Whamcloud - gitweb
235fb0eb88d806041d1797df35b0fc25b2dd10b3
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_compat.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) 2010, Oracle and/or its affiliates. All rights reserved.
28  * Use is subject to license terms.
29  *
30  * Copyright (c) 2012, 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_compat.c
37  *
38  * on-disk structure for managing /O
39  *
40  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
41  */
42
43 /* LUSTRE_VERSION_CODE */
44 #include <lustre_ver.h>
45 /* prerequisite for linux/xattr.h */
46 #include <linux/types.h>
47 /* prerequisite for linux/xattr.h */
48 #include <linux/fs.h>
49 /* XATTR_{REPLACE,CREATE} */
50 #include <linux/xattr.h>
51
52 /*
53  * struct OBD_{ALLOC,FREE}*()
54  * OBD_FAIL_CHECK
55  */
56 #include <obd_support.h>
57
58 #include "osd_internal.h"
59 #include "osd_oi.h"
60
61 static void osd_push_ctxt(const struct osd_device *dev,
62                           struct lvfs_run_ctxt *newctxt,
63                           struct lvfs_run_ctxt *save)
64 {
65         OBD_SET_CTXT_MAGIC(newctxt);
66         newctxt->pwdmnt = dev->od_mnt;
67         newctxt->pwd = dev->od_mnt->mnt_root;
68         newctxt->fs = get_ds();
69         newctxt->umask = current_umask();
70         newctxt->dt = NULL;
71
72         push_ctxt(save, newctxt);
73 }
74
75 /* utility to make a directory */
76 static struct dentry *simple_mkdir(struct dentry *dir, struct vfsmount *mnt,
77                                    const char *name, int mode, int fix)
78 {
79         struct dentry *dchild;
80         int err = 0;
81         ENTRY;
82
83         // ASSERT_KERNEL_CTXT("kernel doing mkdir outside kernel context\n");
84         CDEBUG(D_INODE, "creating directory %.*s\n", (int)strlen(name), name);
85         dchild = ll_lookup_one_len(name, dir, strlen(name));
86         if (IS_ERR(dchild))
87                 GOTO(out_up, dchild);
88
89         if (dchild->d_inode) {
90                 int old_mode = dchild->d_inode->i_mode;
91                 if (!S_ISDIR(old_mode)) {
92                         CERROR("found %s (%lu/%u) is mode %o\n", name,
93                                dchild->d_inode->i_ino,
94                                dchild->d_inode->i_generation, old_mode);
95                         GOTO(out_err, err = -ENOTDIR);
96                 }
97
98                 /* Fixup directory permissions if necessary */
99                 if (fix && (old_mode & S_IALLUGO) != (mode & S_IALLUGO)) {
100                         CDEBUG(D_CONFIG,
101                                "fixing permissions on %s from %o to %o\n",
102                                name, old_mode, mode);
103                         dchild->d_inode->i_mode = (mode & S_IALLUGO) |
104                                                   (old_mode & ~S_IALLUGO);
105                         mark_inode_dirty(dchild->d_inode);
106                 }
107                 GOTO(out_up, dchild);
108         }
109
110         err = vfs_mkdir(dir->d_inode, dchild, mode);
111         if (err)
112                 GOTO(out_err, err);
113
114         RETURN(dchild);
115
116 out_err:
117         dput(dchild);
118         dchild = ERR_PTR(err);
119 out_up:
120         return dchild;
121 }
122
123 static int osd_last_rcvd_subdir_count(struct osd_device *osd)
124 {
125         struct lr_server_data lsd;
126         struct dentry        *dlast;
127         loff_t                off;
128         int                   rc = 0;
129         int                   count = OBJ_SUBDIR_COUNT;
130
131         ENTRY;
132
133         dlast = ll_lookup_one_len(LAST_RCVD, osd_sb(osd)->s_root,
134                                   strlen(LAST_RCVD));
135         if (IS_ERR(dlast))
136                 return PTR_ERR(dlast);
137         else if (dlast->d_inode == NULL)
138                 goto out;
139
140         off = 0;
141         rc = osd_ldiskfs_read(dlast->d_inode, &lsd, sizeof(lsd), &off);
142         if (rc == sizeof(lsd)) {
143                 CDEBUG(D_INFO, "read last_rcvd header, uuid = %s, "
144                        "subdir count = %d\n", lsd.lsd_uuid,
145                        lsd.lsd_subdir_count);
146                 if (le16_to_cpu(lsd.lsd_subdir_count) > 0)
147                         count = le16_to_cpu(lsd.lsd_subdir_count);
148         } else if (rc != 0) {
149                 CERROR("Can't read last_rcvd file, rc = %d\n", rc);
150                 if (rc > 0)
151                         rc = -EFAULT;
152                 dput(dlast);
153                 return rc;
154         }
155 out:
156         dput(dlast);
157         LASSERT(count > 0);
158         return count;
159 }
160
161 static const char remote_parent_dir[] = "REMOTE_PARENT_DIR";
162 static int osd_mdt_init(const struct lu_env *env, struct osd_device *dev)
163 {
164         struct lvfs_run_ctxt    new;
165         struct lvfs_run_ctxt    save;
166         struct dentry           *parent;
167         struct osd_mdobj_map    *omm;
168         struct dentry           *d;
169         struct osd_thread_info  *info = osd_oti_get(env);
170         struct lu_fid           *fid = &info->oti_fid3;
171         int                     rc = 0;
172         ENTRY;
173
174         OBD_ALLOC_PTR(dev->od_mdt_map);
175         if (dev->od_mdt_map == NULL)
176                 RETURN(-ENOMEM);
177
178         omm = dev->od_mdt_map;
179
180         parent = osd_sb(dev)->s_root;
181         osd_push_ctxt(dev, &new, &save);
182
183         d = simple_mkdir(parent, dev->od_mnt, remote_parent_dir,
184                          0755, 1);
185         if (IS_ERR(d))
186                 GOTO(cleanup, rc = PTR_ERR(d));
187
188         omm->omm_remote_parent = d;
189
190         /* Set LMA for remote parent inode */
191         lu_local_obj_fid(fid, REMOTE_PARENT_DIR_OID);
192         rc = osd_ea_fid_set(info, d->d_inode, fid, LMAC_NOT_IN_OI, 0);
193
194         GOTO(cleanup, rc);
195
196 cleanup:
197         pop_ctxt(&save, &new);
198         if (rc) {
199                 if (omm->omm_remote_parent != NULL)
200                         dput(omm->omm_remote_parent);
201                 OBD_FREE_PTR(omm);
202                 dev->od_mdt_map = NULL;
203         }
204         return rc;
205 }
206
207 static void osd_mdt_fini(struct osd_device *osd)
208 {
209         struct osd_mdobj_map *omm = osd->od_mdt_map;
210
211         if (omm == NULL)
212                 return;
213
214         if (omm->omm_remote_parent)
215                 dput(omm->omm_remote_parent);
216
217         OBD_FREE_PTR(omm);
218         osd->od_ost_map = NULL;
219 }
220
221 int osd_add_to_remote_parent(const struct lu_env *env, struct osd_device *osd,
222                              struct osd_object *obj, struct osd_thandle *oh)
223 {
224         struct osd_mdobj_map    *omm = osd->od_mdt_map;
225         struct osd_thread_info  *oti = osd_oti_get(env);
226         struct lustre_mdt_attrs *lma = &oti->oti_mdt_attrs;
227         char                    *name = oti->oti_name;
228         struct dentry           *dentry;
229         struct dentry           *parent;
230         int                     rc;
231
232         /* Set REMOTE_PARENT in lma, so other process like unlink or lfsck
233          * can identify this object quickly */
234         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
235         if (rc != 0)
236                 RETURN(rc);
237
238         lma->lma_incompat |= LMAI_REMOTE_PARENT;
239         lustre_lma_swab(lma);
240         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
241                              sizeof(*lma), XATTR_REPLACE);
242         if (rc != 0)
243                 RETURN(rc);
244
245         parent = omm->omm_remote_parent;
246         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
247         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
248                                            name, strlen(name));
249         mutex_lock(&parent->d_inode->i_mutex);
250         rc = osd_ldiskfs_add_entry(oh->ot_handle, dentry, obj->oo_inode,
251                                    NULL);
252         CDEBUG(D_INODE, "%s: add %s:%lu to remote parent %lu.\n", osd_name(osd),
253                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
254         ldiskfs_inc_count(oh->ot_handle, parent->d_inode);
255         mark_inode_dirty(parent->d_inode);
256         mutex_unlock(&parent->d_inode->i_mutex);
257         RETURN(rc);
258 }
259
260 int osd_delete_from_remote_parent(const struct lu_env *env,
261                                   struct osd_device *osd,
262                                   struct osd_object *obj,
263                                   struct osd_thandle *oh)
264 {
265         struct osd_mdobj_map       *omm = osd->od_mdt_map;
266         struct osd_thread_info     *oti = osd_oti_get(env);
267         struct lustre_mdt_attrs    *lma = &oti->oti_mdt_attrs;
268         char                       *name = oti->oti_name;
269         struct dentry              *dentry;
270         struct dentry              *parent;
271         struct ldiskfs_dir_entry_2 *de;
272         struct buffer_head         *bh;
273         int                        rc;
274
275         /* Check lma to see whether it is remote object */
276         rc = osd_get_lma(oti, obj->oo_inode, &oti->oti_obj_dentry, lma);
277         if (rc != 0)
278                 RETURN(rc);
279
280         if (likely(!(lma->lma_incompat & LMAI_REMOTE_PARENT)))
281                 RETURN(0);
282
283         parent = omm->omm_remote_parent;
284         sprintf(name, DFID_NOBRACE, PFID(lu_object_fid(&obj->oo_dt.do_lu)));
285         dentry = osd_child_dentry_by_inode(env, parent->d_inode,
286                                            name, strlen(name));
287         mutex_lock(&parent->d_inode->i_mutex);
288         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
289                                     NULL, NULL);
290         if (bh == NULL) {
291                 mutex_unlock(&parent->d_inode->i_mutex);
292                 RETURN(-ENOENT);
293         }
294         CDEBUG(D_INODE, "%s: el %s:%lu to remote parent %lu.\n", osd_name(osd),
295                name, obj->oo_inode->i_ino, parent->d_inode->i_ino);
296         rc = ldiskfs_delete_entry(oh->ot_handle, parent->d_inode, de, bh);
297         ldiskfs_dec_count(oh->ot_handle, parent->d_inode);
298         mark_inode_dirty(parent->d_inode);
299         mutex_unlock(&parent->d_inode->i_mutex);
300         brelse(bh);
301
302         /* Get rid of REMOTE_PARENT flag from incompat */
303         lma->lma_incompat &= ~LMAI_REMOTE_PARENT;
304         lustre_lma_swab(lma);
305         rc = __osd_xattr_set(oti, obj->oo_inode, XATTR_NAME_LMA, lma,
306                              sizeof(*lma), XATTR_REPLACE);
307         RETURN(rc);
308 }
309
310 int osd_lookup_in_remote_parent(struct osd_thread_info *oti,
311                                 struct osd_device *osd,
312                                 const struct lu_fid *fid,
313                                 struct osd_inode_id *id)
314 {
315         struct osd_mdobj_map        *omm = osd->od_mdt_map;
316         char                        *name = oti->oti_name;
317         struct dentry               *parent;
318         struct dentry               *dentry;
319         struct ldiskfs_dir_entry_2 *de;
320         struct buffer_head         *bh;
321         int                         rc;
322         ENTRY;
323
324         parent = omm->omm_remote_parent;
325         sprintf(name, DFID_NOBRACE, PFID(fid));
326         dentry = osd_child_dentry_by_inode(oti->oti_env, parent->d_inode,
327                                            name, strlen(name));
328         mutex_lock(&parent->d_inode->i_mutex);
329         bh = osd_ldiskfs_find_entry(parent->d_inode, &dentry->d_name, &de,
330                                     NULL, NULL);
331         if (bh == NULL) {
332                 rc = -ENOENT;
333         } else {
334                 rc = 0;
335                 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
336                 brelse(bh);
337         }
338         mutex_unlock(&parent->d_inode->i_mutex);
339         if (rc == 0)
340                 osd_add_oi_cache(oti, osd, id, fid);
341         RETURN(rc);
342 }
343
344 /*
345  * directory structure on legacy OST:
346  *
347  * O/<seq>/d0-31/<objid>
348  * O/<seq>/LAST_ID
349  * last_rcvd
350  * LAST_GROUP
351  * CONFIGS
352  *
353  */
354 static int osd_ost_init(const struct lu_env *env, struct osd_device *dev)
355 {
356         struct lvfs_run_ctxt     new;
357         struct lvfs_run_ctxt     save;
358         struct dentry           *rootd = osd_sb(dev)->s_root;
359         struct dentry           *d;
360         struct osd_thread_info  *info = osd_oti_get(env);
361         struct inode            *inode;
362         struct lu_fid           *fid = &info->oti_fid3;
363         int                      rc;
364         ENTRY;
365
366         OBD_ALLOC_PTR(dev->od_ost_map);
367         if (dev->od_ost_map == NULL)
368                 RETURN(-ENOMEM);
369
370         /* to get subdir count from last_rcvd */
371         rc = osd_last_rcvd_subdir_count(dev);
372         if (rc < 0) {
373                 OBD_FREE_PTR(dev->od_ost_map);
374                 RETURN(rc);
375         }
376
377         dev->od_ost_map->om_subdir_count = rc;
378         rc = 0;
379
380         INIT_LIST_HEAD(&dev->od_ost_map->om_seq_list);
381         rwlock_init(&dev->od_ost_map->om_seq_list_lock);
382         mutex_init(&dev->od_ost_map->om_dir_init_mutex);
383
384         osd_push_ctxt(dev, &new, &save);
385
386         d = ll_lookup_one_len("O", rootd, strlen("O"));
387         if (IS_ERR(d))
388                 GOTO(cleanup, rc = PTR_ERR(d));
389         if (d->d_inode == NULL) {
390                 dput(d);
391                 /* The lookup() may be called again inside simple_mkdir().
392                  * Since the repeated lookup() only be called for "/O" at
393                  * mount time, it will not affect the whole performance. */
394                 d = simple_mkdir(rootd, dev->od_mnt, "O", 0755, 1);
395                 if (IS_ERR(d))
396                         GOTO(cleanup, rc = PTR_ERR(d));
397
398                 /* It is quite probably that the device is new formatted. */
399                 dev->od_maybe_new = 1;
400         }
401
402         inode = d->d_inode;
403         dev->od_ost_map->om_root = d;
404
405         /* 'What the @fid is' is not imporatant, because the object
406          * has no OI mapping, and only is visible inside the OSD.*/
407         lu_igif_build(fid, inode->i_ino, inode->i_generation);
408         rc = osd_ea_fid_set(info, inode, fid,
409                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
410
411         GOTO(cleanup, rc);
412
413 cleanup:
414         pop_ctxt(&save, &new);
415         if (IS_ERR(d)) {
416                 OBD_FREE_PTR(dev->od_ost_map);
417                 RETURN(PTR_ERR(d));
418         }
419         return rc;
420 }
421
422 static void osd_seq_free(struct osd_obj_map *map,
423                          struct osd_obj_seq *osd_seq)
424 {
425         int j;
426
427         list_del_init(&osd_seq->oos_seq_list);
428
429         if (osd_seq->oos_dirs) {
430                 for (j = 0; j < osd_seq->oos_subdir_count; j++) {
431                         if (osd_seq->oos_dirs[j])
432                                 dput(osd_seq->oos_dirs[j]);
433                 }
434                 OBD_FREE(osd_seq->oos_dirs,
435                          sizeof(struct dentry *) * osd_seq->oos_subdir_count);
436         }
437
438         if (osd_seq->oos_root)
439                 dput(osd_seq->oos_root);
440
441         OBD_FREE_PTR(osd_seq);
442 }
443
444 static void osd_ost_fini(struct osd_device *osd)
445 {
446         struct osd_obj_seq    *osd_seq;
447         struct osd_obj_seq    *tmp;
448         struct osd_obj_map    *map = osd->od_ost_map;
449         ENTRY;
450
451         if (map == NULL)
452                 return;
453
454         write_lock(&map->om_seq_list_lock);
455         list_for_each_entry_safe(osd_seq, tmp, &map->om_seq_list,
456                                  oos_seq_list) {
457                 osd_seq_free(map, osd_seq);
458         }
459         write_unlock(&map->om_seq_list_lock);
460         if (map->om_root)
461                 dput(map->om_root);
462         OBD_FREE_PTR(map);
463         osd->od_ost_map = NULL;
464         EXIT;
465 }
466
467 int osd_obj_map_init(const struct lu_env *env, struct osd_device *dev)
468 {
469         int rc;
470         ENTRY;
471
472         /* prepare structures for OST */
473         rc = osd_ost_init(env, dev);
474         if (rc)
475                 RETURN(rc);
476
477         /* prepare structures for MDS */
478         rc = osd_mdt_init(env, dev);
479
480         RETURN(rc);
481 }
482
483 static struct osd_obj_seq *osd_seq_find_locked(struct osd_obj_map *map, u64 seq)
484 {
485         struct osd_obj_seq *osd_seq;
486
487         list_for_each_entry(osd_seq, &map->om_seq_list, oos_seq_list) {
488                 if (osd_seq->oos_seq == seq)
489                         return osd_seq;
490         }
491         return NULL;
492 }
493
494 static struct osd_obj_seq *osd_seq_find(struct osd_obj_map *map, u64 seq)
495 {
496         struct osd_obj_seq *osd_seq;
497
498         read_lock(&map->om_seq_list_lock);
499         osd_seq = osd_seq_find_locked(map, seq);
500         read_unlock(&map->om_seq_list_lock);
501         return osd_seq;
502 }
503
504 void osd_obj_map_fini(struct osd_device *dev)
505 {
506         osd_ost_fini(dev);
507         osd_mdt_fini(dev);
508 }
509
510 /**
511  * Update the specified OI mapping.
512  *
513  * \retval   1, changed nothing
514  * \retval   0, changed successfully
515  * \retval -ve, on error
516  */
517 static int osd_obj_update_entry(struct osd_thread_info *info,
518                                 struct osd_device *osd,
519                                 struct dentry *dir, const char *name,
520                                 const struct lu_fid *fid,
521                                 const struct osd_inode_id *id,
522                                 handle_t *th)
523 {
524         struct inode               *parent = dir->d_inode;
525         struct dentry              *child;
526         struct ldiskfs_dir_entry_2 *de;
527         struct buffer_head         *bh;
528         struct inode               *inode;
529         struct dentry              *dentry = &info->oti_obj_dentry;
530         struct osd_inode_id        *oi_id  = &info->oti_id3;
531         struct lustre_mdt_attrs    *lma    = &info->oti_mdt_attrs;
532         struct lu_fid              *oi_fid = &lma->lma_self_fid;
533         int                         rc;
534         ENTRY;
535
536         LASSERT(th != NULL);
537         LASSERT(th->h_transaction != NULL);
538
539         child = &info->oti_child_dentry;
540         child->d_parent = dir;
541         child->d_name.hash = 0;
542         child->d_name.name = name;
543         child->d_name.len = strlen(name);
544
545         ll_vfs_dq_init(parent);
546         mutex_lock(&parent->i_mutex);
547         bh = osd_ldiskfs_find_entry(parent, &child->d_name, &de, NULL, NULL);
548         if (bh == NULL)
549                 GOTO(out, rc = -ENOENT);
550
551         if (le32_to_cpu(de->inode) == id->oii_ino)
552                 GOTO(out, rc = 1);
553
554         osd_id_gen(oi_id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
555         inode = osd_iget(info, osd, oi_id);
556         if (IS_ERR(inode)) {
557                 rc = PTR_ERR(inode);
558                 if (rc == -ENOENT || rc == -ESTALE)
559                         goto update;
560                 GOTO(out, rc);
561         }
562
563         rc = osd_get_lma(info, inode, dentry, lma);
564         if (rc == -ENODATA) {
565                 rc = osd_get_idif(info, inode, dentry, oi_fid);
566                 if (rc > 0) {
567                         oi_fid = NULL;
568                         rc = 0;
569                 }
570         }
571         iput(inode);
572
573         /* If the OST-object has neither FID-in-LMA nor FID-in-ff, it is
574          * either a crashed object or a uninitialized one. Replace it. */
575         if (rc == -ENODATA || oi_fid == NULL)
576                 goto update;
577
578         if (rc != 0)
579                 GOTO(out, rc);
580
581         if (lu_fid_eq(fid, oi_fid)) {
582                 CERROR("%s: the FID "DFID" is used by two objects: "
583                        "%u/%u %u/%u\n", osd_name(osd), PFID(fid),
584                        oi_id->oii_ino, oi_id->oii_gen,
585                        id->oii_ino, id->oii_gen);
586                 GOTO(out, rc = -EEXIST);
587         }
588
589 update:
590         /* There may be temporary inconsistency: On one hand, the new
591          * object may be referenced by multiple entries, which is out
592          * of our control unless we traverse the whole /O completely,
593          * which is non-flat order and inefficient, should be avoided;
594          * On the other hand, the old object may become orphan if it
595          * is still valid. Since it was referenced by an invalid entry,
596          * making it as invisible temporary may be not worse. OI scrub
597          * will process it later. */
598         rc = ldiskfs_journal_get_write_access(th, bh);
599         if (rc != 0)
600                 GOTO(out, rc);
601
602         de->inode = cpu_to_le32(id->oii_ino);
603         rc = ldiskfs_handle_dirty_metadata(th, NULL, bh);
604
605         GOTO(out, rc);
606
607 out:
608         brelse(bh);
609         mutex_unlock(&parent->i_mutex);
610         return rc;
611 }
612
613 static int osd_obj_del_entry(struct osd_thread_info *info,
614                              struct osd_device *osd,
615                              struct dentry *dird, char *name,
616                              handle_t *th)
617 {
618         struct ldiskfs_dir_entry_2 *de;
619         struct buffer_head         *bh;
620         struct dentry              *child;
621         struct inode               *dir = dird->d_inode;
622         int                         rc;
623         ENTRY;
624
625         LASSERT(th != NULL);
626         LASSERT(th->h_transaction != NULL);
627
628
629         child = &info->oti_child_dentry;
630         child->d_name.hash = 0;
631         child->d_name.name = name;
632         child->d_name.len = strlen(name);
633         child->d_parent = dird;
634         child->d_inode = NULL;
635
636         ll_vfs_dq_init(dir);
637         mutex_lock(&dir->i_mutex);
638         rc = -ENOENT;
639         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
640         if (bh) {
641                 rc = ldiskfs_delete_entry(th, dir, de, bh);
642                 brelse(bh);
643         }
644         mutex_unlock(&dir->i_mutex);
645
646         RETURN(rc);
647 }
648
649 static int osd_obj_add_entry(struct osd_thread_info *info,
650                              struct osd_device *osd,
651                              struct dentry *dir, char *name,
652                              const struct osd_inode_id *id,
653                              handle_t *th)
654 {
655         struct dentry *child;
656         struct inode *inode;
657         int rc;
658
659         ENTRY;
660
661         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_NO_ENTRY))
662                 RETURN(0);
663
664         LASSERT(th != NULL);
665         LASSERT(th->h_transaction != NULL);
666
667         inode = info->oti_inode;
668         if (unlikely(inode == NULL)) {
669                 OBD_ALLOC_PTR(inode);
670                 if (inode == NULL)
671                         RETURN(-ENOMEM);
672                 info->oti_inode = inode;
673         }
674
675         inode->i_sb = osd_sb(osd);
676         osd_id_to_inode(inode, id);
677         inode->i_mode = S_IFREG; /* for type in ldiskfs dir entry */
678
679         child = &info->oti_child_dentry;
680         child->d_name.hash = 0;
681         child->d_name.name = name;
682         child->d_name.len = strlen(name);
683         child->d_parent = dir;
684         child->d_inode = inode;
685
686         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_COMPAT_INVALID_ENTRY))
687                 inode->i_ino++;
688
689         ll_vfs_dq_init(dir->d_inode);
690         mutex_lock(&dir->d_inode->i_mutex);
691         rc = osd_ldiskfs_add_entry(th, child, inode, NULL);
692         mutex_unlock(&dir->d_inode->i_mutex);
693
694         RETURN(rc);
695 }
696
697 /**
698  * Use LPU64 for legacy OST sequences, but use LPX64i for new
699  * sequences names, so that the O/{seq}/dN/{oid} more closely
700  * follows the DFID/PFID format. This makes it easier to map from
701  * debug messages to objects in the future, and the legacy space
702  * of FID_SEQ_OST_MDT0 will be unused in the future.
703  **/
704 static inline void osd_seq_name(char *seq_name, size_t name_size, u64 seq)
705 {
706         snprintf(seq_name, name_size,
707                  (fid_seq_is_rsvd(seq) ||
708                   fid_seq_is_mdt0(seq)) ? LPU64 : LPX64i,
709                  fid_seq_is_idif(seq) ? 0 : seq);
710 }
711
712 static inline void osd_oid_name(char *name, size_t name_size,
713                                 const struct lu_fid *fid, u64 id)
714 {
715         snprintf(name, name_size,
716                  (fid_seq_is_rsvd(fid_seq(fid)) ||
717                   fid_seq_is_mdt0(fid_seq(fid)) ||
718                   fid_seq_is_idif(fid_seq(fid))) ? LPU64 : LPX64i, id);
719 }
720
721 /* external locking is required */
722 static int osd_seq_load_locked(struct osd_thread_info *info,
723                                struct osd_device *osd,
724                                struct osd_obj_seq *osd_seq)
725 {
726         struct osd_obj_map  *map = osd->od_ost_map;
727         struct dentry       *seq_dir;
728         struct inode        *inode;
729         struct lu_fid       *fid = &info->oti_fid3;
730         int                 rc = 0;
731         int                 i;
732         char                dir_name[32];
733         ENTRY;
734
735         if (osd_seq->oos_root != NULL)
736                 RETURN(0);
737
738         LASSERT(map);
739         LASSERT(map->om_root);
740
741         osd_seq_name(dir_name, sizeof(dir_name), osd_seq->oos_seq);
742
743         seq_dir = simple_mkdir(map->om_root, osd->od_mnt, dir_name, 0755, 1);
744         if (IS_ERR(seq_dir))
745                 GOTO(out_err, rc = PTR_ERR(seq_dir));
746         else if (seq_dir->d_inode == NULL)
747                 GOTO(out_put, rc = -EFAULT);
748
749         inode = seq_dir->d_inode;
750         osd_seq->oos_root = seq_dir;
751
752         /* 'What the @fid is' is not imporatant, because the object
753          * has no OI mapping, and only is visible inside the OSD.*/
754         lu_igif_build(fid, inode->i_ino, inode->i_generation);
755         rc = osd_ea_fid_set(info, inode, fid,
756                             LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
757         if (rc != 0)
758                 GOTO(out_put, rc);
759
760         LASSERT(osd_seq->oos_dirs == NULL);
761         OBD_ALLOC(osd_seq->oos_dirs,
762                   sizeof(seq_dir) * osd_seq->oos_subdir_count);
763         if (osd_seq->oos_dirs == NULL)
764                 GOTO(out_put, rc = -ENOMEM);
765
766         for (i = 0; i < osd_seq->oos_subdir_count; i++) {
767                 struct dentry   *dir;
768
769                 snprintf(dir_name, sizeof(dir_name), "d%u", i);
770                 dir = simple_mkdir(osd_seq->oos_root, osd->od_mnt, dir_name,
771                                    0700, 1);
772                 if (IS_ERR(dir)) {
773                         GOTO(out_free, rc = PTR_ERR(dir));
774                 } else if (dir->d_inode == NULL) {
775                         dput(dir);
776                         GOTO(out_free, rc = -EFAULT);
777                 }
778
779                 inode = dir->d_inode;
780                 osd_seq->oos_dirs[i] = dir;
781
782                 /* 'What the @fid is' is not imporatant, because the object
783                  * has no OI mapping, and only is visible inside the OSD.*/
784                 lu_igif_build(fid, inode->i_ino, inode->i_generation);
785                 rc = osd_ea_fid_set(info, inode, fid,
786                                     LMAC_NOT_IN_OI | LMAC_FID_ON_OST, 0);
787                 if (rc != 0)
788                         GOTO(out_free, rc);
789         }
790
791         if (rc != 0) {
792 out_free:
793                 for (i = 0; i < osd_seq->oos_subdir_count; i++) {
794                         if (osd_seq->oos_dirs[i] != NULL)
795                                 dput(osd_seq->oos_dirs[i]);
796                 }
797                 OBD_FREE(osd_seq->oos_dirs,
798                          sizeof(seq_dir) * osd_seq->oos_subdir_count);
799 out_put:
800                 dput(seq_dir);
801                 osd_seq->oos_root = NULL;
802         }
803 out_err:
804         RETURN(rc);
805 }
806
807 static struct osd_obj_seq *osd_seq_load(struct osd_thread_info *info,
808                                         struct osd_device *osd, u64 seq)
809 {
810         struct osd_obj_map      *map;
811         struct osd_obj_seq      *osd_seq;
812         int                     rc = 0;
813         ENTRY;
814
815         map = osd->od_ost_map;
816         LASSERT(map);
817         LASSERT(map->om_root);
818
819         osd_seq = osd_seq_find(map, seq);
820         if (likely(osd_seq != NULL))
821                 RETURN(osd_seq);
822
823         /* Serializing init process */
824         mutex_lock(&map->om_dir_init_mutex);
825
826         /* Check whether the seq has been added */
827         read_lock(&map->om_seq_list_lock);
828         osd_seq = osd_seq_find_locked(map, seq);
829         if (osd_seq != NULL) {
830                 read_unlock(&map->om_seq_list_lock);
831                 GOTO(cleanup, rc = 0);
832         }
833         read_unlock(&map->om_seq_list_lock);
834
835         OBD_ALLOC_PTR(osd_seq);
836         if (osd_seq == NULL)
837                 GOTO(cleanup, rc = -ENOMEM);
838
839         INIT_LIST_HEAD(&osd_seq->oos_seq_list);
840         osd_seq->oos_seq = seq;
841         /* Init subdir count to be 32, but each seq can have
842          * different subdir count */
843         osd_seq->oos_subdir_count = map->om_subdir_count;
844         rc = osd_seq_load_locked(info, osd, osd_seq);
845         if (rc != 0)
846                 GOTO(cleanup, rc);
847
848         write_lock(&map->om_seq_list_lock);
849         list_add(&osd_seq->oos_seq_list, &map->om_seq_list);
850         write_unlock(&map->om_seq_list_lock);
851
852 cleanup:
853         mutex_unlock(&map->om_dir_init_mutex);
854         if (rc != 0) {
855                 if (osd_seq != NULL)
856                         OBD_FREE_PTR(osd_seq);
857                 RETURN(ERR_PTR(rc));
858         }
859
860         RETURN(osd_seq);
861 }
862
863 int osd_obj_map_lookup(struct osd_thread_info *info, struct osd_device *dev,
864                        const struct lu_fid *fid, struct osd_inode_id *id)
865 {
866         struct osd_obj_map              *map;
867         struct osd_obj_seq              *osd_seq;
868         struct dentry                   *d_seq;
869         struct dentry                   *child;
870         struct ost_id                   *ostid = &info->oti_ostid;
871         int                             dirn;
872         char                            name[32];
873         struct ldiskfs_dir_entry_2      *de;
874         struct buffer_head              *bh;
875         struct inode                    *dir;
876         struct inode                    *inode;
877         ENTRY;
878
879         /* on the very first lookup we find and open directories */
880
881         map = dev->od_ost_map;
882         LASSERT(map);
883         LASSERT(map->om_root);
884
885         fid_to_ostid(fid, ostid);
886         osd_seq = osd_seq_load(info, dev, ostid_seq(ostid));
887         if (IS_ERR(osd_seq))
888                 RETURN(PTR_ERR(osd_seq));
889
890         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
891         d_seq = osd_seq->oos_dirs[dirn];
892         LASSERT(d_seq);
893
894         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
895
896         child = &info->oti_child_dentry;
897         child->d_parent = d_seq;
898         child->d_name.hash = 0;
899         child->d_name.name = name;
900         /* XXX: we can use rc from sprintf() instead of strlen() */
901         child->d_name.len = strlen(name);
902
903         dir = d_seq->d_inode;
904         mutex_lock(&dir->i_mutex);
905         bh = osd_ldiskfs_find_entry(dir, &child->d_name, &de, NULL, NULL);
906         mutex_unlock(&dir->i_mutex);
907
908         if (bh == NULL)
909                 RETURN(-ENOENT);
910
911         osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
912         brelse(bh);
913
914         inode = osd_iget(info, dev, id);
915         if (IS_ERR(inode)) {
916                 int rc = PTR_ERR(inode);
917
918                 RETURN(rc == -ENOENT ? -ESTALE : rc);
919         }
920
921         iput(inode);
922         RETURN(0);
923 }
924
925 int osd_obj_map_insert(struct osd_thread_info *info,
926                        struct osd_device *osd,
927                        const struct lu_fid *fid,
928                        const struct osd_inode_id *id,
929                        handle_t *th)
930 {
931         struct osd_obj_map      *map;
932         struct osd_obj_seq      *osd_seq;
933         struct dentry           *d;
934         struct ost_id           *ostid = &info->oti_ostid;
935         u64                      oid;
936         int                     dirn, rc = 0;
937         char                    name[32];
938         ENTRY;
939
940         map = osd->od_ost_map;
941         LASSERT(map);
942
943         /* map fid to seq:objid */
944         fid_to_ostid(fid, ostid);
945
946         oid = ostid_id(ostid);
947         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
948         if (IS_ERR(osd_seq))
949                 RETURN(PTR_ERR(osd_seq));
950
951         dirn = oid & (osd_seq->oos_subdir_count - 1);
952         d = osd_seq->oos_dirs[dirn];
953         LASSERT(d);
954
955         osd_oid_name(name, sizeof(name), fid, oid);
956
957 again:
958         rc = osd_obj_add_entry(info, osd, d, name, id, th);
959         if (rc == -EEXIST) {
960                 rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
961                 if (unlikely(rc == -ENOENT))
962                         goto again;
963
964                 if (unlikely(rc == 1))
965                         RETURN(0);
966         }
967
968         RETURN(rc);
969 }
970
971 int osd_obj_map_delete(struct osd_thread_info *info, struct osd_device *osd,
972                        const struct lu_fid *fid, handle_t *th)
973 {
974         struct osd_obj_map      *map;
975         struct osd_obj_seq      *osd_seq;
976         struct dentry           *d;
977         struct ost_id           *ostid = &info->oti_ostid;
978         int                     dirn, rc = 0;
979         char                    name[32];
980         ENTRY;
981
982         map = osd->od_ost_map;
983         LASSERT(map);
984
985         /* map fid to seq:objid */
986         fid_to_ostid(fid, ostid);
987
988         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
989         if (IS_ERR(osd_seq))
990                 GOTO(cleanup, rc = PTR_ERR(osd_seq));
991
992         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
993         d = osd_seq->oos_dirs[dirn];
994         LASSERT(d);
995
996         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
997         rc = osd_obj_del_entry(info, osd, d, name, th);
998 cleanup:
999         RETURN(rc);
1000 }
1001
1002 int osd_obj_map_update(struct osd_thread_info *info,
1003                        struct osd_device *osd,
1004                        const struct lu_fid *fid,
1005                        const struct osd_inode_id *id,
1006                        handle_t *th)
1007 {
1008         struct osd_obj_seq      *osd_seq;
1009         struct dentry           *d;
1010         struct ost_id           *ostid = &info->oti_ostid;
1011         int                     dirn, rc = 0;
1012         char                    name[32];
1013         ENTRY;
1014
1015         fid_to_ostid(fid, ostid);
1016         osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1017         if (IS_ERR(osd_seq))
1018                 RETURN(PTR_ERR(osd_seq));
1019
1020         dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1021         d = osd_seq->oos_dirs[dirn];
1022         LASSERT(d);
1023
1024         osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1025         rc = osd_obj_update_entry(info, osd, d, name, fid, id, th);
1026
1027         RETURN(rc);
1028 }
1029
1030 int osd_obj_map_recover(struct osd_thread_info *info,
1031                         struct osd_device *osd,
1032                         struct inode *src_parent,
1033                         struct dentry *src_child,
1034                         const struct lu_fid *fid)
1035 {
1036         struct osd_obj_seq         *osd_seq;
1037         struct dentry              *tgt_parent;
1038         struct dentry              *tgt_child = &info->oti_child_dentry;
1039         struct inode               *dir;
1040         struct inode               *inode     = src_child->d_inode;
1041         struct ost_id              *ostid     = &info->oti_ostid;
1042         handle_t                   *jh;
1043         struct ldiskfs_dir_entry_2 *de;
1044         struct buffer_head         *bh;
1045         char                        name[32];
1046         int                         dirn;
1047         int                         rc        = 0;
1048         ENTRY;
1049
1050         if (fid_is_last_id(fid)) {
1051                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1052                 if (IS_ERR(osd_seq))
1053                         RETURN(PTR_ERR(osd_seq));
1054
1055                 tgt_parent = osd_seq->oos_root;
1056                 tgt_child->d_name.name = "LAST_ID";
1057                 tgt_child->d_name.len = strlen("LAST_ID");
1058         } else {
1059                 fid_to_ostid(fid, ostid);
1060                 osd_seq = osd_seq_load(info, osd, ostid_seq(ostid));
1061                 if (IS_ERR(osd_seq))
1062                         RETURN(PTR_ERR(osd_seq));
1063
1064                 dirn = ostid_id(ostid) & (osd_seq->oos_subdir_count - 1);
1065                 tgt_parent = osd_seq->oos_dirs[dirn];
1066                 osd_oid_name(name, sizeof(name), fid, ostid_id(ostid));
1067                 tgt_child->d_name.name = name;
1068                 tgt_child->d_name.len = strlen(name);
1069         }
1070         LASSERT(tgt_parent != NULL);
1071
1072         dir = tgt_parent->d_inode;
1073         tgt_child->d_name.hash = 0;
1074         tgt_child->d_parent = tgt_parent;
1075         tgt_child->d_inode = inode;
1076
1077         /* The non-initialized src_child may be destroyed. */
1078         jh = osd_journal_start_sb(osd_sb(osd), LDISKFS_HT_MISC,
1079                                 osd_dto_credits_noquota[DTO_INDEX_DELETE] +
1080                                 osd_dto_credits_noquota[DTO_INDEX_INSERT] +
1081                                 osd_dto_credits_noquota[DTO_OBJECT_DELETE]);
1082         if (IS_ERR(jh))
1083                 RETURN(PTR_ERR(jh));
1084
1085         ll_vfs_dq_init(src_parent);
1086         ll_vfs_dq_init(dir);
1087
1088         mutex_lock(&src_parent->i_mutex);
1089         mutex_lock(&dir->i_mutex);
1090         bh = osd_ldiskfs_find_entry(dir, &tgt_child->d_name, &de, NULL, NULL);
1091         if (bh != NULL) {
1092                 /* XXX: If some other object occupied the same slot. And If such
1093                  *      inode is zero-sized and with SUID+SGID, then means it is
1094                  *      a new created one. Maybe we can remove it and insert the
1095                  *      original one back to the /O/<seq>/d<x>. But there are
1096                  *      something to be considered:
1097                  *
1098                  *      1) The OST-object under /lost+found has crashed LMA.
1099                  *         So it should not conflict with the current one.
1100                  *
1101                  *      2) There are race conditions that: someone may just want
1102                  *         to modify the current one. Even if the OI scrub takes
1103                  *         the object lock when remove the current one, it still
1104                  *         cause the modification to be lost becasue the target
1105                  *         has been removed when the RPC service thread waiting
1106                  *         for the lock.
1107                  *
1108                  *      So keep it there before we have suitable solution. */
1109                 brelse(bh);
1110                 mutex_unlock(&dir->i_mutex);
1111                 mutex_unlock(&src_parent->i_mutex);
1112                 ldiskfs_journal_stop(jh);
1113
1114                 rc = -EEXIST;
1115                 /* If the src object has never been modified, then remove it. */
1116                 if (inode->i_size == 0 && inode->i_mode & S_ISUID &&
1117                     inode->i_mode & S_ISGID) {
1118                         rc = ll_vfs_unlink(src_parent, src_child);
1119                         if (unlikely(rc == -ENOENT))
1120                                 rc = 0;
1121                 }
1122                 RETURN(rc);
1123         }
1124
1125         bh = osd_ldiskfs_find_entry(src_parent, &src_child->d_name, &de,
1126                                     NULL, NULL);
1127         if (unlikely(bh == NULL))
1128                 GOTO(unlock, rc = -ENOENT);
1129
1130         rc = ldiskfs_delete_entry(jh, src_parent, de, bh);
1131         brelse(bh);
1132         if (rc != 0)
1133                 GOTO(unlock, rc);
1134
1135         rc = osd_ldiskfs_add_entry(jh, tgt_child, inode, NULL);
1136
1137         GOTO(unlock, rc);
1138
1139 unlock:
1140         mutex_unlock(&dir->i_mutex);
1141         mutex_unlock(&src_parent->i_mutex);
1142         ldiskfs_journal_stop(jh);
1143         return rc;
1144 }
1145
1146 static struct dentry *
1147 osd_object_spec_find(struct osd_thread_info *info, struct osd_device *osd,
1148                      const struct lu_fid *fid, char **name)
1149 {
1150         struct dentry *root = ERR_PTR(-ENOENT);
1151
1152         if (fid_is_last_id(fid)) {
1153                 struct osd_obj_seq *osd_seq;
1154
1155                 /* on creation of LAST_ID we create O/<seq> hierarchy */
1156                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1157                 if (IS_ERR(osd_seq))
1158                         RETURN((struct dentry *)osd_seq);
1159
1160                 *name = "LAST_ID";
1161                 root = osd_seq->oos_root;
1162         } else {
1163                 *name = osd_lf_fid2name(fid);
1164                 if (*name == NULL)
1165                         CWARN("UNKNOWN COMPAT FID "DFID"\n", PFID(fid));
1166                 else if ((*name)[0])
1167                         root = osd_sb(osd)->s_root;
1168         }
1169
1170         return root;
1171 }
1172
1173 int osd_obj_spec_update(struct osd_thread_info *info, struct osd_device *osd,
1174                         const struct lu_fid *fid, const struct osd_inode_id *id,
1175                         handle_t *th)
1176 {
1177         struct dentry   *root;
1178         char            *name = NULL;
1179         int              rc;
1180         ENTRY;
1181
1182         root = osd_object_spec_find(info, osd, fid, &name);
1183         if (!IS_ERR(root)) {
1184                 rc = osd_obj_update_entry(info, osd, root, name, fid, id, th);
1185         } else {
1186                 rc = PTR_ERR(root);
1187                 if (rc == -ENOENT)
1188                         rc = 1;
1189         }
1190
1191         RETURN(rc);
1192 }
1193
1194 int osd_obj_spec_insert(struct osd_thread_info *info, struct osd_device *osd,
1195                         const struct lu_fid *fid, const struct osd_inode_id *id,
1196                         handle_t *th)
1197 {
1198         struct dentry   *root;
1199         char            *name = NULL;
1200         int              rc;
1201         ENTRY;
1202
1203         root = osd_object_spec_find(info, osd, fid, &name);
1204         if (!IS_ERR(root)) {
1205                 rc = osd_obj_add_entry(info, osd, root, name, id, th);
1206         } else {
1207                 rc = PTR_ERR(root);
1208                 if (rc == -ENOENT)
1209                         rc = 0;
1210         }
1211
1212         RETURN(rc);
1213 }
1214
1215 int osd_obj_spec_lookup(struct osd_thread_info *info, struct osd_device *osd,
1216                         const struct lu_fid *fid, struct osd_inode_id *id)
1217 {
1218         struct dentry   *root;
1219         struct dentry   *dentry;
1220         struct inode    *inode;
1221         char            *name = NULL;
1222         int             rc = -ENOENT;
1223         ENTRY;
1224
1225         if (fid_is_last_id(fid)) {
1226                 struct osd_obj_seq *osd_seq;
1227
1228                 osd_seq = osd_seq_load(info, osd, fid_seq(fid));
1229                 if (IS_ERR(osd_seq))
1230                         RETURN(PTR_ERR(osd_seq));
1231                 root = osd_seq->oos_root;
1232                 name = "LAST_ID";
1233         } else {
1234                 root = osd_sb(osd)->s_root;
1235                 name = osd_lf_fid2name(fid);
1236                 if (name == NULL || strlen(name) == 0)
1237                         RETURN(-ENOENT);
1238         }
1239
1240         dentry = ll_lookup_one_len(name, root, strlen(name));
1241         if (!IS_ERR(dentry)) {
1242                 inode = dentry->d_inode;
1243                 if (inode) {
1244                         if (is_bad_inode(inode)) {
1245                                 rc = -EIO;
1246                         } else {
1247                                 osd_id_gen(id, inode->i_ino,
1248                                            inode->i_generation);
1249                                 rc = 0;
1250                         }
1251                 }
1252                 /* if dentry is accessible after osd_compat_spec_insert it
1253                  * will still contain NULL inode, so don't keep it in cache */
1254                 d_invalidate(dentry);
1255                 dput(dentry);
1256         }
1257
1258         RETURN(rc);
1259 }