Whamcloud - gitweb
LU-2725 ost: ost_rw_hpreq_check should return 0 or 1
[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 /* simple_mkdir() */
57 #include <lvfs.h>
58
59 /*
60  * struct OBD_{ALLOC,FREE}*()
61  * OBD_FAIL_CHECK
62  */
63 #include <obd_support.h>
64 /* struct ptlrpc_thread */
65 #include <lustre_net.h>
66 #include <lustre_fid.h>
67
68 #include "osd_internal.h"
69
70 /* llo_* api support */
71 #include <md_object.h>
72 #include <lustre_quota.h>
73
74 int ldiskfs_pdo = 1;
75 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
76                 "ldiskfs with parallel directory operations");
77
78 static const char dot[] = ".";
79 static const char dotdot[] = "..";
80 static const char remote_obj_dir[] = "REM_OBJ_DIR";
81
82 static const struct lu_object_operations      osd_lu_obj_ops;
83 static const struct dt_object_operations      osd_obj_ops;
84 static const struct dt_object_operations      osd_obj_ea_ops;
85 static const struct dt_object_operations      osd_obj_otable_it_ops;
86 static const struct dt_index_operations       osd_index_iam_ops;
87 static const struct dt_index_operations       osd_index_ea_ops;
88
89 #ifdef OSD_TRACK_DECLARES
90 int osd_trans_declare_op2rb[] = {
91         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
92         [OSD_OT_PUNCH]          = OSD_OT_MAX,
93         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
94         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
95         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
96         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
97         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
98         [OSD_OT_WRITE]          = OSD_OT_WRITE,
99         [OSD_OT_INSERT]         = OSD_OT_DELETE,
100         [OSD_OT_DELETE]         = OSD_OT_INSERT,
101         [OSD_OT_QUOTA]          = OSD_OT_MAX,
102 };
103 #endif
104
105 static int osd_has_index(const struct osd_object *obj)
106 {
107         return obj->oo_dt.do_index_ops != NULL;
108 }
109
110 static int osd_object_invariant(const struct lu_object *l)
111 {
112         return osd_invariant(osd_obj(l));
113 }
114
115 /*
116  * Concurrency: doesn't matter
117  */
118 static int osd_read_locked(const struct lu_env *env, struct osd_object *o)
119 {
120         return osd_oti_get(env)->oti_r_locks > 0;
121 }
122
123 /*
124  * Concurrency: doesn't matter
125  */
126 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
127 {
128         struct osd_thread_info *oti = osd_oti_get(env);
129         return oti->oti_w_locks > 0 && o->oo_owner == env;
130 }
131
132 /*
133  * Concurrency: doesn't access mutable data
134  */
135 static int osd_root_get(const struct lu_env *env,
136                         struct dt_device *dev, struct lu_fid *f)
137 {
138         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
139         return 0;
140 }
141
142 /*
143  * OSD object methods.
144  */
145
146 /*
147  * Concurrency: no concurrent access is possible that early in object
148  * life-cycle.
149  */
150 static struct lu_object *osd_object_alloc(const struct lu_env *env,
151                                           const struct lu_object_header *hdr,
152                                           struct lu_device *d)
153 {
154         struct osd_object *mo;
155
156         OBD_ALLOC_PTR(mo);
157         if (mo != NULL) {
158                 struct lu_object *l;
159
160                 l = &mo->oo_dt.do_lu;
161                 dt_object_init(&mo->oo_dt, NULL, d);
162                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
163                 l->lo_ops = &osd_lu_obj_ops;
164                 init_rwsem(&mo->oo_sem);
165                 init_rwsem(&mo->oo_ext_idx_sem);
166                 spin_lock_init(&mo->oo_guard);
167                 return l;
168         } else {
169                 return NULL;
170         }
171 }
172
173 static int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
174                        struct dentry *dentry, struct lustre_mdt_attrs *lma)
175 {
176         int rc;
177
178         dentry->d_inode = inode;
179         rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA, (void *)lma,
180                                    sizeof(*lma));
181         if (rc == -ERANGE) {
182                 /* try with old lma size */
183                 rc = inode->i_op->getxattr(dentry, XATTR_NAME_LMA,
184                                            info->oti_mdt_attrs_old,
185                                            LMA_OLD_SIZE);
186                 if (rc > 0)
187                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
188         }
189         if (rc > 0) {
190                 /* Check LMA compatibility */
191                 if (lma->lma_incompat & ~cpu_to_le32(LMA_INCOMPAT_SUPP)) {
192                         CWARN("%.16s: unsupported incompat LMA feature(s) "
193                               "%lx/%#x\n",
194                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
195                               inode->i_ino, le32_to_cpu(lma->lma_incompat) &
196                                                         ~LMA_INCOMPAT_SUPP);
197                         rc = -ENOSYS;
198                 } else {
199                         lustre_lma_swab(lma);
200                         rc = 0;
201                 }
202         } else if (rc == 0) {
203                 rc = -ENODATA;
204         }
205
206         return rc;
207 }
208
209 /*
210  * retrieve object from backend ext fs.
211  **/
212 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
213                        struct osd_inode_id *id)
214 {
215         struct inode *inode = NULL;
216
217         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
218         if (IS_ERR(inode)) {
219                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
220                        id->oii_ino, PTR_ERR(inode));
221         } else if (id->oii_gen != OSD_OII_NOGEN &&
222                    inode->i_generation != id->oii_gen) {
223                 CDEBUG(D_INODE, "unmatched inode: ino = %u, gen0 = %u, "
224                        "gen1 = %u\n",
225                        id->oii_ino, id->oii_gen, inode->i_generation);
226                 iput(inode);
227                 inode = ERR_PTR(-ESTALE);
228         } else if (inode->i_nlink == 0) {
229                 /* due to parallel readdir and unlink,
230                 * we can have dead inode here. */
231                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
232                 make_bad_inode(inode);
233                 iput(inode);
234                 inode = ERR_PTR(-ESTALE);
235         } else if (is_bad_inode(inode)) {
236                 CWARN("%.16s: bad inode: ino = %u\n",
237                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
238                 iput(inode);
239                 inode = ERR_PTR(-ENOENT);
240         } else {
241                 if (id->oii_gen == OSD_OII_NOGEN)
242                         osd_id_gen(id, inode->i_ino, inode->i_generation);
243
244                 /* Do not update file c/mtime in ldiskfs.
245                  * NB: we don't have any lock to protect this because we don't
246                  * have reference on osd_object now, but contention with
247                  * another lookup + attr_set can't happen in the tiny window
248                  * between if (...) and set S_NOCMTIME. */
249                 if (!(inode->i_flags & S_NOCMTIME))
250                         inode->i_flags |= S_NOCMTIME;
251         }
252         return inode;
253 }
254
255 struct inode *osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
256                            struct osd_inode_id *id, struct lu_fid *fid)
257 {
258         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
259         struct inode            *inode;
260         int                      rc;
261
262         inode = osd_iget(info, dev, id);
263         if (IS_ERR(inode))
264                 return inode;
265
266         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
267         if (rc == 0) {
268                 *fid = lma->lma_self_fid;
269         } else if (rc == -ENODATA) {
270                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
271                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
272                 else
273                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
274         } else {
275                 iput(inode);
276                 inode = ERR_PTR(rc);
277         }
278         return inode;
279 }
280
281 static struct inode *
282 osd_iget_verify(struct osd_thread_info *info, struct osd_device *dev,
283                 struct osd_inode_id *id, const struct lu_fid *fid)
284 {
285         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
286         struct inode            *inode;
287         int                      rc;
288
289         inode = osd_iget(info, dev, id);
290         if (IS_ERR(inode))
291                 return inode;
292
293         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
294         if (rc == -ENODATA)
295                 return inode;
296
297         if (rc != 0) {
298                 iput(inode);
299                 return ERR_PTR(rc);
300         }
301
302         if (!lu_fid_eq(fid, &lma->lma_self_fid)) {
303                 CDEBUG(D_LFSCK, "inconsistent obj: "DFID", %lu, "DFID"\n",
304                        PFID(&lma->lma_self_fid), inode->i_ino, PFID(fid));
305                 iput(inode);
306                 return ERR_PTR(-EREMCHG);
307         }
308
309         return inode;
310 }
311
312 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
313                           const struct lu_fid *fid,
314                           const struct lu_object_conf *conf)
315 {
316         struct osd_thread_info *info;
317         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
318         struct osd_device      *dev;
319         struct osd_idmap_cache *oic;
320         struct osd_inode_id    *id;
321         struct inode           *inode;
322         struct osd_scrub       *scrub;
323         struct scrub_file      *sf;
324         int                     result;
325         int                     verify = 0;
326         ENTRY;
327
328         LINVRNT(osd_invariant(obj));
329         LASSERT(obj->oo_inode == NULL);
330         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID, PFID(fid));
331
332         dev = osd_dev(ldev);
333         scrub = &dev->od_scrub;
334         sf = &scrub->os_file;
335         info = osd_oti_get(env);
336         LASSERT(info);
337         oic = &info->oti_cache;
338
339         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
340                 RETURN(-ENOENT);
341
342         /* Search order: 1. per-thread cache. */
343         if (lu_fid_eq(fid, &oic->oic_fid)) {
344                 id = &oic->oic_lid;
345                 goto iget;
346         }
347
348         id = &info->oti_id;
349         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
350                 /* Search order: 2. OI scrub pending list. */
351                 result = osd_oii_lookup(dev, fid, id);
352                 if (result == 0)
353                         goto iget;
354         }
355
356         if (sf->sf_flags & SF_INCONSISTENT)
357                 verify = 1;
358
359         /*
360          * Objects are created as locking anchors or place holders for objects
361          * yet to be created. No need to osd_oi_lookup() at here because FID
362          * shouldn't never be re-used, if it's really a duplicate FID from
363          * unexpected reason, we should be able to detect it later by calling
364          * do_create->osd_oi_insert()
365          */
366         if (conf != NULL && (conf->loc_flags & LOC_F_NEW) != 0)
367                 GOTO(out, result = 0);
368
369         /* Search order: 3. OI files. */
370         result = osd_oi_lookup(info, dev, fid, id);
371         if (result == -ENOENT) {
372                 if (!fid_is_norm(fid) ||
373                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
374                                       sf->sf_oi_bitmap))
375                         GOTO(out, result = 0);
376
377                 goto trigger;
378         }
379
380         if (result != 0)
381                 GOTO(out, result);
382
383 iget:
384         if (verify == 0)
385                 inode = osd_iget(info, dev, id);
386         else
387                 inode = osd_iget_verify(info, dev, id, fid);
388         if (IS_ERR(inode)) {
389                 result = PTR_ERR(inode);
390                 if (result == -ENOENT || result == -ESTALE) {
391                         fid_zero(&oic->oic_fid);
392                         result = 0;
393                 } else if (result == -EREMCHG) {
394
395 trigger:
396                         if (thread_is_running(&scrub->os_thread)) {
397                                 result = -EINPROGRESS;
398                         } else if (!dev->od_noscrub) {
399                                 result = osd_scrub_start(dev);
400                                 LCONSOLE_ERROR("%.16s: trigger OI scrub by RPC "
401                                                "for "DFID", rc = %d [1]\n",
402                                                LDISKFS_SB(osd_sb(dev))->s_es->\
403                                                s_volume_name,PFID(fid), result);
404                                 if (result == 0 || result == -EALREADY)
405                                         result = -EINPROGRESS;
406                                 else
407                                         result = -EREMCHG;
408                         }
409                 }
410
411                 GOTO(out, result);
412         }
413
414         obj->oo_inode = inode;
415         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
416
417         obj->oo_compat_dot_created = 1;
418         obj->oo_compat_dotdot_created = 1;
419
420         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
421                 GOTO(out, result = 0);
422
423         LASSERT(obj->oo_hl_head == NULL);
424         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
425         if (obj->oo_hl_head == NULL) {
426                 obj->oo_inode = NULL;
427                 iput(inode);
428                 GOTO(out, result = -ENOMEM);
429         }
430         GOTO(out, result = 0);
431
432 out:
433         LINVRNT(osd_invariant(obj));
434         return result;
435 }
436
437 /*
438  * Concurrency: shouldn't matter.
439  */
440 static void osd_object_init0(struct osd_object *obj)
441 {
442         LASSERT(obj->oo_inode != NULL);
443         obj->oo_dt.do_body_ops = &osd_body_ops;
444         obj->oo_dt.do_lu.lo_header->loh_attr |=
445                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
446 }
447
448 /*
449  * Concurrency: no concurrent access is possible that early in object
450  * life-cycle.
451  */
452 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
453                            const struct lu_object_conf *conf)
454 {
455         struct osd_object *obj = osd_obj(l);
456         int result;
457
458         LINVRNT(osd_invariant(obj));
459
460         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
461                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
462                 l->lo_header->loh_attr |= LOHA_EXISTS;
463                 return 0;
464         }
465
466         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
467         obj->oo_dt.do_body_ops = &osd_body_ops_new;
468         if (result == 0 && obj->oo_inode != NULL)
469                 osd_object_init0(obj);
470
471         LINVRNT(osd_invariant(obj));
472         return result;
473 }
474
475 /*
476  * Concurrency: no concurrent access is possible that late in object
477  * life-cycle.
478  */
479 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
480 {
481         struct osd_object *obj = osd_obj(l);
482
483         LINVRNT(osd_invariant(obj));
484
485         dt_object_fini(&obj->oo_dt);
486         if (obj->oo_hl_head != NULL)
487                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
488         OBD_FREE_PTR(obj);
489 }
490
491 /*
492  * Concurrency: no concurrent access is possible that late in object
493  * life-cycle.
494  */
495 static void osd_index_fini(struct osd_object *o)
496 {
497         struct iam_container *bag;
498
499         if (o->oo_dir != NULL) {
500                 bag = &o->oo_dir->od_container;
501                 if (o->oo_inode != NULL) {
502                         if (bag->ic_object == o->oo_inode)
503                                 iam_container_fini(bag);
504                 }
505                 OBD_FREE_PTR(o->oo_dir);
506                 o->oo_dir = NULL;
507         }
508 }
509
510 /*
511  * Concurrency: no concurrent access is possible that late in object
512  * life-cycle (for all existing callers, that is. New callers have to provide
513  * their own locking.)
514  */
515 static int osd_inode_unlinked(const struct inode *inode)
516 {
517         return inode->i_nlink == 0;
518 }
519
520 enum {
521         OSD_TXN_OI_DELETE_CREDITS    = 20,
522         OSD_TXN_INODE_DELETE_CREDITS = 20
523 };
524
525 /*
526  * Journal
527  */
528
529 #if OSD_THANDLE_STATS
530 /**
531  * Set time when the handle is allocated
532  */
533 static void osd_th_alloced(struct osd_thandle *oth)
534 {
535         oth->oth_alloced = cfs_time_current();
536 }
537
538 /**
539  * Set time when the handle started
540  */
541 static void osd_th_started(struct osd_thandle *oth)
542 {
543         oth->oth_started = cfs_time_current();
544 }
545
546 /**
547  * Helper function to convert time interval to microseconds packed in
548  * long int.
549  */
550 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
551 {
552         struct timeval val;
553
554         cfs_duration_usec(cfs_time_sub(end, start), &val);
555         return val.tv_sec * 1000000 + val.tv_usec;
556 }
557
558 /**
559  * Check whether the we deal with this handle for too long.
560  */
561 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
562                                 cfs_time_t alloced, cfs_time_t started,
563                                 cfs_time_t closed)
564 {
565         cfs_time_t now = cfs_time_current();
566
567         LASSERT(dev != NULL);
568
569         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
570                             interval_to_usec(alloced, started));
571         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
572                             interval_to_usec(started, closed));
573         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
574                             interval_to_usec(closed, now));
575
576         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
577                 CWARN("transaction handle %p was open for too long: "
578                       "now "CFS_TIME_T" ,"
579                       "alloced "CFS_TIME_T" ,"
580                       "started "CFS_TIME_T" ,"
581                       "closed "CFS_TIME_T"\n",
582                       oth, now, alloced, started, closed);
583                 libcfs_debug_dumpstack(NULL);
584         }
585 }
586
587 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
588 {                                                                       \
589         cfs_time_t __closed = cfs_time_current();                       \
590         cfs_time_t __alloced = oth->oth_alloced;                        \
591         cfs_time_t __started = oth->oth_started;                        \
592                                                                         \
593         expr;                                                           \
594         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
595 }
596
597 #else /* OSD_THANDLE_STATS */
598
599 #define osd_th_alloced(h)                  do {} while(0)
600 #define osd_th_started(h)                  do {} while(0)
601 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
602
603 #endif /* OSD_THANDLE_STATS */
604
605 /*
606  * Concurrency: doesn't access mutable data.
607  */
608 static int osd_param_is_not_sane(const struct osd_device *dev,
609                                  const struct thandle *th)
610 {
611         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
612
613         return oh->ot_credits > osd_journal(dev)->j_max_transaction_buffers;
614 }
615
616 /*
617  * Concurrency: shouldn't matter.
618  */
619 #ifdef HAVE_LDISKFS_JOURNAL_CALLBACK_ADD
620 static void osd_trans_commit_cb(struct super_block *sb,
621                                 struct journal_callback *jcb, int error)
622 #else
623 static void osd_trans_commit_cb(struct journal_callback *jcb, int error)
624 #endif
625 {
626         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
627         struct thandle     *th  = &oh->ot_super;
628         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
629         struct dt_txn_commit_cb *dcb, *tmp;
630
631         LASSERT(oh->ot_handle == NULL);
632
633         if (error)
634                 CERROR("transaction @0x%p commit error: %d\n", th, error);
635
636         dt_txn_hook_commit(th);
637
638         /* call per-transaction callbacks if any */
639         cfs_list_for_each_entry_safe(dcb, tmp, &oh->ot_dcb_list, dcb_linkage) {
640                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
641                          "commit callback entry: magic=%x name='%s'\n",
642                          dcb->dcb_magic, dcb->dcb_name);
643                 cfs_list_del_init(&dcb->dcb_linkage);
644                 dcb->dcb_func(NULL, th, dcb, error);
645         }
646
647         lu_ref_del_at(&lud->ld_reference, oh->ot_dev_link, "osd-tx", th);
648         lu_device_put(lud);
649         th->th_dev = NULL;
650
651         lu_context_exit(&th->th_ctx);
652         lu_context_fini(&th->th_ctx);
653         OBD_FREE_PTR(oh);
654 }
655
656 static struct thandle *osd_trans_create(const struct lu_env *env,
657                                         struct dt_device *d)
658 {
659         struct osd_thread_info *oti = osd_oti_get(env);
660         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
661         struct osd_thandle     *oh;
662         struct thandle         *th;
663         ENTRY;
664
665         /* on pending IO in this thread should left from prev. request */
666         LASSERT(cfs_atomic_read(&iobuf->dr_numreqs) == 0);
667
668         th = ERR_PTR(-ENOMEM);
669         OBD_ALLOC_GFP(oh, sizeof *oh, CFS_ALLOC_IO);
670         if (oh != NULL) {
671                 oh->ot_quota_trans = &oti->oti_quota_trans;
672                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
673                 th = &oh->ot_super;
674                 th->th_dev = d;
675                 th->th_result = 0;
676                 th->th_tags = LCT_TX_HANDLE;
677                 oh->ot_credits = 0;
678                 oti->oti_dev = osd_dt_dev(d);
679                 CFS_INIT_LIST_HEAD(&oh->ot_dcb_list);
680                 osd_th_alloced(oh);
681
682                 memset(oti->oti_declare_ops, 0, OSD_OT_MAX);
683                 memset(oti->oti_declare_ops_rb, 0, OSD_OT_MAX);
684                 memset(oti->oti_declare_ops_cred, 0, OSD_OT_MAX);
685                 oti->oti_rollback = false;
686         }
687         RETURN(th);
688 }
689
690 /*
691  * Concurrency: shouldn't matter.
692  */
693 int osd_trans_start(const struct lu_env *env, struct dt_device *d,
694                     struct thandle *th)
695 {
696         struct osd_thread_info *oti = osd_oti_get(env);
697         struct osd_device  *dev = osd_dt_dev(d);
698         handle_t           *jh;
699         struct osd_thandle *oh;
700         int rc;
701
702         ENTRY;
703
704         LASSERT(current->journal_info == NULL);
705
706         oh = container_of0(th, struct osd_thandle, ot_super);
707         LASSERT(oh != NULL);
708         LASSERT(oh->ot_handle == NULL);
709
710         rc = dt_txn_hook_start(env, d, th);
711         if (rc != 0)
712                 GOTO(out, rc);
713
714         if (unlikely(osd_param_is_not_sane(dev, th))) {
715                 static unsigned long last_printed;
716                 static int last_credits;
717
718                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
719                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
720                       oh->ot_credits,
721                       osd_journal(dev)->j_max_transaction_buffers);
722 #ifdef OSD_TRACK_DECLARES
723                 CWARN("  create: %u/%u, delete: %u/%u, destroy: %u/%u\n",
724                       oti->oti_declare_ops[OSD_OT_CREATE],
725                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
726                       oti->oti_declare_ops[OSD_OT_DELETE],
727                       oti->oti_declare_ops_cred[OSD_OT_DELETE],
728                       oti->oti_declare_ops[OSD_OT_DESTROY],
729                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
730                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
731                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
732                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
733                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
734                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
735                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
736                       oti->oti_declare_ops[OSD_OT_WRITE],
737                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
738                       oti->oti_declare_ops[OSD_OT_PUNCH],
739                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
740                       oti->oti_declare_ops[OSD_OT_QUOTA],
741                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
742                 CWARN("  insert: %u/%u, delete: %u/%u\n",
743                       oti->oti_declare_ops[OSD_OT_INSERT],
744                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
745                       oti->oti_declare_ops[OSD_OT_DESTROY],
746                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
747                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
748                       oti->oti_declare_ops[OSD_OT_REF_ADD],
749                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
750                       oti->oti_declare_ops[OSD_OT_REF_DEL],
751                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
752
753                 if (last_credits != oh->ot_credits &&
754                     time_after(jiffies, last_printed + 60 * HZ)) {
755                         libcfs_debug_dumpstack(NULL);
756                         last_credits = oh->ot_credits;
757                         last_printed = jiffies;
758                 }
759 #endif
760                 /* XXX Limit the credits to 'max_transaction_buffers', and
761                  *     let the underlying filesystem to catch the error if
762                  *     we really need so many credits.
763                  *
764                  *     This should be removed when we can calculate the
765                  *     credits precisely. */
766                 oh->ot_credits = osd_journal(dev)->j_max_transaction_buffers;
767         }
768
769         /*
770          * XXX temporary stuff. Some abstraction layer should
771          * be used.
772          */
773         jh = ldiskfs_journal_start_sb(osd_sb(dev), oh->ot_credits);
774         osd_th_started(oh);
775         if (!IS_ERR(jh)) {
776                 oh->ot_handle = jh;
777                 LASSERT(oti->oti_txns == 0);
778                 lu_context_init(&th->th_ctx, th->th_tags);
779                 lu_context_enter(&th->th_ctx);
780
781                 lu_device_get(&d->dd_lu_dev);
782                 oh->ot_dev_link = lu_ref_add(&d->dd_lu_dev.ld_reference,
783                                              "osd-tx", th);
784                 oti->oti_txns++;
785                 rc = 0;
786         } else {
787                 rc = PTR_ERR(jh);
788         }
789 out:
790         RETURN(rc);
791 }
792
793 /*
794  * Concurrency: shouldn't matter.
795  */
796 static int osd_trans_stop(const struct lu_env *env, struct thandle *th)
797 {
798         int                     rc = 0;
799         struct osd_thandle     *oh;
800         struct osd_thread_info *oti = osd_oti_get(env);
801         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
802         struct qsd_instance    *qsd = oti->oti_dev->od_quota_slave;
803         ENTRY;
804
805         oh = container_of0(th, struct osd_thandle, ot_super);
806
807         if (qsd != NULL)
808                 /* inform the quota slave device that the transaction is
809                  * stopping */
810                 qsd_op_end(env, qsd, oh->ot_quota_trans);
811         oh->ot_quota_trans = NULL;
812
813         if (oh->ot_handle != NULL) {
814                 handle_t *hdl = oh->ot_handle;
815
816                 /*
817                  * add commit callback
818                  * notice we don't do this in osd_trans_start()
819                  * as underlying transaction can change during truncate
820                  */
821                 osd_journal_callback_set(hdl, osd_trans_commit_cb,
822                                          &oh->ot_jcb);
823
824                 LASSERT(oti->oti_txns == 1);
825                 oti->oti_txns--;
826                 rc = dt_txn_hook_stop(env, th);
827                 if (rc != 0)
828                         CERROR("Failure in transaction hook: %d\n", rc);
829
830                 /* hook functions might modify th_sync */
831                 hdl->h_sync = th->th_sync;
832
833                 oh->ot_handle = NULL;
834                 OSD_CHECK_SLOW_TH(oh, oti->oti_dev,
835                                   rc = ldiskfs_journal_stop(hdl));
836                 if (rc != 0)
837                         CERROR("Failure to stop transaction: %d\n", rc);
838         } else {
839                 OBD_FREE_PTR(oh);
840         }
841
842         /* as we want IO to journal and data IO be concurrent, we don't block
843          * awaiting data IO completion in osd_do_bio(), instead we wait here
844          * once transaction is submitted to the journal. all reqular requests
845          * don't do direct IO (except read/write), thus this wait_event becomes
846          * no-op for them.
847          *
848          * IMPORTANT: we have to wait till any IO submited by the thread is
849          * completed otherwise iobuf may be corrupted by different request
850          */
851         cfs_wait_event(iobuf->dr_wait,
852                        cfs_atomic_read(&iobuf->dr_numreqs) == 0);
853         if (!rc)
854                 rc = iobuf->dr_error;
855
856         RETURN(rc);
857 }
858
859 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
860 {
861         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
862                                                ot_super);
863
864         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
865         LASSERT(&dcb->dcb_func != NULL);
866         cfs_list_add(&dcb->dcb_linkage, &oh->ot_dcb_list);
867
868         return 0;
869 }
870
871 /*
872  * Called just before object is freed. Releases all resources except for
873  * object itself (that is released by osd_object_free()).
874  *
875  * Concurrency: no concurrent access is possible that late in object
876  * life-cycle.
877  */
878 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
879 {
880         struct osd_object *obj   = osd_obj(l);
881         struct inode      *inode = obj->oo_inode;
882
883         LINVRNT(osd_invariant(obj));
884
885         /*
886          * If object is unlinked remove fid->ino mapping from object index.
887          */
888
889         osd_index_fini(obj);
890         if (inode != NULL) {
891                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
892                 qid_t                    uid = inode->i_uid;
893                 qid_t                    gid = inode->i_gid;
894
895                 iput(inode);
896                 obj->oo_inode = NULL;
897
898                 if (qsd != NULL) {
899                         struct osd_thread_info  *info = osd_oti_get(env);
900                         struct lquota_id_info   *qi = &info->oti_qi;
901
902                         /* Release granted quota to master if necessary */
903                         qi->lqi_id.qid_uid = uid;
904                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
905
906                         qi->lqi_id.qid_uid = gid;
907                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
908                 }
909         }
910 }
911
912 /*
913  * Concurrency: ->loo_object_release() is called under site spin-lock.
914  */
915 static void osd_object_release(const struct lu_env *env,
916                                struct lu_object *l)
917 {
918 }
919
920 /*
921  * Concurrency: shouldn't matter.
922  */
923 static int osd_object_print(const struct lu_env *env, void *cookie,
924                             lu_printer_t p, const struct lu_object *l)
925 {
926         struct osd_object *o = osd_obj(l);
927         struct iam_descr  *d;
928
929         if (o->oo_dir != NULL)
930                 d = o->oo_dir->od_container.ic_descr;
931         else
932                 d = NULL;
933         return (*p)(env, cookie,
934                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
935                     o, o->oo_inode,
936                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
937                     o->oo_inode ? o->oo_inode->i_generation : 0,
938                     d ? d->id_ops->id_name : "plain");
939 }
940
941 /*
942  * Concurrency: shouldn't matter.
943  */
944 int osd_statfs(const struct lu_env *env, struct dt_device *d,
945                struct obd_statfs *sfs)
946 {
947         struct osd_device  *osd = osd_dt_dev(d);
948         struct super_block *sb = osd_sb(osd);
949         struct kstatfs     *ksfs;
950         int result = 0;
951
952         if (unlikely(osd->od_mnt == NULL))
953                 return -EINPROGRESS;
954
955         /* osd_lproc.c call this without env, allocate ksfs for that case */
956         if (unlikely(env == NULL)) {
957                 OBD_ALLOC_PTR(ksfs);
958                 if (ksfs == NULL)
959                         return -ENOMEM;
960         } else {
961                 ksfs = &osd_oti_get(env)->oti_ksfs;
962         }
963
964         spin_lock(&osd->od_osfs_lock);
965         /* cache 1 second */
966         if (cfs_time_before_64(osd->od_osfs_age, cfs_time_shift_64(-1))) {
967                 result = sb->s_op->statfs(sb->s_root, ksfs);
968                 if (likely(result == 0)) { /* N.B. statfs can't really fail */
969                         osd->od_osfs_age = cfs_time_current_64();
970                         statfs_pack(&osd->od_statfs, ksfs);
971                         if (sb->s_flags & MS_RDONLY)
972                                 sfs->os_state = OS_STATE_READONLY;
973                 }
974         }
975
976         if (likely(result == 0))
977                 *sfs = osd->od_statfs;
978         spin_unlock(&osd->od_osfs_lock);
979
980         if (unlikely(env == NULL))
981                 OBD_FREE_PTR(ksfs);
982
983         return result;
984 }
985
986 /**
987  * Estimate space needed for file creations. We assume the largest filename
988  * which is 2^64 - 1, hence a filename of 20 chars.
989  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
990  */
991 #ifdef __LDISKFS_DIR_REC_LEN
992 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
993 #else
994 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
995 #endif
996
997 /*
998  * Concurrency: doesn't access mutable data.
999  */
1000 static void osd_conf_get(const struct lu_env *env,
1001                          const struct dt_device *dev,
1002                          struct dt_device_param *param)
1003 {
1004         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1005
1006         /*
1007          * XXX should be taken from not-yet-existing fs abstraction layer.
1008          */
1009         param->ddp_mnt = osd_dt_dev(dev)->od_mnt;
1010         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1011         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1012         param->ddp_block_shift  = sb->s_blocksize_bits;
1013         param->ddp_mount_type     = LDD_MT_LDISKFS;
1014         param->ddp_maxbytes       = sb->s_maxbytes;
1015         /* Overhead estimate should be fairly accurate, so we really take a tiny
1016          * error margin which also avoids fragmenting the filesystem too much */
1017         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1018         /* inode are statically allocated, so per-inode space consumption
1019          * is the space consumed by the directory entry */
1020         param->ddp_inodespace     = PER_OBJ_USAGE;
1021         /* per-fragment overhead to be used by the client code */
1022         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1023         param->ddp_mntopts      = 0;
1024         if (test_opt(sb, XATTR_USER))
1025                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1026         if (test_opt(sb, POSIX_ACL))
1027                 param->ddp_mntopts |= MNTOPT_ACL;
1028
1029 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1030         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1031                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE;
1032         else
1033 #endif
1034                 param->ddp_max_ea_size = sb->s_blocksize;
1035
1036 }
1037
1038 /**
1039  * Helper function to get and fill the buffer with input values.
1040  */
1041 static struct lu_buf *osd_buf_get(const struct lu_env *env, void *area, ssize_t len)
1042 {
1043         struct lu_buf *buf;
1044
1045         buf = &osd_oti_get(env)->oti_buf;
1046         buf->lb_buf = area;
1047         buf->lb_len = len;
1048         return buf;
1049 }
1050
1051 /*
1052  * Concurrency: shouldn't matter.
1053  */
1054 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1055 {
1056         CDEBUG(D_HA, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1057         return ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1058 }
1059
1060 /**
1061  * Start commit for OSD device.
1062  *
1063  * An implementation of dt_commit_async method for OSD device.
1064  * Asychronously starts underlayng fs sync and thereby a transaction
1065  * commit.
1066  *
1067  * \param env environment
1068  * \param d dt device
1069  *
1070  * \see dt_device_operations
1071  */
1072 static int osd_commit_async(const struct lu_env *env,
1073                             struct dt_device *d)
1074 {
1075         struct super_block *s = osd_sb(osd_dt_dev(d));
1076         ENTRY;
1077
1078         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1079         RETURN(s->s_op->sync_fs(s, 0));
1080 }
1081
1082 /*
1083  * Concurrency: shouldn't matter.
1084  */
1085
1086 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1087 {
1088         struct super_block *sb = osd_sb(osd_dt_dev(d));
1089         int rc;
1090         ENTRY;
1091
1092         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1093
1094         rc = __lvfs_set_rdonly(sb->s_bdev, LDISKFS_SB(sb)->journal_bdev);
1095         RETURN(rc);
1096 }
1097
1098 /*
1099  * Concurrency: serialization provided by callers.
1100  */
1101 static int osd_init_capa_ctxt(const struct lu_env *env, struct dt_device *d,
1102                               int mode, unsigned long timeout, __u32 alg,
1103                               struct lustre_capa_key *keys)
1104 {
1105         struct osd_device *dev = osd_dt_dev(d);
1106         ENTRY;
1107
1108         dev->od_fl_capa = mode;
1109         dev->od_capa_timeout = timeout;
1110         dev->od_capa_alg = alg;
1111         dev->od_capa_keys = keys;
1112         RETURN(0);
1113 }
1114
1115 /**
1116  * Note: we do not count into QUOTA here.
1117  * If we mount with --data_journal we may need more.
1118  */
1119 const int osd_dto_credits_noquota[DTO_NR] = {
1120         /**
1121          * Insert/Delete.
1122          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1123          * SINGLEDATA_TRANS_BLOCKS(8)
1124          * XXX Note: maybe iam need more, since iam have more level than
1125          *           EXT3 htree.
1126          */
1127         [DTO_INDEX_INSERT]  = 16,
1128         [DTO_INDEX_DELETE]  = 16,
1129         /**
1130          * Used for OI scrub
1131          */
1132         [DTO_INDEX_UPDATE]  = 16,
1133         /**
1134          * Create a object. The same as create object in EXT3.
1135          * DATA_TRANS_BLOCKS(14) +
1136          * INDEX_EXTRA_BLOCKS(8) +
1137          * 3(inode bits, groups, GDT)
1138          */
1139         [DTO_OBJECT_CREATE] = 25,
1140         /**
1141          * XXX: real credits to be fixed
1142          */
1143         [DTO_OBJECT_DELETE] = 25,
1144         /**
1145          * Attr set credits (inode)
1146          */
1147         [DTO_ATTR_SET_BASE] = 1,
1148         /**
1149          * Xattr set. The same as xattr of EXT3.
1150          * DATA_TRANS_BLOCKS(14)
1151          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1152          * are also counted in. Do not know why?
1153          */
1154         [DTO_XATTR_SET]     = 14,
1155         [DTO_LOG_REC]       = 14,
1156         /**
1157          * credits for inode change during write.
1158          */
1159         [DTO_WRITE_BASE]    = 3,
1160         /**
1161          * credits for single block write.
1162          */
1163         [DTO_WRITE_BLOCK]   = 14,
1164         /**
1165          * Attr set credits for chown.
1166          * This is extra credits for setattr, and it is null without quota
1167          */
1168         [DTO_ATTR_SET_CHOWN]= 0
1169 };
1170
1171 static const struct dt_device_operations osd_dt_ops = {
1172         .dt_root_get       = osd_root_get,
1173         .dt_statfs         = osd_statfs,
1174         .dt_trans_create   = osd_trans_create,
1175         .dt_trans_start    = osd_trans_start,
1176         .dt_trans_stop     = osd_trans_stop,
1177         .dt_trans_cb_add   = osd_trans_cb_add,
1178         .dt_conf_get       = osd_conf_get,
1179         .dt_sync           = osd_sync,
1180         .dt_ro             = osd_ro,
1181         .dt_commit_async   = osd_commit_async,
1182         .dt_init_capa_ctxt = osd_init_capa_ctxt,
1183 };
1184
1185 static void osd_object_read_lock(const struct lu_env *env,
1186                                  struct dt_object *dt, unsigned role)
1187 {
1188         struct osd_object *obj = osd_dt_obj(dt);
1189         struct osd_thread_info *oti = osd_oti_get(env);
1190
1191         LINVRNT(osd_invariant(obj));
1192
1193         LASSERT(obj->oo_owner != env);
1194         down_read_nested(&obj->oo_sem, role);
1195
1196         LASSERT(obj->oo_owner == NULL);
1197         oti->oti_r_locks++;
1198 }
1199
1200 static void osd_object_write_lock(const struct lu_env *env,
1201                                   struct dt_object *dt, unsigned role)
1202 {
1203         struct osd_object *obj = osd_dt_obj(dt);
1204         struct osd_thread_info *oti = osd_oti_get(env);
1205
1206         LINVRNT(osd_invariant(obj));
1207
1208         LASSERT(obj->oo_owner != env);
1209         down_write_nested(&obj->oo_sem, role);
1210
1211         LASSERT(obj->oo_owner == NULL);
1212         obj->oo_owner = env;
1213         oti->oti_w_locks++;
1214 }
1215
1216 static void osd_object_read_unlock(const struct lu_env *env,
1217                                    struct dt_object *dt)
1218 {
1219         struct osd_object *obj = osd_dt_obj(dt);
1220         struct osd_thread_info *oti = osd_oti_get(env);
1221
1222         LINVRNT(osd_invariant(obj));
1223
1224         LASSERT(oti->oti_r_locks > 0);
1225         oti->oti_r_locks--;
1226         up_read(&obj->oo_sem);
1227 }
1228
1229 static void osd_object_write_unlock(const struct lu_env *env,
1230                                     struct dt_object *dt)
1231 {
1232         struct osd_object *obj = osd_dt_obj(dt);
1233         struct osd_thread_info *oti = osd_oti_get(env);
1234
1235         LINVRNT(osd_invariant(obj));
1236
1237         LASSERT(obj->oo_owner == env);
1238         LASSERT(oti->oti_w_locks > 0);
1239         oti->oti_w_locks--;
1240         obj->oo_owner = NULL;
1241         up_write(&obj->oo_sem);
1242 }
1243
1244 static int osd_object_write_locked(const struct lu_env *env,
1245                                    struct dt_object *dt)
1246 {
1247         struct osd_object *obj = osd_dt_obj(dt);
1248
1249         LINVRNT(osd_invariant(obj));
1250
1251         return obj->oo_owner == env;
1252 }
1253
1254 static int capa_is_sane(const struct lu_env *env,
1255                         struct osd_device *dev,
1256                         struct lustre_capa *capa,
1257                         struct lustre_capa_key *keys)
1258 {
1259         struct osd_thread_info *oti = osd_oti_get(env);
1260         struct lustre_capa *tcapa = &oti->oti_capa;
1261         struct obd_capa *oc;
1262         int i, rc = 0;
1263         ENTRY;
1264
1265         oc = capa_lookup(dev->od_capa_hash, capa, 0);
1266         if (oc) {
1267                 if (capa_is_expired(oc)) {
1268                         DEBUG_CAPA(D_ERROR, capa, "expired");
1269                         rc = -ESTALE;
1270                 }
1271                 capa_put(oc);
1272                 RETURN(rc);
1273         }
1274
1275         if (capa_is_expired_sec(capa)) {
1276                 DEBUG_CAPA(D_ERROR, capa, "expired");
1277                 RETURN(-ESTALE);
1278         }
1279
1280         spin_lock(&capa_lock);
1281         for (i = 0; i < 2; i++) {
1282                 if (keys[i].lk_keyid == capa->lc_keyid) {
1283                         oti->oti_capa_key = keys[i];
1284                         break;
1285                 }
1286         }
1287         spin_unlock(&capa_lock);
1288
1289         if (i == 2) {
1290                 DEBUG_CAPA(D_ERROR, capa, "no matched capa key");
1291                 RETURN(-ESTALE);
1292         }
1293
1294         rc = capa_hmac(tcapa->lc_hmac, capa, oti->oti_capa_key.lk_key);
1295         if (rc)
1296                 RETURN(rc);
1297
1298         if (memcmp(tcapa->lc_hmac, capa->lc_hmac, sizeof(capa->lc_hmac))) {
1299                 DEBUG_CAPA(D_ERROR, capa, "HMAC mismatch");
1300                 RETURN(-EACCES);
1301         }
1302
1303         oc = capa_add(dev->od_capa_hash, capa);
1304         capa_put(oc);
1305
1306         RETURN(0);
1307 }
1308
1309 int osd_object_auth(const struct lu_env *env, struct dt_object *dt,
1310                     struct lustre_capa *capa, __u64 opc)
1311 {
1312         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
1313         struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1314         struct md_capainfo *ci;
1315         int rc;
1316
1317         if (!dev->od_fl_capa)
1318                 return 0;
1319
1320         if (capa == BYPASS_CAPA)
1321                 return 0;
1322
1323         ci = md_capainfo(env);
1324         if (unlikely(!ci))
1325                 return 0;
1326
1327         if (ci->mc_auth == LC_ID_NONE)
1328                 return 0;
1329
1330         if (!capa) {
1331                 CERROR("no capability is provided for fid "DFID"\n", PFID(fid));
1332                 return -EACCES;
1333         }
1334
1335         if (!lu_fid_eq(fid, &capa->lc_fid)) {
1336                 DEBUG_CAPA(D_ERROR, capa, "fid "DFID" mismatch with",
1337                            PFID(fid));
1338                 return -EACCES;
1339         }
1340
1341         if (!capa_opc_supported(capa, opc)) {
1342                 DEBUG_CAPA(D_ERROR, capa, "opc "LPX64" not supported by", opc);
1343                 return -EACCES;
1344         }
1345
1346         if ((rc = capa_is_sane(env, dev, capa, dev->od_capa_keys))) {
1347                 DEBUG_CAPA(D_ERROR, capa, "insane (rc %d)", rc);
1348                 return -EACCES;
1349         }
1350
1351         return 0;
1352 }
1353
1354 static struct timespec *osd_inode_time(const struct lu_env *env,
1355                                        struct inode *inode, __u64 seconds)
1356 {
1357         struct osd_thread_info  *oti = osd_oti_get(env);
1358         struct timespec         *t   = &oti->oti_time;
1359
1360         t->tv_sec = seconds;
1361         t->tv_nsec = 0;
1362         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1363         return t;
1364 }
1365
1366
1367 static void osd_inode_getattr(const struct lu_env *env,
1368                               struct inode *inode, struct lu_attr *attr)
1369 {
1370         attr->la_valid      |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1371                                LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1372                                LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE;
1373
1374         attr->la_atime      = LTIME_S(inode->i_atime);
1375         attr->la_mtime      = LTIME_S(inode->i_mtime);
1376         attr->la_ctime      = LTIME_S(inode->i_ctime);
1377         attr->la_mode       = inode->i_mode;
1378         attr->la_size       = i_size_read(inode);
1379         attr->la_blocks     = inode->i_blocks;
1380         attr->la_uid        = inode->i_uid;
1381         attr->la_gid        = inode->i_gid;
1382         attr->la_flags      = LDISKFS_I(inode)->i_flags;
1383         attr->la_nlink      = inode->i_nlink;
1384         attr->la_rdev       = inode->i_rdev;
1385         attr->la_blksize    = 1 << inode->i_blkbits;
1386         attr->la_blkbits    = inode->i_blkbits;
1387 }
1388
1389 static int osd_attr_get(const struct lu_env *env,
1390                         struct dt_object *dt,
1391                         struct lu_attr *attr,
1392                         struct lustre_capa *capa)
1393 {
1394         struct osd_object *obj = osd_dt_obj(dt);
1395
1396         LASSERT(dt_object_exists(dt));
1397         LINVRNT(osd_invariant(obj));
1398
1399         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
1400                 return -EACCES;
1401
1402         spin_lock(&obj->oo_guard);
1403         osd_inode_getattr(env, obj->oo_inode, attr);
1404         spin_unlock(&obj->oo_guard);
1405         return 0;
1406 }
1407
1408 static int osd_declare_attr_set(const struct lu_env *env,
1409                                 struct dt_object *dt,
1410                                 const struct lu_attr *attr,
1411                                 struct thandle *handle)
1412 {
1413         struct osd_thandle     *oh;
1414         struct osd_object      *obj;
1415         struct osd_thread_info *info = osd_oti_get(env);
1416         struct lquota_id_info  *qi = &info->oti_qi;
1417         long long               bspace;
1418         int                     rc = 0;
1419         bool                    allocated;
1420         ENTRY;
1421
1422         LASSERT(dt != NULL);
1423         LASSERT(handle != NULL);
1424
1425         obj = osd_dt_obj(dt);
1426         LASSERT(osd_invariant(obj));
1427
1428         oh = container_of0(handle, struct osd_thandle, ot_super);
1429         LASSERT(oh->ot_handle == NULL);
1430
1431         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1432                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1433
1434         if (attr == NULL || obj->oo_inode == NULL)
1435                 RETURN(rc);
1436
1437         bspace   = obj->oo_inode->i_blocks;
1438         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1439         bspace   = toqb(bspace);
1440
1441         /* Changing ownership is always preformed by super user, it should not
1442          * fail with EDQUOT.
1443          *
1444          * We still need to call the osd_declare_qid() to calculate the journal
1445          * credits for updating quota accounting files and to trigger quota
1446          * space adjustment once the operation is completed.*/
1447         if ((attr->la_valid & LA_UID) != 0 &&
1448              attr->la_uid != obj->oo_inode->i_uid) {
1449                 qi->lqi_type = USRQUOTA;
1450
1451                 /* inode accounting */
1452                 qi->lqi_is_blk = false;
1453
1454                 /* one more inode for the new owner ... */
1455                 qi->lqi_id.qid_uid = attr->la_uid;
1456                 qi->lqi_space      = 1;
1457                 allocated = (attr->la_uid == 0) ? true : false;
1458                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1459                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1460                         rc = 0;
1461                 if (rc)
1462                         RETURN(rc);
1463
1464                 /* and one less inode for the current uid */
1465                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1466                 qi->lqi_space      = -1;
1467                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1468                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1469                         rc = 0;
1470                 if (rc)
1471                         RETURN(rc);
1472
1473                 /* block accounting */
1474                 qi->lqi_is_blk = true;
1475
1476                 /* more blocks for the new owner ... */
1477                 qi->lqi_id.qid_uid = attr->la_uid;
1478                 qi->lqi_space      = bspace;
1479                 allocated = (attr->la_uid == 0) ? true : false;
1480                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1481                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1482                         rc = 0;
1483                 if (rc)
1484                         RETURN(rc);
1485
1486                 /* and finally less blocks for the current owner */
1487                 qi->lqi_id.qid_uid = obj->oo_inode->i_uid;
1488                 qi->lqi_space      = -bspace;
1489                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1490                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1491                         rc = 0;
1492                 if (rc)
1493                         RETURN(rc);
1494         }
1495
1496         if (attr->la_valid & LA_GID &&
1497             attr->la_gid != obj->oo_inode->i_gid) {
1498                 qi->lqi_type = GRPQUOTA;
1499
1500                 /* inode accounting */
1501                 qi->lqi_is_blk = false;
1502
1503                 /* one more inode for the new group owner ... */
1504                 qi->lqi_id.qid_gid = attr->la_gid;
1505                 qi->lqi_space      = 1;
1506                 allocated = (attr->la_gid == 0) ? true : false;
1507                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1508                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1509                         rc = 0;
1510                 if (rc)
1511                         RETURN(rc);
1512
1513                 /* and one less inode for the current gid */
1514                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1515                 qi->lqi_space      = -1;
1516                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1517                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1518                         rc = 0;
1519                 if (rc)
1520                         RETURN(rc);
1521
1522                 /* block accounting */
1523                 qi->lqi_is_blk = true;
1524
1525                 /* more blocks for the new owner ... */
1526                 qi->lqi_id.qid_gid = attr->la_gid;
1527                 qi->lqi_space      = bspace;
1528                 allocated = (attr->la_gid == 0) ? true : false;
1529                 rc = osd_declare_qid(env, oh, qi, allocated, NULL);
1530                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1531                         rc = 0;
1532                 if (rc)
1533                         RETURN(rc);
1534
1535                 /* and finally less blocks for the current owner */
1536                 qi->lqi_id.qid_gid = obj->oo_inode->i_gid;
1537                 qi->lqi_space      = -bspace;
1538                 rc = osd_declare_qid(env, oh, qi, true, NULL);
1539                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1540                         rc = 0;
1541                 if (rc)
1542                         RETURN(rc);
1543         }
1544
1545         RETURN(rc);
1546 }
1547
1548 static int osd_inode_setattr(const struct lu_env *env,
1549                              struct inode *inode, const struct lu_attr *attr)
1550 {
1551         __u64 bits;
1552
1553         bits = attr->la_valid;
1554
1555         LASSERT(!(bits & LA_TYPE)); /* Huh? You want too much. */
1556
1557         if (bits & LA_ATIME)
1558                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1559         if (bits & LA_CTIME)
1560                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1561         if (bits & LA_MTIME)
1562                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1563         if (bits & LA_SIZE) {
1564                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1565                 i_size_write(inode, attr->la_size);
1566         }
1567
1568 #if 0
1569         /* OSD should not change "i_blocks" which is used by quota.
1570          * "i_blocks" should be changed by ldiskfs only. */
1571         if (bits & LA_BLOCKS)
1572                 inode->i_blocks = attr->la_blocks;
1573 #endif
1574         if (bits & LA_MODE)
1575                 inode->i_mode   = (inode->i_mode & S_IFMT) |
1576                         (attr->la_mode & ~S_IFMT);
1577         if (bits & LA_UID)
1578                 inode->i_uid    = attr->la_uid;
1579         if (bits & LA_GID)
1580                 inode->i_gid    = attr->la_gid;
1581         if (bits & LA_NLINK)
1582                 set_nlink(inode, attr->la_nlink);
1583         if (bits & LA_RDEV)
1584                 inode->i_rdev   = attr->la_rdev;
1585
1586         if (bits & LA_FLAGS) {
1587                 /* always keep S_NOCMTIME */
1588                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1589                                  S_NOCMTIME;
1590         }
1591         return 0;
1592 }
1593
1594 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1595 {
1596         if ((attr->la_valid & LA_UID && attr->la_uid != inode->i_uid) ||
1597             (attr->la_valid & LA_GID && attr->la_gid != inode->i_gid)) {
1598                 struct iattr    iattr;
1599                 int             rc;
1600
1601                 iattr.ia_valid = 0;
1602                 if (attr->la_valid & LA_UID)
1603                         iattr.ia_valid |= ATTR_UID;
1604                 if (attr->la_valid & LA_GID)
1605                         iattr.ia_valid |= ATTR_GID;
1606                 iattr.ia_uid = attr->la_uid;
1607                 iattr.ia_gid = attr->la_gid;
1608
1609                 rc = ll_vfs_dq_transfer(inode, &iattr);
1610                 if (rc) {
1611                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1612                                "enforcement enabled on the ldiskfs filesystem?",
1613                                inode->i_sb->s_id, rc);
1614                         return rc;
1615                 }
1616         }
1617         return 0;
1618 }
1619
1620 static int osd_attr_set(const struct lu_env *env,
1621                         struct dt_object *dt,
1622                         const struct lu_attr *attr,
1623                         struct thandle *handle,
1624                         struct lustre_capa *capa)
1625 {
1626         struct osd_object *obj = osd_dt_obj(dt);
1627         struct inode      *inode;
1628         int rc;
1629
1630         LASSERT(handle != NULL);
1631         LASSERT(dt_object_exists(dt));
1632         LASSERT(osd_invariant(obj));
1633
1634         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
1635                 return -EACCES;
1636
1637         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1638
1639         inode = obj->oo_inode;
1640         ll_vfs_dq_init(inode);
1641
1642         rc = osd_quota_transfer(inode, attr);
1643         if (rc)
1644                 return rc;
1645
1646         spin_lock(&obj->oo_guard);
1647         rc = osd_inode_setattr(env, inode, attr);
1648         spin_unlock(&obj->oo_guard);
1649
1650         if (!rc)
1651                 inode->i_sb->s_op->dirty_inode(inode);
1652         return rc;
1653 }
1654
1655 struct dentry *osd_child_dentry_get(const struct lu_env *env,
1656                                     struct osd_object *obj,
1657                                     const char *name, const int namelen)
1658 {
1659         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1660 }
1661
1662 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1663                       cfs_umode_t mode,
1664                       struct dt_allocation_hint *hint,
1665                       struct thandle *th)
1666 {
1667         int result;
1668         struct osd_device  *osd = osd_obj2dev(obj);
1669         struct osd_thandle *oth;
1670         struct dt_object   *parent = NULL;
1671         struct inode       *inode;
1672
1673         LINVRNT(osd_invariant(obj));
1674         LASSERT(obj->oo_inode == NULL);
1675         LASSERT(obj->oo_hl_head == NULL);
1676
1677         if (S_ISDIR(mode) && ldiskfs_pdo) {
1678                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1679                 if (obj->oo_hl_head == NULL)
1680                         return -ENOMEM;
1681         }
1682
1683         oth = container_of(th, struct osd_thandle, ot_super);
1684         LASSERT(oth->ot_handle->h_transaction != NULL);
1685
1686         if (hint && hint->dah_parent)
1687                 parent = hint->dah_parent;
1688
1689         inode = ldiskfs_create_inode(oth->ot_handle,
1690                                      parent ? osd_dt_obj(parent)->oo_inode :
1691                                               osd_sb(osd)->s_root->d_inode,
1692                                      mode);
1693         if (!IS_ERR(inode)) {
1694                 /* Do not update file c/mtime in ldiskfs.
1695                  * NB: don't need any lock because no contention at this
1696                  * early stage */
1697                 inode->i_flags |= S_NOCMTIME;
1698                 inode->i_state |= I_LUSTRE_NOSCRUB;
1699                 obj->oo_inode = inode;
1700                 result = 0;
1701         } else {
1702                 if (obj->oo_hl_head != NULL) {
1703                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1704                         obj->oo_hl_head = NULL;
1705                 }
1706                 result = PTR_ERR(inode);
1707         }
1708         LINVRNT(osd_invariant(obj));
1709         return result;
1710 }
1711
1712 enum {
1713         OSD_NAME_LEN = 255
1714 };
1715
1716 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1717                      struct lu_attr *attr,
1718                      struct dt_allocation_hint *hint,
1719                      struct dt_object_format *dof,
1720                      struct thandle *th)
1721 {
1722         int result;
1723         struct osd_thandle *oth;
1724         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1725
1726         LASSERT(S_ISDIR(attr->la_mode));
1727
1728         oth = container_of(th, struct osd_thandle, ot_super);
1729         LASSERT(oth->ot_handle->h_transaction != NULL);
1730         result = osd_mkfile(info, obj, mode, hint, th);
1731
1732         return result;
1733 }
1734
1735 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1736                         struct lu_attr *attr,
1737                         struct dt_allocation_hint *hint,
1738                         struct dt_object_format *dof,
1739                         struct thandle *th)
1740 {
1741         int result;
1742         struct osd_thandle *oth;
1743         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1744
1745         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1746
1747         LASSERT(S_ISREG(attr->la_mode));
1748
1749         oth = container_of(th, struct osd_thandle, ot_super);
1750         LASSERT(oth->ot_handle->h_transaction != NULL);
1751
1752         result = osd_mkfile(info, obj, mode, hint, th);
1753         if (result == 0) {
1754                 LASSERT(obj->oo_inode != NULL);
1755                 if (feat->dif_flags & DT_IND_VARKEY)
1756                         result = iam_lvar_create(obj->oo_inode,
1757                                                  feat->dif_keysize_max,
1758                                                  feat->dif_ptrsize,
1759                                                  feat->dif_recsize_max,
1760                                                  oth->ot_handle);
1761                 else
1762                         result = iam_lfix_create(obj->oo_inode,
1763                                                  feat->dif_keysize_max,
1764                                                  feat->dif_ptrsize,
1765                                                  feat->dif_recsize_max,
1766                                                  oth->ot_handle);
1767
1768         }
1769         return result;
1770 }
1771
1772 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1773                      struct lu_attr *attr,
1774                      struct dt_allocation_hint *hint,
1775                      struct dt_object_format *dof,
1776                      struct thandle *th)
1777 {
1778         LASSERT(S_ISREG(attr->la_mode));
1779         return osd_mkfile(info, obj, (attr->la_mode &
1780                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1781 }
1782
1783 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1784                      struct lu_attr *attr,
1785                      struct dt_allocation_hint *hint,
1786                      struct dt_object_format *dof,
1787                      struct thandle *th)
1788 {
1789         LASSERT(S_ISLNK(attr->la_mode));
1790         return osd_mkfile(info, obj, (attr->la_mode &
1791                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1792 }
1793
1794 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1795                      struct lu_attr *attr,
1796                      struct dt_allocation_hint *hint,
1797                      struct dt_object_format *dof,
1798                      struct thandle *th)
1799 {
1800         cfs_umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1801         int result;
1802
1803         LINVRNT(osd_invariant(obj));
1804         LASSERT(obj->oo_inode == NULL);
1805         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1806                 S_ISFIFO(mode) || S_ISSOCK(mode));
1807
1808         result = osd_mkfile(info, obj, mode, hint, th);
1809         if (result == 0) {
1810                 LASSERT(obj->oo_inode != NULL);
1811                 /*
1812                  * This inode should be marked dirty for i_rdev.  Currently
1813                  * that is done in the osd_attr_init().
1814                  */
1815                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
1816                                    attr->la_rdev);
1817         }
1818         LINVRNT(osd_invariant(obj));
1819         return result;
1820 }
1821
1822 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
1823                               struct lu_attr *,
1824                               struct dt_allocation_hint *hint,
1825                               struct dt_object_format *dof,
1826                               struct thandle *);
1827
1828 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
1829 {
1830         osd_obj_type_f result;
1831
1832         switch (type) {
1833         case DFT_DIR:
1834                 result = osd_mkdir;
1835                 break;
1836         case DFT_REGULAR:
1837                 result = osd_mkreg;
1838                 break;
1839         case DFT_SYM:
1840                 result = osd_mksym;
1841                 break;
1842         case DFT_NODE:
1843                 result = osd_mknod;
1844                 break;
1845         case DFT_INDEX:
1846                 result = osd_mk_index;
1847                 break;
1848
1849         default:
1850                 LBUG();
1851                 break;
1852         }
1853         return result;
1854 }
1855
1856
1857 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
1858                         struct dt_object *parent, struct dt_object *child,
1859                         cfs_umode_t child_mode)
1860 {
1861         LASSERT(ah);
1862
1863         memset(ah, 0, sizeof(*ah));
1864         ah->dah_parent = parent;
1865         ah->dah_mode = child_mode;
1866 }
1867
1868 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
1869                           struct lu_attr *attr, struct dt_object_format *dof)
1870 {
1871         struct inode   *inode = obj->oo_inode;
1872         __u64           valid = attr->la_valid;
1873         int             result;
1874
1875         attr->la_valid &= ~(LA_TYPE | LA_MODE);
1876
1877         if (dof->dof_type != DFT_NODE)
1878                 attr->la_valid &= ~LA_RDEV;
1879         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
1880                 attr->la_valid &= ~LA_ATIME;
1881         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
1882                 attr->la_valid &= ~LA_CTIME;
1883         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
1884                 attr->la_valid &= ~LA_MTIME;
1885
1886         result = osd_quota_transfer(inode, attr);
1887         if (result)
1888                 return;
1889
1890         if (attr->la_valid != 0) {
1891                 result = osd_inode_setattr(info->oti_env, inode, attr);
1892                 /*
1893                  * The osd_inode_setattr() should always succeed here.  The
1894                  * only error that could be returned is EDQUOT when we are
1895                  * trying to change the UID or GID of the inode. However, this
1896                  * should not happen since quota enforcement is no longer
1897                  * enabled on ldiskfs (lquota takes care of it).
1898                  */
1899                 LASSERTF(result == 0, "%d", result);
1900                 inode->i_sb->s_op->dirty_inode(inode);
1901         }
1902
1903         attr->la_valid = valid;
1904 }
1905
1906 /**
1907  * Helper function for osd_object_create()
1908  *
1909  * \retval 0, on success
1910  */
1911 static int __osd_object_create(struct osd_thread_info *info,
1912                                struct osd_object *obj, struct lu_attr *attr,
1913                                struct dt_allocation_hint *hint,
1914                                struct dt_object_format *dof,
1915                                struct thandle *th)
1916 {
1917         int     result;
1918
1919         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
1920                                                   th);
1921         if (result == 0) {
1922                 osd_attr_init(info, obj, attr, dof);
1923                 osd_object_init0(obj);
1924                 /* bz 24037 */
1925                 if (obj->oo_inode && (obj->oo_inode->i_state & I_NEW))
1926                         unlock_new_inode(obj->oo_inode);
1927         }
1928
1929         return result;
1930 }
1931
1932 /**
1933  * Helper function for osd_object_create()
1934  *
1935  * \retval 0, on success
1936  */
1937 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
1938                            const struct lu_fid *fid, struct thandle *th)
1939 {
1940         struct osd_thread_info *info = osd_oti_get(env);
1941         struct osd_inode_id    *id   = &info->oti_id;
1942         struct osd_device      *osd  = osd_obj2dev(obj);
1943
1944         LASSERT(obj->oo_inode != NULL);
1945
1946         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
1947         return osd_oi_insert(info, osd, fid, id, th);
1948 }
1949
1950 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
1951                    const struct lu_fid *fid, struct lu_seq_range *range)
1952 {
1953         struct seq_server_site  *ss = osd_seq_site(osd);
1954         int                     rc;
1955
1956         if (fid_is_igif(fid)) {
1957                 range->lsr_flags = LU_SEQ_RANGE_MDT;
1958                 range->lsr_index = 0;
1959                 return 0;
1960         }
1961
1962         if (fid_is_idif(fid)) {
1963                 range->lsr_flags = LU_SEQ_RANGE_OST;
1964                 range->lsr_index = fid_idif_ost_idx(fid);
1965                 return 0;
1966         }
1967
1968         if (!fid_is_norm(fid)) {
1969                 range->lsr_flags = LU_SEQ_RANGE_MDT;
1970                 if (ss != NULL)
1971                         /* FIXME: If ss is NULL, it suppose not get lsr_index
1972                          * at all */
1973                         range->lsr_index = ss->ss_node_id;
1974                 return 0;
1975         }
1976
1977         LASSERT(ss != NULL);
1978         range->lsr_flags = -1;
1979         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1980         if (rc != 0) {
1981                 CERROR("%s can not find "DFID": rc = %d\n",
1982                        osd2lu_dev(osd)->ld_obd->obd_name, PFID(fid), rc);
1983         }
1984         return rc;
1985 }
1986
1987
1988 static int osd_declare_object_create(const struct lu_env *env,
1989                                      struct dt_object *dt,
1990                                      struct lu_attr *attr,
1991                                      struct dt_allocation_hint *hint,
1992                                      struct dt_object_format *dof,
1993                                      struct thandle *handle)
1994 {
1995         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
1996         struct osd_thandle      *oh;
1997         int                      rc;
1998         ENTRY;
1999
2000         LASSERT(handle != NULL);
2001
2002         oh = container_of0(handle, struct osd_thandle, ot_super);
2003         LASSERT(oh->ot_handle == NULL);
2004
2005         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2006                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2007         /* XXX: So far, only normal fid needs be inserted into the oi,
2008          *      things could be changed later. Revise following code then. */
2009         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2010             !fid_is_on_ost(osd_oti_get(env), osd_dt_dev(handle->th_dev),
2011                            lu_object_fid(&dt->do_lu))) {
2012                 /* Reuse idle OI block may cause additional one OI block
2013                  * to be changed. */
2014                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2015                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2016         }
2017         /* If this is directory, then we expect . and .. to be inserted as
2018          * well. The one directory block always needs to be created for the
2019          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2020          * block), there is no danger of needing a tree for the first block.
2021          */
2022         if (attr && S_ISDIR(attr->la_mode)) {
2023                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2024                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2025                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2026         }
2027
2028         if (!attr)
2029                 RETURN(0);
2030
2031         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2032                                    false, false, NULL, false);
2033         if (rc != 0)
2034                 RETURN(rc);
2035
2036         /* It does fld look up inside declare, and the result will be
2037          * added to fld cache, so the following fld lookup inside insert
2038          * does not need send RPC anymore, so avoid send rpc with holding
2039          * transaction */
2040         if (fid_is_norm(lu_object_fid(&dt->do_lu)) &&
2041                 !fid_is_last_id(lu_object_fid(&dt->do_lu)))
2042                 osd_fld_lookup(env, osd_dt_dev(handle->th_dev),
2043                                lu_object_fid(&dt->do_lu), range);
2044
2045
2046         RETURN(rc);
2047 }
2048
2049 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2050                              struct lu_attr *attr,
2051                              struct dt_allocation_hint *hint,
2052                              struct dt_object_format *dof,
2053                              struct thandle *th)
2054 {
2055         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2056         struct osd_object      *obj    = osd_dt_obj(dt);
2057         struct osd_thread_info *info   = osd_oti_get(env);
2058         int result;
2059
2060         ENTRY;
2061
2062         LINVRNT(osd_invariant(obj));
2063         LASSERT(!dt_object_exists(dt));
2064         LASSERT(osd_write_locked(env, obj));
2065         LASSERT(th != NULL);
2066
2067         if (unlikely(fid_is_acct(fid)))
2068                 /* Quota files can't be created from the kernel any more,
2069                  * 'tune2fs -O quota' will take care of creating them */
2070                 RETURN(-EPERM);
2071
2072         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2073         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2074
2075         result = __osd_object_create(info, obj, attr, hint, dof, th);
2076         if (result == 0)
2077                 result = __osd_oi_insert(env, obj, fid, th);
2078
2079         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2080         LASSERT(osd_invariant(obj));
2081         RETURN(result);
2082 }
2083
2084 /**
2085  * Called to destroy on-disk representation of the object
2086  *
2087  * Concurrency: must be locked
2088  */
2089 static int osd_declare_object_destroy(const struct lu_env *env,
2090                                       struct dt_object *dt,
2091                                       struct thandle *th)
2092 {
2093         struct osd_object  *obj = osd_dt_obj(dt);
2094         struct inode       *inode = obj->oo_inode;
2095         struct osd_thandle *oh;
2096         int                 rc;
2097         ENTRY;
2098
2099         oh = container_of0(th, struct osd_thandle, ot_super);
2100         LASSERT(oh->ot_handle == NULL);
2101         LASSERT(inode);
2102
2103         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2104                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2105         /* XXX: So far, only normal fid needs to be inserted into the OI,
2106          *      so only normal fid needs to be removed from the OI also.
2107          * Recycle idle OI leaf may cause additional three OI blocks
2108          * to be changed. */
2109         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2110                              fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2111                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3 : 0);
2112
2113         /* one less inode */
2114         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, -1, oh,
2115                                    false, true, NULL, false);
2116         if (rc)
2117                 RETURN(rc);
2118         /* data to be truncated */
2119         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
2120                                    true, true, NULL, false);
2121         RETURN(rc);
2122 }
2123
2124 static int osd_object_destroy(const struct lu_env *env,
2125                               struct dt_object *dt,
2126                               struct thandle *th)
2127 {
2128         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2129         struct osd_object      *obj = osd_dt_obj(dt);
2130         struct inode           *inode = obj->oo_inode;
2131         struct osd_device      *osd = osd_obj2dev(obj);
2132         struct osd_thandle     *oh;
2133         int                     result;
2134         ENTRY;
2135
2136         oh = container_of0(th, struct osd_thandle, ot_super);
2137         LASSERT(oh->ot_handle);
2138         LASSERT(inode);
2139         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2140
2141         if (unlikely(fid_is_acct(fid)))
2142                 RETURN(-EPERM);
2143
2144         /* Parallel control for OI scrub. For most of cases, there is no
2145          * lock contention. So it will not affect unlink performance. */
2146         mutex_lock(&inode->i_mutex);
2147         if (S_ISDIR(inode->i_mode)) {
2148                 LASSERT(osd_inode_unlinked(inode) ||
2149                         inode->i_nlink == 1);
2150                 spin_lock(&obj->oo_guard);
2151                 clear_nlink(inode);
2152                 spin_unlock(&obj->oo_guard);
2153                 inode->i_sb->s_op->dirty_inode(inode);
2154         }
2155
2156         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2157
2158         result = osd_oi_delete(osd_oti_get(env), osd, fid, th);
2159         mutex_unlock(&inode->i_mutex);
2160
2161         /* XXX: add to ext3 orphan list */
2162         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2163
2164         /* not needed in the cache anymore */
2165         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2166
2167         RETURN(0);
2168 }
2169
2170 /**
2171  * Helper function for osd_xattr_set()
2172  */
2173 static int __osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2174                            const struct lu_buf *buf, const char *name, int fl)
2175 {
2176         struct osd_object      *obj      = osd_dt_obj(dt);
2177         struct inode           *inode    = obj->oo_inode;
2178         struct osd_thread_info *info     = osd_oti_get(env);
2179         struct dentry          *dentry   = &info->oti_child_dentry;
2180         int                     fs_flags = 0;
2181         int                     rc;
2182
2183         LASSERT(dt_object_exists(dt));
2184         LASSERT(inode->i_op != NULL && inode->i_op->setxattr != NULL);
2185
2186         if (fl & LU_XATTR_REPLACE)
2187                 fs_flags |= XATTR_REPLACE;
2188
2189         if (fl & LU_XATTR_CREATE)
2190                 fs_flags |= XATTR_CREATE;
2191
2192         ll_vfs_dq_init(inode);
2193         dentry->d_inode = inode;
2194         rc = inode->i_op->setxattr(dentry, name, buf->lb_buf,
2195                                    buf->lb_len, fs_flags);
2196         return rc;
2197 }
2198
2199 /**
2200  * Put the fid into lustre_mdt_attrs, and then place the structure
2201  * inode's ea. This fid should not be altered during the life time
2202  * of the inode.
2203  *
2204  * \retval +ve, on success
2205  * \retval -ve, on error
2206  *
2207  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2208  */
2209 static int osd_ea_fid_set(const struct lu_env *env, struct dt_object *dt,
2210                           const struct lu_fid *fid)
2211 {
2212         struct osd_thread_info  *info      = osd_oti_get(env);
2213         struct lustre_mdt_attrs *mdt_attrs = &info->oti_mdt_attrs;
2214
2215         lustre_lma_init(mdt_attrs, fid);
2216         lustre_lma_swab(mdt_attrs);
2217         return __osd_xattr_set(env, dt,
2218                                osd_buf_get(env, mdt_attrs, sizeof *mdt_attrs),
2219                                XATTR_NAME_LMA, LU_XATTR_CREATE);
2220
2221 }
2222
2223 /**
2224  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2225  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2226  * To have compatilibility with 1.8 ldiskfs driver we need to have
2227  * magic number at start of fid data.
2228  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2229  * its inmemory API.
2230  */
2231 void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2232                                   const struct dt_rec *fid)
2233 {
2234         /* XXX: replace the check with "!fid_is_client_mdt_visible()"
2235          *      when FID in OI file introduced for local object. */
2236         if (!fid_is_norm((const struct lu_fid *)fid) &&
2237             !fid_is_igif((const struct lu_fid *)fid)) {
2238                 param->edp_magic = 0;
2239                 return;
2240         }
2241
2242         param->edp_magic = LDISKFS_LUFID_MAGIC;
2243         param->edp_len =  sizeof(struct lu_fid) + 1;
2244         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2245 }
2246
2247 /**
2248  * Try to read the fid from inode ea into dt_rec, if return value
2249  * i.e. rc is +ve, then we got fid, otherwise we will have to form igif
2250  *
2251  * \param fid object fid.
2252  *
2253  * \retval 0 on success
2254  */
2255 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2256                           __u32 ino, struct lu_fid *fid,
2257                           struct osd_inode_id *id)
2258 {
2259         struct osd_thread_info *info  = osd_oti_get(env);
2260         struct inode           *inode;
2261         ENTRY;
2262
2263         osd_id_gen(id, ino, OSD_OII_NOGEN);
2264         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2265         if (IS_ERR(inode))
2266                 RETURN(PTR_ERR(inode));
2267
2268         iput(inode);
2269         RETURN(0);
2270 }
2271
2272 /**
2273  * OSD layer object create function for interoperability mode (b11826).
2274  * This is mostly similar to osd_object_create(). Only difference being, fid is
2275  * inserted into inode ea here.
2276  *
2277  * \retval   0, on success
2278  * \retval -ve, on error
2279  */
2280 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2281                                 struct lu_attr *attr,
2282                                 struct dt_allocation_hint *hint,
2283                                 struct dt_object_format *dof,
2284                                 struct thandle *th)
2285 {
2286         const struct lu_fid    *fid    = lu_object_fid(&dt->do_lu);
2287         struct osd_object      *obj    = osd_dt_obj(dt);
2288         struct osd_thread_info *info   = osd_oti_get(env);
2289         int                     result;
2290
2291         ENTRY;
2292
2293         LASSERT(osd_invariant(obj));
2294         LASSERT(!dt_object_exists(dt));
2295         LASSERT(osd_write_locked(env, obj));
2296         LASSERT(th != NULL);
2297
2298         if (unlikely(fid_is_acct(fid)))
2299                 /* Quota files can't be created from the kernel any more,
2300                  * 'tune2fs -O quota' will take care of creating them */
2301                 RETURN(-EPERM);
2302
2303         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2304         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2305
2306         result = __osd_object_create(info, obj, attr, hint, dof, th);
2307         /* objects under osd root shld have igif fid, so dont add fid EA */
2308         /* For ost object, the fid will be stored during first write */
2309         if (result == 0 && fid_seq(fid) >= FID_SEQ_NORMAL &&
2310             !fid_is_on_ost(info, osd_dt_dev(th->th_dev), fid))
2311                 result = osd_ea_fid_set(env, dt, fid);
2312
2313         if (result == 0)
2314                 result = __osd_oi_insert(env, obj, fid, th);
2315
2316         LASSERT(ergo(result == 0, dt_object_exists(dt)));
2317         LINVRNT(osd_invariant(obj));
2318         RETURN(result);
2319 }
2320
2321 static int osd_declare_object_ref_add(const struct lu_env *env,
2322                                       struct dt_object *dt,
2323                                       struct thandle *handle)
2324 {
2325         struct osd_thandle       *oh;
2326
2327         /* it's possible that object doesn't exist yet */
2328         LASSERT(handle != NULL);
2329
2330         oh = container_of0(handle, struct osd_thandle, ot_super);
2331         LASSERT(oh->ot_handle == NULL);
2332
2333         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2334                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2335
2336         return 0;
2337 }
2338
2339 /*
2340  * Concurrency: @dt is write locked.
2341  */
2342 static int osd_object_ref_add(const struct lu_env *env,
2343                               struct dt_object *dt, struct thandle *th)
2344 {
2345         struct osd_object *obj = osd_dt_obj(dt);
2346         struct inode      *inode = obj->oo_inode;
2347
2348         LINVRNT(osd_invariant(obj));
2349         LASSERT(dt_object_exists(dt));
2350         LASSERT(osd_write_locked(env, obj));
2351         LASSERT(th != NULL);
2352
2353         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2354
2355         /*
2356          * DIR_NLINK feature is set for compatibility reasons if:
2357          * 1) nlinks > LDISKFS_LINK_MAX, or
2358          * 2) nlinks == 2, since this indicates i_nlink was previously 1.
2359          *
2360          * It is easier to always set this flag (rather than check and set),
2361          * since it has less overhead, and the superblock will be dirtied
2362          * at some point. Both e2fsprogs and any Lustre-supported ldiskfs
2363          * do not actually care whether this flag is set or not.
2364          */
2365         spin_lock(&obj->oo_guard);
2366         /* inc_nlink from 0 may cause WARN_ON */
2367         if(inode->i_nlink == 0)
2368                 set_nlink(inode, 1);
2369         else
2370                 inc_nlink(inode);
2371         if (S_ISDIR(inode->i_mode) && inode->i_nlink > 1) {
2372                 if (inode->i_nlink >= LDISKFS_LINK_MAX ||
2373                     inode->i_nlink == 2)
2374                         set_nlink(inode, 1);
2375         }
2376         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2377         spin_unlock(&obj->oo_guard);
2378         inode->i_sb->s_op->dirty_inode(inode);
2379         LINVRNT(osd_invariant(obj));
2380
2381         return 0;
2382 }
2383
2384 static int osd_declare_object_ref_del(const struct lu_env *env,
2385                                       struct dt_object *dt,
2386                                       struct thandle *handle)
2387 {
2388         struct osd_thandle *oh;
2389
2390         LASSERT(dt_object_exists(dt));
2391         LASSERT(handle != NULL);
2392
2393         oh = container_of0(handle, struct osd_thandle, ot_super);
2394         LASSERT(oh->ot_handle == NULL);
2395
2396         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2397                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2398
2399         return 0;
2400 }
2401
2402 /*
2403  * Concurrency: @dt is write locked.
2404  */
2405 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2406                               struct thandle *th)
2407 {
2408         struct osd_object *obj = osd_dt_obj(dt);
2409         struct inode      *inode = obj->oo_inode;
2410
2411         LINVRNT(osd_invariant(obj));
2412         LASSERT(dt_object_exists(dt));
2413         LASSERT(osd_write_locked(env, obj));
2414         LASSERT(th != NULL);
2415
2416         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2417
2418         spin_lock(&obj->oo_guard);
2419         LASSERT(inode->i_nlink > 0);
2420         drop_nlink(inode);
2421         /* If this is/was a many-subdir directory (nlink > LDISKFS_LINK_MAX)
2422          * then the nlink count is 1. Don't let it be set to 0 or the directory
2423          * inode will be deleted incorrectly. */
2424         if (S_ISDIR(inode->i_mode) && inode->i_nlink == 0)
2425                 set_nlink(inode, 1);
2426         spin_unlock(&obj->oo_guard);
2427         inode->i_sb->s_op->dirty_inode(inode);
2428         LINVRNT(osd_invariant(obj));
2429
2430         return 0;
2431 }
2432
2433 /*
2434  * Get the 64-bit version for an inode.
2435  */
2436 static int osd_object_version_get(const struct lu_env *env,
2437                                   struct dt_object *dt, dt_obj_version_t *ver)
2438 {
2439         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2440
2441         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2442                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2443         *ver = LDISKFS_I(inode)->i_fs_version;
2444         return 0;
2445 }
2446
2447 /*
2448  * Concurrency: @dt is read locked.
2449  */
2450 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2451                          struct lu_buf *buf, const char *name,
2452                          struct lustre_capa *capa)
2453 {
2454         struct osd_object      *obj    = osd_dt_obj(dt);
2455         struct inode           *inode  = obj->oo_inode;
2456         struct osd_thread_info *info   = osd_oti_get(env);
2457         struct dentry          *dentry = &info->oti_obj_dentry;
2458
2459         /* version get is not real XATTR but uses xattr API */
2460         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2461                 /* for version we are just using xattr API but change inode
2462                  * field instead */
2463                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2464                 osd_object_version_get(env, dt, buf->lb_buf);
2465                 return sizeof(dt_obj_version_t);
2466         }
2467
2468         LASSERT(dt_object_exists(dt));
2469         LASSERT(inode->i_op != NULL && inode->i_op->getxattr != NULL);
2470
2471         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2472                 return -EACCES;
2473
2474         dentry->d_inode = inode;
2475         return inode->i_op->getxattr(dentry, name, buf->lb_buf, buf->lb_len);
2476 }
2477
2478
2479 static int osd_declare_xattr_set(const struct lu_env *env,
2480                                  struct dt_object *dt,
2481                                  const struct lu_buf *buf, const char *name,
2482                                  int fl, struct thandle *handle)
2483 {
2484         struct osd_thandle *oh;
2485
2486         LASSERT(handle != NULL);
2487
2488         oh = container_of0(handle, struct osd_thandle, ot_super);
2489         LASSERT(oh->ot_handle == NULL);
2490
2491         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2492                              strcmp(name, XATTR_NAME_VERSION) == 0 ?
2493                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE] :
2494                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2495
2496         return 0;
2497 }
2498
2499 /*
2500  * Set the 64-bit version for object
2501  */
2502 static void osd_object_version_set(const struct lu_env *env,
2503                                    struct dt_object *dt,
2504                                    dt_obj_version_t *new_version)
2505 {
2506         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2507
2508         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2509                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2510
2511         LDISKFS_I(inode)->i_fs_version = *new_version;
2512         /** Version is set after all inode operations are finished,
2513          *  so we should mark it dirty here */
2514         inode->i_sb->s_op->dirty_inode(inode);
2515 }
2516
2517 /*
2518  * Concurrency: @dt is write locked.
2519  */
2520 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2521                          const struct lu_buf *buf, const char *name, int fl,
2522                          struct thandle *handle, struct lustre_capa *capa)
2523 {
2524         LASSERT(handle != NULL);
2525
2526         /* version set is not real XATTR */
2527         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2528                 /* for version we are just using xattr API but change inode
2529                  * field instead */
2530                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2531                 osd_object_version_set(env, dt, buf->lb_buf);
2532                 return sizeof(dt_obj_version_t);
2533         }
2534
2535         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2536                 return -EACCES;
2537
2538         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2539         return __osd_xattr_set(env, dt, buf, name, fl);
2540 }
2541
2542 /*
2543  * Concurrency: @dt is read locked.
2544  */
2545 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2546                           struct lu_buf *buf, struct lustre_capa *capa)
2547 {
2548         struct osd_object      *obj    = osd_dt_obj(dt);
2549         struct inode           *inode  = obj->oo_inode;
2550         struct osd_thread_info *info   = osd_oti_get(env);
2551         struct dentry          *dentry = &info->oti_obj_dentry;
2552
2553         LASSERT(dt_object_exists(dt));
2554         LASSERT(inode->i_op != NULL && inode->i_op->listxattr != NULL);
2555         LASSERT(osd_read_locked(env, obj) || osd_write_locked(env, obj));
2556
2557         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_READ))
2558                 return -EACCES;
2559
2560         dentry->d_inode = inode;
2561         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2562 }
2563
2564 static int osd_declare_xattr_del(const struct lu_env *env,
2565                                  struct dt_object *dt, const char *name,
2566                                  struct thandle *handle)
2567 {
2568         struct osd_thandle *oh;
2569
2570         LASSERT(dt_object_exists(dt));
2571         LASSERT(handle != NULL);
2572
2573         oh = container_of0(handle, struct osd_thandle, ot_super);
2574         LASSERT(oh->ot_handle == NULL);
2575
2576         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2577                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2578
2579         return 0;
2580 }
2581
2582 /*
2583  * Concurrency: @dt is write locked.
2584  */
2585 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2586                          const char *name, struct thandle *handle,
2587                          struct lustre_capa *capa)
2588 {
2589         struct osd_object      *obj    = osd_dt_obj(dt);
2590         struct inode           *inode  = obj->oo_inode;
2591         struct osd_thread_info *info   = osd_oti_get(env);
2592         struct dentry          *dentry = &info->oti_obj_dentry;
2593         int                     rc;
2594
2595         LASSERT(dt_object_exists(dt));
2596         LASSERT(inode->i_op != NULL && inode->i_op->removexattr != NULL);
2597         LASSERT(osd_write_locked(env, obj));
2598         LASSERT(handle != NULL);
2599
2600         if (osd_object_auth(env, dt, capa, CAPA_OPC_META_WRITE))
2601                 return -EACCES;
2602
2603         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2604
2605         ll_vfs_dq_init(inode);
2606         dentry->d_inode = inode;
2607         rc = inode->i_op->removexattr(dentry, name);
2608         return rc;
2609 }
2610
2611 static struct obd_capa *osd_capa_get(const struct lu_env *env,
2612                                      struct dt_object *dt,
2613                                      struct lustre_capa *old,
2614                                      __u64 opc)
2615 {
2616         struct osd_thread_info *info = osd_oti_get(env);
2617         const struct lu_fid *fid = lu_object_fid(&dt->do_lu);
2618         struct osd_object *obj = osd_dt_obj(dt);
2619         struct osd_device *dev = osd_obj2dev(obj);
2620         struct lustre_capa_key *key = &info->oti_capa_key;
2621         struct lustre_capa *capa = &info->oti_capa;
2622         struct obd_capa *oc;
2623         struct md_capainfo *ci;
2624         int rc;
2625         ENTRY;
2626
2627         if (!dev->od_fl_capa)
2628                 RETURN(ERR_PTR(-ENOENT));
2629
2630         LASSERT(dt_object_exists(dt));
2631         LINVRNT(osd_invariant(obj));
2632
2633         /* renewal sanity check */
2634         if (old && osd_object_auth(env, dt, old, opc))
2635                 RETURN(ERR_PTR(-EACCES));
2636
2637         ci = md_capainfo(env);
2638         if (unlikely(!ci))
2639                 RETURN(ERR_PTR(-ENOENT));
2640
2641         switch (ci->mc_auth) {
2642         case LC_ID_NONE:
2643                 RETURN(NULL);
2644         case LC_ID_PLAIN:
2645                 capa->lc_uid = obj->oo_inode->i_uid;
2646                 capa->lc_gid = obj->oo_inode->i_gid;
2647                 capa->lc_flags = LC_ID_PLAIN;
2648                 break;
2649         case LC_ID_CONVERT: {
2650                 __u32 d[4], s[4];
2651
2652                 s[0] = obj->oo_inode->i_uid;
2653                 cfs_get_random_bytes(&(s[1]), sizeof(__u32));
2654                 s[2] = obj->oo_inode->i_gid;
2655                 cfs_get_random_bytes(&(s[3]), sizeof(__u32));
2656                 rc = capa_encrypt_id(d, s, key->lk_key, CAPA_HMAC_KEY_MAX_LEN);
2657                 if (unlikely(rc))
2658                         RETURN(ERR_PTR(rc));
2659
2660                 capa->lc_uid   = ((__u64)d[1] << 32) | d[0];
2661                 capa->lc_gid   = ((__u64)d[3] << 32) | d[2];
2662                 capa->lc_flags = LC_ID_CONVERT;
2663                 break;
2664         }
2665         default:
2666                 RETURN(ERR_PTR(-EINVAL));
2667         }
2668
2669         capa->lc_fid = *fid;
2670         capa->lc_opc = opc;
2671         capa->lc_flags |= dev->od_capa_alg << 24;
2672         capa->lc_timeout = dev->od_capa_timeout;
2673         capa->lc_expiry = 0;
2674
2675         oc = capa_lookup(dev->od_capa_hash, capa, 1);
2676         if (oc) {
2677                 LASSERT(!capa_is_expired(oc));
2678                 RETURN(oc);
2679         }
2680
2681         spin_lock(&capa_lock);
2682         *key = dev->od_capa_keys[1];
2683         spin_unlock(&capa_lock);
2684
2685         capa->lc_keyid = key->lk_keyid;
2686         capa->lc_expiry = cfs_time_current_sec() + dev->od_capa_timeout;
2687
2688         rc = capa_hmac(capa->lc_hmac, capa, key->lk_key);
2689         if (rc) {
2690                 DEBUG_CAPA(D_ERROR, capa, "HMAC failed: %d for", rc);
2691                 RETURN(ERR_PTR(rc));
2692         }
2693
2694         oc = capa_add(dev->od_capa_hash, capa);
2695         RETURN(oc);
2696 }
2697
2698 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
2699 {
2700         struct osd_object       *obj    = osd_dt_obj(dt);
2701         struct inode            *inode  = obj->oo_inode;
2702         struct osd_thread_info  *info   = osd_oti_get(env);
2703         struct dentry           *dentry = &info->oti_obj_dentry;
2704         struct file             *file   = &info->oti_file;
2705         int                     rc;
2706
2707         ENTRY;
2708
2709         dentry->d_inode = inode;
2710         file->f_dentry = dentry;
2711         file->f_mapping = inode->i_mapping;
2712         file->f_op = inode->i_fop;
2713         mutex_lock(&inode->i_mutex);
2714         rc = file->f_op->fsync(file, dentry, 0);
2715         mutex_unlock(&inode->i_mutex);
2716         RETURN(rc);
2717 }
2718
2719 static int osd_data_get(const struct lu_env *env, struct dt_object *dt,
2720                         void **data)
2721 {
2722         struct osd_object *obj = osd_dt_obj(dt);
2723         ENTRY;
2724
2725         *data = (void *)obj->oo_inode;
2726         RETURN(0);
2727 }
2728
2729 /*
2730  * Index operations.
2731  */
2732
2733 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
2734                            const struct dt_index_features *feat)
2735 {
2736         struct iam_descr *descr;
2737
2738         if (osd_object_is_root(o))
2739                 return feat == &dt_directory_features;
2740
2741         LASSERT(o->oo_dir != NULL);
2742
2743         descr = o->oo_dir->od_container.ic_descr;
2744         if (feat == &dt_directory_features) {
2745                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
2746                         return 1;
2747                 else
2748                         return 0;
2749         } else {
2750                 return
2751                         feat->dif_keysize_min <= descr->id_key_size &&
2752                         descr->id_key_size <= feat->dif_keysize_max &&
2753                         feat->dif_recsize_min <= descr->id_rec_size &&
2754                         descr->id_rec_size <= feat->dif_recsize_max &&
2755                         !(feat->dif_flags & (DT_IND_VARKEY |
2756                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
2757                         ergo(feat->dif_flags & DT_IND_UPDATE,
2758                              1 /* XXX check that object (and file system) is
2759                                 * writable */);
2760         }
2761 }
2762
2763 static int osd_iam_container_init(const struct lu_env *env,
2764                                   struct osd_object *obj,
2765                                   struct osd_directory *dir)
2766 {
2767         struct iam_container *bag = &dir->od_container;
2768         int result;
2769
2770         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
2771         if (result != 0)
2772                 return result;
2773
2774         result = iam_container_setup(bag);
2775         if (result == 0)
2776                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
2777         else
2778                 iam_container_fini(bag);
2779
2780         return result;
2781 }
2782
2783
2784 /*
2785  * Concurrency: no external locking is necessary.
2786  */
2787 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
2788                          const struct dt_index_features *feat)
2789 {
2790         int                      result;
2791         int                      skip_iam = 0;
2792         struct osd_object       *obj = osd_dt_obj(dt);
2793
2794         LINVRNT(osd_invariant(obj));
2795         LASSERT(dt_object_exists(dt));
2796
2797         if (osd_object_is_root(obj)) {
2798                 dt->do_index_ops = &osd_index_ea_ops;
2799                 result = 0;
2800         } else if (feat == &dt_directory_features) {
2801                 dt->do_index_ops = &osd_index_ea_ops;
2802                 if (S_ISDIR(obj->oo_inode->i_mode))
2803                         result = 0;
2804                 else
2805                         result = -ENOTDIR;
2806                 skip_iam = 1;
2807         } else if (unlikely(feat == &dt_otable_features)) {
2808                 dt->do_index_ops = &osd_otable_ops;
2809                 return 0;
2810         } else if (unlikely(feat == &dt_acct_features)) {
2811                 dt->do_index_ops = &osd_acct_index_ops;
2812                 result = 0;
2813                 skip_iam = 1;
2814         } else if (!osd_has_index(obj)) {
2815                 struct osd_directory *dir;
2816
2817                 OBD_ALLOC_PTR(dir);
2818                 if (dir != NULL) {
2819
2820                         spin_lock(&obj->oo_guard);
2821                         if (obj->oo_dir == NULL)
2822                                 obj->oo_dir = dir;
2823                         else
2824                                 /*
2825                                  * Concurrent thread allocated container data.
2826                                  */
2827                                 OBD_FREE_PTR(dir);
2828                         spin_unlock(&obj->oo_guard);
2829                         /*
2830                          * Now, that we have container data, serialize its
2831                          * initialization.
2832                          */
2833                         down_write(&obj->oo_ext_idx_sem);
2834                         /*
2835                          * recheck under lock.
2836                          */
2837                         if (!osd_has_index(obj))
2838                                 result = osd_iam_container_init(env, obj, dir);
2839                         else
2840                                 result = 0;
2841                         up_write(&obj->oo_ext_idx_sem);
2842                 } else {
2843                         result = -ENOMEM;
2844                 }
2845         } else {
2846                 result = 0;
2847         }
2848
2849         if (result == 0 && skip_iam == 0) {
2850                 if (!osd_iam_index_probe(env, obj, feat))
2851                         result = -ENOTDIR;
2852         }
2853         LINVRNT(osd_invariant(obj));
2854
2855         if (is_quota_glb_feat(feat))
2856                 result = osd_quota_migration(env, dt, feat);
2857
2858         return result;
2859 }
2860
2861 static int osd_otable_it_attr_get(const struct lu_env *env,
2862                                  struct dt_object *dt,
2863                                  struct lu_attr *attr,
2864                                  struct lustre_capa *capa)
2865 {
2866         attr->la_valid = 0;
2867         return 0;
2868 }
2869
2870 static const struct dt_object_operations osd_obj_ops = {
2871         .do_read_lock         = osd_object_read_lock,
2872         .do_write_lock        = osd_object_write_lock,
2873         .do_read_unlock       = osd_object_read_unlock,
2874         .do_write_unlock      = osd_object_write_unlock,
2875         .do_write_locked      = osd_object_write_locked,
2876         .do_attr_get          = osd_attr_get,
2877         .do_declare_attr_set  = osd_declare_attr_set,
2878         .do_attr_set          = osd_attr_set,
2879         .do_ah_init           = osd_ah_init,
2880         .do_declare_create    = osd_declare_object_create,
2881         .do_create            = osd_object_create,
2882         .do_declare_destroy   = osd_declare_object_destroy,
2883         .do_destroy           = osd_object_destroy,
2884         .do_index_try         = osd_index_try,
2885         .do_declare_ref_add   = osd_declare_object_ref_add,
2886         .do_ref_add           = osd_object_ref_add,
2887         .do_declare_ref_del   = osd_declare_object_ref_del,
2888         .do_ref_del           = osd_object_ref_del,
2889         .do_xattr_get         = osd_xattr_get,
2890         .do_declare_xattr_set = osd_declare_xattr_set,
2891         .do_xattr_set         = osd_xattr_set,
2892         .do_declare_xattr_del = osd_declare_xattr_del,
2893         .do_xattr_del         = osd_xattr_del,
2894         .do_xattr_list        = osd_xattr_list,
2895         .do_capa_get          = osd_capa_get,
2896         .do_object_sync       = osd_object_sync,
2897         .do_data_get          = osd_data_get,
2898 };
2899
2900 /**
2901  * dt_object_operations for interoperability mode
2902  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
2903  */
2904 static const struct dt_object_operations osd_obj_ea_ops = {
2905         .do_read_lock         = osd_object_read_lock,
2906         .do_write_lock        = osd_object_write_lock,
2907         .do_read_unlock       = osd_object_read_unlock,
2908         .do_write_unlock      = osd_object_write_unlock,
2909         .do_write_locked      = osd_object_write_locked,
2910         .do_attr_get          = osd_attr_get,
2911         .do_declare_attr_set  = osd_declare_attr_set,
2912         .do_attr_set          = osd_attr_set,
2913         .do_ah_init           = osd_ah_init,
2914         .do_declare_create    = osd_declare_object_create,
2915         .do_create            = osd_object_ea_create,
2916         .do_declare_destroy   = osd_declare_object_destroy,
2917         .do_destroy           = osd_object_destroy,
2918         .do_index_try         = osd_index_try,
2919         .do_declare_ref_add   = osd_declare_object_ref_add,
2920         .do_ref_add           = osd_object_ref_add,
2921         .do_declare_ref_del   = osd_declare_object_ref_del,
2922         .do_ref_del           = osd_object_ref_del,
2923         .do_xattr_get         = osd_xattr_get,
2924         .do_declare_xattr_set = osd_declare_xattr_set,
2925         .do_xattr_set         = osd_xattr_set,
2926         .do_declare_xattr_del = osd_declare_xattr_del,
2927         .do_xattr_del         = osd_xattr_del,
2928         .do_xattr_list        = osd_xattr_list,
2929         .do_capa_get          = osd_capa_get,
2930         .do_object_sync       = osd_object_sync,
2931         .do_data_get          = osd_data_get,
2932 };
2933
2934 static const struct dt_object_operations osd_obj_otable_it_ops = {
2935         .do_attr_get    = osd_otable_it_attr_get,
2936         .do_index_try   = osd_index_try,
2937 };
2938
2939 static int osd_index_declare_iam_delete(const struct lu_env *env,
2940                                         struct dt_object *dt,
2941                                         const struct dt_key *key,
2942                                         struct thandle *handle)
2943 {
2944         struct osd_thandle    *oh;
2945
2946         oh = container_of0(handle, struct osd_thandle, ot_super);
2947         LASSERT(oh->ot_handle == NULL);
2948
2949         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2950                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
2951
2952         return 0;
2953 }
2954
2955 /**
2956  *      delete a (key, value) pair from index \a dt specified by \a key
2957  *
2958  *      \param  dt      osd index object
2959  *      \param  key     key for index
2960  *      \param  rec     record reference
2961  *      \param  handle  transaction handler
2962  *
2963  *      \retval  0  success
2964  *      \retval -ve   failure
2965  */
2966
2967 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
2968                                 const struct dt_key *key,
2969                                 struct thandle *handle,
2970                                 struct lustre_capa *capa)
2971 {
2972         struct osd_thread_info *oti = osd_oti_get(env);
2973         struct osd_object      *obj = osd_dt_obj(dt);
2974         struct osd_thandle     *oh;
2975         struct iam_path_descr  *ipd;
2976         struct iam_container   *bag = &obj->oo_dir->od_container;
2977         int                     rc;
2978
2979         ENTRY;
2980
2981         LINVRNT(osd_invariant(obj));
2982         LASSERT(dt_object_exists(dt));
2983         LASSERT(bag->ic_object == obj->oo_inode);
2984         LASSERT(handle != NULL);
2985
2986         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
2987                 RETURN(-EACCES);
2988
2989         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
2990
2991         ipd = osd_idx_ipd_get(env, bag);
2992         if (unlikely(ipd == NULL))
2993                 RETURN(-ENOMEM);
2994
2995         oh = container_of0(handle, struct osd_thandle, ot_super);
2996         LASSERT(oh->ot_handle != NULL);
2997         LASSERT(oh->ot_handle->h_transaction != NULL);
2998
2999         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3000                 /* swab quota uid/gid provided by caller */
3001                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3002                 key = (const struct dt_key *)&oti->oti_quota_id;
3003         }
3004
3005         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3006         osd_ipd_put(env, bag, ipd);
3007         LINVRNT(osd_invariant(obj));
3008         RETURN(rc);
3009 }
3010
3011 static int osd_index_declare_ea_delete(const struct lu_env *env,
3012                                        struct dt_object *dt,
3013                                        const struct dt_key *key,
3014                                        struct thandle *handle)
3015 {
3016         struct osd_thandle *oh;
3017         struct inode       *inode;
3018         int                 rc;
3019         ENTRY;
3020
3021         LASSERT(dt_object_exists(dt));
3022         LASSERT(handle != NULL);
3023
3024         oh = container_of0(handle, struct osd_thandle, ot_super);
3025         LASSERT(oh->ot_handle == NULL);
3026
3027         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3028                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3029
3030         inode = osd_dt_obj(dt)->oo_inode;
3031         LASSERT(inode);
3032
3033         rc = osd_declare_inode_qid(env, inode->i_uid, inode->i_gid, 0, oh,
3034                                    true, true, NULL, false);
3035         RETURN(rc);
3036 }
3037
3038 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3039                                           struct dt_rec *fid)
3040 {
3041         struct osd_fid_pack *rec;
3042         int                  rc = -ENODATA;
3043
3044         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3045                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3046                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3047         }
3048         RETURN(rc);
3049 }
3050
3051 /**
3052  * Index delete function for interoperability mode (b11826).
3053  * It will remove the directory entry added by osd_index_ea_insert().
3054  * This entry is needed to maintain name->fid mapping.
3055  *
3056  * \param key,  key i.e. file entry to be deleted
3057  *
3058  * \retval   0, on success
3059  * \retval -ve, on error
3060  */
3061 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3062                                const struct dt_key *key,
3063                                struct thandle *handle,
3064                                struct lustre_capa *capa)
3065 {
3066         struct osd_object          *obj    = osd_dt_obj(dt);
3067         struct inode               *dir    = obj->oo_inode;
3068         struct dentry              *dentry;
3069         struct osd_thandle         *oh;
3070         struct ldiskfs_dir_entry_2 *de;
3071         struct buffer_head         *bh;
3072         struct htree_lock          *hlock = NULL;
3073         int                         rc;
3074
3075         ENTRY;
3076
3077         LINVRNT(osd_invariant(obj));
3078         LASSERT(dt_object_exists(dt));
3079         LASSERT(handle != NULL);
3080
3081         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3082
3083         oh = container_of(handle, struct osd_thandle, ot_super);
3084         LASSERT(oh->ot_handle != NULL);
3085         LASSERT(oh->ot_handle->h_transaction != NULL);
3086
3087         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_DELETE))
3088                 RETURN(-EACCES);
3089
3090         ll_vfs_dq_init(dir);
3091         dentry = osd_child_dentry_get(env, obj,
3092                                       (char *)key, strlen((char *)key));
3093
3094         if (obj->oo_hl_head != NULL) {
3095                 hlock = osd_oti_get(env)->oti_hlock;
3096                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3097                                    dir, LDISKFS_HLOCK_DEL);
3098         } else {
3099                 down_write(&obj->oo_ext_idx_sem);
3100         }
3101
3102         bh = osd_ldiskfs_find_entry(dir, dentry, &de, hlock);
3103         if (bh) {
3104                 rc = ldiskfs_delete_entry(oh->ot_handle,
3105                                           dir, de, bh);
3106                 brelse(bh);
3107         } else {
3108                 rc = -ENOENT;
3109         }
3110         if (hlock != NULL)
3111                 ldiskfs_htree_unlock(hlock);
3112         else
3113                 up_write(&obj->oo_ext_idx_sem);
3114
3115         LASSERT(osd_invariant(obj));
3116         RETURN(rc);
3117 }
3118
3119 /**
3120  *      Lookup index for \a key and copy record to \a rec.
3121  *
3122  *      \param  dt      osd index object
3123  *      \param  key     key for index
3124  *      \param  rec     record reference
3125  *
3126  *      \retval  +ve  success : exact mach
3127  *      \retval  0    return record with key not greater than \a key
3128  *      \retval -ve   failure
3129  */
3130 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3131                                 struct dt_rec *rec, const struct dt_key *key,
3132                                 struct lustre_capa *capa)
3133 {
3134         struct osd_object      *obj = osd_dt_obj(dt);
3135         struct iam_path_descr  *ipd;
3136         struct iam_container   *bag = &obj->oo_dir->od_container;
3137         struct osd_thread_info *oti = osd_oti_get(env);
3138         struct iam_iterator    *it = &oti->oti_idx_it;
3139         struct iam_rec         *iam_rec;
3140         int                     rc;
3141
3142         ENTRY;
3143
3144         LASSERT(osd_invariant(obj));
3145         LASSERT(dt_object_exists(dt));
3146         LASSERT(bag->ic_object == obj->oo_inode);
3147
3148         if (osd_object_auth(env, dt, capa, CAPA_OPC_INDEX_LOOKUP))
3149                 RETURN(-EACCES);
3150
3151         ipd = osd_idx_ipd_get(env, bag);
3152         if (IS_ERR(ipd))
3153                 RETURN(-ENOMEM);
3154
3155         /* got ipd now we can start iterator. */
3156         iam_it_init(it, bag, 0, ipd);
3157
3158         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3159                 /* swab quota uid/gid provided by caller */
3160                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3161                 key = (const struct dt_key *)&oti->oti_quota_id;
3162         }
3163
3164         rc = iam_it_get(it, (struct iam_key *)key);
3165         if (rc >= 0) {
3166                 if (S_ISDIR(obj->oo_inode->i_mode))
3167                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3168                 else
3169                         iam_rec = (struct iam_rec *) rec;
3170
3171                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3172
3173                 if (S_ISDIR(obj->oo_inode->i_mode))
3174                         osd_fid_unpack((struct lu_fid *) rec,
3175                                        (struct osd_fid_pack *)iam_rec);
3176                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3177                         osd_quota_unpack(obj, rec);
3178         }
3179
3180         iam_it_put(it);
3181         iam_it_fini(it);
3182         osd_ipd_put(env, bag, ipd);
3183
3184         LINVRNT(osd_invariant(obj));
3185
3186         RETURN(rc);
3187 }
3188
3189 static int osd_index_declare_iam_insert(const struct lu_env *env,
3190                                         struct dt_object *dt,
3191                                         const struct dt_rec *rec,
3192                                         const struct dt_key *key,
3193                                         struct thandle *handle)
3194 {
3195         struct osd_thandle *oh;
3196
3197         LASSERT(dt_object_exists(dt));
3198         LASSERT(handle != NULL);
3199
3200         oh = container_of0(handle, struct osd_thandle, ot_super);
3201         LASSERT(oh->ot_handle == NULL);
3202
3203         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3204                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3205
3206         return 0;
3207 }
3208
3209 /**
3210  *      Inserts (key, value) pair in \a dt index object.
3211  *
3212  *      \param  dt      osd index object
3213  *      \param  key     key for index
3214  *      \param  rec     record reference
3215  *      \param  th      transaction handler
3216  *
3217  *      \retval  0  success
3218  *      \retval -ve failure
3219  */
3220 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3221                                 const struct dt_rec *rec,
3222                                 const struct dt_key *key, struct thandle *th,
3223                 &n