Whamcloud - gitweb
LU-812 ldiskfs: super_operations->dirty_inode now takes a flag
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_handler.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 only,
8  * as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope that it will be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * General Public License version 2 for more details (a copy is included
14  * in the LICENSE file that accompanied this code).
15  *
16  * You should have received a copy of the GNU General Public License
17  * version 2 along with this program; If not, see
18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19  *
20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21  * CA 95054 USA or visit www.sun.com if you need additional information or
22  * have any questions.
23  *
24  * GPL HEADER END
25  */
26 /*
27  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2011, 2012, Intel Corporation.
31  */
32 /*
33  * This file is part of Lustre, http://www.lustre.org/
34  * Lustre is a trademark of Sun Microsystems, Inc.
35  *
36  * lustre/osd/osd_handler.c
37  *
38  * Top-level entry points into osd module
39  *
40  * Author: Nikita Danilov <nikita@clusterfs.com>
41  *         Pravin Shelar <pravin.shelar@sun.com> : Added fid in dirent
42  */
43
44 #define DEBUG_SUBSYSTEM S_MDS
45
46 #include <linux/module.h>
47
48 /* LUSTRE_VERSION_CODE */
49 #include <lustre_ver.h>
50 /* prerequisite for linux/xattr.h */
51 #include <linux/types.h>
52 /* prerequisite for linux/xattr.h */
53 #include <linux/fs.h>
54 /* XATTR_{REPLACE,CREATE} */
55 #include <linux/xattr.h>
56
57 /*
58  * struct OBD_{ALLOC,FREE}*()
59  * OBD_FAIL_CHECK
60  */
61 #include <obd_support.h>
62 /* struct ptlrpc_thread */
63 #include <lustre_net.h>
64 #include <lustre_fid.h>
65
66 #include "osd_internal.h"
67
68 /* llo_* api support */
69 #include <md_object.h>
70 #include <lustre_quota.h>
71
72 int ldiskfs_pdo = 1;
73 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
74                 "ldiskfs with parallel directory operations");
75
76 static const char dot[] = ".";
77 static const char dotdot[] = "..";
78 static const char remote_obj_dir[] = "REM_OBJ_DIR";
79
80 static const struct lu_object_operations      osd_lu_obj_ops;
81 static const struct dt_object_operations      osd_obj_ops;
82 static const struct dt_object_operations      osd_obj_ea_ops;
83 static const struct dt_object_operations      osd_obj_otable_it_ops;
84 static const struct dt_index_operations       osd_index_iam_ops;
85 static const struct dt_index_operations       osd_index_ea_ops;
86
87 #ifdef OSD_TRACK_DECLARES
88 int osd_trans_declare_op2rb[] = {
89         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
90         [OSD_OT_PUNCH]          = OSD_OT_MAX,
91         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
92         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
93         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
94         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
95         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
96         [OSD_OT_WRITE]          = OSD_OT_WRITE,
97         [OSD_OT_INSERT]         = OSD_OT_DELETE,
98         [OSD_OT_DELETE]         = OSD_OT_INSERT,
99         [OSD_OT_QUOTA]          = OSD_OT_MAX,
100 };
101 #endif
102
103 static int osd_has_index(const struct osd_object *obj)
104 {
105         return obj->oo_dt.do_index_ops != NULL;
106 }
107
108 static int osd_object_invariant(const struct lu_object *l)
109 {
110         return osd_invariant(osd_obj(l));
111 }
112
113 /*
114  * Concurrency: doesn't matter
115  */
116 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
117 {
118         return osd_oti_get(env)->oti_r_locks > 0;
119 }
120
121 /*
122  * Concurrency: doesn't matter
123  */
124 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
125 {
126         struct osd_thread_info *oti = osd_oti_get(env);
127         return oti->oti_w_locks > 0 && o->oo_owner == env;
128 }
129
130 /*
131  * Concurrency: doesn't access mutable data
132  */
133 static int osd_root_get(const struct lu_env *env,
134                         struct dt_device *dev, struct lu_fid *f)
135 {
136         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
137         return 0;
138 }
139
140 /*
141  * OSD object methods.
142  */
143
144 /*
145  * Concurrency: no concurrent access is possible that early in object
146  * life-cycle.
147  */
148 static struct lu_object *osd_object_alloc(const struct lu_env *env,
149                                           const struct lu_object_header *hdr,
150                                           struct lu_device *d)
151 {
152         struct osd_object *mo;
153
154         OBD_ALLOC_PTR(mo);
155         if (mo != NULL) {
156                 struct lu_object *l;
157
158                 l = &mo->oo_dt.do_lu;
159                 dt_object_init(&mo->oo_dt, NULL, d);
160                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
161                 l->lo_ops = &osd_lu_obj_ops;
162                 init_rwsem(&mo->oo_sem);
163                 init_rwsem(&mo->oo_ext_idx_sem);
164                 spin_lock_init(&mo->oo_guard);
165                 return l;
166         } else {
167                 return NULL;
168         }
169 }
170
171 static inline int __osd_xattr_get(struct inode *inode, struct dentry *dentry,
172                                   const char *name, void *buf, int len)
173 {
174         dentry->d_inode = inode;
175         return inode->i_op->getxattr(dentry, name, buf, len);
176 }
177
178 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
179                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
180 {
181         int rc;
182
183         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA, (void *)lma,
184                              sizeof(*lma));
185         if (rc == -ERANGE) {
186                 /* try with old lma size */
187                 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
188                                            info->oti_mdt_attrs_old,
189                                            LMA_OLD_SIZE);
190                 if (rc > 0)
191                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
192         }
193         if (rc > 0) {
194                 /* Check LMA compatibility */
195                 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
196                         CWARN("%.16s: unsupported incompat LMA feature(s) "
197                               "%lx/%#x\n",
198                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
199                               inode->i_ino, le32_to_cpu(lma->lma_incompat) &
200                                                         ~LMA_INCOMPAT_SUPP);
201                         rc = -ENOSYS;
202                 } else {
203                         lustre_lma_swab(lma);
204                         rc = 0;
205                 }
206         } else if (rc == 0) {
207                 rc = -ENODATA;
208         }
209
210         return rc;
211 }
212
213 /*
214  * retrieve object from backend ext fs.
215  **/
216 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
217                        struct osd_inode_id *id)
218 {
219         struct inode *inode = NULL;
220
221         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
222         if (IS_ERR(inode)) {
223                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
224                        id->oii_ino, PTR_ERR(inode));
225         } else if (id->oii_gen != OSD_OII_NOGEN &&
226                    inode->i_generation != id->oii_gen) {
227                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
228                        "gen1 = %u\n",
229                        id->oii_ino, id->oii_gen, inode->i_generation);
230                 iput(inode);
231                 inode = ERR_PTR(-ESTALE);
232         } else if (inode->i_nlink == 0) {
233                 /* due to parallel readdir and unlink,
234                 * we can have dead inode here. */
235                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
236                 make_bad_inode(inode);
237                 iput(inode);
238                 inode = ERR_PTR(-ESTALE);
239         } else if (is_bad_inode(inode)) {
240                 CWARN("%.16s: bad inode: ino = %u\n",
241                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
242                 iput(inode);
243                 inode = ERR_PTR(-ENOENT);
244         } else {
245                 if (id->oii_gen == OSD_OII_NOGEN)
246                         osd_id_gen(id, inode->i_ino, inode->i_generation);
247
248                 /* Do not update file c/mtime in ldiskfs.
249                  * NB: we don't have any lock to protect this because we don't
250                  * have reference on osd_object now, but contention with
251                  * another lookup + attr_set can't happen in the tiny window
252                  * between if (...) and set S_NOCMTIME. */
253                 if (!(inode->i_flags & S_NOCMTIME))
254                         inode->i_flags |= S_NOCMTIME;
255         }
256         return inode;
257 }
258
259 static struct inode *
260 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
261              struct osd_inode_id *id, struct lu_fid *fid)
262 {
263         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
264         struct inode            *inode;
265         int                      rc;
266
267         inode = osd_iget(info, dev, id);
268         if (IS_ERR(inode))
269                 return inode;
270
271         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
272         if (rc == 0) {
273                 *fid = lma->lma_self_fid;
274         } else if (rc == -ENODATA) {
275                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
276                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
277                 else
278                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
279         } else {
280                 iput(inode);
281                 inode = ERR_PTR(rc);
282         }
283         return inode;
284 }
285
286 static struct inode *
287 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
288                 struct osd_inode_id *id, const struct lu_fid *fid)
289 {
290         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
291         struct inode            *inode;
292         int                      rc;
293
294         inode = osd_iget(info, dev, id);
295         if (IS_ERR(inode))
296                 return inode;
297
298         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
299         if (rc == -ENODATA)
300                 return inode;
301
302         if (rc != 0) {
303                 iput(inode);
304                 return ERR_PTR(rc);
305         }
306
307         if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
308                 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
309                        PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
310                 iput(inode);
311                 return ERR_PTR(-EREMCHG);
312         }
313
314         return inode;
315 }
316
317 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
318                           const struct lu_fid *fid,
319                           const struct lu_object_conf *conf)
320 {
321         struct osd_thread_info *info;
322         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
323         struct osd_device      *dev;
324         struct osd_idmap_cache *oic;
325         struct osd_inode_id    *id;
326         struct inode           *inode;
327         struct osd_scrub       *scrub;
328         struct scrub_file      *sf;
329         int                     result;
330         int                     verify = 0;
331         ENTRY;
332
333         LINVRNT(osd_invariant(obj));
334         LASSERT(obj->oo_inode == NULL);
335         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
336
337         dev = osd_dev(ldev);
338         scrub = &dev->od_scrub;
339         sf = &scrub->os_file;
340         info = osd_oti_get(env);
341         LASSERT(info);
342         oic = &info->oti_cache;
343
344         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
345                 RETURN(-ENOENT);
346
347         /* Search order: 1. per-thread cache. */
348         if (lu_fid_eq(fid, &oic->oic_fid)) {
349                 id = &oic->oic_lid;
350                 goto iget;
351         }
352
353         id = &info->oti_id;
354         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
355                 /* Search order: 2. OI scrub pending list. */
356                 result = osd_oii_lookup(dev, fid, id);
357                 if (result == 0)
358                         goto iget;
359         }
360
361         if (sf->sf_flags & SF_INCONSISTENT)
362                 verify = 1;
363
364         /*
365          * Objects are created as locking anchors or place holders for objects
366          * yet to be created. No need to osd_oi_lookup() at here because FID
367          * shouldn't never be re-used, if it's really a duplicate FID from
368          * unexpected reason, we should be able to detect it later by calling
369          * do_create->osd_oi_insert()
370          */
371         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
372                 GOTO(out, result = 0);
373
374         /* Search order: 3. OI files. */
375         result = osd_oi_lookup(info, dev, fid, id, true);
376         if (result == -ENOENT) {
377                 if (!fid_is_norm(fid) || fid_is_on_ost(info, dev, fid) ||
378                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
379                                       sf->sf_oi_bitmap))
380                         GOTO(out, result = 0);
381
382                 goto trigger;
383         }
384
385         if (result != 0)
386                 GOTO(out, result);
387
388 iget:
389         if (verify == 0)
390                 inode = osd_iget(info, dev, id);
391         else
392                 inode = osd_iget_verify(info, dev, id, fid);
393         if (IS_ERR(inode)) {
394                 result = PTR_ERR(inode);
395                 if (result == -ENOENT || result == -ESTALE) {
396                         fid_zero(&oic->oic_fid);
397                         result = 0;
398                 } else if (result == -EREMCHG) {
399
400 trigger:
401                         if (thread_is_running(&scrub->os_thread)) {
402                                 result = -EINPROGRESS;
403                         } else if (!dev->od_noscrub) {
404                                 result = osd_scrub_start(dev);
405                                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
406                                                "for "DFID", rc = %d [1]\n",
407                                                LDISKFS_SB(osd_sb(dev))->s_es->\
408                                                s_volume_name,PFID(fid), result);
409                                 if (result == 0 || result == -EALREADY)
410                                         result = -EINPROGRESS;
411                                 else
412                                         result = -EREMCHG;
413                         }
414                 }
415
416                 GOTO(out, result);
417         }
418
419         obj->oo_inode = inode;
420         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
421
422         obj->oo_compat_dot_created = 1;
423         obj->oo_compat_dotdot_created = 1;
424
425         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
426                 GOTO(out, result = 0);
427
428         LASSERT(obj->oo_hl_head == NULL);
429         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
430         if (obj->oo_hl_head == NULL) {
431                 obj->oo_inode = NULL;
432                 iput(inode);
433                 GOTO(out, result = -ENOMEM);
434         }
435         GOTO(out, result = 0);
436
437 out:
438         LINVRNT(osd_invariant(obj));
439         return result;
440 }
441
442 /*
443  * Concurrency: shouldn't matter.
444  */
445 static void osd_object_init0(struct osd_object *obj)
446 {
447         LASSERT(obj->oo_inode != NULL);
448         obj->oo_dt.do_body_ops = &osd_body_ops;
449         obj->oo_dt.do_lu.lo_header->loh_attr |=
450                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
451 }
452
453 /*
454  * Concurrency: no concurrent access is possible that early in object
455  * life-cycle.
456  */
457 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
458                            const struct lu_object_conf *conf)
459 {
460         struct osd_object *obj = osd_obj(l);
461         int result;
462
463         LINVRNT(osd_invariant(obj));
464
465         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
466                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
467                 l->lo_header->loh_attr |= LOHA_EXISTS;
468                 return 0;
469         }
470
471         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
472         obj->oo_dt.do_body_ops = &osd_body_ops_new;
473         if (result == 0 && obj->oo_inode != NULL)
474                 osd_object_init0(obj);
475
476         LINVRNT(osd_invariant(obj));
477         return result;
478 }
479
480 /*
481  * Concurrency: no concurrent access is possible that late in object
482  * life-cycle.
483  */
484 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
485 {
486         struct osd_object *obj = osd_obj(l);
487
488         LINVRNT(osd_invariant(obj));
489
490         dt_object_fini(&obj->oo_dt);
491         if (obj->oo_hl_head != NULL)
492                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
493         OBD_FREE_PTR(obj);
494 }
495
496 /*
497  * Concurrency: no concurrent access is possible that late in object
498  * life-cycle.
499  */
500 static void osd_index_fini(struct osd_object *o)
501 {
502         struct iam_container *bag;
503
504         if (o->oo_dir != NULL) {
505                 bag = &o->oo_dir->od_container;
506                 if (o->oo_inode != NULL) {
507                         if (bag->ic_object == o->oo_inode)
508                                 iam_container_fini(bag);
509                 }
510                 OBD_FREE_PTR(o->oo_dir);
511                 o->oo_dir = NULL;
512         }
513 }
514
515 /*
516  * Concurrency: no concurrent access is possible that late in object
517  * life-cycle (for all existing callers, that is. New callers have to provide
518  * their own locking.)
519  */
520 static int osd_inode_unlinked(const struct inode *inode)
521 {
522         return inode->i_nlink == 0;
523 }
524
525 enum {
526         OSD_TXN_OI_DELETE_CREDITS    = 20,
527         OSD_TXN_INODE_DELETE_CREDITS = 20
528 };
529
530 /*
531  * Journal
532  */
533
534 #if OSD_THANDLE_STATS
535 /**
536  * Set time when the handle is allocated
537  */
538 static void osd_th_alloced(struct osd_thandle *oth)
539 {
540         oth->oth_alloced = cfs_time_current();
541 }
542
543 /**
544  * Set time when the handle started
545  */
546 static void osd_th_started(struct osd_thandle *oth)
547 {
548         oth->oth_started = cfs_time_current();
549 }
550
551 /**
552  * Helper function to convert time interval to microseconds packed in
553  * long int.
554  */
555 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
556 {
557         struct timeval val;
558
559         cfs_duration_usec(cfs_time_sub(end, start), &val);
560         return val.tv_sec * 1000000 + val.tv_usec;
561 }
562
563 /**
564  * Check whether the we deal with this handle for too long.
565  */
566 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
567                                 cfs_time_t alloced, cfs_time_t started,
568                                 cfs_time_t closed)
569 {
570         cfs_time_t now = cfs_time_current();
571
572         LASSERT(dev != NULL);
573
574         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
575                             interval_to_usec(alloced, started));
576         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
577                             interval_to_usec(started, closed));
578         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
579                             interval_to_usec(closed, now));
580
581         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
582                 CWARN("transaction handle %p was open for too long: "
583                       "now "CFS_TIME_T" ,"
584                       "alloced "CFS_TIME_T" ,"
585                       "started "CFS_TIME_T" ,"
586                       "closed "CFS_TIME_T"\n",
587                       oth, now, alloced, started, closed);
588                 libcfs_debug_dumpstack(NULL);
589         }
590 }
591
592 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
593 {                                                                       \
594         cfs_time_t __closed = cfs_time_current();                       \
595         cfs_time_t __alloced = oth->oth_alloced;                        \
596         cfs_time_t __started = oth->oth_started;                        \
597                                                                         \
598         expr;                                                           \
599         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
600 }
601
602 #else /* OSD_THANDLE_STATS */
603
604 #define osd_th_alloced(h)                  do {} while(0)
605 #define osd_th_started(h)                  do {} while(0)
606 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
607
608 #endif /* OSD_THANDLE_STATS */
609
610 /*
611  * Concurrency: doesn't access mutable data.
612  */
613 static int osd_param_is_not_sane(const struct osd_device *dev,
614                                  const struct thandle *th)
615 {
616         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
617
618         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
619 }
620
621 /*
622  * Concurrency: shouldn't matter.
623  */
624 static void osd_trans_commit_cb(struct super_block *sb,
625                                 struct ldiskfs_journal_cb_entry *jcb, int error)
626 {
627         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
628         struct thandle     *th  = &oh->ot_super;
629         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
630         struct dt_txn_commit_cb *dcb, *tmp;
631
632         LASSERT(oh->ot_handle == NULL);
633
634         if (error)
635                 CERROR("transaction @0x%p commit error: %d\n", th, error);
636
637         dt_txn_hook_commit(th);
638
639         /* call per-transaction callbacks if any */
640         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
641                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
642                          "commit callback entry: magic=%x name='%s'\n",
643                          dcb->dcb_magic, dcb->dcb_name);
644                 cfs_list_del_init(&dcb->dcb_linkage);
645                 dcb->dcb_func(NULL, th, dcb, error);
646         }
647
648         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
649         lu_device_put(lud);
650         th->th_dev = NULL;
651
652         lu_context_exit(&th->th_ctx);
653         lu_context_fini(&th->th_ctx);
654         OBD_FREE_PTR(oh);
655 }
656
657 static struct thandle *osd_trans_create(const struct lu_env *env,
658                                         struct dt_device *d)
659 {
660         struct osd_thread_info *oti = osd_oti_get(env);
661         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
662         struct osd_thandle     *oh;
663         struct thandle         *th;
664         ENTRY;
665
666         /* on pending IO in this thread should left from prev. request */
667         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
668
669         th = ERR_PTR(-ENOMEM);
670         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
671         if (oh != NULL) {
672                 oh->ot_quota_trans = &oti->oti_quota_trans;
673                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
674                 th = &oh->ot_super;
675                 th->th_dev = d;
676                 th->th_result = 0;
677                 th->th_tags = LCT_TX_HANDLE;
678                 oh->ot_credits = 0;
679                 oti->oti_dev = osd_dt_dev(d);
680                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
681                 osd_th_alloced(oh);
682
683                 memset(oti->oti_declare_ops, 0, OSD_OT_MAX);
684                 memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX);
685                 memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX);
686                 oti->oti_rollback = false;
687         }
688         RETURN(th);
689 }
690
691 /*
692  * Concurrency: shouldn't matter.
693  */
694 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
695                     struct thandle *th)
696 {
697         struct osd_thread_info *oti = osd_oti_get(env);
698         struct osd_device  *dev = osd_dt_dev(d);
699         handle_t           *jh;
700         struct osd_thandle *oh;
701         int rc;
702
703         ENTRY;
704
705         LASSERT(current->journal_info == NULL);
706
707         oh = container_of0(th, struct osd_thandle, ot_super);
708         LASSERT(oh != NULL);
709         LASSERT(oh->ot_handle == NULL);
710
711         rc = dt_txn_hook_start(env, d, th);
712         if (rc != 0)
713                 GOTO(out, rc);
714
715         if (unlikely(osd_param_is_not_sane(dev, th))) {
716                 static unsigned long last_printed;
717                 static int last_credits;
718
719                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
720                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
721                       oh->ot_credits,
722                       osd_journal(dev)->j_max_transaction_buffers);
723 #ifdef OSD_TRACK_DECLARES
724                 CWARN("  create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
725                       oti->oti_declare_ops[OSD_OT_CREATE],
726                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
727                       oti->oti_declare_ops[OSD_OT_DELETE],
728                       oti->oti_declare_ops_cred[OSD_OT_DELETE],
729                       oti->oti_declare_ops[OSD_OT_DESTROY],
730                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
731                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
732                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
733                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
734                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
735                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
736                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
737                       oti->oti_declare_ops[OSD_OT_WRITE],
738                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
739                       oti->oti_declare_ops[OSD_OT_PUNCH],
740                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
741                       oti->oti_declare_ops[OSD_OT_QUOTA],
742                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
743                 CWARN("  insert: %u/%u, delete: %u/%u\n",
744                       oti->oti_declare_ops[OSD_OT_INSERT],
745                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
746                       oti->oti_declare_ops[OSD_OT_DESTROY],
747                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
748                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
749                       oti->oti_declare_ops[OSD_OT_REF_ADD],
750                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
751                       oti->oti_declare_ops[OSD_OT_REF_DEL],
752                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
753
754                 if (last_credits != oh->ot_credits &&
755                     time_after(jiffies, last_printed + 60 * HZ)) {
756                         libcfs_debug_dumpstack(NULL);
757                         last_credits = oh->ot_credits;
758                         last_printed = jiffies;
759                 }
760 #endif
761                 /* XXX Limit the credits to 'max_transaction_buffers', and
762                  *     let the underlying filesystem to catch the error if
763                  *     we really need so many credits.
764                  *
765                  *     This should be removed when we can calculate the
766                  *     credits precisely. */
767                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
768         }
769
770         /*
771          * XXX temporary stuff. Some abstraction layer should
772          * be used.
773          */
774         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
775         osd_th_started(oh);
776         if (!IS_ERR(jh)) {
777                 oh->ot_handle = jh;
778                 LASSERT(oti->oti_txns == 0);
779                 lu_context_init(&th->th_ctx, th->th_tags);
780                 lu_context_enter(&th->th_ctx);
781
782                 lu_device_get(&d->dd_lu_dev);
783                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
784                                              "osd-tx", th);
785                 oti->oti_txns++;
786                 rc = 0;
787         } else {
788                 rc = PTR_ERR(jh);
789         }
790 out:
791         RETURN(rc);
792 }
793
794 /*
795  * Concurrency: shouldn't matter.
796  */
797 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
798 {
799         int                     rc = 0;
800         struct osd_thandle     *oh;
801         struct osd_thread_info *oti = osd_oti_get(env);
802         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
803         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
804         ENTRY;
805
806         oh = container_of0(th, struct osd_thandle, ot_super);
807
808         if (qsd != NULL)
809                 /* inform the quota slave device that the transaction is
810                  * stopping */
811                 qsd_op_end(env, qsd, oh->ot_quota_trans);
812         oh->ot_quota_trans = NULL;
813
814         if (oh->ot_handle != NULL) {
815                 handle_t *hdl = oh->ot_handle;
816
817                 /*
818                  * add commit callback
819                  * notice we don't do this in osd_trans_start()
820                  * as underlying transaction can change during truncate
821                  */
822                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
823                                          &oh->ot_jcb);
824
825                 LASSERT(oti->oti_txns == 1);
826                 oti->oti_txns--;
827                 rc = dt_txn_hook_stop(env, th);
828                 if (rc != 0)
829                         CERROR("Failure in transaction hook: %d\n", rc);
830
831                 /* hook functions might modify th_sync */
832                 hdl->h_sync = th->th_sync;
833
834                 oh->ot_handle = NULL;
835                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
836                                   rc = ldiskfs_journal_stop(hdl));
837                 if (rc != 0)
838                         CERROR("Failure to stop transaction: %d\n", rc);
839         } else {
840                 OBD_FREE_PTR(oh);
841         }
842
843         /* as we want IO to journal and data IO be concurrent, we don't block
844          * awaiting data IO completion in osd_do_bio(), instead we wait here
845          * once transaction is submitted to the journal. all reqular requests
846          * don't do direct IO (except read/write), thus this wait_event becomes
847          * no-op for them.
848          *
849          * IMPORTANT: we have to wait till any IO submited by the thread is
850          * completed otherwise iobuf may be corrupted by different request
851          */
852         cfs_wait_event(iobuf->dr_wait,
853                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
854         if (!rc)
855                 rc = iobuf->dr_error;
856
857         RETURN(rc);
858 }
859
860 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
861 {
862         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
863                                                ot_super);
864
865         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
866         LASSERT(&dcb->dcb_func != NULL);
867         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
868
869         return 0;
870 }
871
872 /*
873  * Called just before object is freed. Releases all resources except for
874  * object itself (that is released by osd_object_free()).
875  *
876  * Concurrency: no concurrent access is possible that late in object
877  * life-cycle.
878  */
879 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
880 {
881         struct osd_object *obj   = osd_obj(l);
882         struct inode      *inode = obj->oo_inode;
883
884         LINVRNT(osd_invariant(obj));
885
886         /*
887          * If object is unlinked remove fid->ino mapping from object index.
888          */
889
890         osd_index_fini(obj);
891         if (inode != NULL) {
892                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
893                 qid_t                    uid = inode->i_uid;
894                 qid_t                    gid = inode->i_gid;
895
896                 iput(inode);
897                 obj->oo_inode = NULL;
898
899                 if (qsd != NULL) {
900                         struct osd_thread_info  *info = osd_oti_get(env);
901                         struct lquota_id_info   *qi = &info->oti_qi;
902
903                         /* Release granted quota to master if necessary */
904                         qi->lqi_id.qid_uid = uid;
905                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
906
907                         qi->lqi_id.qid_uid = gid;
908                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
909                 }
910         }
911 }
912
913 /*
914  * Concurrency: ->loo_object_release() is called under site spin-lock.
915  */
916 static void osd_object_release(const struct lu_env *env,
917                                struct lu_object *l)
918 {
919 }
920
921 /*
922  * Concurrency: shouldn't matter.
923  */
924 static int osd_object_print(const struct lu_env *env, void *cookie,
925                             lu_printer_t p, const struct lu_object *l)
926 {
927         struct osd_object *o = osd_obj(l);
928         struct iam_descr  *d;
929
930         if (o->oo_dir != NULL)
931                 d = o->oo_dir->od_container.ic_descr;
932         else
933                 d = NULL;
934         return (*p)(env, cookie,
935                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
936                     o, o->oo_inode,
937                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
938                     o->oo_inode ? o->oo_inode->i_generation : 0,
939                     d ? d->id_ops->id_name : "plain");
940 }
941
942 /*
943  * Concurrency: shouldn't matter.
944  */
945 int osd_statfs(const struct lu_env *env, struct dt_device *d,
946                struct obd_statfs *sfs)
947 {
948         struct osd_device  *osd = osd_dt_dev(d);
949         struct super_block *sb = osd_sb(osd);
950         struct kstatfs     *ksfs;
951         int result = 0;
952
953         if (unlikely(osd->od_mnt == NULL))
954                 return -EINPROGRESS;
955
956         /* osd_lproc.c call this without env, allocate ksfs for that case */
957         if (unlikely(env == NULL)) {
958                 OBD_ALLOC_PTR(ksfs);
959                 if (ksfs == NULL)
960                         return -ENOMEM;
961         } else {
962                 ksfs = &osd_oti_get(env)->oti_ksfs;
963         }
964
965         spin_lock(&osd->od_osfs_lock);
966         /* cache 1 second */
967         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
968                 result = sb->s_op->statfs(sb->s_root, ksfs);
969                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
970                         osd->od_osfs_age = cfs_time_current_64();
971                         statfs_pack(&osd->od_statfs, ksfs);
972                         if (sb->s_flags & MS_RDONLY)
973                                 sfs->os_state = OS_STATE_READONLY;
974                 }
975         }
976
977         if (likely(result == 0))
978                 *sfs = osd->od_statfs;
979         spin_unlock(&osd->od_osfs_lock);
980
981         if (unlikely(env == NULL))
982                 OBD_FREE_PTR(ksfs);
983
984         return result;
985 }
986
987 /**
988  * Estimate space needed for file creations. We assume the largest filename
989  * which is 2^64 - 1, hence a filename of 20 chars.
990  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
991  */
992 #ifdef __LDISKFS_DIR_REC_LEN
993 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
994 #else
995 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
996 #endif
997
998 /*
999  * Concurrency: doesn't access mutable data.
1000  */
1001 static void osd_conf_get(const struct lu_env *env,
1002                          const struct dt_device *dev,
1003                          struct dt_device_param *param)
1004 {
1005         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1006
1007         /*
1008          * XXX should be taken from not-yet-existing fs abstraction layer.
1009          */
1010         param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1011         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1012         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1013         param->ddp_block_shift  = sb->s_blocksize_bits;
1014         param->ddp_mount_type     = LDD_MT_LDISKFS;
1015         param->ddp_maxbytes       = sb->s_maxbytes;
1016         /* Overhead estimate should be fairly accurate, so we really take a tiny
1017          * error margin which also avoids fragmenting the filesystem too much */
1018         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1019         /* inode are statically allocated, so per-inode space consumption
1020          * is the space consumed by the directory entry */
1021         param->ddp_inodespace     = PER_OBJ_USAGE;
1022         /* per-fragment overhead to be used by the client code */
1023         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1024         param->ddp_mntopts      = 0;
1025         if (test_opt(sb, XATTR_USER))
1026                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1027         if (test_opt(sb, POSIX_ACL))
1028                 param->ddp_mntopts |= MNTOPT_ACL;
1029
1030 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1031         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1032                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1033         else
1034 #endif
1035                 param->ddp_max_ea_size = sb->s_blocksize;
1036
1037 }
1038
1039 /*
1040  * Concurrency: shouldn't matter.
1041  */
1042 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1043 {
1044         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1045         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1046 }
1047
1048 /**
1049  * Start commit for OSD device.
1050  *
1051  * An implementation of dt_commit_async method for OSD device.
1052  * Asychronously starts underlayng fs sync and thereby a transaction
1053  * commit.
1054  *
1055  * \param env environment
1056  * \param d dt device
1057  *
1058  * \see dt_device_operations
1059  */
1060 static int osd_commit_async(const struct lu_env *env,
1061                             struct dt_device *d)
1062 {
1063         struct super_block *s = osd_sb(osd_dt_dev(d));
1064         ENTRY;
1065
1066         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1067         RETURN(s->s_op->sync_fs(s, 0));
1068 }
1069
1070 /*
1071  * Concurrency: shouldn't matter.
1072  */
1073
1074 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1075 {
1076         struct super_block *sb = osd_sb(osd_dt_dev(d));
1077         struct block_device *dev = sb->s_bdev;
1078 #ifdef HAVE_DEV_SET_RDONLY
1079         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1080         int rc = 0;
1081 #else
1082         int rc = -EOPNOTSUPP;
1083 #endif
1084         ENTRY;
1085
1086 #ifdef HAVE_DEV_SET_RDONLY
1087         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1088
1089         if (jdev && (jdev != dev)) {
1090                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1091                        (long)jdev);
1092                 dev_set_rdonly(jdev);
1093         }
1094         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1095         dev_set_rdonly(dev);
1096 #else
1097         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1098                osd_dt_dev(d)->od_svname, (long)dev, rc);
1099 #endif
1100         RETURN(rc);
1101 }
1102
1103 /*
1104  * Concurrency: serialization provided by callers.
1105  */
1106 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1107                               int mode, unsigned long timeout, __u32 alg,
1108                               struct lustre_capa_key *keys)
1109 {
1110         struct osd_device *dev = osd_dt_dev(d);
1111         ENTRY;
1112
1113         dev->od_fl_capa = mode;
1114         dev->od_capa_timeout = timeout;
1115         dev->od_capa_alg = alg;
1116         dev->od_capa_keys = keys;
1117         RETURN(0);
1118 }
1119
1120 /**
1121  * Note: we do not count into QUOTA here.
1122  * If we mount with --data_journal we may need more.
1123  */
1124 const int osd_dto_credits_noquota[DTO_NR] = {
1125         /**
1126          * Insert/Delete.
1127          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1128          * SINGLEDATA_TRANS_BLOCKS(8)
1129          * XXX Note: maybe iam need more, since iam have more level than
1130          *           EXT3 htree.
1131          */
1132         [DTO_INDEX_INSERT]  = 16,
1133         [DTO_INDEX_DELETE]  = 16,
1134         /**
1135          * Used for OI scrub
1136          */
1137         [DTO_INDEX_UPDATE]  = 16,
1138         /**
1139          * Create a object. The same as create object in EXT3.
1140          * DATA_TRANS_BLOCKS(14) +
1141          * INDEX_EXTRA_BLOCKS(8) +
1142          * 3(inode bits, groups, GDT)
1143          */
1144         [DTO_OBJECT_CREATE] = 25,
1145         /**
1146          * XXX: real credits to be fixed
1147          */
1148         [DTO_OBJECT_DELETE] = 25,
1149         /**
1150          * Attr set credits (inode)
1151          */
1152         [DTO_ATTR_SET_BASE] = 1,
1153         /**
1154          * Xattr set. The same as xattr of EXT3.
1155          * DATA_TRANS_BLOCKS(14)
1156          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1157          * are also counted in. Do not know why?
1158          */
1159         [DTO_XATTR_SET]     = 14,
1160         [DTO_LOG_REC]       = 14,
1161         /**
1162          * credits for inode change during write.
1163          */
1164         [DTO_WRITE_BASE]    = 3,
1165         /**
1166          * credits for single block write.
1167          */
1168         [DTO_WRITE_BLOCK]   = 14,
1169         /**
1170          * Attr set credits for chown.
1171          * This is extra credits for setattr, and it is null without quota
1172          */
1173         [DTO_ATTR_SET_CHOWN]= 0
1174 };
1175
1176 static const struct dt_device_operations osd_dt_ops = {
1177         .dt_root_get       = osd_root_get,
1178         .dt_statfs         = osd_statfs,
1179         .dt_trans_create   = osd_trans_create,
1180         .dt_trans_start    = osd_trans_start,
1181         .dt_trans_stop     = osd_trans_stop,
1182         .dt_trans_cb_add   = osd_trans_cb_add,
1183         .dt_conf_get       = osd_conf_get,
1184         .dt_sync           = osd_sync,
1185         .dt_ro             = osd_ro,
1186         .dt_commit_async   = osd_commit_async,
1187         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1188 };
1189
1190 static void osd_object_read_lock(const struct lu_env *env,
1191                                  struct dt_object *dt, unsigned role)
1192 {
1193         struct osd_object *obj = osd_dt_obj(dt);
1194         struct osd_thread_info *oti = osd_oti_get(env);
1195
1196         LINVRNT(osd_invariant(obj));
1197
1198         LASSERT(obj->oo_owner != env);
1199         down_read_nested(&obj->oo_sem, role);
1200
1201         LASSERT(obj->oo_owner == NULL);
1202         oti->oti_r_locks++;
1203 }
1204
1205 static void osd_object_write_lock(const struct lu_env *env,
1206                                   struct dt_object *dt, unsigned role)
1207 {
1208         struct osd_object *obj = osd_dt_obj(dt);
1209         struct osd_thread_info *oti = osd_oti_get(env);
1210
1211         LINVRNT(osd_invariant(obj));
1212
1213         LASSERT(obj->oo_owner != env);
1214         down_write_nested(&obj->oo_sem, role);
1215
1216         LASSERT(obj->oo_owner == NULL);
1217         obj->oo_owner = env;
1218         oti->oti_w_locks++;
1219 }
1220
1221 static void osd_object_read_unlock(const struct lu_env *env,
1222                                    struct dt_object *dt)
1223 {
1224         struct osd_object *obj = osd_dt_obj(dt);
1225         struct osd_thread_info *oti = osd_oti_get(env);
1226
1227         LINVRNT(osd_invariant(obj));
1228
1229         LASSERT(oti->oti_r_locks > 0);
1230         oti->oti_r_locks--;
1231         up_read(&obj->oo_sem);
1232 }
1233
1234 static void osd_object_write_unlock(const struct lu_env *env,
1235                                     struct dt_object *dt)
1236 {
1237         struct osd_object *obj = osd_dt_obj(dt);
1238         struct osd_thread_info *oti = osd_oti_get(env);
1239
1240         LINVRNT(osd_invariant(obj));
1241
1242         LASSERT(obj->oo_owner == env);
1243         LASSERT(oti->oti_w_locks > 0);
1244         oti->oti_w_locks--;
1245         obj->oo_owner = NULL;
1246         up_write(&obj->oo_sem);
1247 }
1248
1249 static int osd_object_write_locked(const struct lu_env *env,
1250                                    struct dt_object *dt)
1251 {
1252         struct osd_object *obj = osd_dt_obj(dt);
1253
1254         LINVRNT(osd_invariant(obj));
1255
1256         return obj->oo_owner == env;
1257 }
1258
1259 static int capa_is_sane(const struct lu_env *env,
1260                         struct osd_device *dev,
1261                         struct lustre_capa *capa,
1262                         struct lustre_capa_key *keys)
1263 {
1264         struct osd_thread_info *oti = osd_oti_get(env);
1265         struct lustre_capa *tcapa = &oti->oti_capa;
1266         struct obd_capa *oc;
1267         int i, rc = 0;
1268         ENTRY;
1269
1270         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1271         if (oc) {
1272                 if (capa_is_expired(oc)) {
1273                         DEBUG_CAPA(D_ERROR, capa, "expired");
1274                         rc = -ESTALE;
1275                 }
1276                 capa_put(oc);
1277                 RETURN(rc);
1278         }
1279
1280         if (capa_is_expired_sec(capa)) {
1281                 DEBUG_CAPA(D_ERROR, capa, "expired");
1282                 RETURN(-ESTALE);
1283         }
1284
1285         spin_lock(&capa_lock);
1286         for (i = 0; i < 2; i++) {
1287                 if (keys[i].lk_keyid == capa->lc_keyid) {
1288                         oti->oti_capa_key = keys[i];
1289                         break;
1290                 }
1291         }
1292         spin_unlock(&capa_lock);
1293
1294         if (i == 2) {
1295                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1296                 RETURN(-ESTALE);
1297         }
1298
1299         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1300         if (rc)
1301                 RETURN(rc);
1302
1303         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1304                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1305                 RETURN(-EACCES);
1306         }
1307
1308         oc = capa_add(dev->od_capa_hash, capa);
1309         capa_put(oc);
1310
1311         RETURN(0);
1312 }
1313
1314 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1315                     struct lustre_capa *capa, __u64 opc)
1316 {
1317         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1318         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1319         struct md_capainfo *ci;
1320         int rc;
1321
1322         if (!dev->od_fl_capa)
1323                 return 0;
1324
1325         if (capa == BYPASS_CAPA)
1326                 return 0;
1327
1328         ci = md_capainfo(env);
1329         if (unlikely(!ci))
1330                 return 0;
1331
1332         if (ci->mc_auth == LC_ID_NONE)
1333                 return 0;
1334
1335         if (!capa) {
1336                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1337                 return -EACCES;
1338         }
1339
1340         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1341                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1342                            PFID(fid));
1343                 return -EACCES;
1344         }
1345
1346         if (!capa_opc_supported(capa, opc)) {
1347                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1348                 return -EACCES;
1349         }
1350
1351         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1352                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1353                 return -EACCES;
1354         }
1355
1356         return 0;
1357 }
1358
1359 static struct timespec *osd_inode_time(const struct lu_env *env,
1360                                        struct inode *inode, __u64 seconds)
1361 {
1362         struct osd_thread_info  *oti = osd_oti_get(env);
1363         struct timespec         *t   = &oti->oti_time;
1364
1365         t->tv_sec = seconds;
1366         t->tv_nsec = 0;
1367         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1368         return t;
1369 }
1370
1371
1372 static void osd_inode_getattr(const struct lu_env *env,
1373                               struct inode *inode, struct lu_attr *attr)
1374 {
1375         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1376                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1377                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1378                                LA_TYPE;
1379
1380         attr->la_atime      = LTIME_S(inode->i_atime);
1381         attr->la_mtime      = LTIME_S(inode->i_mtime);
1382         attr->la_ctime      = LTIME_S(inode->i_ctime);
1383         attr->la_mode       = inode->i_mode;
1384         attr->la_size       = i_size_read(inode);
1385         attr->la_blocks     = inode->i_blocks;
1386         attr->la_uid        = inode->i_uid;
1387         attr->la_gid        = inode->i_gid;
1388         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1389         attr->la_nlink      = inode->i_nlink;
1390         attr->la_rdev       = inode->i_rdev;
1391         attr->la_blksize    = 1 << inode->i_blkbits;
1392         attr->la_blkbits    = inode->i_blkbits;
1393 }
1394
1395 static int osd_attr_get(const struct lu_env *env,
1396                         struct dt_object *dt,
1397                         struct lu_attr *attr,
1398                         struct lustre_capa *capa)
1399 {
1400         struct osd_object *obj = osd_dt_obj(dt);
1401
1402         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1403         LINVRNT(osd_invariant(obj));
1404
1405         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1406                 return -EACCES;
1407
1408         spin_lock(&obj->oo_guard);
1409         osd_inode_getattr(env, obj->oo_inode, attr);
1410         spin_unlock(&obj->oo_guard);
1411         return 0;
1412 }
1413
1414 static int osd_declare_attr_set(const struct lu_env *env,
1415                                 struct dt_object *dt,
1416                                 const struct lu_attr *attr,
1417                                 struct thandle *handle)
1418 {
1419         struct osd_thandle     *oh;
1420         struct osd_object      *obj;
1421         struct osd_thread_info *info = osd_oti_get(env);
1422         struct lquota_id_info  *qi = &info->oti_qi;
1423         long long               bspace;
1424         int                     rc = 0;
1425         bool                    allocated;
1426         ENTRY;
1427
1428         LASSERT(dt != NULL);
1429         LASSERT(handle != NULL);
1430
1431         obj = osd_dt_obj(dt);
1432         LASSERT(osd_invariant(obj));
1433
1434         oh = container_of0(handle, struct osd_thandle, ot_super);
1435         LASSERT(oh->ot_handle == NULL);
1436
1437         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1438                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1439
1440         if (attr == NULL || obj->oo_inode == NULL)
1441                 RETURN(rc);
1442
1443         bspace   = obj->oo_inode->i_blocks;
1444         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1445         bspace   = toqb(bspace);
1446
1447         /* Changing ownership is always preformed by super user, it should not
1448          * fail with EDQUOT.
1449          *
1450          * We still need to call the osd_declare_qid() to calculate the journal
1451          * credits for updating quota accounting files and to trigger quota
1452          * space adjustment once the operation is completed.*/
1453         if ((attr->la_valid & LA_UID) != 0 &&
1454              attr->la_uid != obj->oo_inode->i_uid) {
1455                 qi->lqi_type = USRQUOTA;
1456
1457                 /* inode accounting */
1458                 qi->lqi_is_blk = false;
1459
1460                 /* one more inode for the new owner ... */
1461                 qi->lqi_id.qid_uid = attr->la_uid;
1462                 qi->lqi_space      = 1;
1463                 allocated = (attr->la_uid == 0) ? true : false;
1464                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1465                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1466                         rc = 0;
1467                 if (rc)
1468                         RETURN(rc);
1469
1470                 /* and one less inode for the current uid */
1471                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1472                 qi->lqi_space      = -1;
1473                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1474                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1475                         rc = 0;
1476                 if (rc)
1477                         RETURN(rc);
1478
1479                 /* block accounting */
1480                 qi->lqi_is_blk = true;
1481
1482                 /* more blocks for the new owner ... */
1483                 qi->lqi_id.qid_uid = attr->la_uid;
1484                 qi->lqi_space      = bspace;
1485                 allocated = (attr->la_uid == 0) ? true : false;
1486                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1487                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1488                         rc = 0;
1489                 if (rc)
1490                         RETURN(rc);
1491
1492                 /* and finally less blocks for the current owner */
1493                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1494                 qi->lqi_space      = -bspace;
1495                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1496                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1497                         rc = 0;
1498                 if (rc)
1499                         RETURN(rc);
1500         }
1501
1502         if (attr->la_valid & LA_GID &&
1503             attr->la_gid != obj->oo_inode->i_gid) {
1504                 qi->lqi_type = GRPQUOTA;
1505
1506                 /* inode accounting */
1507                 qi->lqi_is_blk = false;
1508
1509                 /* one more inode for the new group owner ... */
1510                 qi->lqi_id.qid_gid = attr->la_gid;
1511                 qi->lqi_space      = 1;
1512                 allocated = (attr->la_gid == 0) ? true : false;
1513                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1514                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1515                         rc = 0;
1516                 if (rc)
1517                         RETURN(rc);
1518
1519                 /* and one less inode for the current gid */
1520                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1521                 qi->lqi_space      = -1;
1522                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1523                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1524                         rc = 0;
1525                 if (rc)
1526                         RETURN(rc);
1527
1528                 /* block accounting */
1529                 qi->lqi_is_blk = true;
1530
1531                 /* more blocks for the new owner ... */
1532                 qi->lqi_id.qid_gid = attr->la_gid;
1533                 qi->lqi_space      = bspace;
1534                 allocated = (attr->la_gid == 0) ? true : false;
1535                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1536                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1537                         rc = 0;
1538                 if (rc)
1539                         RETURN(rc);
1540
1541                 /* and finally less blocks for the current owner */
1542                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1543                 qi->lqi_space      = -bspace;
1544                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1545                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1546                         rc = 0;
1547                 if (rc)
1548                         RETURN(rc);
1549         }
1550
1551         RETURN(rc);
1552 }
1553
1554 static int osd_inode_setattr(const struct lu_env *env,
1555                              struct inode *inode, const struct lu_attr *attr)
1556 {
1557         __u64 bits;
1558
1559         bits = attr->la_valid;
1560
1561         if (bits & LA_ATIME)
1562                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1563         if (bits & LA_CTIME)
1564                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1565         if (bits & LA_MTIME)
1566                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1567         if (bits & LA_SIZE) {
1568                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1569                 i_size_write(inode, attr->la_size);
1570         }
1571
1572 #if 0
1573         /* OSD should not change "i_blocks" which is used by quota.
1574          * "i_blocks" should be changed by ldiskfs only. */
1575         if (bits & LA_BLOCKS)
1576                 inode->i_blocks = attr->la_blocks;
1577 #endif
1578         if (bits & LA_MODE)
1579                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1580                         (attr->la_mode & ~S_IFMT);
1581         if (bits & LA_UID)
1582                 inode->i_uid    = attr->la_uid;
1583         if (bits & LA_GID)
1584                 inode->i_gid    = attr->la_gid;
1585         if (bits & LA_NLINK)
1586                 set_nlink(inode, attr->la_nlink);
1587         if (bits & LA_RDEV)
1588                 inode->i_rdev   = attr->la_rdev;
1589
1590         if (bits & LA_FLAGS) {
1591                 /* always keep S_NOCMTIME */
1592                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1593                                  S_NOCMTIME;
1594         }
1595         return 0;
1596 }
1597
1598 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1599 {
1600         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1601             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1602                 struct iattr    iattr;
1603                 int             rc;
1604
1605                 iattr.ia_valid = 0;
1606                 if (attr->la_valid & LA_UID)
1607                         iattr.ia_valid |= ATTR_UID;
1608                 if (attr->la_valid & LA_GID)
1609                         iattr.ia_valid |= ATTR_GID;
1610                 iattr.ia_uid = attr->la_uid;
1611                 iattr.ia_gid = attr->la_gid;
1612
1613                 rc = ll_vfs_dq_transfer(inode, &iattr);
1614                 if (rc) {
1615                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1616                                "enforcement enabled on the ldiskfs filesystem?",
1617                                inode->i_sb->s_id, rc);
1618                         return rc;
1619                 }
1620         }
1621         return 0;
1622 }
1623
1624 static int osd_attr_set(const struct lu_env *env,
1625                         struct dt_object *dt,
1626                         const struct lu_attr *attr,
1627                         struct thandle *handle,
1628                         struct lustre_capa *capa)
1629 {
1630         struct osd_object *obj = osd_dt_obj(dt);
1631         struct inode      *inode;
1632         int rc;
1633
1634         LASSERT(handle != NULL);
1635         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
1636         LASSERT(osd_invariant(obj));
1637
1638         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1639                 return -EACCES;
1640
1641         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1642
1643         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1644                 struct osd_thread_info  *oti  = osd_oti_get(env);
1645                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1646                 struct lu_fid           *fid1 = &oti->oti_fid;
1647                 struct osd_inode_id     *id   = &oti->oti_id;
1648                 struct iam_path_descr   *ipd;
1649                 struct iam_container    *bag;
1650                 struct osd_thandle      *oh;
1651                 int                      rc;
1652
1653                 fid_cpu_to_be(fid1, fid0);
1654                 memset(id, 1, sizeof(*id));
1655                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1656                                   fid0)->oi_dir.od_container;
1657                 ipd = osd_idx_ipd_get(env, bag);
1658                 if (unlikely(ipd == NULL))
1659                         RETURN(-ENOMEM);
1660
1661                 oh = container_of0(handle, struct osd_thandle, ot_super);
1662                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1663                                 (const struct iam_rec *)id, ipd);
1664                 osd_ipd_put(env, bag, ipd);
1665                 return(rc > 0 ? 0 : rc);
1666         }
1667
1668         inode = obj->oo_inode;
1669         ll_vfs_dq_init(inode);
1670
1671         rc = osd_quota_transfer(inode, attr);
1672         if (rc)
1673                 return rc;
1674
1675         spin_lock(&obj->oo_guard);
1676         rc = osd_inode_setattr(env, inode, attr);
1677         spin_unlock(&obj->oo_guard);
1678
1679         if (!rc)
1680                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1681         return rc;
1682 }
1683
1684 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1685                                     struct osd_object *obj,
1686                                     const char *name, const int namelen)
1687 {
1688         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1689 }
1690
1691 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1692                       cfs_umode_t mode,
1693                       struct dt_allocation_hint *hint,
1694                       struct thandle *th)
1695 {
1696         int result;
1697         struct osd_device  *osd = osd_obj2dev(obj);
1698         struct osd_thandle *oth;
1699         struct dt_object   *parent = NULL;
1700         struct inode       *inode;
1701
1702         LINVRNT(osd_invariant(obj));
1703         LASSERT(obj->oo_inode == NULL);
1704         LASSERT(obj->oo_hl_head == NULL);
1705
1706         if (S_ISDIR(mode) && ldiskfs_pdo) {
1707                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1708                 if (obj->oo_hl_head == NULL)
1709                         return -ENOMEM;
1710         }
1711
1712         oth = container_of(th, struct osd_thandle, ot_super);
1713         LASSERT(oth->ot_handle->h_transaction != NULL);
1714
1715         if (hint && hint->dah_parent)
1716                 parent = hint->dah_parent;
1717
1718         inode = ldiskfs_create_inode(oth->ot_handle,
1719                                      parent ? osd_dt_obj(parent)->oo_inode :
1720                                               osd_sb(osd)->s_root->d_inode,
1721                                      mode);
1722         if (!IS_ERR(inode)) {
1723                 /* Do not update file c/mtime in ldiskfs.
1724                  * NB: don't need any lock because no contention at this
1725                  * early stage */
1726                 inode->i_flags |= S_NOCMTIME;
1727
1728                 /* For new created object, it must be consistent,
1729                  * and it is unnecessary to scrub against it. */
1730                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1731                 obj->oo_inode = inode;
1732                 result = 0;
1733         } else {
1734                 if (obj->oo_hl_head != NULL) {
1735                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1736                         obj->oo_hl_head = NULL;
1737                 }
1738                 result = PTR_ERR(inode);
1739         }
1740         LINVRNT(osd_invariant(obj));
1741         return result;
1742 }
1743
1744 enum {
1745         OSD_NAME_LEN = 255
1746 };
1747
1748 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1749                      struct lu_attr *attr,
1750                      struct dt_allocation_hint *hint,
1751                      struct dt_object_format *dof,
1752                      struct thandle *th)
1753 {
1754         int result;
1755         struct osd_thandle *oth;
1756         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1757
1758         LASSERT(S_ISDIR(attr->la_mode));
1759
1760         oth = container_of(th, struct osd_thandle, ot_super);
1761         LASSERT(oth->ot_handle->h_transaction != NULL);
1762         result = osd_mkfile(info, obj, mode, hint, th);
1763
1764         return result;
1765 }
1766
1767 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1768                         struct lu_attr *attr,
1769                         struct dt_allocation_hint *hint,
1770                         struct dt_object_format *dof,
1771                         struct thandle *th)
1772 {
1773         int result;
1774         struct osd_thandle *oth;
1775         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1776
1777         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1778
1779         LASSERT(S_ISREG(attr->la_mode));
1780
1781         oth = container_of(th, struct osd_thandle, ot_super);
1782         LASSERT(oth->ot_handle->h_transaction != NULL);
1783
1784         result = osd_mkfile(info, obj, mode, hint, th);
1785         if (result == 0) {
1786                 LASSERT(obj->oo_inode != NULL);
1787                 if (feat->dif_flags & DT_IND_VARKEY)
1788                         result = iam_lvar_create(obj->oo_inode,
1789                                                  feat->dif_keysize_max,
1790                                                  feat->dif_ptrsize,
1791                                                  feat->dif_recsize_max,
1792                                                  oth->ot_handle);
1793                 else
1794                         result = iam_lfix_create(obj->oo_inode,
1795                                                  feat->dif_keysize_max,
1796                                                  feat->dif_ptrsize,
1797                                                  feat->dif_recsize_max,
1798                                                  oth->ot_handle);
1799
1800         }
1801         return result;
1802 }
1803
1804 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1805                      struct lu_attr *attr,
1806                      struct dt_allocation_hint *hint,
1807                      struct dt_object_format *dof,
1808                      struct thandle *th)
1809 {
1810         LASSERT(S_ISREG(attr->la_mode));
1811         return osd_mkfile(info, obj, (attr->la_mode &
1812                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1813 }
1814
1815 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1816                      struct lu_attr *attr,
1817                      struct dt_allocation_hint *hint,
1818                      struct dt_object_format *dof,
1819                      struct thandle *th)
1820 {
1821         LASSERT(S_ISLNK(attr->la_mode));
1822         return osd_mkfile(info, obj, (attr->la_mode &
1823                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1824 }
1825
1826 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1827                      struct lu_attr *attr,
1828                      struct dt_allocation_hint *hint,
1829                      struct dt_object_format *dof,
1830                      struct thandle *th)
1831 {
1832         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1833         int result;
1834
1835         LINVRNT(osd_invariant(obj));
1836         LASSERT(obj->oo_inode == NULL);
1837         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1838                 S_ISFIFO(mode) || S_ISSOCK(mode));
1839
1840         result = osd_mkfile(info, obj, mode, hint, th);
1841         if (result == 0) {
1842                 LASSERT(obj->oo_inode != NULL);
1843                 /*
1844                  * This inode should be marked dirty for i_rdev.  Currently
1845                  * that is done in the osd_attr_init().
1846                  */
1847                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1848                                    attr->la_rdev);
1849         }
1850         LINVRNT(osd_invariant(obj));
1851         return result;
1852 }
1853
1854 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1855                               struct lu_attr *,
1856                               struct dt_allocation_hint *hint,
1857                               struct dt_object_format *dof,
1858                               struct thandle *);
1859
1860 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1861 {
1862         osd_obj_type_f result;
1863
1864         switch (type) {
1865         case DFT_DIR:
1866                 result = osd_mkdir;
1867                 break;
1868         case DFT_REGULAR:
1869                 result = osd_mkreg;
1870                 break;
1871         case DFT_SYM:
1872                 result = osd_mksym;
1873                 break;
1874         case DFT_NODE:
1875                 result = osd_mknod;
1876                 break;
1877         case DFT_INDEX:
1878                 result = osd_mk_index;
1879                 break;
1880
1881         default:
1882                 LBUG();
1883                 break;
1884         }
1885         return result;
1886 }
1887
1888
1889 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1890                         struct dt_object *parent, struct dt_object *child,
1891                         cfs_umode_t child_mode)
1892 {
1893         LASSERT(ah);
1894
1895         memset(ah, 0, sizeof(*ah));
1896         ah->dah_parent = parent;
1897         ah->dah_mode = child_mode;
1898 }
1899
1900 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1901                           struct lu_attr *attr, struct dt_object_format *dof)
1902 {
1903         struct inode   *inode = obj->oo_inode;
1904         __u64           valid = attr->la_valid;
1905         int             result;
1906
1907         attr->la_valid &= ~(LA_TYPE | LA_MODE);
1908
1909         if (dof->dof_type != DFT_NODE)
1910                 attr->la_valid &= ~LA_RDEV;
1911         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1912                 attr->la_valid &= ~LA_ATIME;
1913         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1914                 attr->la_valid &= ~LA_CTIME;
1915         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1916                 attr->la_valid &= ~LA_MTIME;
1917
1918         result = osd_quota_transfer(inode, attr);
1919         if (result)
1920                 return;
1921
1922         if (attr->la_valid != 0) {
1923                 result = osd_inode_setattr(info->oti_env, inode, attr);
1924                 /*
1925                  * The osd_inode_setattr() should always succeed here.  The
1926                  * only error that could be returned is EDQUOT when we are
1927                  * trying to change the UID or GID of the inode. However, this
1928                  * should not happen since quota enforcement is no longer
1929                  * enabled on ldiskfs (lquota takes care of it).
1930                  */
1931                 LASSERTF(result == 0, "%d", result);
1932                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1933         }
1934
1935         attr->la_valid = valid;
1936 }
1937
1938 /**
1939  * Helper function for osd_object_create()
1940  *
1941  * \retval 0, on success
1942  */
1943 static int __osd_object_create(struct osd_thread_info *info,
1944                                struct osd_object *obj, struct lu_attr *attr,
1945                                struct dt_allocation_hint *hint,
1946                                struct dt_object_format *dof,
1947                                struct thandle *th)
1948 {
1949         int     result;
1950         __u32   umask;
1951
1952         /* we drop umask so that permissions we pass are not affected */
1953         umask = current->fs->umask;
1954         current->fs->umask = 0;
1955
1956         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1957                                                   th);
1958         if (result == 0) {
1959                 osd_attr_init(info, obj, attr, dof);
1960                 osd_object_init0(obj);
1961                 /* bz 24037 */
1962                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1963                         unlock_new_inode(obj->oo_inode);
1964         }
1965
1966         /* restore previous umask value */
1967         current->fs->umask = umask;
1968
1969         return result;
1970 }
1971
1972 /**
1973  * Helper function for osd_object_create()
1974  *
1975  * \retval 0, on success
1976  */
1977 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1978                            const struct lu_fid *fid, struct thandle *th)
1979 {
1980         struct osd_thread_info *info = osd_oti_get(env);
1981         struct osd_inode_id    *id   = &info->oti_id;
1982         struct osd_device      *osd  = osd_obj2dev(obj);
1983
1984         LASSERT(obj->oo_inode != NULL);
1985
1986         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1987         return osd_oi_insert(info, osd, fid, id, th);
1988 }
1989
1990 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1991                    const struct lu_fid *fid, struct lu_seq_range *range)
1992 {
1993         struct seq_server_site  *ss = osd_seq_site(osd);
1994         int                     rc;
1995
1996         if (fid_is_idif(fid)) {
1997                 range->lsr_flags = LU_SEQ_RANGE_OST;
1998                 range->lsr_index = fid_idif_ost_idx(fid);
1999                 return 0;
2000         }
2001
2002         if (!fid_seq_in_fldb(fid_seq(fid))) {
2003                 range->lsr_flags = LU_SEQ_RANGE_MDT;
2004                 if (ss != NULL)
2005                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2006                          * at all */
2007                         range->lsr_index = ss->ss_node_id;
2008                 return 0;
2009         }
2010
2011         LASSERT(ss != NULL);
2012         range->lsr_flags = -1;
2013         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
2014         if (rc != 0) {
2015                 CERROR("%s can not find "DFID": rc = %d\n",
2016                        osd_name(osd), PFID(fid), rc);
2017         }
2018         return rc;
2019 }
2020
2021 /*
2022  * Concurrency: no external locking is necessary.
2023  */
2024 static int osd_declare_object_create(const struct lu_env *env,
2025                                      struct dt_object *dt,
2026                                      struct lu_attr *attr,
2027                                      struct dt_allocation_hint *hint,
2028                                      struct dt_object_format *dof,
2029                                      struct thandle *handle)
2030 {
2031         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
2032         struct osd_thandle      *oh;
2033         int                      rc;
2034         ENTRY;
2035
2036         LASSERT(handle != NULL);
2037
2038         oh = container_of0(handle, struct osd_thandle, ot_super);
2039         LASSERT(oh->ot_handle == NULL);
2040
2041         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2042                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2043         if (!fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2044                            lu_object_fid(&dt->do_lu)))
2045                 /* Reuse idle OI block may cause additional one OI block
2046                  * to be changed. */
2047                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2048                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2049
2050         /* If this is directory, then we expect . and .. to be inserted as
2051          * well. The one directory block always needs to be created for the
2052          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2053          * block), there is no danger of needing a tree for the first block.
2054          */
2055         if (attr && S_ISDIR(attr->la_mode)) {
2056                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2057                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2058                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2059         }
2060
2061         if (!attr)
2062                 RETURN(0);
2063
2064         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2065                                    false, false, NULL, false);
2066         if (rc != 0)
2067                 RETURN(rc);
2068
2069         /* It does fld look up inside declare, and the result will be
2070          * added to fld cache, so the following fld lookup inside insert
2071          * does not need send RPC anymore, so avoid send rpc with holding
2072          * transaction */
2073         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2074                 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2075                 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2076                                lu_object_fid(&dt->do_lu), range);
2077
2078
2079         RETURN(rc);
2080 }
2081
2082 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2083                              struct lu_attr *attr,
2084                              struct dt_allocation_hint *hint,
2085                              struct dt_object_format *dof,
2086                              struct thandle *th)
2087 {
2088         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2089         struct osd_object      *obj    = osd_dt_obj(dt);
2090         struct osd_thread_info *info   = osd_oti_get(env);
2091         int result;
2092
2093         ENTRY;
2094
2095         LINVRNT(osd_invariant(obj));
2096         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2097         LASSERT(osd_write_locked(env, obj));
2098         LASSERT(th != NULL);
2099
2100         if (unlikely(fid_is_acct(fid)))
2101                 /* Quota files can't be created from the kernel any more,
2102                  * 'tune2fs -O quota' will take care of creating them */
2103                 RETURN(-EPERM);
2104
2105         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2106         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2107
2108         result = __osd_object_create(info, obj, attr, hint, dof, th);
2109         if (result == 0)
2110                 result = __osd_oi_insert(env, obj, fid, th);
2111
2112         LASSERT(ergo(result == 0,
2113                      dt_object_exists(dt) && !dt_object_remote(dt)));
2114
2115         LASSERT(osd_invariant(obj));
2116         RETURN(result);
2117 }
2118
2119 /**
2120  * Called to destroy on-disk representation of the object
2121  *
2122  * Concurrency: must be locked
2123  */
2124 static int osd_declare_object_destroy(const struct lu_env *env,
2125                                       struct dt_object *dt,
2126                                       struct thandle *th)
2127 {
2128         struct osd_object  *obj = osd_dt_obj(dt);
2129         struct inode       *inode = obj->oo_inode;
2130         struct osd_thandle *oh;
2131         int                 rc;
2132         ENTRY;
2133
2134         oh = container_of0(th, struct osd_thandle, ot_super);
2135         LASSERT(oh->ot_handle == NULL);
2136         LASSERT(inode);
2137
2138         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2139                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2140         /* Recycle idle OI leaf may cause additional three OI blocks
2141          * to be changed. */
2142         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2143                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2144
2145         /* one less inode */
2146         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2147                                    false, true, NULL, false);
2148         if (rc)
2149                 RETURN(rc);
2150         /* data to be truncated */
2151         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2152                                    true, true, NULL, false);
2153         RETURN(rc);
2154 }
2155
2156 static int osd_object_destroy(const struct lu_env *env,
2157                               struct dt_object *dt,
2158                               struct thandle *th)
2159 {
2160         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2161         struct osd_object      *obj = osd_dt_obj(dt);
2162         struct inode           *inode = obj->oo_inode;
2163         struct osd_device      *osd = osd_obj2dev(obj);
2164         struct osd_thandle     *oh;
2165         int                     result;
2166         ENTRY;
2167
2168         oh = container_of0(th, struct osd_thandle, ot_super);
2169         LASSERT(oh->ot_handle);
2170         LASSERT(inode);
2171         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2172
2173         if (unlikely(fid_is_acct(fid)))
2174                 RETURN(-EPERM);
2175
2176         /* Parallel control for OI scrub. For most of cases, there is no
2177          * lock contention. So it will not affect unlink performance. */
2178         mutex_lock(&inode->i_mutex);
2179         if (S_ISDIR(inode->i_mode)) {
2180                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1);
2181                 /* it will check/delete the agent inode for every dir
2182                  * destory, how to optimize it? unlink performance
2183                  * impaction XXX */
2184                 result = osd_delete_from_agent(env, osd, obj, oh);
2185                 if (result != 0 && result != -ENOENT) {
2186                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
2187                                osd_name(osd), PFID(fid), result);
2188                 }
2189                 spin_lock(&obj->oo_guard);
2190                 clear_nlink(inode);
2191                 spin_unlock(&obj->oo_guard);
2192                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2193         }
2194
2195         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2196
2197         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2198         mutex_unlock(&inode->i_mutex);
2199
2200         /* XXX: add to ext3 orphan list */
2201         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2202
2203         /* not needed in the cache anymore */
2204         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2205
2206         RETURN(0);
2207 }
2208
2209 static inline int __osd_xattr_set(struct osd_thread_info *info,
2210                                   struct inode *inode, const char *name,
2211                                   const void *buf, int buflen, int fl)
2212 {
2213         struct dentry *dentry = &info->oti_child_dentry;
2214
2215         ll_vfs_dq_init(inode);
2216         dentry->d_inode = inode;
2217         return inode->i_op->setxattr(dentry, name, buf, buflen, fl);
2218 }
2219
2220 /**
2221  * Put the fid into lustre_mdt_attrs, and then place the structure
2222  * inode's ea. This fid should not be altered during the life time
2223  * of the inode.
2224  *
2225  * \retval +ve, on success
2226  * \retval -ve, on error
2227  *
2228  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2229  */
2230 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2231                    const struct lu_fid *fid)
2232 {
2233         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2234         int                      rc;
2235
2236         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2237                 return 0;
2238
2239         if (OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF) && fid_is_client_visible(fid))
2240                 return 0;
2241
2242         lustre_lma_init(lma, fid);
2243         lustre_lma_swab(lma);
2244
2245         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2246                              XATTR_CREATE);
2247         /* Someone may created the EA by race. */
2248         if (unlikely(rc == -EEXIST))
2249                 rc = 0;
2250         return rc;
2251 }
2252
2253 /**
2254  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2255  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2256  * To have compatilibility with 1.8 ldiskfs driver we need to have
2257  * magic number at start of fid data.
2258  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2259  * its inmemory API.
2260  */
2261 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2262                                   const struct dt_rec *fid)
2263 {
2264         if (!fid_is_client_mdt_visible((const struct lu_fid *)fid)) {
2265                 param->edp_magic = 0;
2266                 return;
2267         }
2268
2269         param->edp_magic = LDISKFS_LUFID_MAGIC;
2270         param->edp_len =  sizeof(struct lu_fid) + 1;
2271         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2272 }
2273
2274 /**
2275  * Try to read the fid from inode ea into dt_rec.
2276  *
2277  * \param fid object fid.
2278  *
2279  * \retval 0 on success
2280  */
2281 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2282                           __u32 ino, struct lu_fid *fid,
2283                           struct osd_inode_id *id)
2284 {
2285         struct osd_thread_info *info  = osd_oti_get(env);
2286         struct inode           *inode;
2287         ENTRY;
2288
2289         osd_id_gen(id, ino, OSD_OII_NOGEN);
2290         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2291         if (IS_ERR(inode))
2292                 RETURN(PTR_ERR(inode));
2293
2294         iput(inode);
2295         RETURN(0);
2296 }
2297
2298 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2299                                         struct inode *dir,
2300                                         struct inode  *parent_dir,
2301                                         const struct dt_rec *dot_fid,
2302                                         const struct dt_rec *dot_dot_fid,
2303                                         struct osd_thandle *oth)
2304 {
2305         struct ldiskfs_dentry_param *dot_ldp;
2306         struct ldiskfs_dentry_param *dot_dot_ldp;
2307
2308         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2309         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2310
2311         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2312         dot_ldp->edp_magic = 0;
2313         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2314                                         dir, dot_ldp, dot_dot_ldp);
2315 }
2316
2317 /**
2318  * Create an local inode for remote entry
2319  */
2320 static struct inode *osd_create_remote_inode(const struct lu_env *env,
2321                                              struct osd_device *osd,
2322                                              struct osd_object *pobj,
2323                                              const struct lu_fid *fid,
2324                                              struct thandle *th)
2325 {
2326         struct osd_thread_info  *info = osd_oti_get(env);
2327         struct inode            *local;
2328         struct osd_thandle      *oh;
2329         int                     rc;
2330         ENTRY;
2331
2332         LASSERT(th);
2333         oh = container_of(th, struct osd_thandle, ot_super);
2334         LASSERT(oh->ot_handle->h_transaction != NULL);
2335
2336         /* FIXME: Insert index api needs to know the mode of
2337          * the remote object. Just use S_IFDIR for now */
2338         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, S_IFDIR);
2339         if (IS_ERR(local)) {
2340                 CERROR("%s: create local error %d\n", osd_name(osd),
2341                        (int)PTR_ERR(local));
2342                 RETURN(local);
2343         }
2344
2345         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2346                 (const struct dt_rec *)lu_object_fid(&pobj->oo_dt.do_lu),
2347                 (const struct dt_rec *)fid, oh);
2348         if (rc != 0) {
2349                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2350                         osd_name(osd), PFID(fid), rc);
2351                 RETURN(ERR_PTR(rc));
2352         }
2353
2354         RETURN(local);
2355 }
2356
2357 /**
2358  * Delete local inode for remote entry
2359  */
2360 static int osd_delete_remote_inode(const struct lu_env *env,
2361                                    struct osd_device *osd,
2362                                    const struct lu_fid *fid,
2363                                     __u32 ino, struct osd_thandle *oh)
2364 {
2365         struct osd_thread_info  *oti = osd_oti_get(env);
2366         struct osd_inode_id     *id = &oti->oti_id;
2367         struct inode            *inode;
2368         ENTRY;
2369
2370         id->oii_ino = le32_to_cpu(ino);
2371         id->oii_gen = OSD_OII_NOGEN;
2372         inode = osd_iget(oti, osd, id);
2373         if (IS_ERR(inode)) {
2374                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2375                        PFID(fid), id->oii_ino, id->oii_gen);
2376                 RETURN(PTR_ERR(inode));
2377         }
2378
2379         clear_nlink(inode);
2380         mark_inode_dirty(inode);
2381         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2382                 osd_name(osd), PFID(fid), inode->i_ino);
2383         iput(inode);
2384         RETURN(0);
2385 }
2386
2387 /**
2388  * OSD layer object create function for interoperability mode (b11826).
2389  * This is mostly similar to osd_object_create(). Only difference being, fid is
2390  * inserted into inode ea here.
2391  *
2392  * \retval   0, on success
2393  * \retval -ve, on error
2394  */
2395 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2396                                 struct lu_attr *attr,
2397                                 struct dt_allocation_hint *hint,
2398                                 struct dt_object_format *dof,
2399                                 struct thandle *th)
2400 {
2401         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2402         struct osd_object      *obj    = osd_dt_obj(dt);
2403         struct osd_thread_info *info   = osd_oti_get(env);
2404         int                     result;
2405
2406         ENTRY;
2407
2408         LASSERT(osd_invariant(obj));
2409         LASSERT(!dt_object_exists(dt) && !dt_object_remote(dt));
2410         LASSERT(osd_write_locked(env, obj));
2411         LASSERT(th != NULL);
2412
2413         if (unlikely(fid_is_acct(fid)))
2414                 /* Quota files can't be created from the kernel any more,
2415                  * 'tune2fs -O quota' will take care of creating them */
2416                 RETURN(-EPERM);
2417
2418         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2419         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2420
2421         result = __osd_object_create(info, obj, attr, hint, dof, th);
2422         if ((result == 0) &&
2423             (fid_is_last_id(fid) ||
2424              !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid)))
2425                 result = osd_ea_fid_set(info, obj->oo_inode, fid);
2426
2427         if (result == 0)
2428                 result = __osd_oi_insert(env, obj, fid, th);
2429
2430         LASSERT(ergo(result == 0,
2431                      dt_object_exists(dt) && !dt_object_remote(dt)));
2432         LINVRNT(osd_invariant(obj));
2433         RETURN(result);
2434 }
2435
2436 static int osd_declare_object_ref_add(const struct lu_env *env,
2437                                       struct dt_object *dt,
2438                                       struct thandle *handle)
2439 {
2440         struct osd_thandle       *oh;
2441
2442         /* it's possible that object doesn't exist yet */
2443         LASSERT(handle != NULL);
2444
2445         oh = container_of0(handle, struct osd_thandle, ot_super);
2446         LASSERT(oh->ot_handle == NULL);
2447
2448         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2449                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2450
2451         return 0;
2452 }
2453
2454 /*
2455  * Concurrency: @dt is write locked.
2456  */
2457 static int osd_object_ref_add(const struct lu_env *env,
2458                               struct dt_object *dt, struct thandle *th)
2459 {
2460         struct osd_object *obj = osd_dt_obj(dt);
2461         struct inode      *inode = obj->oo_inode;
2462
2463         LINVRNT(osd_invariant(obj));
2464         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2465         LASSERT(osd_write_locked(env, obj));
2466         LASSERT(th != NULL);
2467
2468         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2469
2470         /*
2471          * DIR_NLINK feature is set for compatibility reasons if:
2472          * 1) nlinks > LDISKFS_LINK_MAX, or
2473          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2474          *
2475          * It is easier to always set this flag (rather than check and set),
2476          * since it has less overhead, and the superblock will be dirtied
2477          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2478          * do not actually care whether this flag is set or not.
2479          */
2480         spin_lock(&obj->oo_guard);
2481         /* inc_nlink from 0 may cause WARN_ON */
2482         if(inode->i_nlink == 0)
2483                 set_nlink(inode, 1);
2484         else
2485                 inc_nlink(inode);
2486         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2487                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2488                     inode->i_nlink == 2)
2489                         set_nlink(inode, 1);
2490         }
2491         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2492         spin_unlock(&obj->oo_guard);
2493         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2494         LINVRNT(osd_invariant(obj));
2495
2496         return 0;
2497 }
2498
2499 static int osd_declare_object_ref_del(const struct lu_env *env,
2500                                       struct dt_object *dt,
2501                                       struct thandle *handle)
2502 {
2503         struct osd_thandle *oh;
2504
2505         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2506         LASSERT(handle != NULL);
2507
2508         oh = container_of0(handle, struct osd_thandle, ot_super);
2509         LASSERT(oh->ot_handle == NULL);
2510
2511         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2512                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2513
2514         return 0;
2515 }
2516
2517 /*
2518  * Concurrency: @dt is write locked.
2519  */
2520 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2521                               struct thandle *th)
2522 {
2523         struct osd_object *obj = osd_dt_obj(dt);
2524         struct inode      *inode = obj->oo_inode;
2525
2526         LINVRNT(osd_invariant(obj));
2527         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2528         LASSERT(osd_write_locked(env, obj));
2529         LASSERT(th != NULL);
2530
2531         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2532
2533         spin_lock(&obj->oo_guard);
2534         LASSERT(inode->i_nlink > 0);
2535         drop_nlink(inode);
2536         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2537          * then the nlink count is 1. Don't let it be set to 0 or the directory
2538          * inode will be deleted incorrectly. */
2539         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2540                 set_nlink(inode, 1);
2541         spin_unlock(&obj->oo_guard);
2542         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2543         LINVRNT(osd_invariant(obj));
2544
2545         return 0;
2546 }
2547
2548 /*
2549  * Get the 64-bit version for an inode.
2550  */
2551 static int osd_object_version_get(const struct lu_env *env,
2552                                   struct dt_object *dt, dt_obj_version_t *ver)
2553 {
2554         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2555
2556         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2557                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2558         *ver = LDISKFS_I(inode)->i_fs_version;
2559         return 0;
2560 }
2561
2562 /*
2563  * Concurrency: @dt is read locked.
2564  */
2565 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2566                          struct lu_buf *buf, const char *name,
2567                          struct lustre_capa *capa)
2568 {
2569         struct osd_object      *obj    = osd_dt_obj(dt);
2570         struct inode           *inode  = obj->oo_inode;
2571         struct osd_thread_info *info   = osd_oti_get(env);
2572         struct dentry          *dentry = &info->oti_obj_dentry;
2573
2574         /* version get is not real XATTR but uses xattr API */
2575         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2576                 /* for version we are just using xattr API but change inode
2577                  * field instead */
2578                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2579                 osd_object_version_get(env, dt, buf->lb_buf);
2580                 return sizeof(dt_obj_version_t);
2581         }
2582
2583         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2584         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2585
2586         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2587                 return -EACCES;
2588
2589         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2590 }
2591
2592
2593 static int osd_declare_xattr_set(const struct lu_env *env,
2594                                  struct dt_object *dt,
2595                                  const struct lu_buf *buf, const char *name,
2596                                  int fl, struct thandle *handle)
2597 {
2598         struct osd_thandle *oh;
2599
2600         LASSERT(handle != NULL);
2601
2602         oh = container_of0(handle, struct osd_thandle, ot_super);
2603         LASSERT(oh->ot_handle == NULL);
2604
2605         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2606                              strcmp(name, XATTR_NAME_VERSION) == 0 ?
2607                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2608                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2609
2610         return 0;
2611 }
2612
2613 /*
2614  * Set the 64-bit version for object
2615  */
2616 static void osd_object_version_set(const struct lu_env *env,
2617                                    struct dt_object *dt,
2618                                    dt_obj_version_t *new_version)
2619 {
2620         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2621
2622         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2623                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2624
2625         LDISKFS_I(inode)->i_fs_version = *new_version;
2626         /** Version is set after all inode operations are finished,
2627          *  so we should mark it dirty here */
2628         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2629 }
2630
2631 /*
2632  * Concurrency: @dt is write locked.
2633  */
2634 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2635                          const struct lu_buf *buf, const char *name, int fl,
2636                          struct thandle *handle, struct lustre_capa *capa)
2637 {
2638         struct osd_object      *obj      = osd_dt_obj(dt);
2639         struct inode           *inode    = obj->oo_inode;
2640         struct osd_thread_info *info     = osd_oti_get(env);
2641         int                     fs_flags = 0;
2642         ENTRY;
2643
2644         LASSERT(handle != NULL);
2645
2646         /* version set is not real XATTR */
2647         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2648                 /* for version we are just using xattr API but change inode
2649                  * field instead */
2650                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2651                 osd_object_version_set(env, dt, buf->lb_buf);
2652                 return sizeof(dt_obj_version_t);
2653         }
2654
2655         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2656                 return -EACCES;
2657
2658         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2659         if (fl & LU_XATTR_REPLACE)
2660                 fs_flags |= XATTR_REPLACE;
2661
2662         if (fl & LU_XATTR_CREATE)
2663                 fs_flags |= XATTR_CREATE;
2664
2665         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2666                                fs_flags);
2667 }
2668
2669 /*
2670  * Concurrency: @dt is read locked.
2671  */
2672 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2673                           struct lu_buf *buf, struct lustre_capa *capa)
2674 {
2675         struct osd_object      *obj    = osd_dt_obj(dt);
2676         struct inode           *inode  = obj->oo_inode;
2677         struct osd_thread_info *info   = osd_oti_get(env);
2678         struct dentry          *dentry = &info->oti_obj_dentry;
2679
2680         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2681         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2682         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2683
2684         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2685                 return -EACCES;
2686
2687         dentry->d_inode = inode;
2688         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2689 }
2690
2691 static int osd_declare_xattr_del(const struct lu_env *env,
2692                                  struct dt_object *dt, const char *name,
2693                                  struct thandle *handle)
2694 {
2695         struct osd_thandle *oh;
2696
2697         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2698         LASSERT(handle != NULL);
2699
2700         oh = container_of0(handle, struct osd_thandle, ot_super);
2701         LASSERT(oh->ot_handle == NULL);
2702
2703         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2704                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2705
2706         return 0;
2707 }
2708
2709 /*
2710  * Concurrency: @dt is write locked.
2711  */
2712 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2713                          const char *name, struct thandle *handle,
2714                          struct lustre_capa *capa)
2715 {
2716         struct osd_object      *obj    = osd_dt_obj(dt);
2717         struct inode           *inode  = obj->oo_inode;
2718         struct osd_thread_info *info   = osd_oti_get(env);
2719         struct dentry          *dentry = &info->oti_obj_dentry;
2720         int                     rc;
2721
2722         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2723         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2724         LASSERT(handle != NULL);
2725
2726         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2727                 return -EACCES;
2728
2729         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2730
2731         ll_vfs_dq_init(inode);
2732         dentry->d_inode = inode;
2733         rc = inode->i_op->removexattr(dentry, name);
2734         return rc;
2735 }
2736
2737 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2738                                      struct dt_object *dt,
2739                                      struct lustre_capa *old,
2740                                      __u64 opc)
2741 {
2742         struct osd_thread_info *info = osd_oti_get(env);
2743         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2744         struct osd_object *obj = osd_dt_obj(dt);
2745         struct osd_device *dev = osd_obj2dev(obj);
2746         struct lustre_capa_key *key = &info->oti_capa_key;
2747         struct lustre_capa *capa = &info->oti_capa;
2748         struct obd_capa *oc;
2749         struct md_capainfo *ci;
2750         int rc;
2751         ENTRY;
2752
2753         if (!dev->od_fl_capa)
2754                 RETURN(ERR_PTR(-ENOENT));
2755
2756         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
2757         LINVRNT(osd_invariant(obj));
2758
2759         /* renewal sanity check */
2760         if (old && osd_object_auth(env, dt, old, opc))
2761                 RETURN(ERR_PTR(-EACCES));
2762
2763         ci = md_capainfo(env);
2764         if (unlikely(!ci))
2765                 RETURN(ERR_PTR(-ENOENT));
2766
2767         switch (ci->mc_auth) {
2768         case LC_ID_NONE:
2769                 RETURN(NULL);
2770         case LC_ID_PLAIN:
2771                 capa->lc_uid = obj->oo_inode->i_uid;
2772                 capa->lc_gid = obj->oo_inode->i_gid;
2773                 capa->lc_flags = LC_ID_PLAIN;
2774                 break;
2775         case LC_ID_CONVERT: {
2776                 __u32 d[4], s[4];
2777
2778                 s[0] = obj->oo_inode->i_uid;
2779                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2780                 s[2] = obj->oo_inode->i_gid;
2781                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2782                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2783                 if (unlikely(rc))
2784                         RETURN(ERR_PTR(rc));
2785
2786                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2787                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2788                 capa->lc_flags = LC_ID_CONVERT;
2789                 break;
2790         }
2791         default:
2792                 RETURN(ERR_PTR(-EINVAL));
2793         }
2794
2795         capa->lc_fid = *fid;
2796         capa->lc_opc = opc;
2797         capa->lc_flags |= dev->od_capa_alg << 24;
2798         capa->lc_timeout = dev->od_capa_timeout;
2799         capa->lc_expiry = 0;
2800
2801         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2802         if (oc) {
2803                 LASSERT(!capa_is_expired(oc));
2804                 RETURN(oc);
2805         }
2806
2807         spin_lock(&capa_lock);
2808         *key = dev->od_capa_keys[1];
2809         spin_unlock(&capa_lock);
2810
2811         capa->lc_keyid = key->lk_keyid;
2812         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2813
2814         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2815         if (rc) {
2816                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2817                 RETURN(ERR_PTR(rc));
2818         }
2819
2820         oc = capa_add(dev->od_capa_hash, capa);
2821         RETURN(oc);
2822 }
2823
2824 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2825 {
2826         struct osd_object       *obj    = osd_dt_obj(dt);
2827         struct inode            *inode  = obj->oo_inode;
2828         struct osd_thread_info  *info   = osd_oti_get(env);
2829         struct dentry           *dentry = &info->oti_obj_dentry;
2830         struct file             *file   = &info->oti_file;
2831         int                     rc;
2832
2833         ENTRY;
2834
2835         dentry->d_inode = inode;
2836         file->f_dentry = dentry;
2837         file->f_mapping = inode->i_mapping;
2838         file->f_op = inode->i_fop;
2839 #ifndef HAVE_FILE_FSYNC_4ARGS
2840         mutex_lock(&inode->i_mutex);
2841 #endif
2842         rc = do_fsync(file, 0);
2843 #ifndef HAVE_FILE_FSYNC_4ARGS
2844         mutex_unlock(&inode->i_mutex);
2845 #endif
2846         RETURN(rc);
2847 }
2848
2849 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2850                         void **data)
2851 {
2852         struct osd_object *obj = osd_dt_obj(dt);
2853         ENTRY;
2854
2855         *data = (void *)obj->oo_inode;
2856         RETURN(0);
2857 }
2858
2859 /*
2860  * Index operations.
2861  */
2862
2863 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2864                            const struct dt_index_features *feat)
2865 {
2866         struct iam_descr *descr;
2867
2868         if (osd_object_is_root(o))
2869                 return feat == &dt_directory_features;
2870
2871         LASSERT(o->oo_dir != NULL);
2872
2873         descr = o->oo_dir->od_container.ic_descr;
2874         if (feat == &dt_directory_features) {
2875                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2876                         return 1;
2877                 else
2878                         return 0;
2879         } else {
2880                 return
2881                         feat->dif_keysize_min <= descr->id_key_size &&
2882                         descr->id_key_size <= feat->dif_keysize_max &&
2883                         feat->dif_recsize_min <= descr->id_rec_size &&
2884                         descr->id_rec_size <= feat->dif_recsize_max &&
2885                         !(feat->dif_flags & (DT_IND_VARKEY |
2886                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2887                         ergo(feat->dif_flags & DT_IND_UPDATE,
2888                              1 /* XXX check that object (and file system) is
2889                                 * writable */);
2890         }
2891 }
2892
2893 static int osd_iam_container_init(const struct lu_env *env,
2894                                   struct osd_object *obj,
2895                                   struct osd_directory *dir)
2896 {
2897         struct iam_container *bag = &dir->od_container;
2898         int result;
2899
2900         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2901         if (result != 0)
2902                 return result;
2903
2904         result = iam_container_setup(bag);
2905         if (result == 0)
2906                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2907         else
2908                 iam_container_fini(bag);
2909
2910         return result;
2911 }
2912
2913
2914 /*
2915  * Concurrency: no external locking is necessary.
2916  */
2917 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2918                          const struct dt_index_features *feat)
2919 {
2920         int                      result;
2921         int                      skip_iam = 0;
2922         struct osd_object       *obj = osd_dt_obj(dt);
2923
2924         LINVRNT(osd_invariant(obj));
2925
2926         if (osd_object_is_root(obj)) {
2927                 dt->do_index_ops = &osd_index_ea_ops;
2928                 result = 0;
2929         } else if (feat == &dt_directory_features) {
2930                 dt->do_index_ops = &osd_index_ea_ops;
2931                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
2932                         result = 0;
2933                 else
2934                         result = -ENOTDIR;
2935                 skip_iam = 1;
2936         } else if (unlikely(feat == &dt_otable_features)) {
2937                 dt->do_index_ops = &osd_otable_ops;
2938                 return 0;
2939         } else if (unlikely(feat == &dt_acct_features)) {
2940                 dt->do_index_ops = &osd_acct_index_ops;
2941                 result = 0;
2942                 skip_iam = 1;
2943         } else if (!osd_has_index(obj)) {
2944                 struct osd_directory *dir;
2945
2946                 OBD_ALLOC_PTR(dir);
2947                 if (dir != NULL) {
2948
2949                         spin_lock(&obj->oo_guard);
2950                         if (obj->oo_dir == NULL)
2951                                 obj->oo_dir = dir;
2952                         else
2953                                 /*
2954                                  * Concurrent thread allocated container data.
2955                                  */
2956                                 OBD_FREE_PTR(dir);
2957                         spin_unlock(&obj->oo_guard);
2958                         /*
2959                          * Now, that we have container data, serialize its
2960                          * initialization.
2961                          */
2962                         down_write(&obj->oo_ext_idx_sem);
2963                         /*
2964                          * recheck under lock.
2965                          */
2966                         if (!osd_has_index(obj))
2967                                 result = osd_iam_container_init(env, obj, dir);
2968                         else
2969                                 result = 0;
2970                         up_write(&obj->oo_ext_idx_sem);
2971                 } else {
2972                         result = -ENOMEM;
2973                 }
2974         } else {
2975                 result = 0;
2976         }
2977
2978         if (result == 0 && skip_iam == 0) {
2979                 if (!osd_iam_index_probe(env, obj, feat))
2980                         result = -ENOTDIR;
2981         }
2982         LINVRNT(osd_invariant(obj));
2983
2984         if (result == 0 && is_quota_glb_feat(feat) &&
2985             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
2986                 result = osd_quota_migration(env, dt, feat);
2987
2988         return result;
2989 }
2990
2991 static int osd_otable_it_attr_get(const struct lu_env *env,
2992                                  struct dt_object *dt,
2993                                  struct lu_attr *attr,
2994                                  struct lustre_capa *capa)
2995 {
2996         attr->la_valid = 0;
2997         return 0;
2998 }
2999
3000 static const struct dt_object_operations osd_obj_ops = {
3001         .do_read_lock         = osd_object_read_lock,
3002         .do_write_lock        = osd_object_write_lock,
3003         .do_read_unlock       = osd_object_read_unlock,
3004         .do_write_unlock      = osd_object_write_unlock,
3005         .do_write_locked      = osd_object_write_locked,
3006         .do_attr_get          = osd_attr_get,
3007         .do_declare_attr_set  = osd_declare_attr_set,
3008         .do_attr_set          = osd_attr_set,
3009         .do_ah_init           = osd_ah_init,
3010         .do_declare_create    = osd_declare_object_create,
3011         .do_create            = osd_object_create,
3012         .do_declare_destroy   = osd_declare_object_destroy,
3013         .do_destroy           = osd_object_destroy,
3014         .do_index_try         = osd_index_try,
3015         .do_declare_ref_add   = osd_declare_object_ref_add,
3016         .do_ref_add           = osd_object_ref_add,
3017         .do_declare_ref_del   = osd_declare_object_ref_del,
3018         .do_ref_del           = osd_object_ref_del,
3019         .do_xattr_get         = osd_xattr_get,
3020         .do_declare_xattr_set = osd_declare_xattr_set,
3021         .do_xattr_set         = osd_xattr_set,
3022         .do_declare_xattr_del = osd_declare_xattr_del,
3023         .do_xattr_del         = osd_xattr_del,
3024         .do_xattr_list        = osd_xattr_list,
3025         .do_capa_get          = osd_capa_get,
3026         .do_object_sync       = osd_object_sync,
3027         .do_data_get          = osd_data_get,
3028 };
3029
3030 /**
3031  * dt_object_operations for interoperability mode
3032  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3033  */
3034 static const struct dt_object_operations osd_obj_ea_ops = {
3035         .do_read_lock         = osd_object_read_lock,
3036         .do_write_lock        = osd_object_write_lock,
3037         .do_read_unlock       = osd_object_read_unlock,
3038         .do_write_unlock      = osd_object_write_unlock,
3039         .do_write_locked      = osd_object_write_locked,
3040         .do_attr_get          = osd_attr_get,
3041         .do_declare_attr_set  = osd_declare_attr_set,
3042         .do_attr_set          = osd_attr_set,
3043         .do_ah_init           = osd_ah_init,
3044         .do_declare_create    = osd_declare_object_create,
3045         .do_create            = osd_object_ea_create,
3046         .do_declare_destroy   = osd_declare_object_destroy,
3047         .do_destroy           = osd_object_destroy,
3048         .do_index_try         = osd_index_try,
3049         .do_declare_ref_add   = osd_declare_object_ref_add,
3050         .do_ref_add           = osd_object_ref_add,
3051         .do_declare_ref_del   = osd_declare_object_ref_del,
3052         .do_ref_del           = osd_object_ref_del,
3053         .do_xattr_get         = osd_xattr_get,
3054         .do_declare_xattr_set = osd_declare_xattr_set,
3055         .do_xattr_set         = osd_xattr_set,
3056         .do_declare_xattr_del = osd_declare_xattr_del,
3057         .do_xattr_del         = osd_xattr_del,
3058         .do_xattr_list        = osd_xattr_list,
3059         .do_capa_get          = osd_capa_get,
3060         .do_object_sync       = osd_object_sync,
3061         .do_data_get          = osd_data_get,
3062 };
3063
3064 static const struct dt_object_operations osd_obj_otable_it_ops = {
3065         .do_attr_get    = osd_otable_it_attr_get,
3066         .do_index_try   = osd_index_try,
3067 };
3068
3069 static int osd_index_declare_iam_delete(const struct lu_env *env,
3070                                         struct dt_object *dt,
3071                                         const struct dt_key *key,
3072                                         struct thandle *handle)
3073 {
3074         struct osd_thandle    *oh;
3075
3076         oh = container_of0(handle, struct osd_thandle, ot_super);
3077         LASSERT(oh->ot_handle == NULL);
3078
3079         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3080                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3081
3082         return 0;
3083 }
3084
3085 /**
3086  *      delete a (key, value) pair from index \a dt specified by \a key
3087  *
3088  *      \param  dt      osd index object
3089  *      \param  key     key for index
3090  *      \param  rec     record reference
3091  *      \param  handle  transaction handler
3092  *
3093  *      \retval  0  success
3094  *      \retval -ve   failure
3095  */
3096
3097 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3098                                 const struct dt_key *key,
3099                                 struct thandle *handle,
3100                                 struct lustre_capa *capa)
3101 {
3102         struct osd_thread_info *oti = osd_oti_get(env);
3103         struct osd_object      *obj = osd_dt_obj(dt);
3104         struct osd_thandle     *oh;
3105         struct iam_path_descr  *ipd;
3106         struct iam_container   *bag = &obj->oo_dir->od_container;
3107         int                     rc;
3108
3109         ENTRY;
3110
3111         LINVRNT(osd_invariant(obj));
3112         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3113         LASSERT(bag->ic_object == obj->oo_inode);
3114         LASSERT(handle != NULL);
3115
3116         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3117                 RETURN(-EACCES);
3118
3119         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3120
3121         ipd = osd_idx_ipd_get(env, bag);
3122         if (unlikely(ipd == NULL))
3123                 RETURN(-ENOMEM);
3124
3125         oh = container_of0(handle, struct osd_thandle, ot_super);
3126         LASSERT(oh->ot_handle != NULL);
3127         LASSERT(oh->ot_handle->h_transaction != NULL);
3128
3129         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3130                 /* swab quota uid/gid provided by caller */
3131                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3132                 key = (const struct dt_key *)&oti->oti_quota_id;
3133         }
3134
3135         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3136         osd_ipd_put(env, bag, ipd);
3137         LINVRNT(osd_invariant(obj));
3138         RETURN(rc);
3139 }
3140
3141 static int osd_index_declare_ea_delete(const struct lu_env *env,
3142                                        struct dt_object *dt,
3143                                        const struct dt_key *key,
3144                                        struct thandle *handle)
3145 {
3146         struct osd_thandle *oh;
3147         struct inode       *inode;
3148         int                 rc;
3149         ENTRY;
3150
3151         LASSERT(dt_object_exists(dt) && !dt_object_remote(dt));
3152         LASSERT(handle != NULL);
3153
3154         oh = container_of0(handle, struct osd_thandle, ot_super);
3155         LASSERT(oh->ot_handle == NULL);
3156
3157         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3158                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3159
3160         inode = osd_dt_obj(dt)->oo_inode;
3161         LASSERT(inode);
3162
3163         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3164                                    true, true, NULL, false);
3165         RETURN(rc);
3166 }
3167
3168 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3169                                           struct dt_rec *fid)
3170 {
3171         struct osd_fid_pack *rec;
3172         int                  rc = -ENODATA;
3173
3174         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3175                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3176                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3177         }
3178         return rc;
3179 }
3180
3181 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3182                           struct lu_fid *fid)
3183 {
3184         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
3185         struct seq_server_site  *ss = osd_seq_site(osd);
3186         int                     rc;
3187         ENTRY;
3188
3189         /* Those FID seqs, which are not in FLDB, must be local seq */
3190         if (unlikely(!fid_seq_in_fldb(fid_seq(fid)) || ss == NULL))
3191                 RETURN(0);
3192
3193         rc = osd_fld_lookup(env, osd, fid, range);
3194         if (rc != 0) {
3195                 CERROR("%s: Can not lookup fld for "DFID"\n",
3196                        osd_name(osd), PFID(fid));
3197                 RETURN(rc);
3198         }
3199
3200         RETURN(ss->ss_node_id != range->lsr_index);
3201 }
3202
3203 /**
3204  * Index delete function for interoperability mode (b11826).
3205  * It will remove the directory entry added by osd_index_ea_insert().
3206  * This entry is needed to maintain name->fid mapping.
3207  *
3208  * \param key,  key i.e. file entry to be deleted
3209  *
3210  * \retval   0, on success
3211  * \retval -ve, on error
3212  */
3213 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3214                                const struct dt_key *key,