Whamcloud - gitweb
669033ffc9b0dadb7a6a8452f6a431a2f32e8df8
[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, 2014, 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_OSD
45
46 #include <linux/module.h>
47 #include <linux/user_namespace.h>
48 #ifdef HAVE_UIDGID_HEADER
49 # include <linux/uidgid.h>
50 #endif
51
52 /* LUSTRE_VERSION_CODE */
53 #include <lustre_ver.h>
54 /* prerequisite for linux/xattr.h */
55 #include <linux/types.h>
56 /* prerequisite for linux/xattr.h */
57 #include <linux/fs.h>
58 /* XATTR_{REPLACE,CREATE} */
59 #include <linux/xattr.h>
60
61 #include <ldiskfs/ldiskfs.h>
62 #include <ldiskfs/xattr.h>
63 #undef ENTRY
64 /*
65  * struct OBD_{ALLOC,FREE}*()
66  * OBD_FAIL_CHECK
67  */
68 #include <obd_support.h>
69 /* struct ptlrpc_thread */
70 #include <lustre_net.h>
71 #include <lustre_fid.h>
72 /* process_config */
73 #include <lustre_param.h>
74
75 #include "osd_internal.h"
76 #include "osd_dynlocks.h"
77
78 /* llo_* api support */
79 #include <md_object.h>
80 #include <lustre_quota.h>
81
82 #include <lustre_linkea.h>
83
84 int ldiskfs_pdo = 1;
85 CFS_MODULE_PARM(ldiskfs_pdo, "i", int, 0644,
86                 "ldiskfs with parallel directory operations");
87
88 int ldiskfs_track_declares_assert;
89 CFS_MODULE_PARM(ldiskfs_track_declares_assert, "i", int, 0644,
90                 "LBUG during tracking of declares");
91
92 /* Slab to allocate dynlocks */
93 struct kmem_cache *dynlock_cachep;
94
95 /* Slab to allocate osd_it_ea */
96 struct kmem_cache *osd_itea_cachep;
97
98 static struct lu_kmem_descr ldiskfs_caches[] = {
99         {
100                 .ckd_cache = &dynlock_cachep,
101                 .ckd_name  = "dynlock_cache",
102                 .ckd_size  = sizeof(struct dynlock_handle)
103         },
104         {
105                 .ckd_cache = &osd_itea_cachep,
106                 .ckd_name  = "osd_itea_cache",
107                 .ckd_size  = sizeof(struct osd_it_ea)
108         },
109         {
110                 .ckd_cache = NULL
111         }
112 };
113
114 static const char dot[] = ".";
115 static const char dotdot[] = "..";
116 static const char remote_obj_dir[] = "REM_OBJ_DIR";
117
118 static const struct lu_object_operations      osd_lu_obj_ops;
119 static const struct dt_object_operations      osd_obj_ops;
120 static const struct dt_object_operations      osd_obj_ea_ops;
121 static const struct dt_object_operations      osd_obj_otable_it_ops;
122 static const struct dt_index_operations       osd_index_iam_ops;
123 static const struct dt_index_operations       osd_index_ea_ops;
124
125 int osd_trans_declare_op2rb[] = {
126         [OSD_OT_ATTR_SET]       = OSD_OT_ATTR_SET,
127         [OSD_OT_PUNCH]          = OSD_OT_MAX,
128         [OSD_OT_XATTR_SET]      = OSD_OT_XATTR_SET,
129         [OSD_OT_CREATE]         = OSD_OT_DESTROY,
130         [OSD_OT_DESTROY]        = OSD_OT_CREATE,
131         [OSD_OT_REF_ADD]        = OSD_OT_REF_DEL,
132         [OSD_OT_REF_DEL]        = OSD_OT_REF_ADD,
133         [OSD_OT_WRITE]          = OSD_OT_WRITE,
134         [OSD_OT_INSERT]         = OSD_OT_DELETE,
135         [OSD_OT_DELETE]         = OSD_OT_INSERT,
136         [OSD_OT_QUOTA]          = OSD_OT_MAX,
137 };
138
139 static int osd_has_index(const struct osd_object *obj)
140 {
141         return obj->oo_dt.do_index_ops != NULL;
142 }
143
144 static int osd_object_invariant(const struct lu_object *l)
145 {
146         return osd_invariant(osd_obj(l));
147 }
148
149 /*
150  * Concurrency: doesn't matter
151  */
152
153 /*
154  * Concurrency: doesn't matter
155  */
156 static int osd_write_locked(const struct lu_env *env, struct osd_object *o)
157 {
158         struct osd_thread_info *oti = osd_oti_get(env);
159         return oti->oti_w_locks > 0 && o->oo_owner == env;
160 }
161
162 /*
163  * Concurrency: doesn't access mutable data
164  */
165 static int osd_root_get(const struct lu_env *env,
166                         struct dt_device *dev, struct lu_fid *f)
167 {
168         lu_local_obj_fid(f, OSD_FS_ROOT_OID);
169         return 0;
170 }
171
172 /*
173  * OSD object methods.
174  */
175
176 /*
177  * Concurrency: no concurrent access is possible that early in object
178  * life-cycle.
179  */
180 static struct lu_object *osd_object_alloc(const struct lu_env *env,
181                                           const struct lu_object_header *hdr,
182                                           struct lu_device *d)
183 {
184         struct osd_object *mo;
185
186         OBD_ALLOC_PTR(mo);
187         if (mo != NULL) {
188                 struct lu_object *l;
189
190                 l = &mo->oo_dt.do_lu;
191                 dt_object_init(&mo->oo_dt, NULL, d);
192                 mo->oo_dt.do_ops = &osd_obj_ea_ops;
193                 l->lo_ops = &osd_lu_obj_ops;
194                 init_rwsem(&mo->oo_sem);
195                 init_rwsem(&mo->oo_ext_idx_sem);
196                 spin_lock_init(&mo->oo_guard);
197                 return l;
198         } else {
199                 return NULL;
200         }
201 }
202
203 int osd_get_lma(struct osd_thread_info *info, struct inode *inode,
204                 struct dentry *dentry, struct lustre_mdt_attrs *lma)
205 {
206         int rc;
207
208         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
209         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
210                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
211         if (rc > 0) {
212                 if ((void *)lma != (void *)info->oti_mdt_attrs_old)
213                         memcpy(lma, info->oti_mdt_attrs_old, sizeof(*lma));
214                 rc = 0;
215                 lustre_lma_swab(lma);
216                 /* Check LMA compatibility */
217                 if (lma->lma_incompat & ~LMA_INCOMPAT_SUPP) {
218                         CWARN("%.16s: unsupported incompat LMA feature(s) %#x "
219                               "for fid = "DFID", ino = %lu\n",
220                               LDISKFS_SB(inode->i_sb)->s_es->s_volume_name,
221                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
222                               PFID(&lma->lma_self_fid), inode->i_ino);
223                         rc = -EOPNOTSUPP;
224                 }
225         } else if (rc == 0) {
226                 rc = -ENODATA;
227         }
228
229         return rc;
230 }
231
232 /*
233  * retrieve object from backend ext fs.
234  **/
235 struct inode *osd_iget(struct osd_thread_info *info, struct osd_device *dev,
236                        struct osd_inode_id *id)
237 {
238         struct inode *inode = NULL;
239
240         inode = ldiskfs_iget(osd_sb(dev), id->oii_ino);
241         if (IS_ERR(inode)) {
242                 CDEBUG(D_INODE, "no inode: ino = %u, rc = %ld\n",
243                        id->oii_ino, PTR_ERR(inode));
244         } else if (id->oii_gen != OSD_OII_NOGEN &&
245                    inode->i_generation != id->oii_gen) {
246                 CDEBUG(D_INODE, "unmatched inode: ino = %u, oii_gen = %u, "
247                        "i_generation = %u\n",
248                        id->oii_ino, id->oii_gen, inode->i_generation);
249                 iput(inode);
250                 inode = ERR_PTR(-ESTALE);
251         } else if (inode->i_nlink == 0) {
252                 /* due to parallel readdir and unlink,
253                 * we can have dead inode here. */
254                 CDEBUG(D_INODE, "stale inode: ino = %u\n", id->oii_ino);
255                 iput(inode);
256                 inode = ERR_PTR(-ESTALE);
257         } else if (is_bad_inode(inode)) {
258                 CWARN("%.16s: bad inode: ino = %u\n",
259                 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, id->oii_ino);
260                 iput(inode);
261                 inode = ERR_PTR(-ENOENT);
262         } else {
263                 if (id->oii_gen == OSD_OII_NOGEN)
264                         osd_id_gen(id, inode->i_ino, inode->i_generation);
265
266                 /* Do not update file c/mtime in ldiskfs.
267                  * NB: we don't have any lock to protect this because we don't
268                  * have reference on osd_object now, but contention with
269                  * another lookup + attr_set can't happen in the tiny window
270                  * between if (...) and set S_NOCMTIME. */
271                 if (!(inode->i_flags & S_NOCMTIME))
272                         inode->i_flags |= S_NOCMTIME;
273         }
274         return inode;
275 }
276
277 static struct inode *
278 osd_iget_fid(struct osd_thread_info *info, struct osd_device *dev,
279              struct osd_inode_id *id, struct lu_fid *fid)
280 {
281         struct lustre_mdt_attrs *lma   = &info->oti_mdt_attrs;
282         struct inode            *inode;
283         int                      rc;
284
285         inode = osd_iget(info, dev, id);
286         if (IS_ERR(inode))
287                 return inode;
288
289         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
290         if (rc == 0) {
291                 *fid = lma->lma_self_fid;
292         } else if (rc == -ENODATA) {
293                 if (unlikely(inode == osd_sb(dev)->s_root->d_inode))
294                         lu_local_obj_fid(fid, OSD_FS_ROOT_OID);
295                 else
296                         lu_igif_build(fid, inode->i_ino, inode->i_generation);
297         } else {
298                 iput(inode);
299                 inode = ERR_PTR(rc);
300         }
301         return inode;
302 }
303
304 /**
305  * \retval +v: new filter_fid, does not contain self-fid
306  * \retval 0:  filter_fid_old, contains self-fid
307  * \retval -v: other failure cases
308  */
309 int osd_get_idif(struct osd_thread_info *info, struct inode *inode,
310                  struct dentry *dentry, struct lu_fid *fid)
311 {
312         struct filter_fid_old   *ff     = &info->oti_ff;
313         struct ost_id           *ostid  = &info->oti_ostid;
314         int                      rc;
315
316         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
317         if (rc == sizeof(*ff)) {
318                 rc = 0;
319                 ostid_set_seq(ostid, le64_to_cpu(ff->ff_seq));
320                 ostid_set_id(ostid, le64_to_cpu(ff->ff_objid));
321                 /* XXX: should use real OST index in the future. LU-3569 */
322                 ostid_to_fid(fid, ostid, 0);
323         } else if (rc == sizeof(struct filter_fid)) {
324                 rc = 1;
325         } else if (rc >= 0) {
326                 rc = -EINVAL;
327         }
328
329         return rc;
330 }
331
332 static int osd_lma_self_repair(struct osd_thread_info *info,
333                                struct osd_device *osd, struct inode *inode,
334                                const struct lu_fid *fid, __u32 compat)
335 {
336         handle_t *jh;
337         int       rc;
338
339         LASSERT(current->journal_info == NULL);
340
341         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
342                                   osd_dto_credits_noquota[DTO_XATTR_SET]);
343         if (IS_ERR(jh)) {
344                 rc = PTR_ERR(jh);
345                 CWARN("%s: cannot start journal for lma_self_repair: rc = %d\n",
346                       osd_name(osd), rc);
347                 return rc;
348         }
349
350         rc = osd_ea_fid_set(info, inode, fid, compat, 0);
351         if (rc != 0)
352                 CWARN("%s: cannot self repair the LMA: rc = %d\n",
353                       osd_name(osd), rc);
354         ldiskfs_journal_stop(jh);
355         return rc;
356 }
357
358 static int osd_check_lma(const struct lu_env *env, struct osd_object *obj)
359 {
360         struct osd_thread_info  *info   = osd_oti_get(env);
361         struct osd_device       *osd    = osd_obj2dev(obj);
362         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
363         struct inode            *inode  = obj->oo_inode;
364         struct dentry           *dentry = &info->oti_obj_dentry;
365         struct lu_fid           *fid    = NULL;
366         const struct lu_fid     *rfid   = lu_object_fid(&obj->oo_dt.do_lu);
367         int                      rc;
368         ENTRY;
369
370         CLASSERT(LMA_OLD_SIZE >= sizeof(*lma));
371         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_LMA,
372                              info->oti_mdt_attrs_old, LMA_OLD_SIZE);
373         if (rc == -ENODATA && !fid_is_igif(rfid) && osd->od_check_ff) {
374                 fid = &lma->lma_self_fid;
375                 rc = osd_get_idif(info, inode, dentry, fid);
376                 if ((rc > 0) || (rc == -ENODATA && osd->od_index_in_idif)) {
377                         /* For the given OST-object, if it has neither LMA nor
378                          * FID in XATTR_NAME_FID, then the given FID (which is
379                          * contained in the @obj, from client RPC for locating
380                          * the OST-object) is trusted. We use it to generate
381                          * the LMA. */
382                         osd_lma_self_repair(info, osd, inode, rfid,
383                                             LMAC_FID_ON_OST);
384                         RETURN(0);
385                 }
386         }
387
388         if (unlikely(rc == -ENODATA))
389                 RETURN(0);
390
391         if (rc < 0)
392                 RETURN(rc);
393
394         if (rc > 0) {
395                 rc = 0;
396                 lustre_lma_swab(lma);
397                 if (unlikely((lma->lma_incompat & ~LMA_INCOMPAT_SUPP) ||
398                              CFS_FAIL_CHECK(OBD_FAIL_OSD_LMA_INCOMPAT))) {
399                         CWARN("%s: unsupported incompat LMA feature(s) %#x for "
400                               "fid = "DFID", ino = %lu\n", osd_name(osd),
401                               lma->lma_incompat & ~LMA_INCOMPAT_SUPP,
402                               PFID(rfid), inode->i_ino);
403                         rc = -EOPNOTSUPP;
404                 } else {
405                         fid = &lma->lma_self_fid;
406                 }
407         }
408
409         if (fid != NULL && unlikely(!lu_fid_eq(rfid, fid))) {
410                 if (fid_is_idif(rfid) && fid_is_idif(fid)) {
411                         struct ost_id   *oi   = &info->oti_ostid;
412                         struct lu_fid   *fid1 = &info->oti_fid3;
413                         __u32            idx  = fid_idif_ost_idx(rfid);
414
415                         /* For old IDIF, the OST index is not part of the IDIF,
416                          * Means that different OSTs may have the same IDIFs.
417                          * Under such case, we need to make some compatible
418                          * check to make sure to trigger OI scrub properly. */
419                         if (idx != 0 && fid_idif_ost_idx(fid) == 0) {
420                                 /* Given @rfid is new, LMA is old. */
421                                 fid_to_ostid(fid, oi);
422                                 ostid_to_fid(fid1, oi, idx);
423                                 if (lu_fid_eq(fid1, rfid)) {
424                                         if (osd->od_index_in_idif)
425                                                 osd_lma_self_repair(info, osd,
426                                                         inode, rfid,
427                                                         LMAC_FID_ON_OST);
428                                         RETURN(0);
429                                 }
430                         }
431                 }
432
433                 rc = -EREMCHG;
434         }
435
436         RETURN(rc);
437 }
438
439 static int osd_fid_lookup(const struct lu_env *env, struct osd_object *obj,
440                           const struct lu_fid *fid,
441                           const struct lu_object_conf *conf)
442 {
443         struct osd_thread_info *info;
444         struct lu_device       *ldev   = obj->oo_dt.do_lu.lo_dev;
445         struct osd_device      *dev;
446         struct osd_idmap_cache *oic;
447         struct osd_inode_id    *id;
448         struct inode           *inode;
449         struct osd_scrub       *scrub;
450         struct scrub_file      *sf;
451         int                     result;
452         int                     saved  = 0;
453         bool                    in_oi  = false;
454         bool                    in_cache = false;
455         bool                    triggered = false;
456         ENTRY;
457
458         LINVRNT(osd_invariant(obj));
459         LASSERT(obj->oo_inode == NULL);
460         LASSERTF(fid_is_sane(fid) || fid_is_idif(fid), DFID"\n", PFID(fid));
461
462         dev = osd_dev(ldev);
463         scrub = &dev->od_scrub;
464         sf = &scrub->os_file;
465         info = osd_oti_get(env);
466         LASSERT(info);
467         oic = &info->oti_cache;
468
469         if (OBD_FAIL_CHECK(OBD_FAIL_OST_ENOENT))
470                 RETURN(-ENOENT);
471
472         /* For the object is created as locking anchor, or for the object to
473          * be created on disk. No need to osd_oi_lookup() at here because FID
474          * shouldn't never be re-used, if it's really a duplicate FID from
475          * unexpected reason, we should be able to detect it later by calling
476          * do_create->osd_oi_insert(). */
477         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
478                 GOTO(out, result = 0);
479
480         /* Search order: 1. per-thread cache. */
481         if (lu_fid_eq(fid, &oic->oic_fid) &&
482             likely(oic->oic_dev == dev)) {
483                 id = &oic->oic_lid;
484                 in_cache = true;
485                 goto iget;
486         }
487
488         id = &info->oti_id;
489         if (!list_empty(&scrub->os_inconsistent_items)) {
490                 /* Search order: 2. OI scrub pending list. */
491                 result = osd_oii_lookup(dev, fid, id);
492                 if (result == 0)
493                         goto iget;
494         }
495
496         /* Search order: 3. OI files. */
497         result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
498         if (result == -ENOENT) {
499                 if (!(fid_is_norm(fid) || fid_is_igif(fid)) ||
500                     !ldiskfs_test_bit(osd_oi_fid2idx(dev,fid),
501                                       sf->sf_oi_bitmap))
502                         GOTO(out, result = 0);
503
504                 goto trigger;
505         }
506
507         if (result != 0)
508                 GOTO(out, result);
509
510         in_oi = true;
511
512 iget:
513         inode = osd_iget(info, dev, id);
514         if (IS_ERR(inode)) {
515                 result = PTR_ERR(inode);
516                 if (result != -ENOENT && result != -ESTALE)
517                         GOTO(out, result);
518
519                 if (in_cache)
520                         fid_zero(&oic->oic_fid);
521
522                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
523                 if (result != 0)
524                         GOTO(out, result = (result == -ENOENT ? 0 : result));
525
526                 /* The OI mapping is there, but the inode is NOT there.
527                  * Two possible cases for that:
528                  *
529                  * 1) Backup/restore caused the OI invalid.
530                  * 2) Someone unlinked the object but NOT removed
531                  *    the OI mapping, such as mount target device
532                  *    as ldiskfs, and modify something directly.
533                  *
534                  * Generally, when the device is mounted, it will
535                  * auto check whether the system is restored from
536                  * file-level backup or not. We trust such detect
537                  * to distinguish the 1st case from the 2nd case. */
538                 if (!(scrub->os_file.sf_flags & SF_INCONSISTENT))
539                         GOTO(out, result = 0);
540
541 trigger:
542                 if (unlikely(triggered))
543                         GOTO(out, result = saved);
544
545                 triggered = true;
546                 if (thread_is_running(&scrub->os_thread)) {
547                         result = -EINPROGRESS;
548                 } else if (!dev->od_noscrub) {
549                         /* Since we do not know the right OI mapping, we have
550                          * to trigger OI scrub to scan the whole device. */
551                         result = osd_scrub_start(dev, SS_AUTO_FULL |
552                                 SS_CLEAR_DRYRUN | SS_CLEAR_FAILOUT);
553                         CDEBUG(D_LFSCK | D_CONSOLE, "%.16s: trigger OI "
554                                "scrub by RPC for "DFID", rc = %d [1]\n",
555                                osd_name(dev), PFID(fid), result);
556                         if (result == 0 || result == -EALREADY)
557                                 result = -EINPROGRESS;
558                         else
559                                 result = -EREMCHG;
560                 }
561
562                 /* We still have chance to get the valid inode: for the
563                  * object which is referenced by remote name entry, the
564                  * object on the local MDT will be linked under the dir
565                  * of "/REMOTE_PARENT_DIR" with its FID string as name.
566                  *
567                  * We do not know whether the object for the given FID
568                  * is referenced by some remote name entry or not, and
569                  * especially for DNE II, a multiple-linked object may
570                  * have many name entries reside on many MDTs.
571                  *
572                  * To simplify the operation, OSD will not distinguish
573                  * more, just lookup "/REMOTE_PARENT_DIR". Usually, it
574                  * only happened for the RPC from other MDT during the
575                  * OI scrub, or for the client side RPC with FID only,
576                  * such as FID to path, or from old connected client. */
577                 saved = result;
578                 result = osd_lookup_in_remote_parent(info, dev, fid, id);
579                 if (result == 0) {
580                         in_oi = false;
581                         goto iget;
582                 }
583
584                 GOTO(out, result = saved);
585         }
586
587         obj->oo_inode = inode;
588         LASSERT(obj->oo_inode->i_sb == osd_sb(dev));
589
590         result = osd_check_lma(env, obj);
591         if (result != 0) {
592                 iput(inode);
593                 obj->oo_inode = NULL;
594
595                 if (result != -EREMCHG)
596                         GOTO(out, result);
597
598                 if (in_cache)
599                         fid_zero(&oic->oic_fid);
600
601                 result = osd_oi_lookup(info, dev, fid, id, OI_CHECK_FLD);
602                 if (result == 0)
603                         goto trigger;
604
605                 if (result != -ENOENT)
606                         GOTO(out, result);
607
608                 if (!in_oi && (fid_is_norm(fid) || fid_is_igif(fid)) &&
609                     ldiskfs_test_bit(osd_oi_fid2idx(dev, fid),
610                                      sf->sf_oi_bitmap))
611                         goto trigger;
612
613                 GOTO(out, result = 0);
614         }
615
616         obj->oo_compat_dot_created = 1;
617         obj->oo_compat_dotdot_created = 1;
618
619         if (!S_ISDIR(inode->i_mode) || !ldiskfs_pdo) /* done */
620                 GOTO(out, result = 0);
621
622         LASSERT(obj->oo_hl_head == NULL);
623         obj->oo_hl_head = ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
624         if (obj->oo_hl_head == NULL) {
625                 obj->oo_inode = NULL;
626                 iput(inode);
627                 GOTO(out, result = -ENOMEM);
628         }
629         GOTO(out, result = 0);
630
631 out:
632         LINVRNT(osd_invariant(obj));
633         return result;
634 }
635
636 /*
637  * Concurrency: shouldn't matter.
638  */
639 static void osd_object_init0(struct osd_object *obj)
640 {
641         LASSERT(obj->oo_inode != NULL);
642         obj->oo_dt.do_body_ops = &osd_body_ops;
643         obj->oo_dt.do_lu.lo_header->loh_attr |=
644                 (LOHA_EXISTS | (obj->oo_inode->i_mode & S_IFMT));
645 }
646
647 /*
648  * Concurrency: no concurrent access is possible that early in object
649  * life-cycle.
650  */
651 static int osd_object_init(const struct lu_env *env, struct lu_object *l,
652                            const struct lu_object_conf *conf)
653 {
654         struct osd_object *obj = osd_obj(l);
655         int result;
656
657         LINVRNT(osd_invariant(obj));
658
659         if (fid_is_otable_it(&l->lo_header->loh_fid)) {
660                 obj->oo_dt.do_ops = &osd_obj_otable_it_ops;
661                 l->lo_header->loh_attr |= LOHA_EXISTS;
662                 return 0;
663         }
664
665         result = osd_fid_lookup(env, obj, lu_object_fid(l), conf);
666         obj->oo_dt.do_body_ops = &osd_body_ops_new;
667         if (result == 0 && obj->oo_inode != NULL)
668                 osd_object_init0(obj);
669
670         LINVRNT(osd_invariant(obj));
671         return result;
672 }
673
674 /*
675  * Concurrency: no concurrent access is possible that late in object
676  * life-cycle.
677  */
678 static void osd_object_free(const struct lu_env *env, struct lu_object *l)
679 {
680         struct osd_object *obj = osd_obj(l);
681
682         LINVRNT(osd_invariant(obj));
683
684         dt_object_fini(&obj->oo_dt);
685         if (obj->oo_hl_head != NULL)
686                 ldiskfs_htree_lock_head_free(obj->oo_hl_head);
687         OBD_FREE_PTR(obj);
688 }
689
690 /*
691  * Concurrency: no concurrent access is possible that late in object
692  * life-cycle.
693  */
694 static void osd_index_fini(struct osd_object *o)
695 {
696         struct iam_container *bag;
697
698         if (o->oo_dir != NULL) {
699                 bag = &o->oo_dir->od_container;
700                 if (o->oo_inode != NULL) {
701                         if (bag->ic_object == o->oo_inode)
702                                 iam_container_fini(bag);
703                 }
704                 OBD_FREE_PTR(o->oo_dir);
705                 o->oo_dir = NULL;
706         }
707 }
708
709 /*
710  * Concurrency: no concurrent access is possible that late in object
711  * life-cycle (for all existing callers, that is. New callers have to provide
712  * their own locking.)
713  */
714 static int osd_inode_unlinked(const struct inode *inode)
715 {
716         return inode->i_nlink == 0;
717 }
718
719 enum {
720         OSD_TXN_OI_DELETE_CREDITS    = 20,
721         OSD_TXN_INODE_DELETE_CREDITS = 20
722 };
723
724 /*
725  * Journal
726  */
727
728 #if OSD_THANDLE_STATS
729 /**
730  * Set time when the handle is allocated
731  */
732 static void osd_th_alloced(struct osd_thandle *oth)
733 {
734         oth->oth_alloced = cfs_time_current();
735 }
736
737 /**
738  * Set time when the handle started
739  */
740 static void osd_th_started(struct osd_thandle *oth)
741 {
742         oth->oth_started = cfs_time_current();
743 }
744
745 /**
746  * Helper function to convert time interval to microseconds packed in
747  * long int.
748  */
749 static long interval_to_usec(cfs_time_t start, cfs_time_t end)
750 {
751         struct timeval val;
752
753         cfs_duration_usec(cfs_time_sub(end, start), &val);
754         return val.tv_sec * 1000000 + val.tv_usec;
755 }
756
757 /**
758  * Check whether the we deal with this handle for too long.
759  */
760 static void __osd_th_check_slow(void *oth, struct osd_device *dev,
761                                 cfs_time_t alloced, cfs_time_t started,
762                                 cfs_time_t closed)
763 {
764         cfs_time_t now = cfs_time_current();
765
766         LASSERT(dev != NULL);
767
768         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_STARTING,
769                             interval_to_usec(alloced, started));
770         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_OPEN,
771                             interval_to_usec(started, closed));
772         lprocfs_counter_add(dev->od_stats, LPROC_OSD_THANDLE_CLOSING,
773                             interval_to_usec(closed, now));
774
775         if (cfs_time_before(cfs_time_add(alloced, cfs_time_seconds(30)), now)) {
776                 CWARN("transaction handle %p was open for too long: "
777                       "now "CFS_TIME_T" ,"
778                       "alloced "CFS_TIME_T" ,"
779                       "started "CFS_TIME_T" ,"
780                       "closed "CFS_TIME_T"\n",
781                       oth, now, alloced, started, closed);
782                 libcfs_debug_dumpstack(NULL);
783         }
784 }
785
786 #define OSD_CHECK_SLOW_TH(oth, dev, expr)                               \
787 {                                                                       \
788         cfs_time_t __closed = cfs_time_current();                       \
789         cfs_time_t __alloced = oth->oth_alloced;                        \
790         cfs_time_t __started = oth->oth_started;                        \
791                                                                         \
792         expr;                                                           \
793         __osd_th_check_slow(oth, dev, __alloced, __started, __closed);  \
794 }
795
796 #else /* OSD_THANDLE_STATS */
797
798 #define osd_th_alloced(h)                  do {} while(0)
799 #define osd_th_started(h)                  do {} while(0)
800 #define OSD_CHECK_SLOW_TH(oth, dev, expr)  expr
801
802 #endif /* OSD_THANDLE_STATS */
803
804 /*
805  * Concurrency: doesn't access mutable data.
806  */
807 static int osd_param_is_not_sane(const struct osd_device *dev,
808                                  const struct thandle *th)
809 {
810         struct osd_thandle *oh = container_of(th, typeof(*oh), ot_super);
811
812         return oh->ot_credits > osd_transaction_size(dev);
813 }
814
815 /*
816  * Concurrency: shouldn't matter.
817  */
818 static void osd_trans_commit_cb(struct super_block *sb,
819                                 struct ldiskfs_journal_cb_entry *jcb, int error)
820 {
821         struct osd_thandle *oh = container_of0(jcb, struct osd_thandle, ot_jcb);
822         struct thandle     *th  = &oh->ot_super;
823         struct lu_device   *lud = &th->th_dev->dd_lu_dev;
824         struct dt_txn_commit_cb *dcb, *tmp;
825
826         LASSERT(oh->ot_handle == NULL);
827
828         if (error)
829                 CERROR("transaction @0x%p commit error: %d\n", th, error);
830
831         dt_txn_hook_commit(th);
832
833         /* call per-transaction callbacks if any */
834         list_for_each_entry_safe(dcb, tmp, &oh->ot_commit_dcb_list,
835                                  dcb_linkage) {
836                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
837                          "commit callback entry: magic=%x name='%s'\n",
838                          dcb->dcb_magic, dcb->dcb_name);
839                 list_del_init(&dcb->dcb_linkage);
840                 dcb->dcb_func(NULL, th, dcb, error);
841         }
842
843         lu_ref_del_at(&lud->ld_reference, &oh->ot_dev_link, "osd-tx", th);
844         lu_device_put(lud);
845         th->th_dev = NULL;
846
847         lu_context_exit(&th->th_ctx);
848         lu_context_fini(&th->th_ctx);
849         OBD_FREE_PTR(oh);
850 }
851
852 static struct thandle *osd_trans_create(const struct lu_env *env,
853                                         struct dt_device *d)
854 {
855         struct osd_thread_info  *oti = osd_oti_get(env);
856         struct osd_iobuf        *iobuf = &oti->oti_iobuf;
857         struct osd_thandle      *oh;
858         struct thandle          *th;
859         ENTRY;
860
861         /* on pending IO in this thread should left from prev. request */
862         LASSERT(atomic_read(&iobuf->dr_numreqs) == 0);
863
864         th = ERR_PTR(-ENOMEM);
865         OBD_ALLOC_GFP(oh, sizeof *oh, GFP_NOFS);
866         if (oh != NULL) {
867                 oh->ot_quota_trans = &oti->oti_quota_trans;
868                 memset(oh->ot_quota_trans, 0, sizeof(*oh->ot_quota_trans));
869                 th = &oh->ot_super;
870                 th->th_dev = d;
871                 th->th_result = 0;
872                 th->th_tags = LCT_TX_HANDLE;
873                 oh->ot_credits = 0;
874                 INIT_LIST_HEAD(&oh->ot_commit_dcb_list);
875                 INIT_LIST_HEAD(&oh->ot_stop_dcb_list);
876                 osd_th_alloced(oh);
877
878                 memset(oti->oti_declare_ops, 0,
879                                         sizeof(oti->oti_declare_ops));
880                 memset(oti->oti_declare_ops_rb, 0,
881                                         sizeof(oti->oti_declare_ops_rb));
882                 memset(oti->oti_declare_ops_cred, 0,
883                                         sizeof(oti->oti_declare_ops_cred));
884                 oti->oti_rollback = false;
885         }
886         RETURN(th);
887 }
888
889 /*
890  * Concurrency: shouldn't matter.
891  */
892 static int osd_trans_start(const struct lu_env *env, struct dt_device *d,
893                            struct thandle *th)
894 {
895         struct osd_thread_info *oti = osd_oti_get(env);
896         struct osd_device  *dev = osd_dt_dev(d);
897         handle_t           *jh;
898         struct osd_thandle *oh;
899         int rc;
900
901         ENTRY;
902
903         LASSERT(current->journal_info == NULL);
904
905         oh = container_of0(th, struct osd_thandle, ot_super);
906         LASSERT(oh != NULL);
907         LASSERT(oh->ot_handle == NULL);
908
909         rc = dt_txn_hook_start(env, d, th);
910         if (rc != 0)
911                 GOTO(out, rc);
912
913         if (unlikely(osd_param_is_not_sane(dev, th))) {
914                 static unsigned long last_printed;
915                 static int last_credits;
916
917                 CWARN("%.16s: too many transaction credits (%d > %d)\n",
918                       LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name,
919                       oh->ot_credits,
920                       osd_journal(dev)->j_max_transaction_buffers);
921                 CWARN("  create: %u/%u, destroy: %u/%u\n",
922                       oti->oti_declare_ops[OSD_OT_CREATE],
923                       oti->oti_declare_ops_cred[OSD_OT_CREATE],
924                       oti->oti_declare_ops[OSD_OT_DESTROY],
925                       oti->oti_declare_ops_cred[OSD_OT_DESTROY]);
926                 CWARN("  attr_set: %u/%u, xattr_set: %u/%u\n",
927                       oti->oti_declare_ops[OSD_OT_ATTR_SET],
928                       oti->oti_declare_ops_cred[OSD_OT_ATTR_SET],
929                       oti->oti_declare_ops[OSD_OT_XATTR_SET],
930                       oti->oti_declare_ops_cred[OSD_OT_XATTR_SET]);
931                 CWARN("  write: %u/%u, punch: %u/%u, quota %u/%u\n",
932                       oti->oti_declare_ops[OSD_OT_WRITE],
933                       oti->oti_declare_ops_cred[OSD_OT_WRITE],
934                       oti->oti_declare_ops[OSD_OT_PUNCH],
935                       oti->oti_declare_ops_cred[OSD_OT_PUNCH],
936                       oti->oti_declare_ops[OSD_OT_QUOTA],
937                       oti->oti_declare_ops_cred[OSD_OT_QUOTA]);
938                 CWARN("  insert: %u/%u, delete: %u/%u\n",
939                       oti->oti_declare_ops[OSD_OT_INSERT],
940                       oti->oti_declare_ops_cred[OSD_OT_INSERT],
941                       oti->oti_declare_ops[OSD_OT_DELETE],
942                       oti->oti_declare_ops_cred[OSD_OT_DELETE]);
943                 CWARN("  ref_add: %u/%u, ref_del: %u/%u\n",
944                       oti->oti_declare_ops[OSD_OT_REF_ADD],
945                       oti->oti_declare_ops_cred[OSD_OT_REF_ADD],
946                       oti->oti_declare_ops[OSD_OT_REF_DEL],
947                       oti->oti_declare_ops_cred[OSD_OT_REF_DEL]);
948
949                 if (last_credits != oh->ot_credits &&
950                     time_after(jiffies, last_printed +
951                                msecs_to_jiffies(60 * MSEC_PER_SEC))) {
952                         libcfs_debug_dumpstack(NULL);
953                         last_credits = oh->ot_credits;
954                         last_printed = jiffies;
955                 }
956                 /* XXX Limit the credits to 'max_transaction_buffers', and
957                  *     let the underlying filesystem to catch the error if
958                  *     we really need so many credits.
959                  *
960                  *     This should be removed when we can calculate the
961                  *     credits precisely. */
962                 oh->ot_credits = osd_transaction_size(dev);
963         }
964
965         /*
966          * XXX temporary stuff. Some abstraction layer should
967          * be used.
968          */
969         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, oh->ot_credits);
970         osd_th_started(oh);
971         if (!IS_ERR(jh)) {
972                 oh->ot_handle = jh;
973                 LASSERT(oti->oti_txns == 0);
974                 lu_context_init(&th->th_ctx, th->th_tags);
975                 lu_context_enter(&th->th_ctx);
976
977                 lu_device_get(&d->dd_lu_dev);
978                 lu_ref_add_at(&d->dd_lu_dev.ld_reference, &oh->ot_dev_link,
979                               "osd-tx", th);
980                 oti->oti_txns++;
981                 rc = 0;
982         } else {
983                 rc = PTR_ERR(jh);
984         }
985 out:
986         RETURN(rc);
987 }
988
989 static int osd_seq_exists(const struct lu_env *env,
990                           struct osd_device *osd, u64 seq)
991 {
992         struct lu_seq_range     *range = &osd_oti_get(env)->oti_seq_range;
993         struct seq_server_site  *ss = osd_seq_site(osd);
994         int                     rc;
995         ENTRY;
996
997         LASSERT(ss != NULL);
998         LASSERT(ss->ss_server_fld != NULL);
999
1000         rc = osd_fld_lookup(env, osd, seq, range);
1001         if (rc != 0) {
1002                 if (rc != -ENOENT)
1003                         CERROR("%s: can't lookup FLD sequence "LPX64
1004                                ": rc = %d\n", osd_name(osd), seq, rc);
1005                 RETURN(0);
1006         }
1007
1008         RETURN(ss->ss_node_id == range->lsr_index);
1009 }
1010
1011 static void osd_trans_stop_cb(struct osd_thandle *oth, int result)
1012 {
1013         struct dt_txn_commit_cb *dcb;
1014         struct dt_txn_commit_cb *tmp;
1015
1016         /* call per-transaction stop callbacks if any */
1017         list_for_each_entry_safe(dcb, tmp, &oth->ot_stop_dcb_list,
1018                                  dcb_linkage) {
1019                 LASSERTF(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC,
1020                          "commit callback entry: magic=%x name='%s'\n",
1021                          dcb->dcb_magic, dcb->dcb_name);
1022                 list_del_init(&dcb->dcb_linkage);
1023                 dcb->dcb_func(NULL, &oth->ot_super, dcb, result);
1024         }
1025 }
1026
1027 /*
1028  * Concurrency: shouldn't matter.
1029  */
1030 static int osd_trans_stop(const struct lu_env *env, struct dt_device *dt,
1031                           struct thandle *th)
1032 {
1033         int                     rc = 0;
1034         struct osd_thandle     *oh;
1035         struct osd_thread_info *oti = osd_oti_get(env);
1036         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
1037         struct osd_device      *osd = osd_dt_dev(th->th_dev);
1038         struct qsd_instance    *qsd = osd->od_quota_slave;
1039         struct lquota_trans    *qtrans;
1040         ENTRY;
1041
1042         oh = container_of0(th, struct osd_thandle, ot_super);
1043
1044         qtrans = oh->ot_quota_trans;
1045         oh->ot_quota_trans = NULL;
1046
1047         if (oh->ot_handle != NULL) {
1048                 handle_t *hdl = oh->ot_handle;
1049
1050                 /*
1051                  * add commit callback
1052                  * notice we don't do this in osd_trans_start()
1053                  * as underlying transaction can change during truncate
1054                  */
1055                 ldiskfs_journal_callback_add(hdl, osd_trans_commit_cb,
1056                                          &oh->ot_jcb);
1057
1058                 LASSERT(oti->oti_txns == 1);
1059                 oti->oti_txns--;
1060
1061                 rc = dt_txn_hook_stop(env, th);
1062                 if (rc != 0)
1063                         CERROR("%s: failed in transaction hook: rc = %d\n",
1064                                osd_name(osd), rc);
1065
1066                 osd_trans_stop_cb(oh, rc);
1067                 /* hook functions might modify th_sync */
1068                 hdl->h_sync = th->th_sync;
1069
1070                 oh->ot_handle = NULL;
1071                 OSD_CHECK_SLOW_TH(oh, osd, rc = ldiskfs_journal_stop(hdl));
1072                 if (rc != 0)
1073                         CERROR("%s: failed to stop transaction: rc = %d\n",
1074                                osd_name(osd), rc);
1075         } else {
1076                 osd_trans_stop_cb(oh, th->th_result);
1077                 OBD_FREE_PTR(oh);
1078         }
1079
1080         /* inform the quota slave device that the transaction is stopping */
1081         qsd_op_end(env, qsd, qtrans);
1082
1083         /* as we want IO to journal and data IO be concurrent, we don't block
1084          * awaiting data IO completion in osd_do_bio(), instead we wait here
1085          * once transaction is submitted to the journal. all reqular requests
1086          * don't do direct IO (except read/write), thus this wait_event becomes
1087          * no-op for them.
1088          *
1089          * IMPORTANT: we have to wait till any IO submited by the thread is
1090          * completed otherwise iobuf may be corrupted by different request
1091          */
1092         wait_event(iobuf->dr_wait,
1093                        atomic_read(&iobuf->dr_numreqs) == 0);
1094         osd_fini_iobuf(osd, iobuf);
1095         if (!rc)
1096                 rc = iobuf->dr_error;
1097
1098         RETURN(rc);
1099 }
1100
1101 static int osd_trans_cb_add(struct thandle *th, struct dt_txn_commit_cb *dcb)
1102 {
1103         struct osd_thandle *oh = container_of0(th, struct osd_thandle,
1104                                                ot_super);
1105
1106         LASSERT(dcb->dcb_magic == TRANS_COMMIT_CB_MAGIC);
1107         LASSERT(&dcb->dcb_func != NULL);
1108         if (dcb->dcb_flags & DCB_TRANS_STOP)
1109                 list_add(&dcb->dcb_linkage, &oh->ot_stop_dcb_list);
1110         else
1111                 list_add(&dcb->dcb_linkage, &oh->ot_commit_dcb_list);
1112
1113         return 0;
1114 }
1115
1116 /*
1117  * Called just before object is freed. Releases all resources except for
1118  * object itself (that is released by osd_object_free()).
1119  *
1120  * Concurrency: no concurrent access is possible that late in object
1121  * life-cycle.
1122  */
1123 static void osd_object_delete(const struct lu_env *env, struct lu_object *l)
1124 {
1125         struct osd_object *obj   = osd_obj(l);
1126         struct inode      *inode = obj->oo_inode;
1127
1128         LINVRNT(osd_invariant(obj));
1129
1130         /*
1131          * If object is unlinked remove fid->ino mapping from object index.
1132          */
1133
1134         osd_index_fini(obj);
1135         if (inode != NULL) {
1136                 struct qsd_instance     *qsd = osd_obj2dev(obj)->od_quota_slave;
1137                 qid_t                    uid = i_uid_read(inode);
1138                 qid_t                    gid = i_gid_read(inode);
1139
1140                 iput(inode);
1141                 obj->oo_inode = NULL;
1142
1143                 if (qsd != NULL) {
1144                         struct osd_thread_info  *info = osd_oti_get(env);
1145                         struct lquota_id_info   *qi = &info->oti_qi;
1146
1147                         /* Release granted quota to master if necessary */
1148                         qi->lqi_id.qid_uid = uid;
1149                         qsd_op_adjust(env, qsd, &qi->lqi_id, USRQUOTA);
1150
1151                         qi->lqi_id.qid_uid = gid;
1152                         qsd_op_adjust(env, qsd, &qi->lqi_id, GRPQUOTA);
1153                 }
1154         }
1155 }
1156
1157 /*
1158  * Concurrency: ->loo_object_release() is called under site spin-lock.
1159  */
1160 static void osd_object_release(const struct lu_env *env,
1161                                struct lu_object *l)
1162 {
1163 }
1164
1165 /*
1166  * Concurrency: shouldn't matter.
1167  */
1168 static int osd_object_print(const struct lu_env *env, void *cookie,
1169                             lu_printer_t p, const struct lu_object *l)
1170 {
1171         struct osd_object *o = osd_obj(l);
1172         struct iam_descr  *d;
1173
1174         if (o->oo_dir != NULL)
1175                 d = o->oo_dir->od_container.ic_descr;
1176         else
1177                 d = NULL;
1178         return (*p)(env, cookie,
1179                     LUSTRE_OSD_LDISKFS_NAME"-object@%p(i:%p:%lu/%u)[%s]",
1180                     o, o->oo_inode,
1181                     o->oo_inode ? o->oo_inode->i_ino : 0UL,
1182                     o->oo_inode ? o->oo_inode->i_generation : 0,
1183                     d ? d->id_ops->id_name : "plain");
1184 }
1185
1186 #define GRANT_FOR_LOCAL_OIDS 32 /* 128kB for last_rcvd, quota files, ... */
1187
1188 /*
1189  * Concurrency: shouldn't matter.
1190  */
1191 int osd_statfs(const struct lu_env *env, struct dt_device *d,
1192                struct obd_statfs *sfs)
1193 {
1194         struct osd_device  *osd = osd_dt_dev(d);
1195         struct super_block *sb = osd_sb(osd);
1196         struct kstatfs     *ksfs;
1197         int result = 0;
1198
1199         if (unlikely(osd->od_mnt == NULL))
1200                 return -EINPROGRESS;
1201
1202         /* osd_lproc.c call this without env, allocate ksfs for that case */
1203         if (unlikely(env == NULL)) {
1204                 OBD_ALLOC_PTR(ksfs);
1205                 if (ksfs == NULL)
1206                         return -ENOMEM;
1207         } else {
1208                 ksfs = &osd_oti_get(env)->oti_ksfs;
1209         }
1210
1211         spin_lock(&osd->od_osfs_lock);
1212         result = sb->s_op->statfs(sb->s_root, ksfs);
1213         if (likely(result == 0)) { /* N.B. statfs can't really fail */
1214                 statfs_pack(sfs, ksfs);
1215                 if (sb->s_flags & MS_RDONLY)
1216                         sfs->os_state = OS_STATE_READONLY;
1217         }
1218
1219         spin_unlock(&osd->od_osfs_lock);
1220
1221         if (unlikely(env == NULL))
1222                 OBD_FREE_PTR(ksfs);
1223
1224         /* Reserve a small amount of space for local objects like last_rcvd,
1225          * llog, quota files, ... */
1226         if (sfs->os_bavail <= GRANT_FOR_LOCAL_OIDS) {
1227                 sfs->os_bavail = 0;
1228         } else {
1229                 sfs->os_bavail -= GRANT_FOR_LOCAL_OIDS;
1230                 /** Take out metadata overhead for indirect blocks */
1231                 sfs->os_bavail -= sfs->os_bavail >> (sb->s_blocksize_bits - 3);
1232         }
1233
1234         return result;
1235 }
1236
1237 /**
1238  * Estimate space needed for file creations. We assume the largest filename
1239  * which is 2^64 - 1, hence a filename of 20 chars.
1240  * This is 28 bytes per object which is 28MB for 1M objects ... no so bad.
1241  */
1242 #ifdef __LDISKFS_DIR_REC_LEN
1243 #define PER_OBJ_USAGE __LDISKFS_DIR_REC_LEN(20)
1244 #else
1245 #define PER_OBJ_USAGE LDISKFS_DIR_REC_LEN(20)
1246 #endif
1247
1248 /*
1249  * Concurrency: doesn't access mutable data.
1250  */
1251 static void osd_conf_get(const struct lu_env *env,
1252                          const struct dt_device *dev,
1253                          struct dt_device_param *param)
1254 {
1255         struct super_block *sb = osd_sb(osd_dt_dev(dev));
1256         int                ea_overhead;
1257
1258         /*
1259          * XXX should be taken from not-yet-existing fs abstraction layer.
1260          */
1261         param->ddp_max_name_len = LDISKFS_NAME_LEN;
1262         param->ddp_max_nlink    = LDISKFS_LINK_MAX;
1263         param->ddp_block_shift  = sb->s_blocksize_bits;
1264         param->ddp_mount_type     = LDD_MT_LDISKFS;
1265         param->ddp_maxbytes       = sb->s_maxbytes;
1266         /* Overhead estimate should be fairly accurate, so we really take a tiny
1267          * error margin which also avoids fragmenting the filesystem too much */
1268         param->ddp_grant_reserved = 2; /* end up to be 1.9% after conversion */
1269         /* inode are statically allocated, so per-inode space consumption
1270          * is the space consumed by the directory entry */
1271         param->ddp_inodespace     = PER_OBJ_USAGE;
1272         /* per-fragment overhead to be used by the client code */
1273         param->ddp_grant_frag     = 6 * LDISKFS_BLOCK_SIZE(sb);
1274         param->ddp_mntopts      = 0;
1275         if (test_opt(sb, XATTR_USER))
1276                 param->ddp_mntopts |= MNTOPT_USERXATTR;
1277         if (test_opt(sb, POSIX_ACL))
1278                 param->ddp_mntopts |= MNTOPT_ACL;
1279
1280         /* LOD might calculate the max stripe count based on max_ea_size,
1281          * so we need take account in the overhead as well,
1282          * xattr_header + magic + xattr_entry_head */
1283         ea_overhead = sizeof(struct ldiskfs_xattr_header) + sizeof(__u32) +
1284                       LDISKFS_XATTR_LEN(XATTR_NAME_MAX_LEN);
1285
1286 #if defined(LDISKFS_FEATURE_INCOMPAT_EA_INODE)
1287         if (LDISKFS_HAS_INCOMPAT_FEATURE(sb, LDISKFS_FEATURE_INCOMPAT_EA_INODE))
1288                 param->ddp_max_ea_size = LDISKFS_XATTR_MAX_LARGE_EA_SIZE -
1289                                                                 ea_overhead;
1290         else
1291 #endif
1292                 param->ddp_max_ea_size = sb->s_blocksize - ea_overhead;
1293 }
1294
1295 /*
1296  * Concurrency: shouldn't matter.
1297  */
1298 static int osd_sync(const struct lu_env *env, struct dt_device *d)
1299 {
1300         int rc;
1301
1302         CDEBUG(D_CACHE, "syncing OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1303
1304         rc = ldiskfs_force_commit(osd_sb(osd_dt_dev(d)));
1305
1306         CDEBUG(D_CACHE, "synced OSD %s: rc = %d\n",
1307                LUSTRE_OSD_LDISKFS_NAME, rc);
1308
1309         return rc;
1310 }
1311
1312 /**
1313  * Start commit for OSD device.
1314  *
1315  * An implementation of dt_commit_async method for OSD device.
1316  * Asychronously starts underlayng fs sync and thereby a transaction
1317  * commit.
1318  *
1319  * \param env environment
1320  * \param d dt device
1321  *
1322  * \see dt_device_operations
1323  */
1324 static int osd_commit_async(const struct lu_env *env,
1325                             struct dt_device *d)
1326 {
1327         struct super_block *s = osd_sb(osd_dt_dev(d));
1328         ENTRY;
1329
1330         CDEBUG(D_HA, "async commit OSD %s\n", LUSTRE_OSD_LDISKFS_NAME);
1331         RETURN(s->s_op->sync_fs(s, 0));
1332 }
1333
1334 /*
1335  * Concurrency: shouldn't matter.
1336  */
1337
1338 static int osd_ro(const struct lu_env *env, struct dt_device *d)
1339 {
1340         struct super_block *sb = osd_sb(osd_dt_dev(d));
1341         struct block_device *dev = sb->s_bdev;
1342 #ifdef HAVE_DEV_SET_RDONLY
1343         struct block_device *jdev = LDISKFS_SB(sb)->journal_bdev;
1344         int rc = 0;
1345 #else
1346         int rc = -EOPNOTSUPP;
1347 #endif
1348         ENTRY;
1349
1350 #ifdef HAVE_DEV_SET_RDONLY
1351         CERROR("*** setting %s read-only ***\n", osd_dt_dev(d)->od_svname);
1352
1353         if (jdev && (jdev != dev)) {
1354                 CDEBUG(D_IOCTL | D_HA, "set journal dev %lx rdonly\n",
1355                        (long)jdev);
1356                 dev_set_rdonly(jdev);
1357         }
1358         CDEBUG(D_IOCTL | D_HA, "set dev %lx rdonly\n", (long)dev);
1359         dev_set_rdonly(dev);
1360 #else
1361         CERROR("%s: %lx CANNOT BE SET READONLY: rc = %d\n",
1362                osd_dt_dev(d)->od_svname, (long)dev, rc);
1363 #endif
1364         RETURN(rc);
1365 }
1366
1367 /**
1368  * Note: we do not count into QUOTA here.
1369  * If we mount with --data_journal we may need more.
1370  */
1371 const int osd_dto_credits_noquota[DTO_NR] = {
1372         /**
1373          * Insert.
1374          * INDEX_EXTRA_TRANS_BLOCKS(8) +
1375          * SINGLEDATA_TRANS_BLOCKS(8)
1376          * XXX Note: maybe iam need more, since iam have more level than
1377          *           EXT3 htree.
1378          */
1379         [DTO_INDEX_INSERT]  = 16,
1380         /**
1381          * Delete
1382          * just modify a single entry, probably merge few within a block
1383          */
1384         [DTO_INDEX_DELETE]  = 1,
1385         /**
1386          * Used for OI scrub
1387          */
1388         [DTO_INDEX_UPDATE]  = 16,
1389         /**
1390          * 4(inode, inode bits, groups, GDT)
1391          *   notice: OI updates are counted separately with DTO_INDEX_INSERT
1392          */
1393         [DTO_OBJECT_CREATE] = 4,
1394         /**
1395          * 4(inode, inode bits, groups, GDT)
1396          *   notice: OI updates are counted separately with DTO_INDEX_DELETE
1397          */
1398         [DTO_OBJECT_DELETE] = 4,
1399         /**
1400          * Attr set credits (inode)
1401          */
1402         [DTO_ATTR_SET_BASE] = 1,
1403         /**
1404          * Xattr set. The same as xattr of EXT3.
1405          * DATA_TRANS_BLOCKS(14)
1406          * XXX Note: in original MDS implmentation INDEX_EXTRA_TRANS_BLOCKS
1407          * are also counted in. Do not know why?
1408          */
1409         [DTO_XATTR_SET]     = 14,
1410         /**
1411          * credits for inode change during write.
1412          */
1413         [DTO_WRITE_BASE]    = 3,
1414         /**
1415          * credits for single block write.
1416          */
1417         [DTO_WRITE_BLOCK]   = 14,
1418         /**
1419          * Attr set credits for chown.
1420          * This is extra credits for setattr, and it is null without quota
1421          */
1422         [DTO_ATTR_SET_CHOWN] = 0
1423 };
1424
1425 static const struct dt_device_operations osd_dt_ops = {
1426         .dt_root_get       = osd_root_get,
1427         .dt_statfs         = osd_statfs,
1428         .dt_trans_create   = osd_trans_create,
1429         .dt_trans_start    = osd_trans_start,
1430         .dt_trans_stop     = osd_trans_stop,
1431         .dt_trans_cb_add   = osd_trans_cb_add,
1432         .dt_conf_get       = osd_conf_get,
1433         .dt_sync           = osd_sync,
1434         .dt_ro             = osd_ro,
1435         .dt_commit_async   = osd_commit_async,
1436 };
1437
1438 static void osd_object_read_lock(const struct lu_env *env,
1439                                  struct dt_object *dt, unsigned role)
1440 {
1441         struct osd_object *obj = osd_dt_obj(dt);
1442         struct osd_thread_info *oti = osd_oti_get(env);
1443
1444         LINVRNT(osd_invariant(obj));
1445
1446         LASSERT(obj->oo_owner != env);
1447         down_read_nested(&obj->oo_sem, role);
1448
1449         LASSERT(obj->oo_owner == NULL);
1450         oti->oti_r_locks++;
1451 }
1452
1453 static void osd_object_write_lock(const struct lu_env *env,
1454                                   struct dt_object *dt, unsigned role)
1455 {
1456         struct osd_object *obj = osd_dt_obj(dt);
1457         struct osd_thread_info *oti = osd_oti_get(env);
1458
1459         LINVRNT(osd_invariant(obj));
1460
1461         LASSERT(obj->oo_owner != env);
1462         down_write_nested(&obj->oo_sem, role);
1463
1464         LASSERT(obj->oo_owner == NULL);
1465         obj->oo_owner = env;
1466         oti->oti_w_locks++;
1467 }
1468
1469 static void osd_object_read_unlock(const struct lu_env *env,
1470                                    struct dt_object *dt)
1471 {
1472         struct osd_object *obj = osd_dt_obj(dt);
1473         struct osd_thread_info *oti = osd_oti_get(env);
1474
1475         LINVRNT(osd_invariant(obj));
1476
1477         LASSERT(oti->oti_r_locks > 0);
1478         oti->oti_r_locks--;
1479         up_read(&obj->oo_sem);
1480 }
1481
1482 static void osd_object_write_unlock(const struct lu_env *env,
1483                                     struct dt_object *dt)
1484 {
1485         struct osd_object *obj = osd_dt_obj(dt);
1486         struct osd_thread_info *oti = osd_oti_get(env);
1487
1488         LINVRNT(osd_invariant(obj));
1489
1490         LASSERT(obj->oo_owner == env);
1491         LASSERT(oti->oti_w_locks > 0);
1492         oti->oti_w_locks--;
1493         obj->oo_owner = NULL;
1494         up_write(&obj->oo_sem);
1495 }
1496
1497 static int osd_object_write_locked(const struct lu_env *env,
1498                                    struct dt_object *dt)
1499 {
1500         struct osd_object *obj = osd_dt_obj(dt);
1501
1502         LINVRNT(osd_invariant(obj));
1503
1504         return obj->oo_owner == env;
1505 }
1506
1507 static struct timespec *osd_inode_time(const struct lu_env *env,
1508                                        struct inode *inode, __u64 seconds)
1509 {
1510         struct osd_thread_info  *oti = osd_oti_get(env);
1511         struct timespec         *t   = &oti->oti_time;
1512
1513         t->tv_sec = seconds;
1514         t->tv_nsec = 0;
1515         *t = timespec_trunc(*t, inode->i_sb->s_time_gran);
1516         return t;
1517 }
1518
1519
1520 static void osd_inode_getattr(const struct lu_env *env,
1521                               struct inode *inode, struct lu_attr *attr)
1522 {
1523         attr->la_valid  |= LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1524                            LA_SIZE | LA_BLOCKS | LA_UID | LA_GID |
1525                            LA_FLAGS | LA_NLINK | LA_RDEV | LA_BLKSIZE |
1526                            LA_TYPE;
1527
1528         attr->la_atime   = LTIME_S(inode->i_atime);
1529         attr->la_mtime   = LTIME_S(inode->i_mtime);
1530         attr->la_ctime   = LTIME_S(inode->i_ctime);
1531         attr->la_mode    = inode->i_mode;
1532         attr->la_size    = i_size_read(inode);
1533         attr->la_blocks  = inode->i_blocks;
1534         attr->la_uid     = i_uid_read(inode);
1535         attr->la_gid     = i_gid_read(inode);
1536         attr->la_flags   = LDISKFS_I(inode)->i_flags;
1537         attr->la_nlink   = inode->i_nlink;
1538         attr->la_rdev    = inode->i_rdev;
1539         attr->la_blksize = 1 << inode->i_blkbits;
1540         attr->la_blkbits = inode->i_blkbits;
1541 }
1542
1543 static int osd_attr_get(const struct lu_env *env,
1544                         struct dt_object *dt,
1545                         struct lu_attr *attr)
1546 {
1547         struct osd_object *obj = osd_dt_obj(dt);
1548
1549         if (!dt_object_exists(dt))
1550                 return -ENOENT;
1551
1552         LASSERT(!dt_object_remote(dt));
1553         LINVRNT(osd_invariant(obj));
1554
1555         spin_lock(&obj->oo_guard);
1556         osd_inode_getattr(env, obj->oo_inode, attr);
1557         spin_unlock(&obj->oo_guard);
1558         return 0;
1559 }
1560
1561 static int osd_declare_attr_set(const struct lu_env *env,
1562                                 struct dt_object *dt,
1563                                 const struct lu_attr *attr,
1564                                 struct thandle *handle)
1565 {
1566         struct osd_thandle     *oh;
1567         struct osd_object      *obj;
1568         struct osd_thread_info *info = osd_oti_get(env);
1569         struct lquota_id_info  *qi = &info->oti_qi;
1570         qid_t                   uid;
1571         qid_t                   gid;
1572         long long               bspace;
1573         int                     rc = 0;
1574         bool                    enforce;
1575         ENTRY;
1576
1577         LASSERT(dt != NULL);
1578         LASSERT(handle != NULL);
1579
1580         obj = osd_dt_obj(dt);
1581         LASSERT(osd_invariant(obj));
1582
1583         oh = container_of0(handle, struct osd_thandle, ot_super);
1584         LASSERT(oh->ot_handle == NULL);
1585
1586         osd_trans_declare_op(env, oh, OSD_OT_ATTR_SET,
1587                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
1588
1589         if (attr == NULL || obj->oo_inode == NULL)
1590                 RETURN(rc);
1591
1592         bspace   = obj->oo_inode->i_blocks;
1593         bspace <<= obj->oo_inode->i_sb->s_blocksize_bits;
1594         bspace   = toqb(bspace);
1595
1596         /* Changing ownership is always preformed by super user, it should not
1597          * fail with EDQUOT.
1598          *
1599          * We still need to call the osd_declare_qid() to calculate the journal
1600          * credits for updating quota accounting files and to trigger quota
1601          * space adjustment once the operation is completed.*/
1602         if (attr->la_valid & LA_UID || attr->la_valid & LA_GID) {
1603                 /* USERQUOTA */
1604                 uid = i_uid_read(obj->oo_inode);
1605                 qi->lqi_type = USRQUOTA;
1606                 enforce = (attr->la_valid & LA_UID) && (attr->la_uid != uid);
1607                 /* inode accounting */
1608                 qi->lqi_is_blk = false;
1609
1610                 /* one more inode for the new uid ... */
1611                 qi->lqi_id.qid_uid = attr->la_uid;
1612                 qi->lqi_space      = 1;
1613                 /* Reserve credits for the new uid */
1614                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1615                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1616                         rc = 0;
1617                 if (rc)
1618                         RETURN(rc);
1619
1620                 /* and one less inode for the current uid */
1621                 qi->lqi_id.qid_uid = uid;
1622                 qi->lqi_space      = -1;
1623                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1624                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1625                         rc = 0;
1626                 if (rc)
1627                         RETURN(rc);
1628
1629                 /* block accounting */
1630                 qi->lqi_is_blk = true;
1631
1632                 /* more blocks for the new uid ... */
1633                 qi->lqi_id.qid_uid = attr->la_uid;
1634                 qi->lqi_space      = bspace;
1635                 /*
1636                  * Credits for the new uid has been reserved, re-use "obj"
1637                  * to save credit reservation.
1638                  */
1639                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1640                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1641                         rc = 0;
1642                 if (rc)
1643                         RETURN(rc);
1644
1645                 /* and finally less blocks for the current uid */
1646                 qi->lqi_id.qid_uid = uid;
1647                 qi->lqi_space      = -bspace;
1648                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1649                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1650                         rc = 0;
1651                 if (rc)
1652                         RETURN(rc);
1653
1654                 /* GROUP QUOTA */
1655                 gid = i_gid_read(obj->oo_inode);
1656                 qi->lqi_type = GRPQUOTA;
1657                 enforce = (attr->la_valid & LA_GID) && (attr->la_gid != gid);
1658
1659                 /* inode accounting */
1660                 qi->lqi_is_blk = false;
1661
1662                 /* one more inode for the new gid ... */
1663                 qi->lqi_id.qid_gid = attr->la_gid;
1664                 qi->lqi_space      = 1;
1665                 rc = osd_declare_qid(env, oh, qi, NULL, enforce, NULL);
1666                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1667                         rc = 0;
1668                 if (rc)
1669                         RETURN(rc);
1670
1671                 /* and one less inode for the current gid */
1672                 qi->lqi_id.qid_gid = gid;
1673                 qi->lqi_space      = -1;
1674                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1675                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1676                         rc = 0;
1677                 if (rc)
1678                         RETURN(rc);
1679
1680                 /* block accounting */
1681                 qi->lqi_is_blk = true;
1682
1683                 /* more blocks for the new gid ... */
1684                 qi->lqi_id.qid_gid = attr->la_gid;
1685                 qi->lqi_space      = bspace;
1686                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1687                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1688                         rc = 0;
1689                 if (rc)
1690                         RETURN(rc);
1691
1692                 /* and finally less blocks for the current gid */
1693                 qi->lqi_id.qid_gid = gid;
1694                 qi->lqi_space      = -bspace;
1695                 rc = osd_declare_qid(env, oh, qi, obj, enforce, NULL);
1696                 if (rc == -EDQUOT || rc == -EINPROGRESS)
1697                         rc = 0;
1698                 if (rc)
1699                         RETURN(rc);
1700         }
1701
1702         RETURN(rc);
1703 }
1704
1705 static int osd_inode_setattr(const struct lu_env *env,
1706                              struct inode *inode, const struct lu_attr *attr)
1707 {
1708         __u64 bits = attr->la_valid;
1709
1710         /* Only allow set size for regular file */
1711         if (!S_ISREG(inode->i_mode))
1712                 bits &= ~(LA_SIZE | LA_BLOCKS);
1713
1714         if (bits == 0)
1715                 return 0;
1716
1717         if (bits & LA_ATIME)
1718                 inode->i_atime  = *osd_inode_time(env, inode, attr->la_atime);
1719         if (bits & LA_CTIME)
1720                 inode->i_ctime  = *osd_inode_time(env, inode, attr->la_ctime);
1721         if (bits & LA_MTIME)
1722                 inode->i_mtime  = *osd_inode_time(env, inode, attr->la_mtime);
1723         if (bits & LA_SIZE) {
1724                 LDISKFS_I(inode)->i_disksize = attr->la_size;
1725                 i_size_write(inode, attr->la_size);
1726         }
1727
1728 #if 0
1729         /* OSD should not change "i_blocks" which is used by quota.
1730          * "i_blocks" should be changed by ldiskfs only. */
1731         if (bits & LA_BLOCKS)
1732                 inode->i_blocks = attr->la_blocks;
1733 #endif
1734         if (bits & LA_MODE)
1735                 inode->i_mode = (inode->i_mode & S_IFMT) |
1736                                 (attr->la_mode & ~S_IFMT);
1737         if (bits & LA_UID)
1738                 i_uid_write(inode, attr->la_uid);
1739         if (bits & LA_GID)
1740                 i_gid_write(inode, attr->la_gid);
1741         if (bits & LA_NLINK)
1742                 set_nlink(inode, attr->la_nlink);
1743         if (bits & LA_RDEV)
1744                 inode->i_rdev = attr->la_rdev;
1745
1746         if (bits & LA_FLAGS) {
1747                 /* always keep S_NOCMTIME */
1748                 inode->i_flags = ll_ext_to_inode_flags(attr->la_flags) |
1749                                  S_NOCMTIME;
1750         }
1751         return 0;
1752 }
1753
1754 static int osd_quota_transfer(struct inode *inode, const struct lu_attr *attr)
1755 {
1756         if ((attr->la_valid & LA_UID && attr->la_uid != i_uid_read(inode)) ||
1757             (attr->la_valid & LA_GID && attr->la_gid != i_gid_read(inode))) {
1758                 struct iattr    iattr;
1759                 int             rc;
1760
1761                 ll_vfs_dq_init(inode);
1762                 iattr.ia_valid = 0;
1763                 if (attr->la_valid & LA_UID)
1764                         iattr.ia_valid |= ATTR_UID;
1765                 if (attr->la_valid & LA_GID)
1766                         iattr.ia_valid |= ATTR_GID;
1767                 iattr.ia_uid = make_kuid(&init_user_ns, attr->la_uid);
1768                 iattr.ia_gid = make_kgid(&init_user_ns, attr->la_gid);
1769
1770                 rc = ll_vfs_dq_transfer(inode, &iattr);
1771                 if (rc) {
1772                         CERROR("%s: quota transfer failed: rc = %d. Is quota "
1773                                "enforcement enabled on the ldiskfs "
1774                                "filesystem?\n", inode->i_sb->s_id, rc);
1775                         return rc;
1776                 }
1777         }
1778         return 0;
1779 }
1780
1781 static int osd_attr_set(const struct lu_env *env,
1782                         struct dt_object *dt,
1783                         const struct lu_attr *attr,
1784                         struct thandle *handle)
1785 {
1786         struct osd_object *obj = osd_dt_obj(dt);
1787         struct inode      *inode;
1788         int rc;
1789
1790         if (!dt_object_exists(dt))
1791                 return -ENOENT;
1792
1793         LASSERT(handle != NULL);
1794         LASSERT(!dt_object_remote(dt));
1795         LASSERT(osd_invariant(obj));
1796
1797         osd_trans_exec_op(env, handle, OSD_OT_ATTR_SET);
1798
1799         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_FID_MAPPING)) {
1800                 struct osd_thread_info  *oti  = osd_oti_get(env);
1801                 const struct lu_fid     *fid0 = lu_object_fid(&dt->do_lu);
1802                 struct lu_fid           *fid1 = &oti->oti_fid;
1803                 struct osd_inode_id     *id   = &oti->oti_id;
1804                 struct iam_path_descr   *ipd;
1805                 struct iam_container    *bag;
1806                 struct osd_thandle      *oh;
1807                 int                      rc;
1808
1809                 fid_cpu_to_be(fid1, fid0);
1810                 memset(id, 1, sizeof(*id));
1811                 bag = &osd_fid2oi(osd_dev(dt->do_lu.lo_dev),
1812                                   fid0)->oi_dir.od_container;
1813                 ipd = osd_idx_ipd_get(env, bag);
1814                 if (unlikely(ipd == NULL))
1815                         RETURN(-ENOMEM);
1816
1817                 oh = container_of0(handle, struct osd_thandle, ot_super);
1818                 rc = iam_update(oh->ot_handle, bag, (const struct iam_key *)fid1,
1819                                 (const struct iam_rec *)id, ipd);
1820                 osd_ipd_put(env, bag, ipd);
1821                 return(rc > 0 ? 0 : rc);
1822         }
1823
1824         inode = obj->oo_inode;
1825
1826         rc = osd_quota_transfer(inode, attr);
1827         if (rc)
1828                 return rc;
1829
1830         spin_lock(&obj->oo_guard);
1831         rc = osd_inode_setattr(env, inode, attr);
1832         spin_unlock(&obj->oo_guard);
1833
1834         if (!rc)
1835                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
1836         return rc;
1837 }
1838
1839 static struct dentry *osd_child_dentry_get(const struct lu_env *env,
1840                                            struct osd_object *obj,
1841                                            const char *name, const int namelen)
1842 {
1843         return osd_child_dentry_by_inode(env, obj->oo_inode, name, namelen);
1844 }
1845
1846 static int osd_mkfile(struct osd_thread_info *info, struct osd_object *obj,
1847                       umode_t mode, struct dt_allocation_hint *hint,
1848                       struct thandle *th)
1849 {
1850         int result;
1851         struct osd_device  *osd = osd_obj2dev(obj);
1852         struct osd_thandle *oth;
1853         struct dt_object   *parent = NULL;
1854         struct inode       *inode;
1855
1856         LINVRNT(osd_invariant(obj));
1857         LASSERT(obj->oo_inode == NULL);
1858         LASSERT(obj->oo_hl_head == NULL);
1859
1860         if (S_ISDIR(mode) && ldiskfs_pdo) {
1861                 obj->oo_hl_head =ldiskfs_htree_lock_head_alloc(HTREE_HBITS_DEF);
1862                 if (obj->oo_hl_head == NULL)
1863                         return -ENOMEM;
1864         }
1865
1866         oth = container_of(th, struct osd_thandle, ot_super);
1867         LASSERT(oth->ot_handle->h_transaction != NULL);
1868
1869         if (hint && hint->dah_parent)
1870                 parent = hint->dah_parent;
1871
1872         inode = ldiskfs_create_inode(oth->ot_handle,
1873                                      parent ? osd_dt_obj(parent)->oo_inode :
1874                                               osd_sb(osd)->s_root->d_inode,
1875                                      mode);
1876         if (!IS_ERR(inode)) {
1877                 /* Do not update file c/mtime in ldiskfs. */
1878                 inode->i_flags |= S_NOCMTIME;
1879
1880                 /* For new created object, it must be consistent,
1881                  * and it is unnecessary to scrub against it. */
1882                 ldiskfs_set_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
1883
1884                 obj->oo_inode = inode;
1885                 result = 0;
1886         } else {
1887                 if (obj->oo_hl_head != NULL) {
1888                         ldiskfs_htree_lock_head_free(obj->oo_hl_head);
1889                         obj->oo_hl_head = NULL;
1890                 }
1891                 result = PTR_ERR(inode);
1892         }
1893         LINVRNT(osd_invariant(obj));
1894         return result;
1895 }
1896
1897 enum {
1898         OSD_NAME_LEN = 255
1899 };
1900
1901 static int osd_mkdir(struct osd_thread_info *info, struct osd_object *obj,
1902                      struct lu_attr *attr,
1903                      struct dt_allocation_hint *hint,
1904                      struct dt_object_format *dof,
1905                      struct thandle *th)
1906 {
1907         int result;
1908         struct osd_thandle *oth;
1909         __u32 mode = (attr->la_mode & (S_IFMT | S_IRWXUGO | S_ISVTX));
1910
1911         LASSERT(S_ISDIR(attr->la_mode));
1912
1913         oth = container_of(th, struct osd_thandle, ot_super);
1914         LASSERT(oth->ot_handle->h_transaction != NULL);
1915         result = osd_mkfile(info, obj, mode, hint, th);
1916
1917         return result;
1918 }
1919
1920 static int osd_mk_index(struct osd_thread_info *info, struct osd_object *obj,
1921                         struct lu_attr *attr,
1922                         struct dt_allocation_hint *hint,
1923                         struct dt_object_format *dof,
1924                         struct thandle *th)
1925 {
1926         int result;
1927         struct osd_thandle *oth;
1928         const struct dt_index_features *feat = dof->u.dof_idx.di_feat;
1929
1930         __u32 mode = (attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX));
1931
1932         LASSERT(S_ISREG(attr->la_mode));
1933
1934         oth = container_of(th, struct osd_thandle, ot_super);
1935         LASSERT(oth->ot_handle->h_transaction != NULL);
1936
1937         result = osd_mkfile(info, obj, mode, hint, th);
1938         if (result == 0) {
1939                 LASSERT(obj->oo_inode != NULL);
1940                 if (feat->dif_flags & DT_IND_VARKEY)
1941                         result = iam_lvar_create(obj->oo_inode,
1942                                                  feat->dif_keysize_max,
1943                                                  feat->dif_ptrsize,
1944                                                  feat->dif_recsize_max,
1945                                                  oth->ot_handle);
1946                 else
1947                         result = iam_lfix_create(obj->oo_inode,
1948                                                  feat->dif_keysize_max,
1949                                                  feat->dif_ptrsize,
1950                                                  feat->dif_recsize_max,
1951                                                  oth->ot_handle);
1952
1953         }
1954         return result;
1955 }
1956
1957 static int osd_mkreg(struct osd_thread_info *info, struct osd_object *obj,
1958                      struct lu_attr *attr,
1959                      struct dt_allocation_hint *hint,
1960                      struct dt_object_format *dof,
1961                      struct thandle *th)
1962 {
1963         LASSERT(S_ISREG(attr->la_mode));
1964         return osd_mkfile(info, obj, (attr->la_mode &
1965                                (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1966 }
1967
1968 static int osd_mksym(struct osd_thread_info *info, struct osd_object *obj,
1969                      struct lu_attr *attr,
1970                      struct dt_allocation_hint *hint,
1971                      struct dt_object_format *dof,
1972                      struct thandle *th)
1973 {
1974         LASSERT(S_ISLNK(attr->la_mode));
1975         return osd_mkfile(info, obj, (attr->la_mode &
1976                               (S_IFMT | S_IALLUGO | S_ISVTX)), hint, th);
1977 }
1978
1979 static int osd_mknod(struct osd_thread_info *info, struct osd_object *obj,
1980                      struct lu_attr *attr,
1981                      struct dt_allocation_hint *hint,
1982                      struct dt_object_format *dof,
1983                      struct thandle *th)
1984 {
1985         umode_t mode = attr->la_mode & (S_IFMT | S_IALLUGO | S_ISVTX);
1986         int result;
1987
1988         LINVRNT(osd_invariant(obj));
1989         LASSERT(obj->oo_inode == NULL);
1990         LASSERT(S_ISCHR(mode) || S_ISBLK(mode) ||
1991                 S_ISFIFO(mode) || S_ISSOCK(mode));
1992
1993         result = osd_mkfile(info, obj, mode, hint, th);
1994         if (result == 0) {
1995                 LASSERT(obj->oo_inode != NULL);
1996                 /*
1997                  * This inode should be marked dirty for i_rdev.  Currently
1998                  * that is done in the osd_attr_init().
1999                  */
2000                 init_special_inode(obj->oo_inode, obj->oo_inode->i_mode,
2001                                    attr->la_rdev);
2002         }
2003         LINVRNT(osd_invariant(obj));
2004         return result;
2005 }
2006
2007 typedef int (*osd_obj_type_f)(struct osd_thread_info *, struct osd_object *,
2008                               struct lu_attr *,
2009                               struct dt_allocation_hint *hint,
2010                               struct dt_object_format *dof,
2011                               struct thandle *);
2012
2013 static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
2014 {
2015         osd_obj_type_f result;
2016
2017         switch (type) {
2018         case DFT_DIR:
2019                 result = osd_mkdir;
2020                 break;
2021         case DFT_REGULAR:
2022                 result = osd_mkreg;
2023                 break;
2024         case DFT_SYM:
2025                 result = osd_mksym;
2026                 break;
2027         case DFT_NODE:
2028                 result = osd_mknod;
2029                 break;
2030         case DFT_INDEX:
2031                 result = osd_mk_index;
2032                 break;
2033
2034         default:
2035                 LBUG();
2036                 break;
2037         }
2038         return result;
2039 }
2040
2041
2042 static void osd_ah_init(const struct lu_env *env, struct dt_allocation_hint *ah,
2043                         struct dt_object *parent, struct dt_object *child,
2044                         umode_t child_mode)
2045 {
2046         LASSERT(ah);
2047
2048         ah->dah_parent = parent;
2049         ah->dah_mode = child_mode;
2050 }
2051
2052 static void osd_attr_init(struct osd_thread_info *info, struct osd_object *obj,
2053                           struct lu_attr *attr, struct dt_object_format *dof)
2054 {
2055         struct inode   *inode = obj->oo_inode;
2056         __u64           valid = attr->la_valid;
2057         int             result;
2058
2059         attr->la_valid &= ~(LA_TYPE | LA_MODE);
2060
2061         if (dof->dof_type != DFT_NODE)
2062                 attr->la_valid &= ~LA_RDEV;
2063         if ((valid & LA_ATIME) && (attr->la_atime == LTIME_S(inode->i_atime)))
2064                 attr->la_valid &= ~LA_ATIME;
2065         if ((valid & LA_CTIME) && (attr->la_ctime == LTIME_S(inode->i_ctime)))
2066                 attr->la_valid &= ~LA_CTIME;
2067         if ((valid & LA_MTIME) && (attr->la_mtime == LTIME_S(inode->i_mtime)))
2068                 attr->la_valid &= ~LA_MTIME;
2069
2070         result = osd_quota_transfer(inode, attr);
2071         if (result)
2072                 return;
2073
2074         if (attr->la_valid != 0) {
2075                 result = osd_inode_setattr(info->oti_env, inode, attr);
2076                 /*
2077                  * The osd_inode_setattr() should always succeed here.  The
2078                  * only error that could be returned is EDQUOT when we are
2079                  * trying to change the UID or GID of the inode. However, this
2080                  * should not happen since quota enforcement is no longer
2081                  * enabled on ldiskfs (lquota takes care of it).
2082                  */
2083                 LASSERTF(result == 0, "%d\n", result);
2084                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2085         }
2086
2087         attr->la_valid = valid;
2088 }
2089
2090 /**
2091  * Helper function for osd_object_create()
2092  *
2093  * \retval 0, on success
2094  */
2095 static int __osd_object_create(struct osd_thread_info *info,
2096                                struct osd_object *obj, struct lu_attr *attr,
2097                                struct dt_allocation_hint *hint,
2098                                struct dt_object_format *dof,
2099                                struct thandle *th)
2100 {
2101         int     result;
2102         __u32   umask;
2103
2104         /* we drop umask so that permissions we pass are not affected */
2105         umask = current->fs->umask;
2106         current->fs->umask = 0;
2107
2108         result = osd_create_type_f(dof->dof_type)(info, obj, attr, hint, dof,
2109                                                   th);
2110         if (likely(obj->oo_inode != NULL)) {
2111                 LASSERT(obj->oo_inode->i_state & I_NEW);
2112
2113                 /* Unlock the inode before attr initialization to avoid
2114                  * unnecessary dqget operations. LU-6378 */
2115                 unlock_new_inode(obj->oo_inode);
2116         }
2117
2118         if (likely(result == 0)) {
2119                 osd_attr_init(info, obj, attr, dof);
2120                 osd_object_init0(obj);
2121         }
2122
2123         /* restore previous umask value */
2124         current->fs->umask = umask;
2125
2126         return result;
2127 }
2128
2129 /**
2130  * Helper function for osd_object_create()
2131  *
2132  * \retval 0, on success
2133  */
2134 static int __osd_oi_insert(const struct lu_env *env, struct osd_object *obj,
2135                            const struct lu_fid *fid, struct thandle *th)
2136 {
2137         struct osd_thread_info *info = osd_oti_get(env);
2138         struct osd_inode_id    *id   = &info->oti_id;
2139         struct osd_device      *osd  = osd_obj2dev(obj);
2140         struct osd_thandle     *oh;
2141
2142         LASSERT(obj->oo_inode != NULL);
2143
2144         oh = container_of0(th, struct osd_thandle, ot_super);
2145         LASSERT(oh->ot_handle);
2146
2147         osd_id_gen(id, obj->oo_inode->i_ino, obj->oo_inode->i_generation);
2148         return osd_oi_insert(info, osd, fid, id, oh->ot_handle, OI_CHECK_FLD);
2149 }
2150
2151 int osd_fld_lookup(const struct lu_env *env, struct osd_device *osd,
2152                    u64 seq, struct lu_seq_range *range)
2153 {
2154         struct seq_server_site  *ss = osd_seq_site(osd);
2155
2156         if (fid_seq_is_idif(seq)) {
2157                 fld_range_set_ost(range);
2158                 range->lsr_index = idif_ost_idx(seq);
2159                 return 0;
2160         }
2161
2162         if (!fid_seq_in_fldb(seq)) {
2163                 fld_range_set_mdt(range);
2164                 if (ss != NULL)
2165                         /* FIXME: If ss is NULL, it suppose not get lsr_index
2166                          * at all */
2167                         range->lsr_index = ss->ss_node_id;
2168                 return 0;
2169         }
2170
2171         LASSERT(ss != NULL);
2172         fld_range_set_any(range);
2173         /* OSD will only do local fld lookup */
2174         return fld_local_lookup(env, ss->ss_server_fld, seq, range);
2175 }
2176
2177 /*
2178  * Concurrency: no external locking is necessary.
2179  */
2180 static int osd_declare_object_create(const struct lu_env *env,
2181                                      struct dt_object *dt,
2182                                      struct lu_attr *attr,
2183                                      struct dt_allocation_hint *hint,
2184                                      struct dt_object_format *dof,
2185                                      struct thandle *handle)
2186 {
2187         struct osd_thandle      *oh;
2188         int                      rc;
2189         ENTRY;
2190
2191         LASSERT(handle != NULL);
2192
2193         oh = container_of0(handle, struct osd_thandle, ot_super);
2194         LASSERT(oh->ot_handle == NULL);
2195
2196         osd_trans_declare_op(env, oh, OSD_OT_CREATE,
2197                              osd_dto_credits_noquota[DTO_OBJECT_CREATE]);
2198         /* Reuse idle OI block may cause additional one OI block
2199          * to be changed. */
2200         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2201                              osd_dto_credits_noquota[DTO_INDEX_INSERT] + 1);
2202
2203         /* If this is directory, then we expect . and .. to be inserted as
2204          * well. The one directory block always needs to be created for the
2205          * directory, so we could use DTO_WRITE_BASE here (GDT, block bitmap,
2206          * block), there is no danger of needing a tree for the first block.
2207          */
2208         if (attr && S_ISDIR(attr->la_mode)) {
2209                 osd_trans_declare_op(env, oh, OSD_OT_INSERT,
2210                                      osd_dto_credits_noquota[DTO_WRITE_BASE]);
2211                 osd_trans_declare_op(env, oh, OSD_OT_INSERT, 0);
2212         }
2213
2214         if (!attr)
2215                 RETURN(0);
2216
2217         rc = osd_declare_inode_qid(env, attr->la_uid, attr->la_gid, 1, oh,
2218                                    osd_dt_obj(dt), false, NULL, false);
2219         if (rc != 0)
2220                 RETURN(rc);
2221
2222         RETURN(rc);
2223 }
2224
2225 static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
2226                              struct lu_attr *attr,
2227                              struct dt_allocation_hint *hint,
2228                              struct dt_object_format *dof, struct thandle *th)
2229 {
2230         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2231         struct osd_object       *obj    = osd_dt_obj(dt);
2232         struct osd_thread_info  *info   = osd_oti_get(env);
2233         int result;
2234         ENTRY;
2235
2236         if (dt_object_exists(dt))
2237                 return -EEXIST;
2238
2239         LINVRNT(osd_invariant(obj));
2240         LASSERT(!dt_object_remote(dt));
2241         LASSERT(osd_write_locked(env, obj));
2242         LASSERT(th != NULL);
2243
2244         if (unlikely(fid_is_acct(fid)))
2245                 /* Quota files can't be created from the kernel any more,
2246                  * 'tune2fs -O quota' will take care of creating them */
2247                 RETURN(-EPERM);
2248
2249         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2250         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2251
2252         result = __osd_object_create(info, obj, attr, hint, dof, th);
2253         if (result == 0)
2254                 result = __osd_oi_insert(env, obj, fid, th);
2255
2256         LASSERT(ergo(result == 0,
2257                      dt_object_exists(dt) && !dt_object_remote(dt)));
2258
2259         LASSERT(osd_invariant(obj));
2260         RETURN(result);
2261 }
2262
2263 /**
2264  * Called to destroy on-disk representation of the object
2265  *
2266  * Concurrency: must be locked
2267  */
2268 static int osd_declare_object_destroy(const struct lu_env *env,
2269                                       struct dt_object *dt,
2270                                       struct thandle *th)
2271 {
2272         struct osd_object  *obj = osd_dt_obj(dt);
2273         struct inode       *inode = obj->oo_inode;
2274         struct osd_thandle *oh;
2275         int                 rc;
2276         ENTRY;
2277
2278         oh = container_of0(th, struct osd_thandle, ot_super);
2279         LASSERT(oh->ot_handle == NULL);
2280         LASSERT(inode);
2281
2282         osd_trans_declare_op(env, oh, OSD_OT_DESTROY,
2283                              osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
2284         /* Recycle idle OI leaf may cause additional three OI blocks
2285          * to be changed. */
2286         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
2287                              osd_dto_credits_noquota[DTO_INDEX_DELETE] + 3);
2288         /* one less inode */
2289         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2290                                    -1, oh, obj, false, NULL, false);
2291         if (rc)
2292                 RETURN(rc);
2293         /* data to be truncated */
2294         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
2295                                    0, oh, obj, true, NULL, false);
2296         RETURN(rc);
2297 }
2298
2299 static int osd_object_destroy(const struct lu_env *env,
2300                               struct dt_object *dt,
2301                               struct thandle *th)
2302 {
2303         const struct lu_fid    *fid = lu_object_fid(&dt->do_lu);
2304         struct osd_object      *obj = osd_dt_obj(dt);
2305         struct inode           *inode = obj->oo_inode;
2306         struct osd_device      *osd = osd_obj2dev(obj);
2307         struct osd_thandle     *oh;
2308         int                     result;
2309         ENTRY;
2310
2311         oh = container_of0(th, struct osd_thandle, ot_super);
2312         LASSERT(oh->ot_handle);
2313         LASSERT(inode);
2314         LASSERT(!lu_object_is_dying(dt->do_lu.lo_header));
2315
2316         if (unlikely(fid_is_acct(fid)))
2317                 RETURN(-EPERM);
2318
2319         if (S_ISDIR(inode->i_mode)) {
2320                 LASSERT(osd_inode_unlinked(inode) || inode->i_nlink == 1 ||
2321                         inode->i_nlink == 2);
2322                 /* it will check/delete the inode from remote parent,
2323                  * how to optimize it? unlink performance impaction XXX */
2324                 result = osd_delete_from_remote_parent(env, osd, obj, oh);
2325                 if (result != 0 && result != -ENOENT) {
2326                         CERROR("%s: delete inode "DFID": rc = %d\n",
2327                                osd_name(osd), PFID(fid), result);
2328                 }
2329                 spin_lock(&obj->oo_guard);
2330                 clear_nlink(inode);
2331                 spin_unlock(&obj->oo_guard);
2332                 ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2333         }
2334
2335         osd_trans_exec_op(env, th, OSD_OT_DESTROY);
2336
2337         result = osd_oi_delete(osd_oti_get(env), osd, fid, oh->ot_handle,
2338                                OI_CHECK_FLD);
2339
2340         /* XXX: add to ext3 orphan list */
2341         /* rc = ext3_orphan_add(handle_t *handle, struct inode *inode) */
2342
2343         /* not needed in the cache anymore */
2344         set_bit(LU_OBJECT_HEARD_BANSHEE, &dt->do_lu.lo_header->loh_flags);
2345
2346         RETURN(0);
2347 }
2348
2349 /**
2350  * Put the fid into lustre_mdt_attrs, and then place the structure
2351  * inode's ea. This fid should not be altered during the life time
2352  * of the inode.
2353  *
2354  * \retval +ve, on success
2355  * \retval -ve, on error
2356  *
2357  * FIXME: It is good to have/use ldiskfs_xattr_set_handle() here
2358  */
2359 int osd_ea_fid_set(struct osd_thread_info *info, struct inode *inode,
2360                    const struct lu_fid *fid, __u32 compat, __u32 incompat)
2361 {
2362         struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2363         int                      rc;
2364         ENTRY;
2365
2366         if (OBD_FAIL_CHECK(OBD_FAIL_FID_INLMA))
2367                 RETURN(0);
2368
2369         lustre_lma_init(lma, fid, compat, incompat);
2370         lustre_lma_swab(lma);
2371
2372         rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma, sizeof(*lma),
2373                              XATTR_CREATE);
2374         /* LMA may already exist, but we need to check that all the
2375          * desired compat/incompat flags have been added. */
2376         if (unlikely(rc == -EEXIST)) {
2377                 if (compat == 0 && incompat == 0)
2378                         RETURN(0);
2379
2380                 rc = __osd_xattr_get(inode, &info->oti_obj_dentry,
2381                                      XATTR_NAME_LMA, info->oti_mdt_attrs_old,
2382                                      LMA_OLD_SIZE);
2383                 if (rc <= 0)
2384                         RETURN(-EINVAL);
2385
2386                 lustre_lma_swab(lma);
2387                 if (!(~lma->lma_compat & compat) &&
2388                     !(~lma->lma_incompat & incompat))
2389                         RETURN(0);
2390
2391                 lma->lma_compat |= compat;
2392                 lma->lma_incompat |= incompat;
2393                 lustre_lma_swab(lma);
2394                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2395                                      sizeof(*lma), XATTR_REPLACE);
2396         }
2397
2398         RETURN(rc);
2399 }
2400
2401 /**
2402  * ldiskfs supports fid in dirent, it is passed in dentry->d_fsdata.
2403  * lustre 1.8 also uses d_fsdata for passing other info to ldiskfs.
2404  * To have compatilibility with 1.8 ldiskfs driver we need to have
2405  * magic number at start of fid data.
2406  * \ldiskfs_dentry_param is used only to pass fid from osd to ldiskfs.
2407  * its inmemory API.
2408  */
2409 static void osd_get_ldiskfs_dirent_param(struct ldiskfs_dentry_param *param,
2410                                          const struct lu_fid *fid)
2411 {
2412         if (!fid_is_namespace_visible(fid) ||
2413             OBD_FAIL_CHECK(OBD_FAIL_FID_IGIF)) {
2414                 param->edp_magic = 0;
2415                 return;
2416         }
2417
2418         param->edp_magic = LDISKFS_LUFID_MAGIC;
2419         param->edp_len =  sizeof(struct lu_fid) + 1;
2420         fid_cpu_to_be((struct lu_fid *)param->edp_data, (struct lu_fid *)fid);
2421 }
2422
2423 /**
2424  * Try to read the fid from inode ea into dt_rec.
2425  *
2426  * \param fid object fid.
2427  *
2428  * \retval 0 on success
2429  */
2430 static int osd_ea_fid_get(const struct lu_env *env, struct osd_object *obj,
2431                           __u32 ino, struct lu_fid *fid,
2432                           struct osd_inode_id *id)
2433 {
2434         struct osd_thread_info *info  = osd_oti_get(env);
2435         struct inode           *inode;
2436         ENTRY;
2437
2438         osd_id_gen(id, ino, OSD_OII_NOGEN);
2439         inode = osd_iget_fid(info, osd_obj2dev(obj), id, fid);
2440         if (IS_ERR(inode))
2441                 RETURN(PTR_ERR(inode));
2442
2443         iput(inode);
2444         RETURN(0);
2445 }
2446
2447 static int osd_add_dot_dotdot_internal(struct osd_thread_info *info,
2448                                         struct inode *dir,
2449                                         struct inode *parent_dir,
2450                                         const struct lu_fid *dot_fid,
2451                                         const struct lu_fid *dot_dot_fid,
2452                                         struct osd_thandle *oth)
2453 {
2454         struct ldiskfs_dentry_param *dot_ldp;
2455         struct ldiskfs_dentry_param *dot_dot_ldp;
2456
2457         dot_dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp2;
2458         osd_get_ldiskfs_dirent_param(dot_dot_ldp, dot_dot_fid);
2459
2460         dot_ldp = (struct ldiskfs_dentry_param *)info->oti_ldp;
2461         dot_ldp->edp_magic = 0;
2462         return ldiskfs_add_dot_dotdot(oth->ot_handle, parent_dir,
2463                                         dir, dot_ldp, dot_dot_ldp);
2464 }
2465
2466 /**
2467  * Create an local agent inode for remote entry
2468  */
2469 static struct inode *osd_create_local_agent_inode(const struct lu_env *env,
2470                                                   struct osd_device *osd,
2471                                                   struct osd_object *pobj,
2472                                                   const struct lu_fid *fid,
2473                                                   __u32 type,
2474                                                   struct thandle *th)
2475 {
2476         struct osd_thread_info  *info = osd_oti_get(env);
2477         struct inode            *local;
2478         struct osd_thandle      *oh;
2479         int                     rc;
2480         ENTRY;
2481
2482         LASSERT(th);
2483         oh = container_of(th, struct osd_thandle, ot_super);
2484         LASSERT(oh->ot_handle->h_transaction != NULL);
2485
2486         local = ldiskfs_create_inode(oh->ot_handle, pobj->oo_inode, type);
2487         if (IS_ERR(local)) {
2488                 CERROR("%s: create local error %d\n", osd_name(osd),
2489                        (int)PTR_ERR(local));
2490                 RETURN(local);
2491         }
2492
2493         ldiskfs_set_inode_state(local, LDISKFS_STATE_LUSTRE_NOSCRUB);
2494         unlock_new_inode(local);
2495
2496         /* Set special LMA flag for local agent inode */
2497         rc = osd_ea_fid_set(info, local, fid, 0, LMAI_AGENT);
2498         if (rc != 0) {
2499                 CERROR("%s: set LMA for "DFID" remote inode failed: rc = %d\n",
2500                        osd_name(osd), PFID(fid), rc);
2501                 RETURN(ERR_PTR(rc));
2502         }
2503
2504         if (!S_ISDIR(type))
2505                 RETURN(local);
2506
2507         rc = osd_add_dot_dotdot_internal(info, local, pobj->oo_inode,
2508                                          lu_object_fid(&pobj->oo_dt.do_lu),
2509                                          fid, oh);
2510         if (rc != 0) {
2511                 CERROR("%s: "DFID" add dot dotdot error: rc = %d\n",
2512                         osd_name(osd), PFID(fid), rc);
2513                 RETURN(ERR_PTR(rc));
2514         }
2515
2516         RETURN(local);
2517 }
2518
2519 /**
2520  * Delete local agent inode for remote entry
2521  */
2522 static int osd_delete_local_agent_inode(const struct lu_env *env,
2523                                         struct osd_device *osd,
2524                                         const struct lu_fid *fid,
2525                                         __u32 ino, struct osd_thandle *oh)
2526 {
2527         struct osd_thread_info  *oti = osd_oti_get(env);
2528         struct osd_inode_id     *id = &oti->oti_id;
2529         struct inode            *inode;
2530         ENTRY;
2531
2532         id->oii_ino = le32_to_cpu(ino);
2533         id->oii_gen = OSD_OII_NOGEN;
2534         inode = osd_iget(oti, osd, id);
2535         if (IS_ERR(inode)) {
2536                 CERROR("%s: iget error "DFID" id %u:%u\n", osd_name(osd),
2537                        PFID(fid), id->oii_ino, id->oii_gen);
2538                 RETURN(PTR_ERR(inode));
2539         }
2540
2541         clear_nlink(inode);
2542         mark_inode_dirty(inode);
2543         CDEBUG(D_INODE, "%s: delete remote inode "DFID" %lu\n",
2544                 osd_name(osd), PFID(fid), inode->i_ino);
2545         iput(inode);
2546         RETURN(0);
2547 }
2548
2549 /**
2550  * OSD layer object create function for interoperability mode (b11826).
2551  * This is mostly similar to osd_object_create(). Only difference being, fid is
2552  * inserted into inode ea here.
2553  *
2554  * \retval   0, on success
2555  * \retval -ve, on error
2556  */
2557 static int osd_object_ea_create(const struct lu_env *env, struct dt_object *dt,
2558                                 struct lu_attr *attr,
2559                                 struct dt_allocation_hint *hint,
2560                                 struct dt_object_format *dof,
2561                                 struct thandle *th)
2562 {
2563         const struct lu_fid     *fid    = lu_object_fid(&dt->do_lu);
2564         struct osd_object       *obj    = osd_dt_obj(dt);
2565         struct osd_thread_info  *info   = osd_oti_get(env);
2566         int                      result;
2567
2568         ENTRY;
2569
2570         if (dt_object_exists(dt))
2571                 RETURN(-EEXIST);
2572
2573         LASSERT(osd_invariant(obj));
2574         LASSERT(!dt_object_remote(dt));
2575         LASSERT(osd_write_locked(env, obj));
2576         LASSERT(th != NULL);
2577
2578         if (unlikely(fid_is_acct(fid)))
2579                 /* Quota files can't be created from the kernel any more,
2580                  * 'tune2fs -O quota' will take care of creating them */
2581                 RETURN(-EPERM);
2582
2583         osd_trans_exec_op(env, th, OSD_OT_CREATE);
2584         osd_trans_declare_rb(env, th, OSD_OT_REF_ADD);
2585
2586         result = __osd_object_create(info, obj, attr, hint, dof, th);
2587         if (result == 0) {
2588                 if (fid_is_idif(fid) &&
2589                     !osd_dev(dt->do_lu.lo_dev)->od_index_in_idif) {
2590                         struct lu_fid *tfid = &info->oti_fid;
2591                         struct ost_id *oi   = &info->oti_ostid;
2592
2593                         fid_to_ostid(fid, oi);
2594                         ostid_to_fid(tfid, oi, 0);
2595                         result = osd_ea_fid_set(info, obj->oo_inode, tfid,
2596                                                 LMAC_FID_ON_OST, 0);
2597                 } else {
2598                         result = osd_ea_fid_set(info, obj->oo_inode, fid,
2599                                 fid_is_on_ost(info, osd_obj2dev(obj),
2600                                               fid, OI_CHECK_FLD) ?
2601                                 LMAC_FID_ON_OST : 0, 0);
2602                 }
2603         }
2604
2605         if (result == 0)
2606                 result = __osd_oi_insert(env, obj, fid, th);
2607
2608         LASSERT(ergo(result == 0,
2609                      dt_object_exists(dt) && !dt_object_remote(dt)));
2610         LINVRNT(osd_invariant(obj));
2611         RETURN(result);
2612 }
2613
2614 static int osd_declare_object_ref_add(const struct lu_env *env,
2615                                       struct dt_object *dt,
2616                                       struct thandle *handle)
2617 {
2618         struct osd_thandle       *oh;
2619
2620         /* it's possible that object doesn't exist yet */
2621         LASSERT(handle != NULL);
2622
2623         oh = container_of0(handle, struct osd_thandle, ot_super);
2624         LASSERT(oh->ot_handle == NULL);
2625
2626         osd_trans_declare_op(env, oh, OSD_OT_REF_ADD,
2627                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2628
2629         return 0;
2630 }
2631
2632 /*
2633  * Concurrency: @dt is write locked.
2634  */
2635 static int osd_object_ref_add(const struct lu_env *env,
2636                               struct dt_object *dt, struct thandle *th)
2637 {
2638         struct osd_object  *obj = osd_dt_obj(dt);
2639         struct inode       *inode = obj->oo_inode;
2640         struct osd_thandle *oh;
2641         int                 rc = 0;
2642
2643         if (!dt_object_exists(dt))
2644                 return -ENOENT;
2645
2646         LINVRNT(osd_invariant(obj));
2647         LASSERT(!dt_object_remote(dt));
2648         LASSERT(osd_write_locked(env, obj));
2649         LASSERT(th != NULL);
2650
2651         oh = container_of0(th, struct osd_thandle, ot_super);
2652         LASSERT(oh->ot_handle != NULL);
2653
2654         osd_trans_exec_op(env, th, OSD_OT_REF_ADD);
2655
2656         CDEBUG(D_INODE, DFID" increase nlink %d\n",
2657                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2658         /*
2659          * The DIR_NLINK feature allows directories to exceed LDISKFS_LINK_MAX
2660          * (65000) subdirectories by storing "1" in i_nlink if the link count
2661          * would otherwise overflow. Directory tranversal tools understand
2662          * that (st_nlink == 1) indicates that the filesystem dose not track
2663          * hard links count on the directory, and will not abort subdirectory
2664          * scanning early once (st_nlink - 2) subdirs have been found.
2665          *
2666          * This also has to properly handle the case of inodes with nlink == 0
2667          * in case they are being linked into the PENDING directory
2668          */
2669         spin_lock(&obj->oo_guard);
2670         if (unlikely(inode->i_nlink == 0))
2671                 /* inc_nlink from 0 may cause WARN_ON */
2672                 set_nlink(inode, 1);
2673         else {
2674                 ldiskfs_inc_count(oh->ot_handle, inode);
2675                 if (!S_ISDIR(inode->i_mode))
2676                         LASSERT(inode->i_nlink <= LDISKFS_LINK_MAX);
2677         }
2678         spin_unlock(&obj->oo_guard);
2679
2680         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2681         LINVRNT(osd_invariant(obj));
2682
2683         return rc;
2684 }
2685
2686 static int osd_declare_object_ref_del(const struct lu_env *env,
2687                                       struct dt_object *dt,
2688                                       struct thandle *handle)
2689 {
2690         struct osd_thandle *oh;
2691
2692         LASSERT(!dt_object_remote(dt));
2693         LASSERT(handle != NULL);
2694
2695         oh = container_of0(handle, struct osd_thandle, ot_super);
2696         LASSERT(oh->ot_handle == NULL);
2697
2698         osd_trans_declare_op(env, oh, OSD_OT_REF_DEL,
2699                              osd_dto_credits_noquota[DTO_ATTR_SET_BASE]);
2700
2701         return 0;
2702 }
2703
2704 /*
2705  * Concurrency: @dt is write locked.
2706  */
2707 static int osd_object_ref_del(const struct lu_env *env, struct dt_object *dt,
2708                               struct thandle *th)
2709 {
2710         struct osd_object       *obj = osd_dt_obj(dt);
2711         struct inode            *inode = obj->oo_inode;
2712         struct osd_device       *osd = osd_dev(dt->do_lu.lo_dev);
2713         struct osd_thandle      *oh;
2714
2715         if (!dt_object_exists(dt))
2716                 return -ENOENT;
2717
2718         LINVRNT(osd_invariant(obj));
2719         LASSERT(!dt_object_remote(dt));
2720         LASSERT(osd_write_locked(env, obj));
2721         LASSERT(th != NULL);
2722
2723         oh = container_of0(th, struct osd_thandle, ot_super);
2724         LASSERT(oh->ot_handle != NULL);
2725
2726         osd_trans_exec_op(env, th, OSD_OT_REF_DEL);
2727
2728         spin_lock(&obj->oo_guard);
2729         /* That can be result of upgrade from old Lustre version and
2730          * applied only to local files.  Just skip this ref_del call.
2731          * ext4_unlink() only treats this as a warning, don't LASSERT here.*/
2732         if (inode->i_nlink == 0) {
2733                 CDEBUG_LIMIT(fid_is_norm(lu_object_fid(&dt->do_lu)) ?
2734                              D_ERROR : D_INODE, "%s: nlink == 0 on "DFID
2735                              ", maybe an upgraded file? (LU-3915)\n",
2736                              osd_name(osd), PFID(lu_object_fid(&dt->do_lu)));
2737                 spin_unlock(&obj->oo_guard);
2738                 return 0;
2739         }
2740
2741         CDEBUG(D_INODE, DFID" decrease nlink %d\n",
2742                PFID(lu_object_fid(&dt->do_lu)), inode->i_nlink);
2743
2744         ldiskfs_dec_count(oh->ot_handle, inode);
2745         spin_unlock(&obj->oo_guard);
2746
2747         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2748         LINVRNT(osd_invariant(obj));
2749
2750         return 0;
2751 }
2752
2753 /*
2754  * Get the 64-bit version for an inode.
2755  */
2756 static int osd_object_version_get(const struct lu_env *env,
2757                                   struct dt_object *dt, dt_obj_version_t *ver)
2758 {
2759         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2760
2761         CDEBUG(D_INODE, "Get version "LPX64" for inode %lu\n",
2762                LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2763         *ver = LDISKFS_I(inode)->i_fs_version;
2764         return 0;
2765 }
2766
2767 /*
2768  * Concurrency: @dt is read locked.
2769  */
2770 static int osd_xattr_get(const struct lu_env *env, struct dt_object *dt,
2771                          struct lu_buf *buf, const char *name)
2772 {
2773         struct osd_object      *obj    = osd_dt_obj(dt);
2774         struct inode           *inode  = obj->oo_inode;
2775         struct osd_thread_info *info   = osd_oti_get(env);
2776         struct dentry          *dentry = &info->oti_obj_dentry;
2777
2778         /* version get is not real XATTR but uses xattr API */
2779         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2780                 /* for version we are just using xattr API but change inode
2781                  * field instead */
2782                 if (buf->lb_len == 0)
2783                         return sizeof(dt_obj_version_t);
2784
2785                 if (buf->lb_len < sizeof(dt_obj_version_t))
2786                         return -ERANGE;
2787
2788                 osd_object_version_get(env, dt, buf->lb_buf);
2789
2790                 return sizeof(dt_obj_version_t);
2791         }
2792
2793         if (!dt_object_exists(dt))
2794                 return -ENOENT;
2795
2796         LASSERT(!dt_object_remote(dt));
2797         LASSERT(inode->i_op != NULL);
2798         LASSERT(inode->i_op->getxattr != NULL);
2799
2800         return __osd_xattr_get(inode, dentry, name, buf->lb_buf, buf->lb_len);
2801 }
2802
2803
2804 static int osd_declare_xattr_set(const struct lu_env *env,
2805                                  struct dt_object *dt,
2806                                  const struct lu_buf *buf, const char *name,
2807                                  int fl, struct thandle *handle)
2808 {
2809         struct osd_thandle *oh;
2810         int credits;
2811         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2812
2813         LASSERT(handle != NULL);
2814
2815         oh = container_of0(handle, struct osd_thandle, ot_super);
2816         LASSERT(oh->ot_handle == NULL);
2817
2818         /* optimistic optimization: LMA is set first and usually fit inode */
2819         if (strcmp(name, XATTR_NAME_LMA) == 0) {
2820                 if (dt_object_exists(dt))
2821                         credits = 0;
2822                 else
2823                         credits = 1;
2824         } else if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2825                 credits = 1;
2826         } else {
2827                 credits = osd_dto_credits_noquota[DTO_XATTR_SET];
2828                 if (buf && buf->lb_len > sb->s_blocksize) {
2829                         credits *= (buf->lb_len + sb->s_blocksize - 1) >>
2830                                         sb->s_blocksize_bits;
2831                 }
2832                 /*
2833                  * xattr set may involve inode quota change, reserve credits for
2834                  * dquot_initialize()
2835                  */
2836                 oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2837         }
2838
2839         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET, credits);
2840
2841         return 0;
2842 }
2843
2844 /*
2845  * Set the 64-bit version for object
2846  */
2847 static void osd_object_version_set(const struct lu_env *env,
2848                                    struct dt_object *dt,
2849                                    dt_obj_version_t *new_version)
2850 {
2851         struct inode *inode = osd_dt_obj(dt)->oo_inode;
2852
2853         CDEBUG(D_INODE, "Set version "LPX64" (old "LPX64") for inode %lu\n",
2854                *new_version, LDISKFS_I(inode)->i_fs_version, inode->i_ino);
2855
2856         LDISKFS_I(inode)->i_fs_version = *new_version;
2857         /** Version is set after all inode operations are finished,
2858          *  so we should mark it dirty here */
2859         ll_dirty_inode(inode, I_DIRTY_DATASYNC);
2860 }
2861
2862 /*
2863  * Concurrency: @dt is write locked.
2864  */
2865 static int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
2866                          const struct lu_buf *buf, const char *name, int fl,
2867                          struct thandle *handle)
2868 {
2869         struct osd_object      *obj      = osd_dt_obj(dt);
2870         struct inode           *inode    = obj->oo_inode;
2871         struct osd_thread_info *info     = osd_oti_get(env);
2872         int                     fs_flags = 0;
2873         ENTRY;
2874
2875         LASSERT(handle != NULL);
2876
2877         /* version set is not real XATTR */
2878         if (strcmp(name, XATTR_NAME_VERSION) == 0) {
2879                 /* for version we are just using xattr API but change inode
2880                  * field instead */
2881                 LASSERT(buf->lb_len == sizeof(dt_obj_version_t));
2882                 osd_object_version_set(env, dt, buf->lb_buf);
2883                 return sizeof(dt_obj_version_t);
2884         }
2885
2886         CDEBUG(D_INODE, DFID" set xattr '%s' with size %zu\n",
2887                PFID(lu_object_fid(&dt->do_lu)), name, buf->lb_len);
2888
2889         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2890         if (fl & LU_XATTR_REPLACE)
2891                 fs_flags |= XATTR_REPLACE;
2892
2893         if (fl & LU_XATTR_CREATE)
2894                 fs_flags |= XATTR_CREATE;
2895
2896         if (strcmp(name, XATTR_NAME_LMV) == 0) {
2897                 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2898                 int                      rc;
2899
2900                 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2901                 if (rc != 0)
2902                         RETURN(rc);
2903
2904                 lma->lma_incompat |= LMAI_STRIPED;
2905                 lustre_lma_swab(lma);
2906                 rc = __osd_xattr_set(info, inode, XATTR_NAME_LMA, lma,
2907                                      sizeof(*lma), XATTR_REPLACE);
2908                 if (rc != 0)
2909                         RETURN(rc);
2910         }
2911
2912         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_LINKEA_OVERFLOW) &&
2913             strcmp(name, XATTR_NAME_LINK) == 0)
2914                 return -ENOSPC;
2915
2916         return __osd_xattr_set(info, inode, name, buf->lb_buf, buf->lb_len,
2917                                fs_flags);
2918 }
2919
2920 /*
2921  * Concurrency: @dt is read locked.
2922  */
2923 static int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
2924                           const struct lu_buf *buf)
2925 {
2926         struct osd_object      *obj    = osd_dt_obj(dt);
2927         struct inode           *inode  = obj->oo_inode;
2928         struct osd_thread_info *info   = osd_oti_get(env);
2929         struct dentry          *dentry = &info->oti_obj_dentry;
2930
2931         if (!dt_object_exists(dt))
2932                 return -ENOENT;
2933
2934         LASSERT(!dt_object_remote(dt));
2935         LASSERT(inode->i_op != NULL);
2936         LASSERT(inode->i_op->listxattr != NULL);
2937
2938         dentry->d_inode = inode;
2939         dentry->d_sb = inode->i_sb;
2940         return inode->i_op->listxattr(dentry, buf->lb_buf, buf->lb_len);
2941 }
2942
2943 static int osd_declare_xattr_del(const struct lu_env *env,
2944                                  struct dt_object *dt, const char *name,
2945                                  struct thandle *handle)
2946 {
2947         struct osd_thandle *oh;
2948         struct super_block *sb = osd_sb(osd_dev(dt->do_lu.lo_dev));
2949
2950         LASSERT(!dt_object_remote(dt));
2951         LASSERT(handle != NULL);
2952
2953         oh = container_of0(handle, struct osd_thandle, ot_super);
2954         LASSERT(oh->ot_handle == NULL);
2955
2956         osd_trans_declare_op(env, oh, OSD_OT_XATTR_SET,
2957                              osd_dto_credits_noquota[DTO_XATTR_SET]);
2958         /*
2959          * xattr del may involve inode quota change, reserve credits for
2960          * dquot_initialize()
2961          */
2962         oh->ot_credits += LDISKFS_MAXQUOTAS_INIT_BLOCKS(sb);
2963
2964         return 0;
2965 }
2966
2967 /*
2968  * Concurrency: @dt is write locked.
2969  */
2970 static int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
2971                          const char *name, struct thandle *handle)
2972 {
2973         struct osd_object      *obj    = osd_dt_obj(dt);
2974         struct inode           *inode  = obj->oo_inode;
2975         struct osd_thread_info *info   = osd_oti_get(env);
2976         struct dentry          *dentry = &info->oti_obj_dentry;
2977         int                     rc;
2978
2979         if (!dt_object_exists(dt))
2980                 return -ENOENT;
2981
2982         LASSERT(!dt_object_remote(dt));
2983         LASSERT(inode->i_op != NULL);
2984         LASSERT(inode->i_op->removexattr != NULL);
2985         LASSERT(handle != NULL);
2986
2987         osd_trans_exec_op(env, handle, OSD_OT_XATTR_SET);
2988
2989         ll_vfs_dq_init(inode);
2990         dentry->d_inode = inode;
2991         dentry->d_sb = inode->i_sb;
2992         rc = inode->i_op->removexattr(dentry, name);
2993         return rc;
2994 }
2995
2996 static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
2997                            __u64 start, __u64 end)
2998 {
2999         struct osd_object       *obj    = osd_dt_obj(dt);
3000         struct inode            *inode  = obj->oo_inode;
3001         struct osd_thread_info  *info   = osd_oti_get(env);
3002         struct dentry           *dentry = &info->oti_obj_dentry;
3003         struct file             *file   = &info->oti_file;
3004         int                     rc;
3005
3006         ENTRY;
3007
3008         dentry->d_inode = inode;
3009         dentry->d_sb = inode->i_sb;
3010         file->f_path.dentry = dentry;
3011         file->f_mapping = inode->i_mapping;
3012         file->f_op = inode->i_fop;
3013         set_file_inode(file, inode);
3014
3015         rc = ll_vfs_fsync_range(file, start, end, 0);
3016
3017         RETURN(rc);
3018 }
3019
3020 /*
3021  * Index operations.
3022  */
3023
3024 static int osd_iam_index_probe(const struct lu_env *env, struct osd_object *o,
3025                            const struct dt_index_features *feat)
3026 {
3027         struct iam_descr *descr;
3028
3029         if (osd_object_is_root(o))
3030                 return feat == &dt_directory_features;
3031
3032         LASSERT(o->oo_dir != NULL);
3033
3034         descr = o->oo_dir->od_container.ic_descr;
3035         if (feat == &dt_directory_features) {
3036                 if (descr->id_rec_size == sizeof(struct osd_fid_pack))
3037                         return 1;
3038                 else
3039                         return 0;
3040         } else {
3041                 return
3042                         feat->dif_keysize_min <= descr->id_key_size &&
3043                         descr->id_key_size <= feat->dif_keysize_max &&
3044                         feat->dif_recsize_min <= descr->id_rec_size &&
3045                         descr->id_rec_size <= feat->dif_recsize_max &&
3046                         !(feat->dif_flags & (DT_IND_VARKEY |
3047                                              DT_IND_VARREC | DT_IND_NONUNQ)) &&
3048                         ergo(feat->dif_flags & DT_IND_UPDATE,
3049                              1 /* XXX check that object (and file system) is
3050                                 * writable */);
3051         }
3052 }
3053
3054 static int osd_iam_container_init(const struct lu_env *env,
3055                                   struct osd_object *obj,
3056                                   struct osd_directory *dir)
3057 {
3058         struct iam_container *bag = &dir->od_container;
3059         int result;
3060
3061         result = iam_container_init(bag, &dir->od_descr, obj->oo_inode);
3062         if (result != 0)
3063                 return result;
3064
3065         result = iam_container_setup(bag);
3066         if (result == 0)
3067                 obj->oo_dt.do_index_ops = &osd_index_iam_ops;
3068         else
3069                 iam_container_fini(bag);
3070
3071         return result;
3072 }
3073
3074
3075 /*
3076  * Concurrency: no external locking is necessary.
3077  */
3078 static int osd_index_try(const struct lu_env *env, struct dt_object *dt,
3079                          const struct dt_index_features *feat)
3080 {
3081         int                      result;
3082         int                      skip_iam = 0;
3083         struct osd_object       *obj = osd_dt_obj(dt);
3084
3085         LINVRNT(osd_invariant(obj));
3086
3087         if (osd_object_is_root(obj)) {
3088                 dt->do_index_ops = &osd_index_ea_ops;
3089                 result = 0;
3090         } else if (feat == &dt_directory_features) {
3091                 dt->do_index_ops = &osd_index_ea_ops;
3092                 if (obj->oo_inode != NULL && S_ISDIR(obj->oo_inode->i_mode))
3093                         result = 0;
3094                 else
3095                         result = -ENOTDIR;
3096                 skip_iam = 1;
3097         } else if (unlikely(feat == &dt_otable_features)) {
3098                 dt->do_index_ops = &osd_otable_ops;
3099                 return 0;
3100         } else if (unlikely(feat == &dt_acct_features)) {
3101                 dt->do_index_ops = &osd_acct_index_ops;
3102                 result = 0;
3103                 skip_iam = 1;
3104         } else if (!osd_has_index(obj)) {
3105                 struct osd_directory *dir;
3106
3107                 OBD_ALLOC_PTR(dir);
3108                 if (dir != NULL) {
3109
3110                         spin_lock(&obj->oo_guard);
3111                         if (obj->oo_dir == NULL)
3112                                 obj->oo_dir = dir;
3113                         else
3114                                 /*
3115                                  * Concurrent thread allocated container data.
3116                                  */
3117                                 OBD_FREE_PTR(dir);
3118                         spin_unlock(&obj->oo_guard);
3119                         /*
3120                          * Now, that we have container data, serialize its
3121                          * initialization.
3122                          */
3123                         down_write(&obj->oo_ext_idx_sem);
3124                         /*
3125                          * recheck under lock.
3126                          */
3127                         if (!osd_has_index(obj))
3128                                 result = osd_iam_container_init(env, obj,
3129                                                                 obj->oo_dir);
3130                         else
3131                                 result = 0;
3132                         up_write(&obj->oo_ext_idx_sem);
3133                 } else {
3134                         result = -ENOMEM;
3135                 }
3136         } else {
3137                 result = 0;
3138         }
3139
3140         if (result == 0 && skip_iam == 0) {
3141                 if (!osd_iam_index_probe(env, obj, feat))
3142                         result = -ENOTDIR;
3143         }
3144         LINVRNT(osd_invariant(obj));
3145
3146         if (result == 0 && feat == &dt_quota_glb_features &&
3147             fid_seq(lu_object_fid(&dt->do_lu)) == FID_SEQ_QUOTA_GLB)
3148                 result = osd_quota_migration(env, dt);
3149
3150         return result;
3151 }
3152
3153 static int osd_otable_it_attr_get(const struct lu_env *env,
3154                                  struct dt_object *dt,
3155                                  struct lu_attr *attr)
3156 {
3157         attr->la_valid = 0;
3158         return 0;
3159 }
3160
3161 static const struct dt_object_operations osd_obj_ops = {
3162         .do_read_lock         = osd_object_read_lock,
3163         .do_write_lock        = osd_object_write_lock,
3164         .do_read_unlock       = osd_object_read_unlock,
3165         .do_write_unlock      = osd_object_write_unlock,
3166         .do_write_locked      = osd_object_write_locked,
3167         .do_attr_get          = osd_attr_get,
3168         .do_declare_attr_set  = osd_declare_attr_set,
3169         .do_attr_set          = osd_attr_set,
3170         .do_ah_init           = osd_ah_init,
3171         .do_declare_create    = osd_declare_object_create,
3172         .do_create            = osd_object_create,
3173         .do_declare_destroy   = osd_declare_object_destroy,
3174         .do_destroy           = osd_object_destroy,
3175         .do_index_try         = osd_index_try,
3176         .do_declare_ref_add   = osd_declare_object_ref_add,
3177         .do_ref_add           = osd_object_ref_add,
3178         .do_declare_ref_del   = osd_declare_object_ref_del,
3179         .do_ref_del           = osd_object_ref_del,
3180         .do_xattr_get         = osd_xattr_get,
3181         .do_declare_xattr_set = osd_declare_xattr_set,
3182         .do_xattr_set         = osd_xattr_set,
3183         .do_declare_xattr_del = osd_declare_xattr_del,
3184         .do_xattr_del         = osd_xattr_del,
3185         .do_xattr_list        = osd_xattr_list,
3186         .do_object_sync       = osd_object_sync,
3187 };
3188
3189 /**
3190  * dt_object_operations for interoperability mode
3191  * (i.e. to run 2.0 mds on 1.8 disk) (b11826)
3192  */
3193 static const struct dt_object_operations osd_obj_ea_ops = {
3194         .do_read_lock         = osd_object_read_lock,
3195         .do_write_lock        = osd_object_write_lock,
3196         .do_read_unlock       = osd_object_read_unlock,
3197         .do_write_unlock      = osd_object_write_unlock,
3198         .do_write_locked      = osd_object_write_locked,
3199         .do_attr_get          = osd_attr_get,
3200         .do_declare_attr_set  = osd_declare_attr_set,
3201         .do_attr_set          = osd_attr_set,
3202         .do_ah_init           = osd_ah_init,
3203         .do_declare_create    = osd_declare_object_create,
3204         .do_create            = osd_object_ea_create,
3205         .do_declare_destroy   = osd_declare_object_destroy,
3206         .do_destroy           = osd_object_destroy,
3207         .do_index_try         = osd_index_try,
3208         .do_declare_ref_add   = osd_declare_object_ref_add,
3209         .do_ref_add           = osd_object_ref_add,
3210         .do_declare_ref_del   = osd_declare_object_ref_del,
3211         .do_ref_del           = osd_object_ref_del,
3212         .do_xattr_get         = osd_xattr_get,
3213         .do_declare_xattr_set = osd_declare_xattr_set,
3214         .do_xattr_set         = osd_xattr_set,
3215         .do_declare_xattr_del = osd_declare_xattr_del,
3216         .do_xattr_del         = osd_xattr_del,
3217         .do_xattr_list        = osd_xattr_list,
3218         .do_object_sync       = osd_object_sync,
3219 };
3220
3221 static const struct dt_object_operations osd_obj_otable_it_ops = {
3222         .do_attr_get    = osd_otable_it_attr_get,
3223         .do_index_try   = osd_index_try,
3224 };
3225
3226 static int osd_index_declare_iam_delete(const struct lu_env *env,
3227                                         struct dt_object *dt,
3228                                         const struct dt_key *key,
3229                                         struct thandle *handle)
3230 {
3231         struct osd_thandle    *oh;
3232
3233         oh = container_of0(handle, struct osd_thandle, ot_super);
3234         LASSERT(oh->ot_handle == NULL);
3235
3236         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3237                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3238
3239         return 0;
3240 }
3241
3242 /**
3243  *      delete a (key, value) pair from index \a dt specified by \a key
3244  *
3245  *      \param  dt      osd index object
3246  *      \param  key     key for index
3247  *      \param  rec     record reference
3248  *      \param  handle  transaction handler
3249  *
3250  *      \retval  0  success
3251  *      \retval -ve   failure
3252  */
3253 static int osd_index_iam_delete(const struct lu_env *env, struct dt_object *dt,
3254                                 const struct dt_key *key,
3255                                 struct thandle *handle)
3256 {
3257         struct osd_thread_info *oti = osd_oti_get(env);
3258         struct osd_object      *obj = osd_dt_obj(dt);
3259         struct osd_thandle     *oh;
3260         struct iam_path_descr  *ipd;
3261         struct iam_container   *bag = &obj->oo_dir->od_container;
3262         int                     rc;
3263         ENTRY;
3264
3265         if (!dt_object_exists(dt))
3266                 RETURN(-ENOENT);
3267
3268         LINVRNT(osd_invariant(obj));
3269         LASSERT(!dt_object_remote(dt));
3270         LASSERT(bag->ic_object == obj->oo_inode);
3271         LASSERT(handle != NULL);
3272
3273         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3274
3275         ipd = osd_idx_ipd_get(env, bag);
3276         if (unlikely(ipd == NULL))
3277                 RETURN(-ENOMEM);
3278
3279         oh = container_of0(handle, struct osd_thandle, ot_super);
3280         LASSERT(oh->ot_handle != NULL);
3281         LASSERT(oh->ot_handle->h_transaction != NULL);
3282
3283         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3284                 /* swab quota uid/gid provided by caller */
3285                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3286                 key = (const struct dt_key *)&oti->oti_quota_id;
3287         }
3288
3289         rc = iam_delete(oh->ot_handle, bag, (const struct iam_key *)key, ipd);
3290         osd_ipd_put(env, bag, ipd);
3291         LINVRNT(osd_invariant(obj));
3292         RETURN(rc);
3293 }
3294
3295 static int osd_index_declare_ea_delete(const struct lu_env *env,
3296                                        struct dt_object *dt,
3297                                        const struct dt_key *key,
3298                                        struct thandle *handle)
3299 {
3300         struct osd_thandle *oh;
3301         struct inode       *inode;
3302         int                 rc;
3303         ENTRY;
3304
3305         LASSERT(!dt_object_remote(dt));
3306         LASSERT(handle != NULL);
3307
3308         oh = container_of0(handle, struct osd_thandle, ot_super);
3309         LASSERT(oh->ot_handle == NULL);
3310
3311         osd_trans_declare_op(env, oh, OSD_OT_DELETE,
3312                              osd_dto_credits_noquota[DTO_INDEX_DELETE]);
3313
3314         inode = osd_dt_obj(dt)->oo_inode;
3315         LASSERT(inode);
3316
3317         rc = osd_declare_inode_qid(env, i_uid_read(inode), i_gid_read(inode),
3318                                    0, oh, osd_dt_obj(dt), true, NULL, false);
3319         RETURN(rc);
3320 }
3321
3322 static inline int osd_get_fid_from_dentry(struct ldiskfs_dir_entry_2 *de,
3323                                           struct dt_rec *fid)
3324 {
3325         struct osd_fid_pack *rec;
3326         int                  rc = -ENODATA;
3327
3328         if (de->file_type & LDISKFS_DIRENT_LUFID) {
3329                 rec = (struct osd_fid_pack *) (de->name + de->name_len + 1);
3330                 rc = osd_fid_unpack((struct lu_fid *)fid, rec);
3331                 if (rc == 0 && unlikely(!fid_is_sane((struct lu_fid *)fid)))
3332                         rc = -EINVAL;
3333         }
3334         return rc;
3335 }
3336
3337 static int osd_remote_fid(const struct lu_env *env, struct osd_device *osd,
3338                           const struct lu_fid *fid)
3339 {
3340         struct seq_server_site  *ss = osd_seq_site(osd);
3341         ENTRY;
3342
3343         /* FID seqs not in FLDB, must be local seq */
3344         if (unlikely(!fid_seq_in_fldb(fid_seq(fid))))
3345                 RETURN(0);
3346
3347         /* If FLD is not being initialized yet, it only happens during the
3348          * initialization, likely during mgs initialization, and we assume
3349          * this is local FID. */
3350         if (ss == NULL || ss->ss_server_fld == NULL)
3351                 RETURN(0);
3352
3353         /* Only check the local FLDB here */
3354         if (osd_seq_exists(env, osd, fid_seq(fid)))
3355                 RETURN(0);
3356
3357         RETURN(1);
3358 }
3359
3360 /**
3361  * Index delete function for interoperability mode (b11826).
3362  * It will remove the directory entry added by osd_index_ea_insert().
3363  * This entry is needed to maintain name->fid mapping.
3364  *
3365  * \param key,  key i.e. file entry to be deleted
3366  *
3367  * \retval   0, on success
3368  * \retval -ve, on error
3369  */
3370 static int osd_index_ea_delete(const struct lu_env *env, struct dt_object *dt,
3371                                const struct dt_key *key, struct thandle *handle)
3372 {
3373         struct osd_object          *obj = osd_dt_obj(dt);
3374         struct inode               *dir = obj->oo_inode;
3375         struct dentry              *dentry;
3376         struct osd_thandle         *oh;
3377         struct ldiskfs_dir_entry_2 *de = NULL;
3378         struct buffer_head         *bh;
3379         struct htree_lock          *hlock = NULL;
3380         struct lu_fid              *fid = &osd_oti_get(env)->oti_fid;
3381         struct osd_device          *osd = osd_dev(dt->do_lu.lo_dev);
3382         int                        rc;
3383         ENTRY;
3384
3385         if (!dt_object_exists(dt))
3386                 RETURN(-ENOENT);
3387
3388         LINVRNT(osd_invariant(obj));
3389         LASSERT(!dt_object_remote(dt));
3390         LASSERT(handle != NULL);
3391
3392         osd_trans_exec_op(env, handle, OSD_OT_DELETE);
3393
3394         oh = container_of(handle, struct osd_thandle, ot_super);
3395         LASSERT(oh->ot_handle != NULL);
3396         LASSERT(oh->ot_handle->h_transaction != NULL);
3397
3398         ll_vfs_dq_init(dir);
3399         dentry = osd_child_dentry_get(env, obj,
3400                                       (char *)key, strlen((char *)key));
3401
3402         if (obj->oo_hl_head != NULL) {
3403                 hlock = osd_oti_get(env)->oti_hlock;
3404                 ldiskfs_htree_lock(hlock, obj->oo_hl_head,
3405                                    dir, LDISKFS_HLOCK_DEL);
3406         } else {
3407                 down_write(&obj->oo_ext_idx_sem);
3408         }
3409
3410         bh = osd_ldiskfs_find_entry(dir, &dentry->d_name, &de, NULL, hlock);
3411         if (bh) {
3412                 __u32 ino = 0;
3413
3414                 /* If this is not the ".." entry, it might be a remote DNE
3415                  * entry and  we need to check if the FID is for a remote
3416                  * MDT.  If the FID is  not in the directory entry (e.g.
3417                  * upgraded 1.8 filesystem without dirdata enabled) then
3418                  * we need to get the FID from the LMA. For a remote directory
3419                  * there HAS to be an LMA, it cannot be an IGIF inode in this
3420                  * case.
3421                  *
3422                  * Delete the entry before the agent inode in order to
3423                  * simplify error handling.  At worst an error after deleting
3424                  * the entry first might leak the agent inode afterward. The
3425                  * reverse would need filesystem abort in case of error deleting
3426                  * the entry after the agent had been removed, or leave a
3427                  * dangling entry pointing at a random inode. */
3428                 if (strcmp((char *)key, dotdot) != 0) {
3429                         LASSERT(de != NULL);
3430                         rc = osd_get_fid_from_dentry(de, (struct dt_rec *)fid);
3431                         /* If Fid is not in dentry, try to get it from LMA */
3432                         if (rc == -ENODATA) {
3433                                 struct osd_inode_id *id;
3434                                 struct inode *inode;
3435
3436                                 /* Before trying to get fid from the inode,
3437                                  * check whether the inode is valid.
3438                                  *
3439                                  * If the inode has been deleted, do not go
3440                                  * ahead to do osd_ea_fid_get, which will set
3441                                  * the inode to bad inode, which might cause
3442                                  * the inode to be deleted uncorrectly */
3443                                 inode = ldiskfs_iget(osd_sb(osd),
3444                                                      le32_to_cpu(de->inode));
3445                                 if (IS_ERR(inode)) {
3446                                         CDEBUG(D_INODE, "%s: "DFID"get inode"
3447                                                "error.\n", osd_name(osd),
3448                                                PFID(fid));
3449                                         rc = PTR_ERR(inode);
3450                                 } else {
3451                                         if (likely(inode->i_nlink != 0)) {
3452                                                 id = &osd_oti_get(env)->oti_id;
3453                                                 rc = osd_ea_fid_get(env, obj,
3454                                                         le32_to_cpu(de->inode),
3455                                                                     fid, id);
3456                                         } else {
3457                                                 CDEBUG(D_INFO, "%s: %u "DFID
3458                                                        "deleted.\n",
3459                                                        osd_name(osd),
3460                                                        le32_to_cpu(de->inode),
3461                                                        PFID(fid));
3462                                                 rc = -ESTALE;
3463                                         }
3464                                         iput(inode);
3465                                 }
3466                         }
3467                         if (rc == 0 &&
3468                             unlikely(osd_remote_fid(env, osd, fid)))
3469                                 /* Need to delete agent inode */
3470                                 ino = le32_to_cpu(de->inode);
3471                 }
3472                 rc = ldiskfs_delete_entry(oh->ot_handle, dir, de, bh);
3473                 brelse(bh);
3474                 if (rc == 0 && unlikely(ino != 0)) {
3475                         rc = osd_delete_local_agent_inode(env, osd, fid, ino,
3476                                                           oh);
3477                         if (rc != 0)
3478                                 CERROR("%s: del local inode "DFID": rc = %d\n",
3479                                        osd_name(osd), PFID(fid), rc);
3480                 }
3481         } else {
3482                 rc = -ENOENT;
3483         }
3484         if (hlock != NULL)
3485                 ldiskfs_htree_unlock(hlock);
3486         else
3487                 up_write(&obj->oo_ext_idx_sem);
3488
3489         if (rc != 0)
3490                 GOTO(out, rc);
3491
3492         /* For inode on the remote MDT, .. will point to
3493          * /Agent directory, Check whether it needs to delete
3494          * from agent directory */
3495         if (unlikely(strcmp((char *)key, dotdot) == 0)) {
3496                 rc = osd_delete_from_remote_parent(env, osd_obj2dev(obj), obj,
3497                                                    oh);
3498                 if (rc != 0 && rc != -ENOENT) {
3499                         CERROR("%s: delete agent inode "DFID": rc = %d\n",
3500                                osd_name(osd), PFID(fid), rc);
3501                 }
3502
3503                 if (rc == -ENOENT)
3504                         rc = 0;
3505
3506                 GOTO(out, rc);
3507         }
3508 out:
3509
3510         LASSERT(osd_invariant(obj));
3511         RETURN(rc);
3512 }
3513
3514 /**
3515  *      Lookup index for \a key and copy record to \a rec.
3516  *
3517  *      \param  dt      osd index object
3518  *      \param  key     key for index
3519  *      \param  rec     record reference
3520  *
3521  *      \retval  +ve  success : exact mach
3522  *      \retval  0    return record with key not greater than \a key
3523  *      \retval -ve   failure
3524  */
3525 static int osd_index_iam_lookup(const struct lu_env *env, struct dt_object *dt,
3526                                 struct dt_rec *rec, const struct dt_key *key)
3527 {
3528         struct osd_object      *obj = osd_dt_obj(dt);
3529         struct iam_path_descr  *ipd;
3530         struct iam_container   *bag = &obj->oo_dir->od_container;
3531         struct osd_thread_info *oti = osd_oti_get(env);
3532         struct iam_iterator    *it = &oti->oti_idx_it;
3533         struct iam_rec         *iam_rec;
3534         int                     rc;
3535         ENTRY;
3536
3537         if (!dt_object_exists(dt))
3538                 RETURN(-ENOENT);
3539
3540         LASSERT(osd_invariant(obj));
3541         LASSERT(!dt_object_remote(dt));
3542         LASSERT(bag->ic_object == obj->oo_inode);
3543
3544         ipd = osd_idx_ipd_get(env, bag);
3545         if (IS_ERR(ipd))
3546                 RETURN(-ENOMEM);
3547
3548         /* got ipd now we can start iterator. */
3549         iam_it_init(it, bag, 0, ipd);
3550
3551         if (fid_is_quota(lu_object_fid(&dt->do_lu))) {
3552                 /* swab quota uid/gid provided by caller */
3553                 oti->oti_quota_id = cpu_to_le64(*((__u64 *)key));
3554                 key = (const struct dt_key *)&oti->oti_quota_id;
3555         }
3556
3557         rc = iam_it_get(it, (struct iam_key *)key);
3558         if (rc >= 0) {
3559                 if (S_ISDIR(obj->oo_inode->i_mode))
3560                         iam_rec = (struct iam_rec *)oti->oti_ldp;
3561                 else
3562                         iam_rec = (struct iam_rec *) rec;
3563
3564                 iam_reccpy(&it->ii_path.ip_leaf, (struct iam_rec *)iam_rec);
3565
3566                 if (S_ISDIR(obj->oo_inode->i_mode))
3567                         osd_fid_unpack((struct lu_fid *) rec,
3568                                        (struct osd_fid_pack *)iam_rec);
3569                 else if (fid_is_quota(lu_object_fid(&dt->do_lu)))
3570                         osd_quota_unpack(obj, rec);
3571         }
3572
3573         iam_it_put(it);
3574         iam_it_fini(it);
3575         osd_ipd_put(env, bag, ipd);
3576
3577         LINVRNT(osd_invariant(obj));
3578
3579         RETURN(rc);
3580 }
3581
3582 static int osd_index_declare_iam_insert(const struct lu_env *env,
3583                                         struct dt_object *dt,
3584                                         const struct dt_rec *rec,
3585                                         const struct dt_key *key,
3586                                         struct thandle *handle)
3587 {
3588         struct osd_thandle *oh;
3589
3590         LASSERT(handle != NULL);
3591
3592         oh = container_of0(handle, struct osd_thandle, ot_super);
3593         LASSERT(oh->ot_handle == NULL);
3594
3595         osd_trans_declare_op(env, oh, OSD_OT_INSERT,
3596                              osd_dto_credits_noquota[DTO_INDEX_INSERT]);
3597
3598         return 0;
3599 }
3600
3601 /**
3602  *      Inserts (key, value) pair in \a dt index object.
3603  *
3604  *      \param  dt      osd index object
3605  *      \param  key     key for index
3606  *      \param  rec     record reference
3607  *      \param  th      transaction handler
3608  *
3609  *      \retval  0  success
3610  *      \retval -ve failure
3611  */
3612 static int osd_index_iam_insert(const struct lu_env *env, struct dt_object *dt,
3613                                 const struct dt_rec *rec,
3614                                 const struct dt_key *key, struct thandle *th,
3615                                 int ignore_quota)
3616 {
3617         struct osd_object     *obj = osd_dt_obj(dt);
3618         struct iam_path_descr *ipd;
3619         struct osd_thandle    *oh;
3620         struct iam_container  *bag;
3621         struct osd_thread_info *oti = osd_oti_get(env);
3622         struct iam_rec         *iam_rec;
3623         int                     rc;
3624         ENTRY;
3625
3626         if (!dt_object_exists(dt))
3627                 RETURN(-ENOENT);
3628
3629         LINVRNT(osd_invariant(obj));
3630         LASSERT(!dt_object_remote(dt));
3631
3632         bag = &obj->oo_dir->od_container;
3633         LASSERT(bag->ic_object == obj->oo_inode);
3634         LASSERT(th != NULL);
3635
3636         osd_trans_exec_op(env, th, OSD_OT_INSERT);
3637
3638         ipd = osd_idx_ipd_get(env, bag);
3639         if (unlikely(ipd == NULL))
3640                 RETURN(-ENOMEM);
3641
3642         oh = container_of0(th, struct osd_thandle, ot_super);
3643         LASSERT(oh->ot_handle != NULL);
3644         LASSERT(oh->ot_handle->h_transaction != NULL);
3645         if (S_ISDIR(obj->oo_inode->i_mode)) {
3646                 iam_rec =&n