif
2932         RETURN(rc);
2933 }
2934
2935 static int osd_inode_setattr(const struct lu_env *env,
2936                              struct inode *inode, const struct lu_attr *attr)
2937 {
2938         __u64 bits = attr->la_valid;
2939
2940         /* Only allow set size for regular file */
2941         if (!S_ISREG(inode->i_mode))
2942                 bits &= ~(LA_SIZE | LA_BLOCKS);
2943
2944         if (bits == 0)
2945                 return 0;
2946
2947         if (bits & LA_ATIME)
2948                 inode->i_atime = osd_inode_time(inode, attr->la_atime);
2949         if (bits & LA_CTIME)
2950                 inode->i_ctime = osd_inode_time(inode, attr->la_ctime);
2951         if (bits & LA_MTIME)
2952                 inode->i_mtime = osd_inode_time(inode, attr->la_mtime);
2953         if (bits & LA_SIZE) {
2954                 spin_lock(&inode->i_lock);
2955                 LDISKFS_I(inode)->i_disksize = attr->la_size;
2956                 i_size_write(inode, attr->la_size);
2957                 spin_unlock(&inode->i_lock);
2958         }
2959
2960         /*
2961          * OSD should not change "i_blocks" which is used by quota.
2962          * "i_blocks" should be changed by ldiskfs only.
2963          */
2964         if (bits & LA_MODE)
2965                 inode->i_mode = (inode->i_mode & S_IFMT) |
2966                                 (attr->la_mode & ~S_IFMT);
2967         if (bits & LA_UID)
2968                 i_uid_write(inode, attr->la_uid);
2969         if (bits & LA_GID)
2970                 i_gid_write(inode, attr->la_gid);
2971         if (bits & LA_PROJID)
2972                 i_projid_write(inode, attr->la_projid);
2973         if (bits & LA_NLINK)
2974                 set_nlink(inode, attr->la_nlink);
2975         if (bits & LA_RDEV)
2976                 inode->i_rdev = attr->la_rdev;
2977
2978         if (bits & LA_FLAGS) {
2979                 /* always keep S_NOCMTIME */
2980                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
2981                                  S_NOCMTIME;
2982 #if defined(S_ENCRYPTED)
2983                 /* Always remove S_ENCRYPTED, because ldiskfs must not be
2984                  * aware of encryption status. It is just stored into LMA
2985                  * so that it can be forwared to client side.
2986                  */
2987                 inode->i_flags &= ~S_ENCRYPTED;
2988 #endif
2989                 /*
2990                  * Ext4 did not transfer inherit flags from
2991                  * @inode->i_flags to raw inode i_flags when writing
2992                  * flags, we do it explictly here.
2993                  */
2994                 if (attr->la_flags & LUSTRE_PROJINHERIT_FL)
2995                         LDISKFS_I(inode)->i_flags |= LUSTRE_PROJINHERIT_FL;
2996                 else
2997                         LDISKFS_I(inode)->i_flags &= ~LUSTRE_PROJINHERIT_FL;
2998         }
2999         return 0;
3000 }
3001
3002 #ifdef HAVE_PROJECT_QUOTA
3003 static int osd_transfer_project(struct inode *inode, __u32 projid,
3004                                 struct thandle *handle)
3005 {
3006         struct super_block *sb = inode->i_sb;
3007         struct ldiskfs_inode_info *ei = LDISKFS_I(inode);
3008         int err;
3009         kprojid_t kprojid;
3010         struct ldiskfs_iloc iloc;
3011         struct ldiskfs_inode *raw_inode;
3012         struct dquot *transfer_to[LDISKFS_MAXQUOTAS] = { };
3013
3014         if (!ldiskfs_has_feature_project(sb)) {
3015                 LASSERT(__kprojid_val(LDISKFS_I(inode)->i_projid)
3016                         == LDISKFS_DEF_PROJID);
3017                 if (projid != LDISKFS_DEF_PROJID)
3018                         return -EOPNOTSUPP;
3019                 else
3020                         return 0;
3021         }
3022
3023         if (LDISKFS_INODE_SIZE(sb) <= LDISKFS_GOOD_OLD_INODE_SIZE)
3024                 return -EOPNOTSUPP;
3025
3026         kprojid = make_kprojid(&init_user_ns, (projid_t)projid);
3027         if (projid_eq(kprojid, LDISKFS_I(inode)->i_projid))
3028                 return 0;
3029
3030         err = ldiskfs_get_inode_loc(inode, &iloc);
3031         if (err)
3032                 return err;
3033
3034         raw_inode = ldiskfs_raw_inode(&iloc);
3035         if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3036                 struct osd_thandle *oh = container_of(handle,
3037                                                       struct osd_thandle,
3038                                                       ot_super);
3039                 /**
3040                  * try to expand inode size automatically.
3041                  */
3042                 ldiskfs_mark_inode_dirty(oh->ot_handle, inode);
3043                 if (!LDISKFS_FITS_IN_INODE(raw_inode, ei, i_projid)) {
3044                         err = -EOVERFLOW;
3045                         brelse(iloc.bh);
3046                         return err;
3047                 }
3048         }
3049         brelse(iloc.bh);
3050
3051         dquot_initialize(inode);
3052         transfer_to[PRJQUOTA] = dqget(sb, make_kqid_projid(kprojid));
3053         if (transfer_to[PRJQUOTA]) {
3054                 err = __dquot_transfer(inode, transfer_to);
3055                 dqput(transfer_to[PRJQUOTA]);
3056                 if (err)
3057                         return err;
3058         }
3059
3060         return err;
3061 }
3062 #endif
3063
3064 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr,
3065                               struct thandle *handle)
3066 {
3067         int rc;
3068
3069         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
3070             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
3071                 struct iattr iattr;
3072
3073                 CDEBUG(D_QUOTA,
3074                        "executing dquot_transfer inode %ld uid %d -> %d gid %d -> %d\n",
3075                        inode->i_ino, i_uid_read(inode), attr->la_uid,
3076                        i_gid_read(inode), attr->la_gid);
3077
3078                 dquot_initialize(inode);
3079                 iattr.ia_valid = 0;
3080                 if (attr->la_valid & LA_UID)
3081                         iattr.ia_valid |= ATTR_UID;
3082                 if (attr->la_valid & LA_GID)
3083                         iattr.ia_valid |= ATTR_GID;
3084                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
3085                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
3086
3087                 rc = dquot_transfer(inode, &iattr);
3088                 if (rc) {
3089                         CERROR("%s: quota transfer failed. Is quota enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3090                                osd_ino2name(inode), rc);
3091                         return rc;
3092                 }
3093         }
3094
3095         /* Handle project id transfer here properly */
3096         if (attr->la_valid & LA_PROJID &&
3097             attr->la_projid != i_projid_read(inode)) {
3098 #ifdef HAVE_PROJECT_QUOTA
3099                 rc = osd_transfer_project(inode, attr->la_projid, handle);
3100 #else
3101                 rc = -ENOTSUPP;
3102 #endif
3103                 if (rc) {
3104                         CERROR("%s: quota transfer failed. Is project enforcement enabled on the ldiskfs filesystem? rc = %d\n",
3105                                osd_ino2name(inode), rc);
3106                         return rc;
3107                 }
3108         }
3109         return 0;
3110 }
3111
3112 static int osd_attr_set(const struct lu_env *env,
3113                         struct dt_object *dt,
3114                         const struct lu_attr *attr,
3115                         struct thandle *handle)
3116 {
3117         struct osd_object *obj = osd_dt_obj(dt);
3118         struct inode *inode;
3119         int rc;
3120
3121         if (!dt_object_exists(dt))
3122                 return -ENOENT;
3123
3124         LASSERT(handle != NULL);
3125         LASSERT(!dt_object_remote(dt));
3126         LASSERT(osd_invariant(obj));
3127
3128         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
3129
3130         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING) &&
3131             !osd_obj2dev(obj)->od_is_ost) {
3132                 struct osd_thread_info *oti = osd_oti_get(env);
3133                 const struct lu_fid *fid0 = lu_object_fid(&dt->do_lu);
3134                 struct lu_fid *fid1 = &oti->oti_fid;
3135                 struct osd_inode_id *id = &oti->oti_id;
3136                 struct iam_path_descr *ipd;
3137                 struct iam_container *bag;
3138                 struct osd_thandle *oh;
3139                 int rc;
3140
3141                 fid_cpu_to_be(fid1, fid0);
3142                 memset(id, 1, sizeof(*id));
3143                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
3144                                   fid0)->oi_dir.od_container;
3145                 ipd = osd_idx_ipd_get(env, bag);
3146                 if (unlikely(ipd == NULL))
3147                         RETURN(-ENOMEM);
3148
3149                 oh = container_of(handle, struct osd_thandle, ot_super);
3150                 rc = iam_update(oh->ot_handle, bag,
3151                                 (const struct iam_key *)fid1,
3152                                 (const struct iam_rec *)id, ipd);
3153                 osd_ipd_put(env, bag, ipd);
3154                 return(rc > 0 ? 0 : rc);
3155         }
3156
3157         inode = obj->oo_inode;
3158
3159         rc = osd_quota_transfer(inode, attr, handle);
3160         if (rc)
3161                 return rc;
3162
3163         spin_lock(&obj->oo_guard);
3164         rc = osd_inode_setattr(env, inode, attr);
3165         spin_unlock(&obj->oo_guard);
3166         if (rc != 0)
3167                 GOTO(out, rc);
3168
3169         osd_dirty_inode(inode, I_DIRTY_DATASYNC);
3170
3171         osd_trans_exec_check(env, handle, OSD_OT_ATTR_SET);
3172
3173         if (!(attr->la_valid & LA_FLAGS))
3174                 GOTO(out, rc);
3175
3176         /* Let's check if there are extra flags need to be set into LMA */
3177         if (attr->la_flags & LUSTRE_LMA_FL_MASKS) {
3178                 struct osd_thread_info *info = osd_oti_get(env);
3179                 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
3180
3181                 LASSERT(!obj->oo_pfid_in_lma);
3182
3183