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