Whamcloud - gitweb
LU-7608 kernel: kernel upgrade [SLES12 SP1 3.12.51-60.25]
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_scrub.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,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  * GNU General Public License version 2 for more details.  A copy is
14  * included in the COPYING file that accompanied this code.
15
16  * You should have received a copy of the GNU General Public License
17  * along with this program; if not, write to the Free Software
18  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2012, 2015, Intel Corporation.
24  */
25 /*
26  * lustre/osd-ldiskfs/osd_scrub.c
27  *
28  * Top-level entry points into osd module
29  *
30  * The OI scrub is used for rebuilding Object Index files when restores MDT from
31  * file-level backup.
32  *
33  * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
34  *
35  * Author: Fan Yong <yong.fan@whamcloud.com>
36  */
37
38 #define DEBUG_SUBSYSTEM S_LFSCK
39
40 #include <linux/kthread.h>
41 #include <lustre/lustre_idl.h>
42 #include <lustre_disk.h>
43 #include <dt_object.h>
44 #include <linux/xattr.h>
45
46 #include "osd_internal.h"
47 #include "osd_oi.h"
48 #include "osd_scrub.h"
49
50 #define HALF_SEC        msecs_to_jiffies(MSEC_PER_SEC >> 1)
51
52 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
53
54 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
55 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
56 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
57 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
58 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
59 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
60 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
61 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
62 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
63 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
64                                     * flags in LMA */
65
66 /* misc functions */
67
68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
69 {
70         return container_of0(scrub, struct osd_device, od_scrub);
71 }
72
73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
74 {
75         return osd_sb(osd_scrub2dev(scrub));
76 }
77
78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
79                                        struct osd_otable_cache *ooc)
80 {
81         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
82 }
83
84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
85 {
86         return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
87 }
88
89 /**
90  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
91  *
92  * \retval   1, changed nothing
93  * \retval   0, changed successfully
94  * \retval -ve, on error
95  */
96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
97                                      struct osd_device *dev,
98                                      const struct lu_fid *fid,
99                                      const struct osd_inode_id *id,
100                                      int ops, bool force,
101                                      enum oi_check_flags flags)
102 {
103         handle_t *th;
104         int       rc;
105         ENTRY;
106
107         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
108                 RETURN(0);
109
110         /* DTO_INDEX_INSERT is enough for other two ops:
111          * delete/update, but save stack. */
112         th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
113                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
114         if (IS_ERR(th)) {
115                 rc = PTR_ERR(th);
116                 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
117                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
118                        PFID(fid), id->oii_ino, id->oii_gen, rc);
119                 RETURN(rc);
120         }
121
122         switch (ops) {
123         case DTO_INDEX_UPDATE:
124                 rc = osd_oi_update(info, dev, fid, id, th, flags);
125                 if (unlikely(rc == -ENOENT)) {
126                         /* Some unlink thread may removed the OI mapping. */
127                         rc = 1;
128                 }
129                 break;
130         case DTO_INDEX_INSERT:
131                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
132                 if (unlikely(rc == -EEXIST)) {
133                         rc = 1;
134                         /* XXX: There are trouble things when adding OI
135                          *      mapping for IGIF object, which may cause
136                          *      multiple objects to be mapped to the same
137                          *      IGIF formatted FID. Consider the following
138                          *      situations:
139                          *
140                          *      1) The MDT is upgrading from 1.8 device.
141                          *      The OI scrub generates IGIF FID1 for the
142                          *      OBJ1 and adds the OI mapping.
143                          *
144                          *      2) For some reason, the OI scrub does not
145                          *      process all the IGIF objects completely.
146                          *
147                          *      3) The MDT is backuped and restored against
148                          *      this device.
149                          *
150                          *      4) When the MDT mounts up, the OI scrub will
151                          *      try to rebuild the OI files. For some IGIF
152                          *      object, OBJ2, which was not processed by the
153                          *      OI scrub before the backup/restore, and the
154                          *      new generated IGIF formatted FID may be just
155                          *      the FID1, the same as OBJ1.
156                          *
157                          *      Under such case, the OI scrub cannot know how
158                          *      to generate new FID for the OBJ2.
159                          *
160                          *      Currently, we do nothing for that. One possible
161                          *      solution is to generate new normal FID for the
162                          *      conflict object.
163                          *
164                          *      Anyway, it is rare, only exists in theory. */
165                 }
166                 break;
167         case DTO_INDEX_DELETE:
168                 rc = osd_oi_delete(info, dev, fid, th, flags);
169                 if (rc == -ENOENT) {
170                         /* It is normal that the unlink thread has removed the
171                          * OI mapping already. */
172                         rc = 1;
173                 }
174                 break;
175         default:
176                 LASSERTF(0, "Unexpected ops %d\n", ops);
177                 break;
178         }
179
180         ldiskfs_journal_stop(th);
181         if (rc < 0)
182                 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
183                        DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
184                        PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
185                        rc);
186
187         RETURN(rc);
188 }
189
190 /* OI_scrub file ops */
191
192 static void osd_scrub_file_to_cpu(struct scrub_file *des,
193                                   struct scrub_file *src)
194 {
195         memcpy(des->sf_uuid, src->sf_uuid, 16);
196         des->sf_flags   = le64_to_cpu(src->sf_flags);
197         des->sf_magic   = le32_to_cpu(src->sf_magic);
198         des->sf_status  = le16_to_cpu(src->sf_status);
199         des->sf_param   = le16_to_cpu(src->sf_param);
200         des->sf_time_last_complete      =
201                                 le64_to_cpu(src->sf_time_last_complete);
202         des->sf_time_latest_start       =
203                                 le64_to_cpu(src->sf_time_latest_start);
204         des->sf_time_last_checkpoint    =
205                                 le64_to_cpu(src->sf_time_last_checkpoint);
206         des->sf_pos_latest_start        =
207                                 le64_to_cpu(src->sf_pos_latest_start);
208         des->sf_pos_last_checkpoint     =
209                                 le64_to_cpu(src->sf_pos_last_checkpoint);
210         des->sf_pos_first_inconsistent  =
211                                 le64_to_cpu(src->sf_pos_first_inconsistent);
212         des->sf_items_checked           =
213                                 le64_to_cpu(src->sf_items_checked);
214         des->sf_items_updated           =
215                                 le64_to_cpu(src->sf_items_updated);
216         des->sf_items_failed            =
217                                 le64_to_cpu(src->sf_items_failed);
218         des->sf_items_updated_prior     =
219                                 le64_to_cpu(src->sf_items_updated_prior);
220         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
221         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
222         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
223         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
224         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
225 }
226
227 static void osd_scrub_file_to_le(struct scrub_file *des,
228                                  struct scrub_file *src)
229 {
230         memcpy(des->sf_uuid, src->sf_uuid, 16);
231         des->sf_flags   = cpu_to_le64(src->sf_flags);
232         des->sf_magic   = cpu_to_le32(src->sf_magic);
233         des->sf_status  = cpu_to_le16(src->sf_status);
234         des->sf_param   = cpu_to_le16(src->sf_param);
235         des->sf_time_last_complete      =
236                                 cpu_to_le64(src->sf_time_last_complete);
237         des->sf_time_latest_start       =
238                                 cpu_to_le64(src->sf_time_latest_start);
239         des->sf_time_last_checkpoint    =
240                                 cpu_to_le64(src->sf_time_last_checkpoint);
241         des->sf_pos_latest_start        =
242                                 cpu_to_le64(src->sf_pos_latest_start);
243         des->sf_pos_last_checkpoint     =
244                                 cpu_to_le64(src->sf_pos_last_checkpoint);
245         des->sf_pos_first_inconsistent  =
246                                 cpu_to_le64(src->sf_pos_first_inconsistent);
247         des->sf_items_checked           =
248                                 cpu_to_le64(src->sf_items_checked);
249         des->sf_items_updated           =
250                                 cpu_to_le64(src->sf_items_updated);
251         des->sf_items_failed            =
252                                 cpu_to_le64(src->sf_items_failed);
253         des->sf_items_updated_prior     =
254                                 cpu_to_le64(src->sf_items_updated_prior);
255         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
256         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
257         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
258         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
259         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
260 }
261
262 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
263 {
264         struct scrub_file *sf = &scrub->os_file;
265
266         memset(sf, 0, sizeof(*sf));
267         memcpy(sf->sf_uuid, uuid, 16);
268         sf->sf_magic = SCRUB_MAGIC_V1;
269         sf->sf_status = SS_INIT;
270 }
271
272 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
273 {
274         struct scrub_file *sf = &scrub->os_file;
275
276         CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
277                LPX64", add flags = "LPX64"\n",
278                osd_scrub2name(scrub), sf->sf_flags, flags);
279
280         memcpy(sf->sf_uuid, uuid, 16);
281         sf->sf_status = SS_INIT;
282         sf->sf_flags |= flags;
283         sf->sf_flags &= ~SF_AUTO;
284         sf->sf_run_time = 0;
285         sf->sf_time_latest_start = 0;
286         sf->sf_time_last_checkpoint = 0;
287         sf->sf_pos_latest_start = 0;
288         sf->sf_pos_last_checkpoint = 0;
289         sf->sf_pos_first_inconsistent = 0;
290         sf->sf_items_checked = 0;
291         sf->sf_items_updated = 0;
292         sf->sf_items_failed = 0;
293         if (!scrub->os_in_join)
294                 sf->sf_items_updated_prior = 0;
295
296         sf->sf_items_noscrub = 0;
297         sf->sf_items_igif = 0;
298 }
299
300 static int osd_scrub_file_load(struct osd_scrub *scrub)
301 {
302         loff_t  pos  = 0;
303         int     len  = sizeof(scrub->os_file_disk);
304         int     rc;
305
306         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
307         if (rc == len) {
308                 struct scrub_file *sf = &scrub->os_file;
309
310                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
311                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
312                         CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
313                                "0x%x != 0x%x\n", osd_scrub2name(scrub),
314                                sf->sf_magic, SCRUB_MAGIC_V1);
315                         /* Process it as new scrub file. */
316                         rc = -ENOENT;
317                 } else {
318                         rc = 0;
319                 }
320         } else if (rc != 0) {
321                 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
322                        "expected = %d: rc = %d\n",
323                        osd_scrub2name(scrub), len, rc);
324                 if (rc > 0)
325                         rc = -EFAULT;
326         } else {
327                 /* return -ENOENT for empty scrub file case. */
328                 rc = -ENOENT;
329         }
330
331         return rc;
332 }
333
334 int osd_scrub_file_store(struct osd_scrub *scrub)
335 {
336         struct osd_device *dev;
337         handle_t          *jh;
338         loff_t             pos     = 0;
339         int                len     = sizeof(scrub->os_file_disk);
340         int                credits;
341         int                rc;
342
343         dev = container_of0(scrub, struct osd_device, od_scrub);
344         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
345                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
346         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
347         if (IS_ERR(jh)) {
348                 rc = PTR_ERR(jh);
349                 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
350                        "rc = %d\n", osd_scrub2name(scrub), rc);
351                 return rc;
352         }
353
354         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
355         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
356                                       len, 0, &pos, jh);
357         ldiskfs_journal_stop(jh);
358         if (rc != 0)
359                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
360                        "expected = %d: rc = %d\n",
361                        osd_scrub2name(scrub), len, rc);
362
363         scrub->os_time_last_checkpoint = cfs_time_current();
364         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
365                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
366         return rc;
367 }
368
369 static int
370 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
371                      struct inode *inode, const struct lu_fid *fid)
372 {
373         struct filter_fid_old   *ff      = &info->oti_ff;
374         struct dentry           *dentry  = &info->oti_obj_dentry;
375         struct lu_fid           *tfid    = &info->oti_fid;
376         handle_t                *jh;
377         int                      size    = 0;
378         int                      rc;
379         bool                     removed = false;
380         bool                     reset   = true;
381         ENTRY;
382
383         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
384                 RETURN(0);
385
386         if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
387                 struct ost_id *oi = &info->oti_ostid;
388
389                 fid_to_ostid(fid, oi);
390                 ostid_to_fid(tfid, oi, 0);
391         } else {
392                 *tfid = *fid;
393         }
394
395         /* We want the LMA to fit into the 256-byte OST inode, so operate
396          * as following:
397          * 1) read old XATTR_NAME_FID and save the parent FID;
398          * 2) delete the old XATTR_NAME_FID;
399          * 3) make new LMA and add it;
400          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
401          *
402          * Making the LMA to fit into the 256-byte OST inode can save time for
403          * normal osd_check_lma() and for other OI scrub scanning in future.
404          * So it is worth to make some slow conversion here. */
405         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
406                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
407         if (IS_ERR(jh)) {
408                 rc = PTR_ERR(jh);
409                 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
410                        DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
411                 RETURN(rc);
412         }
413
414         /* 1) read old XATTR_NAME_FID and save the parent FID */
415         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
416         if (rc == sizeof(*ff)) {
417                 /* 2) delete the old XATTR_NAME_FID */
418                 ll_vfs_dq_init(inode);
419                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
420                 if (rc != 0)
421                         GOTO(stop, rc);
422
423                 removed = true;
424         } else if (unlikely(rc == -ENODATA)) {
425                 reset = false;
426         } else if (rc != sizeof(struct filter_fid)) {
427                 GOTO(stop, rc = -EINVAL);
428         }
429
430         /* 3) make new LMA and add it */
431         rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
432         if (rc == 0 && reset)
433                 size = sizeof(struct filter_fid);
434         else if (rc != 0 && removed)
435                 /* If failed, we should try to add the old back. */
436                 size = sizeof(struct filter_fid_old);
437
438         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
439         if (size > 0) {
440                 int rc1;
441
442                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
443                                       XATTR_CREATE);
444                 if (rc1 != 0 && rc == 0)
445                         rc = rc1;
446         }
447
448         GOTO(stop, rc);
449
450 stop:
451         ldiskfs_journal_stop(jh);
452         if (rc < 0)
453                 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
454                        osd_name(dev), PFID(tfid), rc);
455         return rc;
456 }
457
458 static int
459 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
460                        struct osd_idmap_cache *oic, int val)
461 {
462         struct osd_scrub             *scrub  = &dev->od_scrub;
463         struct scrub_file            *sf     = &scrub->os_file;
464         struct lu_fid                *fid    = &oic->oic_fid;
465         struct osd_inode_id          *lid    = &oic->oic_lid;
466         struct osd_inode_id          *lid2   = &info->oti_id;
467         struct osd_inconsistent_item *oii    = NULL;
468         struct inode                 *inode  = NULL;
469         int                           ops    = DTO_INDEX_UPDATE;
470         int                           idx;
471         int                           rc;
472         bool                          converted = false;
473         ENTRY;
474
475         down_write(&scrub->os_rwsem);
476         scrub->os_new_checked++;
477         if (val < 0)
478                 GOTO(out, rc = val);
479
480         if (scrub->os_in_prior)
481                 oii = list_entry(oic, struct osd_inconsistent_item,
482                                  oii_cache);
483
484         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
485                 GOTO(out, rc = 0);
486
487         if (fid_is_igif(fid))
488                 sf->sf_items_igif++;
489
490         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
491                 inode = osd_iget(info, dev, lid);
492                 if (IS_ERR(inode)) {
493                         rc = PTR_ERR(inode);
494                         /* Someone removed the inode. */
495                         if (rc == -ENOENT || rc == -ESTALE)
496                                 rc = 0;
497                         GOTO(out, rc);
498                 }
499
500                 /* The inode has been reused as EA inode, ignore it. */
501                 if (unlikely(osd_is_ea_inode(inode)))
502                         GOTO(out, rc = 0);
503
504                 sf->sf_flags |= SF_UPGRADE;
505                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
506                 dev->od_check_ff = 1;
507                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
508                 if (rc != 0)
509                         GOTO(out, rc);
510
511                 converted = true;
512         }
513
514         if ((val == SCRUB_NEXT_NOLMA) &&
515             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
516                 GOTO(out, rc = 0);
517
518         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
519                 ops = DTO_INDEX_INSERT;
520
521                 goto iget;
522         }
523
524         rc = osd_oi_lookup(info, dev, fid, lid2,
525                 (val == SCRUB_NEXT_OSTOBJ ||
526                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
527         if (rc != 0) {
528                 if (rc == -ENOENT)
529                         ops = DTO_INDEX_INSERT;
530                 else if (rc != -ESTALE)
531                         GOTO(out, rc);
532
533 iget:
534                 if (inode == NULL) {
535                         inode = osd_iget(info, dev, lid);
536                         if (IS_ERR(inode)) {
537                                 rc = PTR_ERR(inode);
538                                 /* Someone removed the inode. */
539                                 if (rc == -ENOENT || rc == -ESTALE)
540                                         rc = 0;
541                                 GOTO(out, rc);
542                         }
543
544                         /* The inode has been reused as EA inode, ignore it. */
545                         if (unlikely(osd_is_ea_inode(inode)))
546                                 GOTO(out, rc = 0);
547                 }
548
549                 if (!scrub->os_partial_scan)
550                         scrub->os_full_speed = 1;
551
552                 idx = osd_oi_fid2idx(dev, fid);
553                 switch (val) {
554                 case SCRUB_NEXT_NOLMA:
555                         sf->sf_flags |= SF_UPGRADE;
556                         if (!(sf->sf_param & SP_DRYRUN)) {
557                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
558                                 if (rc != 0)
559                                         GOTO(out, rc);
560                         }
561
562                         if (!(sf->sf_flags & SF_INCONSISTENT))
563                                 dev->od_igif_inoi = 0;
564                         break;
565                 case SCRUB_NEXT_OSTOBJ:
566                         sf->sf_flags |= SF_INCONSISTENT;
567                 case SCRUB_NEXT_OSTOBJ_OLD:
568                         break;
569                 default:
570                         sf->sf_flags |= SF_RECREATED;
571                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
572                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
573                         break;
574                 }
575         } else if (osd_id_eq(lid, lid2)) {
576                 if (converted)
577                         sf->sf_items_updated++;
578
579                 GOTO(out, rc = 0);
580         } else {
581                 if (!scrub->os_partial_scan)
582                         scrub->os_full_speed = 1;
583
584                 sf->sf_flags |= SF_INCONSISTENT;
585
586                 /* XXX: If the device is restored from file-level backup, then
587                  *      some IGIFs may have been already in OI files, and some
588                  *      may be not yet. Means upgrading from 1.8 may be partly
589                  *      processed, but some clients may hold some immobilized
590                  *      IGIFs, and use them to access related objects. Under
591                  *      such case, OSD does not know whether an given IGIF has
592                  *      been processed or to be processed, and it also cannot
593                  *      generate local ino#/gen# directly from the immobilized
594                  *      IGIF because of the backup/restore. Then force OSD to
595                  *      lookup the given IGIF in OI files, and if no entry,
596                  *      then ask the client to retry after upgrading completed.
597                  *      No better choice. */
598                 dev->od_igif_inoi = 1;
599         }
600
601         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
602                         (val == SCRUB_NEXT_OSTOBJ ||
603                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
604         if (rc == 0) {
605                 if (scrub->os_in_prior)
606                         sf->sf_items_updated_prior++;
607                 else
608                         sf->sf_items_updated++;
609         }
610
611         GOTO(out, rc);
612
613 out:
614         if (rc < 0) {
615                 sf->sf_items_failed++;
616                 if (sf->sf_pos_first_inconsistent == 0 ||
617                     sf->sf_pos_first_inconsistent > lid->oii_ino)
618                         sf->sf_pos_first_inconsistent = lid->oii_ino;
619         } else {
620                 rc = 0;
621         }
622
623         /* There may be conflict unlink during the OI scrub,
624          * if happend, then remove the new added OI mapping. */
625         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
626             unlikely(ldiskfs_test_inode_state(inode,
627                                               LDISKFS_STATE_LUSTRE_DESTROY)))
628                 osd_scrub_refresh_mapping(info, dev, fid, lid,
629                                 DTO_INDEX_DELETE, false,
630                                 (val == SCRUB_NEXT_OSTOBJ ||
631                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
632                                 OI_KNOWN_ON_OST : 0);
633         up_write(&scrub->os_rwsem);
634
635         if (inode != NULL && !IS_ERR(inode))
636                 iput(inode);
637
638         if (oii != NULL) {
639                 LASSERT(list_empty(&oii->oii_list));
640
641                 OBD_FREE_PTR(oii);
642         }
643
644         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
645 }
646
647 /* OI scrub APIs */
648
649 static int osd_scrub_prep(struct osd_device *dev)
650 {
651         struct osd_scrub     *scrub  = &dev->od_scrub;
652         struct ptlrpc_thread *thread = &scrub->os_thread;
653         struct scrub_file    *sf     = &scrub->os_file;
654         __u32                 flags  = scrub->os_start_flags;
655         int                   rc;
656         bool                  drop_dryrun = false;
657         ENTRY;
658
659         CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
660                osd_scrub2name(scrub), flags);
661
662         down_write(&scrub->os_rwsem);
663         if (flags & SS_SET_FAILOUT)
664                 sf->sf_param |= SP_FAILOUT;
665         else if (flags & SS_CLEAR_FAILOUT)
666                 sf->sf_param &= ~SP_FAILOUT;
667
668         if (flags & SS_SET_DRYRUN) {
669                 sf->sf_param |= SP_DRYRUN;
670         } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
671                 sf->sf_param &= ~SP_DRYRUN;
672                 drop_dryrun = true;
673         }
674
675         if (flags & SS_RESET)
676                 osd_scrub_file_reset(scrub,
677                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
678
679         if (flags & SS_AUTO_FULL) {
680                 scrub->os_full_speed = 1;
681                 scrub->os_partial_scan = 0;
682                 sf->sf_flags |= SF_AUTO;
683         } else if (flags & SS_AUTO_PARTIAL) {
684                 scrub->os_full_speed = 0;
685                 scrub->os_partial_scan = 1;
686                 sf->sf_flags |= SF_AUTO;
687         } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
688                                    SF_UPGRADE)) {
689                 scrub->os_full_speed = 1;
690                 scrub->os_partial_scan = 0;
691         } else {
692                 scrub->os_full_speed = 0;
693                 scrub->os_partial_scan = 0;
694         }
695
696         spin_lock(&scrub->os_lock);
697         scrub->os_in_prior = 0;
698         scrub->os_waiting = 0;
699         scrub->os_paused = 0;
700         scrub->os_in_join = 0;
701         scrub->os_full_scrub = 0;
702         spin_unlock(&scrub->os_lock);
703         scrub->os_new_checked = 0;
704         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
705                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
706         else if (sf->sf_pos_last_checkpoint != 0)
707                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
708         else
709                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
710
711         scrub->os_pos_current = sf->sf_pos_latest_start;
712         sf->sf_status = SS_SCANNING;
713         sf->sf_time_latest_start = cfs_time_current_sec();
714         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
715         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
716         rc = osd_scrub_file_store(scrub);
717         if (rc == 0) {
718                 spin_lock(&scrub->os_lock);
719                 thread_set_flags(thread, SVC_RUNNING);
720                 spin_unlock(&scrub->os_lock);
721                 wake_up_all(&thread->t_ctl_waitq);
722         }
723         up_write(&scrub->os_rwsem);
724
725         RETURN(rc);
726 }
727
728 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
729 {
730         struct scrub_file *sf = &scrub->os_file;
731         int                rc;
732
733         if (likely(cfs_time_before(cfs_time_current(),
734                                    scrub->os_time_next_checkpoint) ||
735                    scrub->os_new_checked == 0))
736                 return 0;
737
738         down_write(&scrub->os_rwsem);
739         sf->sf_items_checked += scrub->os_new_checked;
740         scrub->os_new_checked = 0;
741         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
742         sf->sf_time_last_checkpoint = cfs_time_current_sec();
743         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
744                                             scrub->os_time_last_checkpoint);
745         rc = osd_scrub_file_store(scrub);
746         up_write(&scrub->os_rwsem);
747
748         return rc;
749 }
750
751 static int osd_scrub_post(struct osd_scrub *scrub, int result)
752 {
753         struct scrub_file *sf = &scrub->os_file;
754         int rc;
755         ENTRY;
756
757         CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
758                osd_scrub2name(scrub), result);
759
760         down_write(&scrub->os_rwsem);
761         spin_lock(&scrub->os_lock);
762         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
763         spin_unlock(&scrub->os_lock);
764         if (scrub->os_new_checked > 0) {
765                 sf->sf_items_checked += scrub->os_new_checked;
766                 scrub->os_new_checked = 0;
767                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
768         }
769         sf->sf_time_last_checkpoint = cfs_time_current_sec();
770         if (result > 0) {
771                 struct osd_device *dev =
772                         container_of0(scrub, struct osd_device, od_scrub);
773
774                 dev->od_igif_inoi = 1;
775                 dev->od_check_ff = 0;
776                 sf->sf_status = SS_COMPLETED;
777                 if (!(sf->sf_param & SP_DRYRUN)) {
778                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
779                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
780                                           SF_UPGRADE | SF_AUTO);
781                 }
782                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
783                 sf->sf_success_count++;
784         } else if (result == 0) {
785                 if (scrub->os_paused)
786                         sf->sf_status = SS_PAUSED;
787                 else
788                         sf->sf_status = SS_STOPPED;
789         } else {
790                 sf->sf_status = SS_FAILED;
791         }
792         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
793                                             scrub->os_time_last_checkpoint);
794         rc = osd_scrub_file_store(scrub);
795         up_write(&scrub->os_rwsem);
796
797         RETURN(rc < 0 ? rc : result);
798 }
799
800 /* iteration engine */
801
802 struct osd_iit_param {
803         struct super_block *sb;
804         struct buffer_head *bitmap;
805         ldiskfs_group_t bg;
806         __u32 gbase;
807         __u32 offset;
808 };
809
810 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
811                                    struct osd_device *dev,
812                                    struct osd_iit_param *param,
813                                    struct osd_idmap_cache **oic,
814                                    const bool noslot);
815
816 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
817                                    struct osd_device *dev,
818                                    struct osd_iit_param *param,
819                                    struct osd_idmap_cache *oic,
820                                    bool *noslot, int rc);
821
822 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
823 {
824         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
825                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
826         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
827                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
828                 return SCRUB_NEXT_BREAK;
829         } else {
830                 *pos = param->gbase + param->offset;
831                 return 0;
832         }
833 }
834
835 /**
836  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
837  * \retval 0: FID-on-MDT
838  */
839 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
840                                       struct osd_device *dev,
841                                       struct lu_fid *fid)
842 {
843         /* XXX: The initial OI scrub will scan the top level /O to generate
844          *      a small local FLDB according to the <seq>. If the given FID
845          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
846          *      quite possible for FID-on-MDT. */
847         if (dev->od_is_ost)
848                 return SCRUB_NEXT_OSTOBJ_OLD;
849         else
850                 return 0;
851 }
852
853 static int osd_scrub_get_fid(struct osd_thread_info *info,
854                              struct osd_device *dev, struct inode *inode,
855                              struct lu_fid *fid, bool scrub)
856 {
857         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
858         int                      rc;
859         bool                     has_lma = false;
860
861         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
862         if (rc == 0) {
863                 has_lma = true;
864                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
865                     lma->lma_incompat & LMAI_AGENT)
866                         return SCRUB_NEXT_CONTINUE;
867
868                 *fid = lma->lma_self_fid;
869                 if (!scrub)
870                         return 0;
871
872                 if (lma->lma_compat & LMAC_FID_ON_OST)
873                         return SCRUB_NEXT_OSTOBJ;
874
875                 if (fid_is_idif(fid))
876                         return SCRUB_NEXT_OSTOBJ_OLD;
877
878                 /* For local object. */
879                 if (fid_is_internal(fid))
880                         return 0;
881
882                 /* For external visible MDT-object with non-normal FID. */
883                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
884                         return 0;
885
886                 /* For the object with normal FID, it may be MDT-object,
887                  * or may be 2.4 OST-object, need further distinguish.
888                  * Fall through to next section. */
889         }
890
891         if (rc == -ENODATA || rc == 0) {
892                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
893                 if (rc == 0) {
894                         if (scrub)
895                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
896                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
897                         return rc;
898                 }
899
900                 if (rc > 0) {
901                         if (!has_lma)
902                                 /* It is FID-on-OST, but we do not know how
903                                  * to generate its FID, ignore it directly. */
904                                 rc = SCRUB_NEXT_CONTINUE;
905                         else
906                                 /* It is 2.4 OST-object. */
907                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
908                         return rc;
909                 }
910
911                 if (rc != -ENODATA)
912                         return rc;
913
914                 if (!has_lma) {
915                         if (dev->od_scrub.os_convert_igif) {
916                                 lu_igif_build(fid, inode->i_ino,
917                                               inode->i_generation);
918                                 if (scrub)
919                                         rc = SCRUB_NEXT_NOLMA;
920                                 else
921                                         rc = 0;
922                         } else {
923                                 /* It may be FID-on-OST, or may be FID for
924                                  * non-MDT0, anyway, we do not know how to
925                                  * generate its FID, ignore it directly. */
926                                 rc = SCRUB_NEXT_CONTINUE;
927                         }
928                         return rc;
929                 }
930
931                 /* For OI scrub case only: the object has LMA but has no ff
932                  * (or ff crashed). It may be MDT-object, may be OST-object
933                  * with crashed ff. The last check is local FLDB. */
934                 rc = osd_scrub_check_local_fldb(info, dev, fid);
935         }
936
937         return rc;
938 }
939
940 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
941                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
942                         struct super_block *sb, bool scrub)
943 {
944         struct inode *inode;
945         int           rc;
946         ENTRY;
947
948         /* Not handle the backend root object and agent parent object.
949          * They are neither visible to namespace nor have OI mappings. */
950         if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
951                      pos == osd_remote_parent_ino(dev)))
952                 RETURN(SCRUB_NEXT_CONTINUE);
953
954         osd_id_gen(lid, pos, OSD_OII_NOGEN);
955         inode = osd_iget(info, dev, lid);
956         if (IS_ERR(inode)) {
957                 rc = PTR_ERR(inode);
958                 /* The inode may be removed after bitmap searching, or the
959                  * file is new created without inode initialized yet. */
960                 if (rc == -ENOENT || rc == -ESTALE)
961                         RETURN(SCRUB_NEXT_CONTINUE);
962
963                 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
964                        "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
965                        pos, rc);
966                 RETURN(rc);
967         }
968
969         /* It is an EA inode, no OI mapping for it, skip it. */
970         if (osd_is_ea_inode(inode))
971                 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
972
973         if (scrub &&
974             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
975                 /* Only skip it for the first OI scrub accessing. */
976                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
977                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
978         }
979
980         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
981
982         GOTO(put, rc);
983
984 put:
985         iput(inode);
986         return rc;
987 }
988
989 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
990                           struct osd_iit_param *param,
991                           struct osd_idmap_cache **oic, const bool noslot)
992 {
993         struct osd_scrub     *scrub  = &dev->od_scrub;
994         struct ptlrpc_thread *thread = &scrub->os_thread;
995         struct lu_fid        *fid;
996         struct osd_inode_id  *lid;
997         int                   rc;
998
999         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1000                 struct l_wait_info lwi;
1001
1002                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1003                 if (likely(lwi.lwi_timeout > 0))
1004                         l_wait_event(thread->t_ctl_waitq,
1005                                 !list_empty(&scrub->os_inconsistent_items) ||
1006                                 !thread_is_running(thread),
1007                                 &lwi);
1008         }
1009
1010         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1011                 spin_lock(&scrub->os_lock);
1012                 thread_set_flags(thread, SVC_STOPPING);
1013                 spin_unlock(&scrub->os_lock);
1014                 return SCRUB_NEXT_CRASH;
1015         }
1016
1017         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1018                 return SCRUB_NEXT_FATAL;
1019
1020         if (unlikely(!thread_is_running(thread)))
1021                 return SCRUB_NEXT_EXIT;
1022
1023         if (!list_empty(&scrub->os_inconsistent_items)) {
1024                 spin_lock(&scrub->os_lock);
1025                 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1026                         struct osd_inconsistent_item *oii;
1027
1028                         oii = list_entry(scrub->os_inconsistent_items.next,
1029                                 struct osd_inconsistent_item, oii_list);
1030                         list_del_init(&oii->oii_list);
1031                         spin_unlock(&scrub->os_lock);
1032
1033                         *oic = &oii->oii_cache;
1034                         scrub->os_in_prior = 1;
1035
1036                         return 0;
1037                 }
1038                 spin_unlock(&scrub->os_lock);
1039         }
1040
1041         if (noslot)
1042                 return SCRUB_NEXT_WAIT;
1043
1044         rc = osd_iit_next(param, &scrub->os_pos_current);
1045         if (rc != 0)
1046                 return rc;
1047
1048         *oic = &scrub->os_oic;
1049         fid = &(*oic)->oic_fid;
1050         lid = &(*oic)->oic_lid;
1051         rc = osd_iit_iget(info, dev, fid, lid,
1052                           scrub->os_pos_current, param->sb, true);
1053         return rc;
1054 }
1055
1056 static int osd_preload_next(struct osd_thread_info *info,
1057                             struct osd_device *dev, struct osd_iit_param *param,
1058                             struct osd_idmap_cache **oic, const bool noslot)
1059 {
1060         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
1061         struct osd_scrub        *scrub;
1062         struct ptlrpc_thread    *thread;
1063         int                      rc;
1064
1065         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1066         if (rc != 0)
1067                 return rc;
1068
1069         scrub = &dev->od_scrub;
1070         thread = &scrub->os_thread;
1071         if (thread_is_running(thread) &&
1072             ooc->ooc_pos_preload >= scrub->os_pos_current)
1073                 return SCRUB_NEXT_EXIT;
1074
1075         rc = osd_iit_iget(info, dev,
1076                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1077                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1078                           ooc->ooc_pos_preload, param->sb, false);
1079         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1080          * ignore the failure, so it still need to skip the inode next time. */
1081         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1082         return rc;
1083 }
1084
1085 static inline int
1086 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1087 {
1088         spin_lock(&scrub->os_lock);
1089         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1090             !list_empty(&scrub->os_inconsistent_items) ||
1091             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1092                 scrub->os_waiting = 0;
1093         else
1094                 scrub->os_waiting = 1;
1095         spin_unlock(&scrub->os_lock);
1096
1097         return !scrub->os_waiting;
1098 }
1099
1100 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1101                           struct osd_iit_param *param,
1102                           struct osd_idmap_cache *oic, bool *noslot, int rc)
1103 {
1104         struct l_wait_info       lwi    = { 0 };
1105         struct osd_scrub        *scrub  = &dev->od_scrub;
1106         struct scrub_file       *sf     = &scrub->os_file;
1107         struct ptlrpc_thread    *thread = &scrub->os_thread;
1108         struct osd_otable_it    *it     = dev->od_otable_it;
1109         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1110
1111         switch (rc) {
1112         case SCRUB_NEXT_CONTINUE:
1113                 goto next;
1114         case SCRUB_NEXT_WAIT:
1115                 goto wait;
1116         case SCRUB_NEXT_NOSCRUB:
1117                 down_write(&scrub->os_rwsem);
1118                 scrub->os_new_checked++;
1119                 sf->sf_items_noscrub++;
1120                 up_write(&scrub->os_rwsem);
1121                 goto next;
1122         }
1123
1124         rc = osd_scrub_check_update(info, dev, oic, rc);
1125         if (rc != 0) {
1126                 scrub->os_in_prior = 0;
1127                 return rc;
1128         }
1129
1130         rc = osd_scrub_checkpoint(scrub);
1131         if (rc != 0) {
1132                 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1133                        "rc = %d\n", osd_scrub2name(scrub),
1134                        scrub->os_pos_current, rc);
1135                 /* Continue, as long as the scrub itself can go ahead. */
1136         }
1137
1138         if (scrub->os_in_prior) {
1139                 scrub->os_in_prior = 0;
1140                 return 0;
1141         }
1142
1143 next:
1144         scrub->os_pos_current = param->gbase + ++(param->offset);
1145
1146 wait:
1147         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1148             ooc->ooc_pos_preload < scrub->os_pos_current) {
1149                 spin_lock(&scrub->os_lock);
1150                 it->ooi_waiting = 0;
1151                 wake_up_all(&thread->t_ctl_waitq);
1152                 spin_unlock(&scrub->os_lock);
1153         }
1154
1155         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1156                 return 0;
1157
1158         if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1159                 *noslot = false;
1160                 return 0;
1161         }
1162
1163         if (it != NULL)
1164                 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1165                              &lwi);
1166
1167         if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1168                 *noslot = false;
1169         else
1170                 *noslot = true;
1171         return 0;
1172 }
1173
1174 static int osd_preload_exec(struct osd_thread_info *info,
1175                             struct osd_device *dev, struct osd_iit_param *param,
1176                             struct osd_idmap_cache *oic, bool *noslot, int rc)
1177 {
1178         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1179
1180         if (rc == 0) {
1181                 ooc->ooc_cached_items++;
1182                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1183                                         ~OSD_OTABLE_IT_CACHE_MASK;
1184         }
1185         return rc > 0 ? 0 : rc;
1186 }
1187
1188 #define SCRUB_IT_ALL    1
1189 #define SCRUB_IT_CRASH  2
1190
1191 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1192                            bool inconsistent)
1193 {
1194         struct osd_scrub     *scrub  = &dev->od_scrub;
1195         struct ptlrpc_thread *thread = &scrub->os_thread;
1196         struct scrub_file    *sf     = &scrub->os_file;
1197         int                   rc;
1198         ENTRY;
1199
1200         LASSERT(!(flags & SS_AUTO_PARTIAL));
1201
1202         down_write(&scrub->os_rwsem);
1203         scrub->os_in_join = 1;
1204         if (flags & SS_SET_FAILOUT)
1205                 sf->sf_param |= SP_FAILOUT;
1206         else if (flags & SS_CLEAR_FAILOUT)
1207                 sf->sf_param &= ~SP_FAILOUT;
1208
1209         if (flags & SS_SET_DRYRUN)
1210                 sf->sf_param |= SP_DRYRUN;
1211         else if (flags & SS_CLEAR_DRYRUN)
1212                 sf->sf_param &= ~SP_DRYRUN;
1213
1214         if (flags & SS_RESET) {
1215                 osd_scrub_file_reset(scrub,
1216                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1217                         inconsistent ? SF_INCONSISTENT : 0);
1218                 sf->sf_status = SS_SCANNING;
1219         }
1220
1221         if (flags & SS_AUTO_FULL) {
1222                 sf->sf_flags |= SF_AUTO;
1223                 scrub->os_full_speed = 1;
1224         }
1225
1226         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1227                 scrub->os_full_speed = 1;
1228         else
1229                 scrub->os_full_speed = 0;
1230
1231         scrub->os_new_checked = 0;
1232         if (sf->sf_pos_last_checkpoint != 0)
1233                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1234         else
1235                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1236
1237         scrub->os_pos_current = sf->sf_pos_latest_start;
1238         sf->sf_time_latest_start = cfs_time_current_sec();
1239         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1240         sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1241         rc = osd_scrub_file_store(scrub);
1242         if (rc != 0)
1243                 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1244                        "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1245
1246         spin_lock(&scrub->os_lock);
1247         scrub->os_waiting = 0;
1248         scrub->os_paused = 0;
1249         scrub->os_partial_scan = 0;
1250         scrub->os_in_join = 0;
1251         scrub->os_full_scrub = 0;
1252         spin_unlock(&scrub->os_lock);
1253         wake_up_all(&thread->t_ctl_waitq);
1254         up_write(&scrub->os_rwsem);
1255
1256         EXIT;
1257 }
1258
1259 static int osd_inode_iteration(struct osd_thread_info *info,
1260                                struct osd_device *dev, __u32 max, bool preload)
1261 {
1262         struct osd_scrub     *scrub  = &dev->od_scrub;
1263         struct ptlrpc_thread *thread = &scrub->os_thread;
1264         struct scrub_file    *sf     = &scrub->os_file;
1265         osd_iit_next_policy   next;
1266         osd_iit_exec_policy   exec;
1267         __u32                *pos;
1268         __u32                *count;
1269         struct osd_iit_param  param  = { NULL };
1270         struct l_wait_info    lwi    = { 0 };
1271         __u32                 limit;
1272         int                   rc;
1273         bool                  noslot = true;
1274         ENTRY;
1275
1276         param.sb = osd_sb(dev);
1277         if (preload)
1278                 goto full;
1279
1280         while (scrub->os_partial_scan && !scrub->os_in_join) {
1281                 struct osd_idmap_cache *oic = NULL;
1282
1283                 rc = osd_scrub_next(info, dev, &param, &oic, noslot);
1284                 switch (rc) {
1285                 case SCRUB_NEXT_EXIT:
1286                         RETURN(0);
1287                 case SCRUB_NEXT_CRASH:
1288                         RETURN(SCRUB_IT_CRASH);
1289                 case SCRUB_NEXT_FATAL:
1290                         RETURN(-EINVAL);
1291                 case SCRUB_NEXT_WAIT: {
1292                         struct kstatfs *ksfs = &info->oti_ksfs;
1293                         __u64 saved_flags;
1294
1295                         if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1296                             unlikely(sf->sf_items_updated_prior == 0))
1297                                 goto wait;
1298
1299                         if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1300                             scrub->os_full_scrub) {
1301                                 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1302                                                true);
1303                                 goto full;
1304                         }
1305
1306                         rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1307                         if (rc == 0) {
1308                                 __u64 used = ksfs->f_files - ksfs->f_ffree;
1309
1310                                 do_div(used, sf->sf_items_updated_prior);
1311                                 /* If we hit too much inconsistent OI
1312                                  * mappings during the partial scan,
1313                                  * then scan the device completely. */
1314                                 if (used < dev->od_full_scrub_ratio) {
1315                                         osd_scrub_join(dev,
1316                                                 SS_AUTO_FULL | SS_RESET, true);
1317                                         goto full;
1318                                 }
1319                         }
1320
1321 wait:
1322                         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1323                             cfs_fail_val > 0)
1324                                 continue;
1325
1326                         saved_flags = sf->sf_flags;
1327                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1328                                           SF_UPGRADE | SF_AUTO);
1329                         sf->sf_status = SS_COMPLETED;
1330                         l_wait_event(thread->t_ctl_waitq,
1331                                      !thread_is_running(thread) ||
1332                                      !scrub->os_partial_scan ||
1333                                      scrub->os_in_join ||
1334                                      !list_empty(&scrub->os_inconsistent_items),
1335                                      &lwi);
1336                         sf->sf_flags = saved_flags;
1337                         sf->sf_status = SS_SCANNING;
1338
1339                         if (unlikely(!thread_is_running(thread)))
1340                                 RETURN(0);
1341
1342                         if (!scrub->os_partial_scan || scrub->os_in_join)
1343                                 goto full;
1344
1345                         continue;
1346                 }
1347                 default:
1348                         LASSERTF(rc == 0, "rc = %d\n", rc);
1349
1350                         osd_scrub_exec(info, dev, &param, oic, &noslot, rc);
1351                         break;
1352                 }
1353         }
1354
1355 full:
1356         if (!preload) {
1357                 l_wait_event(thread->t_ctl_waitq,
1358                              !thread_is_running(thread) || !scrub->os_in_join,
1359                              &lwi);
1360
1361                 if (unlikely(!thread_is_running(thread)))
1362                         RETURN(0);
1363         }
1364
1365         noslot = false;
1366         if (!preload) {
1367                 next = osd_scrub_next;
1368                 exec = osd_scrub_exec;
1369                 pos = &scrub->os_pos_current;
1370                 count = &scrub->os_new_checked;
1371         } else {
1372                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1373
1374                 next = osd_preload_next;
1375                 exec = osd_preload_exec;
1376                 pos = &ooc->ooc_pos_preload;
1377                 count = &ooc->ooc_cached_items;
1378         }
1379         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1380
1381         while (*pos <= limit && *count < max) {
1382                 struct osd_idmap_cache *oic = NULL;
1383                 struct ldiskfs_group_desc *desc;
1384
1385                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1386                 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1387                 if (desc == NULL)
1388                         RETURN(-EIO);
1389
1390                 ldiskfs_lock_group(param.sb, param.bg);
1391                 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1392                         ldiskfs_unlock_group(param.sb, param.bg);
1393                         *pos = 1 + (param.bg + 1) *
1394                                 LDISKFS_INODES_PER_GROUP(param.sb);
1395                         continue;
1396                 }
1397                 ldiskfs_unlock_group(param.sb, param.bg);
1398
1399                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1400                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1401                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1402                 if (param.bitmap == NULL) {
1403                         CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1404                                "scrub will stop, urgent mode\n",
1405                                osd_scrub2name(scrub), (__u32)param.bg);
1406                         RETURN(-EIO);
1407                 }
1408
1409                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1410                        *count < max) {
1411                         if (param.offset +
1412                                 ldiskfs_itable_unused_count(param.sb, desc) >
1413                             LDISKFS_INODES_PER_GROUP(param.sb))
1414                                 goto next_group;
1415
1416                         rc = next(info, dev, &param, &oic, noslot);
1417                         switch (rc) {
1418                         case SCRUB_NEXT_BREAK:
1419                                 goto next_group;
1420                         case SCRUB_NEXT_EXIT:
1421                                 brelse(param.bitmap);
1422                                 RETURN(0);
1423                         case SCRUB_NEXT_CRASH:
1424                                 brelse(param.bitmap);
1425                                 RETURN(SCRUB_IT_CRASH);
1426                         case SCRUB_NEXT_FATAL:
1427                                 brelse(param.bitmap);
1428                                 RETURN(-EINVAL);
1429                         }
1430
1431                         rc = exec(info, dev, &param, oic, &noslot, rc);
1432                         if (rc != 0) {
1433                                 brelse(param.bitmap);
1434                                 RETURN(rc);
1435                         }
1436                 }
1437
1438 next_group:
1439                 brelse(param.bitmap);
1440         }
1441
1442         if (*pos > limit)
1443                 RETURN(SCRUB_IT_ALL);
1444         RETURN(0);
1445 }
1446
1447 static int osd_otable_it_preload(const struct lu_env *env,
1448                                  struct osd_otable_it *it)
1449 {
1450         struct osd_device       *dev   = it->ooi_dev;
1451         struct osd_scrub        *scrub = &dev->od_scrub;
1452         struct osd_otable_cache *ooc   = &it->ooi_cache;
1453         int                      rc;
1454         ENTRY;
1455
1456         rc = osd_inode_iteration(osd_oti_get(env), dev,
1457                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1458         if (rc == SCRUB_IT_ALL)
1459                 it->ooi_all_cached = 1;
1460
1461         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1462                 spin_lock(&scrub->os_lock);
1463                 scrub->os_waiting = 0;
1464                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1465                 spin_unlock(&scrub->os_lock);
1466         }
1467
1468         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1469 }
1470
1471 static int osd_scrub_main(void *args)
1472 {
1473         struct lu_env         env;
1474         struct osd_device    *dev    = (struct osd_device *)args;
1475         struct osd_scrub     *scrub  = &dev->od_scrub;
1476         struct ptlrpc_thread *thread = &scrub->os_thread;
1477         int                   rc;
1478         ENTRY;
1479
1480         rc = lu_env_init(&env, LCT_LOCAL);
1481         if (rc != 0) {
1482                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1483                        osd_scrub2name(scrub), rc);
1484                 GOTO(noenv, rc);
1485         }
1486
1487         rc = osd_scrub_prep(dev);
1488         if (rc != 0) {
1489                 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1490                        osd_scrub2name(scrub), rc);
1491                 GOTO(out, rc);
1492         }
1493
1494         if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1495                 struct l_wait_info lwi = { 0 };
1496                 struct osd_otable_it *it = dev->od_otable_it;
1497                 struct osd_otable_cache *ooc = &it->ooi_cache;
1498
1499                 l_wait_event(thread->t_ctl_waitq,
1500                              it->ooi_user_ready || !thread_is_running(thread),
1501                              &lwi);
1502                 if (unlikely(!thread_is_running(thread)))
1503                         GOTO(post, rc = 0);
1504
1505                 scrub->os_pos_current = ooc->ooc_pos_preload;
1506         }
1507
1508         CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1509                osd_scrub2name(scrub), scrub->os_start_flags,
1510                scrub->os_pos_current);
1511
1512         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1513         if (unlikely(rc == SCRUB_IT_CRASH))
1514                 GOTO(out, rc = -EINVAL);
1515         GOTO(post, rc);
1516
1517 post:
1518         rc = osd_scrub_post(scrub, rc);
1519         CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1520                osd_scrub2name(scrub), scrub->os_pos_current, rc);
1521
1522 out:
1523         while (!list_empty(&scrub->os_inconsistent_items)) {
1524                 struct osd_inconsistent_item *oii;
1525
1526                 oii = list_entry(scrub->os_inconsistent_items.next,
1527                                      struct osd_inconsistent_item, oii_list);
1528                 list_del_init(&oii->oii_list);
1529                 OBD_FREE_PTR(oii);
1530         }
1531         lu_env_fini(&env);
1532
1533 noenv:
1534         spin_lock(&scrub->os_lock);
1535         thread_set_flags(thread, SVC_STOPPED);
1536         wake_up_all(&thread->t_ctl_waitq);
1537         spin_unlock(&scrub->os_lock);
1538         return rc;
1539 }
1540
1541 /* initial OI scrub */
1542
1543 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1544                          struct dentry *, filldir_t filldir);
1545
1546 #ifdef HAVE_FILLDIR_USE_CTX
1547 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1548                                int namelen, loff_t offset, __u64 ino,
1549                                unsigned d_type);
1550 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1551                            int namelen, loff_t offset, __u64 ino,
1552                            unsigned d_type);
1553 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1554                            int namelen, loff_t offset, __u64 ino,
1555                            unsigned d_type);
1556 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1557                             int namelen, loff_t offset, __u64 ino,
1558                             unsigned d_type);
1559 #else
1560 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1561                                loff_t offset, __u64 ino, unsigned d_type);
1562 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1563                            loff_t offset, __u64 ino, unsigned d_type);
1564 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1565                            loff_t offset, __u64 ino, unsigned d_type);
1566 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1567                             loff_t offset, __u64 ino, unsigned d_type);
1568 #endif
1569
1570 static int
1571 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1572                      struct dentry *dentry, filldir_t filldir);
1573 static int
1574 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1575                   struct dentry *dentry, filldir_t filldir);
1576
1577 static int
1578 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1579                      struct dentry *dentry, filldir_t filldir);
1580
1581 enum osd_lf_flags {
1582         OLF_SCAN_SUBITEMS       = 0x0001,
1583         OLF_HIDE_FID            = 0x0002,
1584         OLF_SHOW_NAME           = 0x0004,
1585         OLF_NO_OI               = 0x0008,
1586         OLF_IDX_IN_FID          = 0x0010,
1587 };
1588
1589 struct osd_lf_map {
1590         char            *olm_name;
1591         struct lu_fid    olm_fid;
1592         __u16            olm_flags;
1593         __u16            olm_namelen;
1594         scandir_t        olm_scandir;
1595         filldir_t        olm_filldir;
1596 };
1597
1598 /* Add the new introduced local files in the list in the future. */
1599 static const struct osd_lf_map osd_lf_maps[] = {
1600         /* CATALOGS */
1601         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1602                 sizeof(CATLIST) - 1, NULL, NULL },
1603
1604         /* CONFIGS */
1605         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1606                 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1607                 osd_ios_general_scan, osd_ios_varfid_fill },
1608
1609         /* NIDTBL_VERSIONS */
1610         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1611                 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1612                 osd_ios_varfid_fill },
1613
1614         /* PENDING */
1615         { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1616
1617         /* ROOT */
1618         { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1619                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1620                 osd_ios_ROOT_scan, NULL },
1621
1622         /* changelog_catalog */
1623         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1624                 NULL, NULL },
1625
1626         /* changelog_users */
1627         { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1628                 NULL, NULL },
1629
1630         /* fld */
1631         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1632                 sizeof("fld") - 1, NULL, NULL },
1633
1634         /* last_rcvd */
1635         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1636                 sizeof(LAST_RCVD) - 1, NULL, NULL },
1637
1638         /* reply_data */
1639         { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1640                 sizeof(REPLY_DATA) - 1, NULL, NULL },
1641
1642         /* lov_objid */
1643         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1644                 sizeof(LOV_OBJID) - 1, NULL, NULL },
1645
1646         /* lov_objseq */
1647         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1648                 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1649
1650         /* quota_master */
1651         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1652                 osd_ios_general_scan, osd_ios_varfid_fill },
1653
1654         /* quota_slave */
1655         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1656                 osd_ios_general_scan, osd_ios_varfid_fill },
1657
1658         /* seq_ctl */
1659         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1660                 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1661
1662         /* seq_srv */
1663         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1664                 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1665
1666         /* health_check */
1667         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1668                 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1669
1670         /* LFSCK */
1671         { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1672                 osd_ios_general_scan, osd_ios_varfid_fill },
1673
1674         /* lfsck_bookmark */
1675         { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1676                 NULL, NULL },
1677
1678         /* lfsck_layout */
1679         { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1680                 NULL, NULL },
1681
1682         /* lfsck_namespace */
1683         { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1684                 NULL, NULL },
1685
1686         /* OBJECTS, upgrade from old device */
1687         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1688                 osd_ios_OBJECTS_scan, NULL },
1689
1690         /* lquota_v2.user, upgrade from old device */
1691         { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1692                 NULL, NULL },
1693
1694         /* lquota_v2.group, upgrade from old device */
1695         { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1696                 NULL, NULL },
1697
1698         /* LAST_GROUP, upgrade from old device */
1699         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1700                 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1701
1702         /* committed batchid for cross-MDT operation */
1703         { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1704                 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1705
1706         /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1707          * and f_oid = index for their log files.  See lu_update_log{_dir}_fid()
1708          * for more details. */
1709
1710         /* update_log */
1711         { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1712                 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1713                 NULL, NULL },
1714
1715         /* update_log_dir */
1716         { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1717                 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1718                 sizeof("update_log_dir") - 1,
1719                 osd_ios_general_scan, osd_ios_uld_fill },
1720
1721         /* lost+found */
1722         { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1723                 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1724                 osd_ios_general_scan, osd_ios_lf_fill },
1725
1726         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1727 };
1728
1729 /* Add the new introduced files under .lustre/ in the list in the future. */
1730 static const struct osd_lf_map osd_dl_maps[] = {
1731         /* .lustre/fid */
1732         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1733                 sizeof("fid") - 1, NULL, NULL },
1734
1735         /* .lustre/lost+found */
1736         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1737                 sizeof("lost+found") - 1, NULL, NULL },
1738
1739         { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1740 };
1741
1742 struct osd_ios_item {
1743         struct list_head oii_list;
1744         struct dentry   *oii_dentry;
1745         scandir_t        oii_scandir;
1746         filldir_t        oii_filldir;
1747 };
1748
1749 struct osd_ios_filldir_buf {
1750 #ifdef HAVE_DIR_CONTEXT
1751         /* please keep it as first member */
1752         struct dir_context       ctx;
1753 #endif
1754         struct osd_thread_info  *oifb_info;
1755         struct osd_device       *oifb_dev;
1756         struct dentry           *oifb_dentry;
1757 };
1758
1759 static inline struct dentry *
1760 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1761 {
1762         struct dentry *dentry;
1763
1764         dentry = ll_lookup_one_len(name, parent, namelen);
1765         if (IS_ERR(dentry)) {
1766                 int rc = PTR_ERR(dentry);
1767
1768                 if (rc != -ENOENT)
1769                         CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1770                                namelen, name, parent->d_name.len,
1771                                parent->d_name.name, parent->d_inode->i_ino,
1772                                parent->d_inode->i_generation, rc);
1773
1774                 return dentry;
1775         }
1776
1777         if (dentry->d_inode == NULL) {
1778                 dput(dentry);
1779                 return ERR_PTR(-ENOENT);
1780         }
1781
1782         return dentry;
1783 }
1784
1785 static int
1786 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1787                  scandir_t scandir, filldir_t filldir)
1788 {
1789         struct osd_ios_item *item;
1790         ENTRY;
1791
1792         OBD_ALLOC_PTR(item);
1793         if (item == NULL)
1794                 RETURN(-ENOMEM);
1795
1796         INIT_LIST_HEAD(&item->oii_list);
1797         item->oii_dentry = dget(dentry);
1798         item->oii_scandir = scandir;
1799         item->oii_filldir = filldir;
1800         list_add_tail(&item->oii_list, &dev->od_ios_list);
1801
1802         RETURN(0);
1803 }
1804
1805 /**
1806  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1807  *
1808  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1809  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1810  * reference the inode, or fixed if it is missing or references another inode.
1811  */
1812 static int
1813 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1814                  struct inode *inode, const struct lu_fid *fid, int flags)
1815 {
1816         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1817         struct osd_inode_id     *id     = &info->oti_id;
1818         struct osd_inode_id     *id2    = &info->oti_id2;
1819         struct osd_scrub        *scrub  = &dev->od_scrub;
1820         struct scrub_file       *sf     = &scrub->os_file;
1821         struct lu_fid            tfid;
1822         int                      rc;
1823         ENTRY;
1824
1825         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1826         if (rc != 0 && rc != -ENODATA) {
1827                 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1828                        "rc = %d\n", osd_name(dev), rc);
1829
1830                 RETURN(rc);
1831         }
1832
1833         osd_id_gen(id, inode->i_ino, inode->i_generation);
1834         if (rc == -ENODATA) {
1835                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1836                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1837                 } else {
1838                         tfid = *fid;
1839                         if (flags & OLF_IDX_IN_FID) {
1840                                 LASSERT(dev->od_index >= 0);
1841
1842                                 tfid.f_oid = dev->od_index;
1843                         }
1844                 }
1845                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1846                 if (rc != 0) {
1847                         CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1848                               "scrub: rc = %d\n", osd_name(dev), rc);
1849
1850                         RETURN(rc);
1851                 }
1852         } else {
1853                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1854                         RETURN(0);
1855
1856                 tfid = lma->lma_self_fid;
1857         }
1858
1859         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1860         if (rc != 0) {
1861                 if (rc != -ENOENT)
1862                         RETURN(rc);
1863
1864                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1865                                                DTO_INDEX_INSERT, true, 0);
1866                 if (rc > 0)
1867                         rc = 0;
1868
1869                 RETURN(rc);
1870         }
1871
1872         if (osd_id_eq_strict(id, id2))
1873                 RETURN(0);
1874
1875         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1876                 osd_scrub_file_reset(scrub,
1877                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1878                                      SF_INCONSISTENT);
1879                 rc = osd_scrub_file_store(scrub);
1880                 if (rc != 0)
1881                         RETURN(rc);
1882         }
1883
1884         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1885                                        DTO_INDEX_UPDATE, true, 0);
1886         if (rc > 0)
1887                 rc = 0;
1888
1889         RETURN(rc);
1890 }
1891
1892 /**
1893  * It scans the /lost+found, and for the OST-object (with filter_fid
1894  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1895  */
1896 #ifdef HAVE_FILLDIR_USE_CTX
1897 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1898                            int namelen,
1899 #else
1900 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1901 #endif
1902                            loff_t offset, __u64 ino, unsigned d_type)
1903 {
1904         struct osd_ios_filldir_buf *fill_buf =
1905                 (struct osd_ios_filldir_buf *)buf;
1906         struct osd_thread_info     *info     = fill_buf->oifb_info;
1907         struct osd_device          *dev      = fill_buf->oifb_dev;
1908         struct lu_fid              *fid      = &info->oti_fid;
1909         struct osd_scrub           *scrub    = &dev->od_scrub;
1910         struct dentry              *parent   = fill_buf->oifb_dentry;
1911         struct dentry              *child;
1912         struct inode               *dir      = parent->d_inode;
1913         struct inode               *inode;
1914         int                         rc;
1915         ENTRY;
1916
1917         /* skip any '.' started names */
1918         if (name[0] == '.')
1919                 RETURN(0);
1920
1921         scrub->os_lf_scanned++;
1922         child = osd_ios_lookup_one_len(name, parent, namelen);
1923         if (IS_ERR(child)) {
1924                 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1925                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1926                 RETURN(0);
1927         }
1928
1929         inode = child->d_inode;
1930         if (S_ISDIR(inode->i_mode)) {
1931                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1932                                       osd_ios_lf_fill);
1933                 if (rc != 0)
1934                         CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1935                               "rc = %d\n", osd_name(dev), namelen, name, rc);
1936                 GOTO(put, rc);
1937         }
1938
1939         if (!S_ISREG(inode->i_mode))
1940                 GOTO(put, rc = 0);
1941
1942         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1943         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1944                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1945                 if (rc == 0) {
1946                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1947                                "/lost+found.\n", namelen, name, PFID(fid));
1948                         scrub->os_lf_repaired++;
1949                 } else {
1950                         CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1951                                DFID": rc = %d\n",
1952                                osd_name(dev), namelen, name, PFID(fid), rc);
1953                 }
1954         }
1955
1956         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1957          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1958          *      can process them in furtuer. */
1959
1960         GOTO(put, rc);
1961
1962 put:
1963         if (rc < 0)
1964                 scrub->os_lf_failed++;
1965         dput(child);
1966         /* skip the failure to make the scanning to continue. */
1967         return 0;
1968 }
1969
1970 #ifdef HAVE_FILLDIR_USE_CTX
1971 static int osd_ios_varfid_fill(struct dir_context  *buf, const char *name,
1972                                int namelen,
1973 #else
1974 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1975 #endif
1976                                loff_t offset, __u64 ino, unsigned d_type)
1977 {
1978         struct osd_ios_filldir_buf *fill_buf =
1979                 (struct osd_ios_filldir_buf *)buf;
1980         struct osd_device          *dev      = fill_buf->oifb_dev;
1981         struct dentry              *child;
1982         int                         rc;
1983         ENTRY;
1984
1985         /* skip any '.' started names */
1986         if (name[0] == '.')
1987                 RETURN(0);
1988
1989         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1990         if (IS_ERR(child))
1991                 RETURN(PTR_ERR(child));
1992
1993         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1994                               NULL, 0);
1995         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1996                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1997                                       osd_ios_varfid_fill);
1998         dput(child);
1999
2000         RETURN(rc);
2001 }
2002
2003 #ifdef HAVE_FILLDIR_USE_CTX
2004 static int osd_ios_dl_fill(struct dir_context  *buf, const char *name,
2005                            int namelen,
2006 #else
2007 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
2008 #endif
2009                            loff_t offset, __u64 ino, unsigned d_type)
2010 {
2011         struct osd_ios_filldir_buf *fill_buf =
2012                 (struct osd_ios_filldir_buf *)buf;
2013         struct osd_device          *dev      = fill_buf->oifb_dev;
2014         const struct osd_lf_map    *map;
2015         struct dentry              *child;
2016         int                         rc       = 0;
2017         ENTRY;
2018
2019         /* skip any '.' started names */
2020         if (name[0] == '.')
2021                 RETURN(0);
2022
2023         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2024                 if (map->olm_namelen != namelen)
2025                         continue;
2026
2027                 if (strncmp(map->olm_name, name, namelen) == 0)
2028                         break;
2029         }
2030
2031         if (map->olm_name == NULL)
2032                 RETURN(0);
2033
2034         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2035         if (IS_ERR(child))
2036                 RETURN(PTR_ERR(child));
2037
2038         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2039                               &map->olm_fid, map->olm_flags);
2040         dput(child);
2041
2042         RETURN(rc);
2043 }
2044
2045 #ifdef HAVE_FILLDIR_USE_CTX
2046 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
2047                             int namelen,
2048 #else
2049 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
2050 #endif
2051                             loff_t offset, __u64 ino, unsigned d_type)
2052 {
2053         struct osd_ios_filldir_buf *fill_buf =
2054                 (struct osd_ios_filldir_buf *)buf;
2055         struct dentry              *child;
2056         struct lu_fid               tfid;
2057         int                         rc       = 0;
2058         ENTRY;
2059
2060         /* skip any non-DFID format name */
2061         if (name[0] != '[')
2062                 RETURN(0);
2063
2064         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2065         if (IS_ERR(child))
2066                 RETURN(PTR_ERR(child));
2067
2068         /* skip the start '[' */
2069         sscanf(&name[1], SFID, RFID(&tfid));
2070         if (fid_is_sane(&tfid))
2071                 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2072                                       child->d_inode, &tfid, 0);
2073         else
2074                 rc = -EIO;
2075         dput(child);
2076
2077         RETURN(rc);
2078 }
2079
2080 #ifdef HAVE_FILLDIR_USE_CTX
2081 static int osd_ios_root_fill(struct dir_context *buf, const char *name,
2082                              int namelen,
2083 #else
2084 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2085 #endif
2086                              loff_t offset, __u64 ino, unsigned d_type)
2087 {
2088         struct osd_ios_filldir_buf *fill_buf =
2089                 (struct osd_ios_filldir_buf *)buf;
2090         struct osd_device          *dev      = fill_buf->oifb_dev;
2091         const struct osd_lf_map    *map;
2092         struct dentry              *child;
2093         int                         rc       = 0;
2094         ENTRY;
2095
2096         /* skip any '.' started names */
2097         if (name[0] == '.')
2098                 RETURN(0);
2099
2100         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2101                 if (map->olm_namelen != namelen)
2102                         continue;
2103
2104                 if (strncmp(map->olm_name, name, namelen) == 0)
2105                         break;
2106         }
2107
2108         if (map->olm_name == NULL)
2109                 RETURN(0);
2110
2111         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2112         if (IS_ERR(child))
2113                 RETURN(PTR_ERR(child));
2114
2115         if (!(map->olm_flags & OLF_NO_OI))
2116                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2117                                       &map->olm_fid, map->olm_flags);
2118         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2119                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2120                                       map->olm_filldir);
2121         dput(child);
2122
2123         RETURN(rc);
2124 }
2125
2126 static int
2127 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2128                      struct dentry *dentry, filldir_t filldir)
2129 {
2130         struct osd_ios_filldir_buf    buf   = {
2131 #ifdef HAVE_DIR_CONTEXT
2132                                                 .ctx.actor = filldir,
2133 #endif
2134                                                 .oifb_info = info,
2135                                                 .oifb_dev = dev,
2136                                                 .oifb_dentry = dentry };
2137         struct file                  *filp  = &info->oti_file;
2138         struct inode                 *inode = dentry->d_inode;
2139         const struct file_operations *fops  = inode->i_fop;
2140         int                           rc;
2141         ENTRY;
2142
2143         LASSERT(filldir != NULL);
2144
2145         filp->f_pos = 0;
2146         filp->f_path.dentry = dentry;
2147         filp->f_mode = FMODE_64BITHASH;
2148         filp->f_mapping = inode->i_mapping;
2149         filp->f_op = fops;
2150         filp->private_data = NULL;
2151         set_file_inode(filp, inode);
2152
2153 #ifdef HAVE_DIR_CONTEXT
2154         buf.ctx.pos = filp->f_pos;
2155         rc = fops->iterate(filp, &buf.ctx);
2156         filp->f_pos = buf.ctx.pos;
2157 #else
2158         rc = fops->readdir(filp, &buf, filldir);
2159 #endif
2160         fops->release(inode, filp);
2161
2162         RETURN(rc);
2163 }
2164
2165 static int
2166 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2167                   struct dentry *dentry, filldir_t filldir)
2168 {
2169         struct osd_scrub  *scrub  = &dev->od_scrub;
2170         struct scrub_file *sf     = &scrub->os_file;
2171         struct dentry     *child;
2172         int                rc;
2173         ENTRY;
2174
2175         /* It is existing MDT0 device. We only allow the case of object without
2176          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2177          * can generate IGIF mode FID for the object and related OI mapping. If
2178          * it is on other MDTs, then becuase file-level backup/restore, related
2179          * OI mapping may be invalid already, we do not know which is the right
2180          * FID for the object. We only allow IGIF objects to reside on the MDT0.
2181          *
2182          * XXX: For the case of object on non-MDT0 device with neither LMA nor
2183          *      "fid" xattr, then something crashed. We cannot re-generate the
2184          *      FID directly, instead, the OI scrub will scan the OI structure
2185          *      and try to re-generate the LMA from the OI mapping. But if the
2186          *      OI mapping crashed or lost also, then we have to give up under
2187          *      double failure cases. */
2188         scrub->os_convert_igif = 1;
2189         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2190                                        strlen(dot_lustre_name));
2191         if (IS_ERR(child)) {
2192                 rc = PTR_ERR(child);
2193                 if (rc == -ENOENT) {
2194                         /* It is 1.8 MDT device. */
2195                         if (!(sf->sf_flags & SF_UPGRADE)) {
2196                                 osd_scrub_file_reset(scrub,
2197                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2198                                         SF_UPGRADE);
2199                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2200                                 rc = osd_scrub_file_store(scrub);
2201                         } else {
2202                                 rc = 0;
2203                         }
2204                 }
2205         } else {
2206                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2207                  * so the client will get IGIF for the ".lustre" object when
2208                  * the MDT restart.
2209                  *
2210                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2211                  * it does not know whether there are some old clients cached
2212                  * the ".lustre" IGIF during the upgrading. Two choices:
2213                  *
2214                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2215                  *    It will allow the old connected clients to access the
2216                  *    ".lustre" with cached IGIF. But it will cause others
2217                  *    on the MDT failed to check "fid_is_dot_lustre()".
2218                  *
2219                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2220                  *    for ".lustre" in spite of whether there are some clients
2221                  *    cached the ".lustre" IGIF or not. It enables the check
2222                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
2223                  *    that the old connected clients cannot access the ".lustre"
2224                  *    with the cached IGIF.
2225                  *
2226                  * Usually, it is rare case for the old connected clients
2227                  * to access the ".lustre" with cached IGIF. So we prefer
2228                  * to the solution 2). */
2229                 rc = osd_ios_scan_one(info, dev, child->d_inode,
2230                                       &LU_DOT_LUSTRE_FID, 0);
2231                 if (rc == 0)
2232                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2233                                               osd_ios_dl_fill);
2234                 dput(child);
2235         }
2236
2237         RETURN(rc);
2238 }
2239
2240 static int
2241 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2242                      struct dentry *dentry, filldir_t filldir)
2243 {
2244         struct osd_scrub  *scrub  = &dev->od_scrub;
2245         struct scrub_file *sf     = &scrub->os_file;
2246         struct dentry     *child;
2247         int                rc;
2248         ENTRY;
2249
2250         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2251                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2252                 rc = osd_scrub_file_store(scrub);
2253                 if (rc != 0)
2254                         RETURN(rc);
2255         }
2256
2257         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2258         if (!IS_ERR(child)) {
2259                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2260                 dput(child);
2261         } else {
2262                 rc = PTR_ERR(child);
2263         }
2264
2265         if (rc != 0 && rc != -ENOENT)
2266                 RETURN(rc);
2267
2268         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2269         if (!IS_ERR(child)) {
2270                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2271                 dput(child);
2272         } else {
2273                 rc = PTR_ERR(child);
2274         }
2275
2276         if (rc == -ENOENT)
2277                 rc = 0;
2278
2279         RETURN(rc);
2280 }
2281
2282 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2283                                 struct osd_device *dev)
2284 {
2285         struct osd_ios_item     *item    = NULL;
2286         scandir_t                scandir = osd_ios_general_scan;
2287         filldir_t                filldir = osd_ios_root_fill;
2288         struct dentry           *dentry  = osd_sb(dev)->s_root;
2289         const struct osd_lf_map *map     = osd_lf_maps;
2290         int                      rc;
2291         ENTRY;
2292
2293         /* Lookup IGIF in OI by force for initial OI scrub. */
2294         dev->od_igif_inoi = 1;
2295
2296         while (1) {
2297                 rc = scandir(info, dev, dentry, filldir);
2298                 if (item != NULL) {
2299                         dput(item->oii_dentry);
2300                         OBD_FREE_PTR(item);
2301                 }
2302
2303                 if (rc != 0)
2304                         break;
2305
2306                 if (list_empty(&dev->od_ios_list))
2307                         break;
2308
2309                 item = list_entry(dev->od_ios_list.next,
2310                                   struct osd_ios_item, oii_list);
2311                 list_del_init(&item->oii_list);
2312
2313                 LASSERT(item->oii_scandir != NULL);
2314                 scandir = item->oii_scandir;
2315                 filldir = item->oii_filldir;
2316                 dentry = item->oii_dentry;
2317         }
2318
2319         while (!list_empty(&dev->od_ios_list)) {
2320                 item = list_entry(dev->od_ios_list.next,
2321                                   struct osd_ios_item, oii_list);
2322                 list_del_init(&item->oii_list);
2323                 dput(item->oii_dentry);
2324                 OBD_FREE_PTR(item);
2325         }
2326
2327         if (rc != 0)
2328                 RETURN(rc);
2329
2330         /* There maybe the case that the object has been removed, but its OI
2331          * mapping is still in the OI file, such as the "CATALOGS" after MDT
2332          * file-level backup/restore. So here cleanup the stale OI mappings. */
2333         while (map->olm_name != NULL) {
2334                 struct dentry *child;
2335
2336                 if (fid_is_zero(&map->olm_fid)) {
2337                         map++;
2338                         continue;
2339                 }
2340
2341                 child = osd_ios_lookup_one_len(map->olm_name,
2342                                                osd_sb(dev)->s_root,
2343                                                map->olm_namelen);
2344                 if (!IS_ERR(child))
2345                         dput(child);
2346                 else if (PTR_ERR(child) == -ENOENT)
2347                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2348                                                   NULL, DTO_INDEX_DELETE,
2349                                                   true, 0);
2350                 map++;
2351         }
2352
2353         RETURN(0);
2354 }
2355
2356 char *osd_lf_fid2name(const struct lu_fid *fid)
2357 {
2358         const struct osd_lf_map *map = osd_lf_maps;
2359
2360         while (map->olm_name != NULL) {
2361                 if (!lu_fid_eq(fid, &map->olm_fid)) {
2362                         map++;
2363                         continue;
2364                 }
2365
2366                 if (map->olm_flags & OLF_SHOW_NAME)
2367                         return map->olm_name;
2368                 else
2369                         return "";
2370         }
2371
2372         return NULL;
2373 }
2374
2375 /* OI scrub start/stop */
2376
2377 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2378 {
2379         struct osd_scrub     *scrub  = &dev->od_scrub;
2380         struct ptlrpc_thread *thread = &scrub->os_thread;
2381         struct l_wait_info    lwi    = { 0 };
2382         struct task_struct   *task;
2383         int                   rc;
2384         ENTRY;
2385
2386         /* os_lock: sync status between stop and scrub thread */
2387         spin_lock(&scrub->os_lock);
2388
2389 again:
2390         if (thread_is_running(thread)) {
2391                 spin_unlock(&scrub->os_lock);
2392                 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2393                      (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2394                         RETURN(-EALREADY);
2395
2396                 osd_scrub_join(dev, flags, false);
2397                 spin_lock(&scrub->os_lock);
2398                 if (!thread_is_running(thread))
2399                         goto again;
2400
2401                 spin_unlock(&scrub->os_lock);
2402                 RETURN(0);
2403         }
2404
2405         if (unlikely(thread_is_stopping(thread))) {
2406                 spin_unlock(&scrub->os_lock);
2407                 l_wait_event(thread->t_ctl_waitq,
2408                              thread_is_stopped(thread),
2409                              &lwi);
2410                 spin_lock(&scrub->os_lock);
2411                 goto again;
2412         }
2413         spin_unlock(&scrub->os_lock);
2414
2415         if (scrub->os_file.sf_status == SS_COMPLETED) {
2416                 if (!(flags & SS_SET_FAILOUT))
2417                         flags |= SS_CLEAR_FAILOUT;
2418
2419                 if (!(flags & SS_SET_DRYRUN))
2420                         flags |= SS_CLEAR_DRYRUN;
2421
2422                 flags |= SS_RESET;
2423         }
2424
2425         scrub->os_start_flags = flags;
2426         thread_set_flags(thread, 0);
2427         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2428         if (IS_ERR(task)) {
2429                 rc = PTR_ERR(task);
2430                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2431                        osd_scrub2name(scrub), rc);
2432                 RETURN(rc);
2433         }
2434
2435         l_wait_event(thread->t_ctl_waitq,
2436                      thread_is_running(thread) || thread_is_stopped(thread),
2437                      &lwi);
2438
2439         RETURN(0);
2440 }
2441
2442 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2443 {
2444         int rc;
2445         ENTRY;
2446
2447         /* od_otable_mutex: prevent curcurrent start/stop */
2448         mutex_lock(&dev->od_otable_mutex);
2449         rc = do_osd_scrub_start(dev, flags);
2450         mutex_unlock(&dev->od_otable_mutex);
2451
2452         RETURN(rc == -EALREADY ? 0 : rc);
2453 }
2454
2455 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2456 {
2457         struct ptlrpc_thread *thread = &scrub->os_thread;
2458         struct l_wait_info    lwi    = { 0 };
2459
2460         /* os_lock: sync status between stop and scrub thread */
2461         spin_lock(&scrub->os_lock);
2462         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2463                 thread_set_flags(thread, SVC_STOPPING);
2464                 spin_unlock(&scrub->os_lock);
2465                 wake_up_all(&thread->t_ctl_waitq);
2466                 l_wait_event(thread->t_ctl_waitq,
2467                              thread_is_stopped(thread),
2468                              &lwi);
2469                 /* Do not skip the last lock/unlock, which can guarantee that
2470                  * the caller cannot return until the OI scrub thread exit. */
2471                 spin_lock(&scrub->os_lock);
2472         }
2473         spin_unlock(&scrub->os_lock);
2474 }
2475
2476 static void osd_scrub_stop(struct osd_device *dev)
2477 {
2478         /* od_otable_mutex: prevent curcurrent start/stop */
2479         mutex_lock(&dev->od_otable_mutex);
2480         dev->od_scrub.os_paused = 1;
2481         do_osd_scrub_stop(&dev->od_scrub);
2482         mutex_unlock(&dev->od_otable_mutex);
2483 }
2484
2485 /* OI scrub setup/cleanup */
2486
2487 static const char osd_scrub_name[] = "OI_scrub";
2488
2489 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2490 {
2491         struct osd_thread_info     *info   = osd_oti_get(env);
2492         struct osd_scrub           *scrub  = &dev->od_scrub;
2493         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2494         struct scrub_file          *sf     = &scrub->os_file;
2495         struct super_block         *sb     = osd_sb(dev);
2496         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2497         struct lvfs_run_ctxt        saved;
2498         struct file                *filp;
2499         struct inode               *inode;
2500         struct lu_fid              *fid    = &info->oti_fid;
2501         bool                        dirty  = false;
2502         bool                        restored = false;
2503         int                         rc     = 0;
2504         ENTRY;
2505
2506         memset(scrub, 0, sizeof(*scrub));
2507         OBD_SET_CTXT_MAGIC(ctxt);
2508         ctxt->pwdmnt = dev->od_mnt;
2509         ctxt->pwd = dev->od_mnt->mnt_root;
2510         ctxt->fs = get_ds();
2511
2512         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2513         init_rwsem(&scrub->os_rwsem);
2514         spin_lock_init(&scrub->os_lock);
2515         INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2516
2517         push_ctxt(&saved, ctxt);
2518         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2519         if (IS_ERR(filp)) {
2520                 pop_ctxt(&saved, ctxt);
2521                 RETURN(PTR_ERR(filp));
2522         }
2523
2524         inode = filp->f_path.dentry->d_inode;
2525         /* 'What the @fid is' is not imporatant, because the object
2526          * has no OI mapping, and only is visible inside the OSD.*/
2527         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2528         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2529         if (rc != 0) {
2530                 filp_close(filp, NULL);
2531                 pop_ctxt(&saved, ctxt);
2532                 RETURN(rc);
2533         }
2534
2535         scrub->os_inode = igrab(inode);
2536         filp_close(filp, NULL);
2537         pop_ctxt(&saved, ctxt);
2538
2539         rc = osd_scrub_file_load(scrub);
2540         if (rc == -ENOENT) {
2541                 osd_scrub_file_init(scrub, es->s_uuid);
2542                 /* If the "/O" dir does not exist when mount (indicated by
2543                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2544                  * then it is quite probably that the device is a new one,
2545                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2546                  *
2547                  * For the rare case that "/O" and "OI_scrub" both lost on
2548                  * an old device, it can be found and cleared later.
2549                  *
2550                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2551                  * need to check "filter_fid_old" and to convert it to
2552                  * "filter_fid" for each object, and all the IGIF should
2553                  * have their FID mapping in OI files already. */
2554                 if (dev->od_maybe_new)
2555                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2556                 dirty = true;
2557         } else if (rc != 0) {
2558                 GOTO(cleanup_inode, rc);
2559         } else {
2560                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2561                         struct obd_uuid *old_uuid;
2562                         struct obd_uuid *new_uuid;
2563
2564                         OBD_ALLOC_PTR(old_uuid);
2565                         OBD_ALLOC_PTR(new_uuid);
2566                         if (old_uuid == NULL || new_uuid == NULL) {
2567                                 CERROR("%.16s: UUID has been changed, but"
2568                                        "failed to allocate RAM for report\n",
2569                                        LDISKFS_SB(sb)->s_es->s_volume_name);
2570                         } else {
2571                                 class_uuid_unparse(sf->sf_uuid, old_uuid);
2572                                 class_uuid_unparse(es->s_uuid, new_uuid);
2573                                 CERROR("%.16s: UUID has been changed from "
2574                                        "%s to %s\n",
2575                                        LDISKFS_SB(sb)->s_es->s_volume_name,
2576                                        old_uuid->uuid, new_uuid->uuid);
2577                         }
2578                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2579                         dirty = true;
2580                         restored = true;
2581                         if (old_uuid != NULL)
2582                                 OBD_FREE_PTR(old_uuid);
2583                         if (new_uuid != NULL)
2584                                 OBD_FREE_PTR(new_uuid);
2585                 } else if (sf->sf_status == SS_SCANNING) {
2586                         sf->sf_status = SS_CRASHED;
2587                         dirty = true;
2588                 }
2589         }
2590
2591         if (sf->sf_pos_last_checkpoint != 0)
2592                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2593         else
2594                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2595
2596         if (dirty) {
2597                 rc = osd_scrub_file_store(scrub);
2598                 if (rc != 0)
2599                         GOTO(cleanup_inode, rc);
2600         }
2601
2602         /* Initialize OI files. */
2603         rc = osd_oi_init(info, dev, restored);
2604         if (rc < 0)
2605                 GOTO(cleanup_inode, rc);
2606
2607         rc = osd_initial_OI_scrub(info, dev);
2608         if (rc != 0)
2609                 GOTO(cleanup_oi, rc);
2610
2611         if (sf->sf_flags & SF_UPGRADE ||
2612             !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2613               sf->sf_success_count > 0)) {
2614                 dev->od_igif_inoi = 0;
2615                 dev->od_check_ff = dev->od_is_ost;
2616         } else {
2617                 dev->od_igif_inoi = 1;
2618                 dev->od_check_ff = 0;
2619         }
2620
2621         if (sf->sf_flags & SF_INCONSISTENT)
2622                 /* The 'od_igif_inoi' will be set under the
2623                  * following cases:
2624                  * 1) new created system, or
2625                  * 2) restored from file-level backup, or
2626                  * 3) the upgrading completed.
2627                  *
2628                  * The 'od_igif_inoi' may be cleared by OI scrub
2629                  * later if found that the system is upgrading. */
2630                 dev->od_igif_inoi = 1;
2631
2632         if (!dev->od_noscrub &&
2633             ((sf->sf_status == SS_PAUSED) ||
2634              (sf->sf_status == SS_CRASHED &&
2635               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2636                               SF_UPGRADE | SF_AUTO)) ||
2637              (sf->sf_status == SS_INIT &&
2638               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2639                               SF_UPGRADE))))
2640                 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2641
2642         if (rc != 0)
2643                 GOTO(cleanup_oi, rc);
2644
2645         /* it is possible that dcache entries may keep objects after they are
2646          * deleted by OSD. While it looks safe this can cause object data to
2647          * stay until umount causing failures in tests calculating free space,
2648          * e.g. replay-ost-single. Since those dcache entries are not used
2649          * anymore let's just free them after use here */
2650         shrink_dcache_sb(sb);
2651
2652         RETURN(0);
2653 cleanup_oi:
2654         osd_oi_fini(info, dev);
2655 cleanup_inode:
2656         iput(scrub->os_inode);
2657         scrub->os_inode = NULL;
2658
2659         return rc;
2660 }
2661
2662 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2663 {
2664         struct osd_scrub *scrub = &dev->od_scrub;
2665
2666         LASSERT(dev->od_otable_it == NULL);
2667
2668         if (scrub->os_inode != NULL) {
2669                 osd_scrub_stop(dev);
2670                 iput(scrub->os_inode);
2671                 scrub->os_inode = NULL;
2672         }
2673         if (dev->od_oi_table != NULL)
2674                 osd_oi_fini(osd_oti_get(env), dev);
2675 }
2676
2677 /* object table based iteration APIs */
2678
2679 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2680                                        struct dt_object *dt, __u32 attr)
2681 {
2682         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2683         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2684         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2685         struct osd_scrub       *scrub = &dev->od_scrub;
2686         struct osd_otable_it   *it;
2687         __u32                   start = 0;
2688         int                     rc;
2689         ENTRY;
2690
2691         /* od_otable_mutex: prevent curcurrent init/fini */
2692         mutex_lock(&dev->od_otable_mutex);
2693         if (dev->od_otable_it != NULL)
2694                 GOTO(out, it = ERR_PTR(-EALREADY));
2695
2696         OBD_ALLOC_PTR(it);
2697         if (it == NULL)
2698                 GOTO(out, it = ERR_PTR(-ENOMEM));
2699
2700         dev->od_otable_it = it;
2701         it->ooi_dev = dev;
2702         it->ooi_cache.ooc_consumer_idx = -1;
2703         if (flags & DOIF_OUTUSED)
2704                 it->ooi_used_outside = 1;
2705
2706         if (flags & DOIF_RESET)
2707                 start |= SS_RESET;
2708
2709         if (valid & DOIV_ERROR_HANDLE) {
2710                 if (flags & DOIF_FAILOUT)
2711                         start |= SS_SET_FAILOUT;
2712                 else
2713                         start |= SS_CLEAR_FAILOUT;
2714         }
2715
2716         if (valid & DOIV_DRYRUN) {
2717                 if (flags & DOIF_DRYRUN)
2718                         start |= SS_SET_DRYRUN;
2719                 else
2720                         start |= SS_CLEAR_DRYRUN;
2721         }
2722
2723         rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2724         if (rc < 0 && rc != -EALREADY) {
2725                 dev->od_otable_it = NULL;
2726                 OBD_FREE_PTR(it);
2727                 GOTO(out, it = ERR_PTR(rc));
2728         }
2729
2730         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2731
2732         GOTO(out, it);
2733
2734 out:
2735         mutex_unlock(&dev->od_otable_mutex);
2736         return (struct dt_it *)it;
2737 }
2738
2739 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2740 {
2741         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2742         struct osd_device    *dev = it->ooi_dev;
2743
2744         /* od_otable_mutex: prevent curcurrent init/fini */
2745         mutex_lock(&dev->od_otable_mutex);
2746         do_osd_scrub_stop(&dev->od_scrub);
2747         LASSERT(dev->od_otable_it == it);
2748
2749         dev->od_otable_it = NULL;
2750         mutex_unlock(&dev->od_otable_mutex);
2751         OBD_FREE_PTR(it);
2752 }
2753
2754 static int osd_otable_it_get(const struct lu_env *env,
2755                              struct dt_it *di, const struct dt_key *key)
2756 {
2757         return 0;
2758 }
2759
2760 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2761 {
2762 }
2763
2764 static inline int
2765 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2766 {
2767         spin_lock(&scrub->os_lock);
2768         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2769             scrub->os_waiting ||
2770             !thread_is_running(&scrub->os_thread))
2771                 it->ooi_waiting = 0;
2772         else
2773                 it->ooi_waiting = 1;
2774         spin_unlock(&scrub->os_lock);
2775
2776         return !it->ooi_waiting;
2777 }
2778
2779 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2780 {
2781         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2782         struct osd_device       *dev    = it->ooi_dev;
2783         struct osd_scrub        *scrub  = &dev->od_scrub;
2784         struct osd_otable_cache *ooc    = &it->ooi_cache;
2785         struct ptlrpc_thread    *thread = &scrub->os_thread;
2786         struct l_wait_info       lwi    = { 0 };
2787         int                      rc;
2788         ENTRY;
2789
2790         LASSERT(it->ooi_user_ready);
2791
2792 again:
2793         if (!thread_is_running(thread) && !it->ooi_used_outside)
2794                 RETURN(1);
2795
2796         if (ooc->ooc_cached_items > 0) {
2797                 ooc->ooc_cached_items--;
2798                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2799                                         ~OSD_OTABLE_IT_CACHE_MASK;
2800                 RETURN(0);
2801         }
2802
2803         if (it->ooi_all_cached) {
2804                 l_wait_event(thread->t_ctl_waitq,
2805                              !thread_is_running(thread),
2806                              &lwi);
2807                 RETURN(1);
2808         }
2809
2810         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2811                 spin_lock(&scrub->os_lock);
2812                 scrub->os_waiting = 0;
2813                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2814                 spin_unlock(&scrub->os_lock);
2815         }
2816
2817         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2818                 l_wait_event(thread->t_ctl_waitq,
2819                              osd_otable_it_wakeup(scrub, it),
2820                              &lwi);
2821
2822         if (!thread_is_running(thread) && !it->ooi_used_outside)
2823                 RETURN(1);
2824
2825         rc = osd_otable_it_preload(env, it);
2826         if (rc >= 0)
2827                 goto again;
2828
2829         RETURN(rc);
2830 }
2831
2832 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2833                                         const struct dt_it *di)
2834 {
2835         return NULL;
2836 }
2837
2838 static int osd_otable_it_key_size(const struct lu_env *env,
2839                                   const struct dt_it *di)
2840 {
2841         return sizeof(__u64);
2842 }
2843
2844 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2845                              struct dt_rec *rec, __u32 attr)
2846 {
2847         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2848         struct osd_otable_cache *ooc = &it->ooi_cache;
2849
2850         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2851
2852         /* Filter out Invald FID already. */
2853         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2854                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2855                  PFID((struct lu_fid *)rec),
2856                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2857
2858         return 0;
2859 }
2860
2861 static __u64 osd_otable_it_store(const struct lu_env *env,
2862                                  const struct dt_it *di)
2863 {
2864         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2865         struct osd_otable_cache *ooc = &it->ooi_cache;
2866         __u64                    hash;
2867
2868         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2869                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2870         else
2871                 hash = ooc->ooc_pos_preload;
2872         return hash;
2873 }
2874
2875 /**
2876  * Set the OSD layer iteration start position as the specified hash.
2877  */
2878 static int osd_otable_it_load(const struct lu_env *env,
2879                               const struct dt_it *di, __u64 hash)
2880 {
2881         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2882         struct osd_device       *dev   = it->ooi_dev;
2883         struct osd_otable_cache *ooc   = &it->ooi_cache;
2884         struct osd_scrub        *scrub = &dev->od_scrub;
2885         int                      rc;
2886         ENTRY;
2887
2888         /* Forbid to set iteration position after iteration started. */
2889         if (it->ooi_user_ready)
2890                 RETURN(-EPERM);
2891
2892         if (hash > OSD_OTABLE_MAX_HASH)
2893                 hash = OSD_OTABLE_MAX_HASH;
2894
2895         ooc->ooc_pos_preload = hash;
2896         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2897                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2898
2899         it->ooi_user_ready = 1;
2900         if (!scrub->os_full_speed)
2901                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2902
2903         /* Unplug OSD layer iteration by the first next() call. */
2904         rc = osd_otable_it_next(env, (struct dt_it *)it);
2905
2906         RETURN(rc);
2907 }
2908
2909 static int osd_otable_it_key_rec(const struct lu_env *env,
2910                                  const struct dt_it *di, void *key_rec)
2911 {
2912         return 0;
2913 }
2914
2915 const struct dt_index_operations osd_otable_ops = {
2916         .dio_it = {
2917                 .init     = osd_otable_it_init,
2918                 .fini     = osd_otable_it_fini,
2919                 .get      = osd_otable_it_get,
2920                 .put      = osd_otable_it_put,
2921                 .next     = osd_otable_it_next,
2922                 .key      = osd_otable_it_key,
2923                 .key_size = osd_otable_it_key_size,
2924                 .rec      = osd_otable_it_rec,
2925                 .store    = osd_otable_it_store,
2926                 .load     = osd_otable_it_load,
2927                 .key_rec  = osd_otable_it_key_rec,
2928         }
2929 };
2930
2931 /* high priority inconsistent items list APIs */
2932
2933 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL  60
2934
2935 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2936                    int insert)
2937 {
2938         struct osd_inconsistent_item *oii;
2939         struct osd_scrub             *scrub  = &dev->od_scrub;
2940         struct ptlrpc_thread         *thread = &scrub->os_thread;
2941         int                           wakeup = 0;
2942         ENTRY;
2943
2944         OBD_ALLOC_PTR(oii);
2945         if (unlikely(oii == NULL))
2946                 RETURN(-ENOMEM);
2947
2948         INIT_LIST_HEAD(&oii->oii_list);
2949         oii->oii_cache = *oic;
2950         oii->oii_insert = insert;
2951
2952         if (scrub->os_partial_scan) {
2953                 __u64 now = cfs_time_current_sec();
2954
2955                 /* If there haven't been errors in a long time,
2956                  * decay old count until either the errors are
2957                  * gone or we reach the current interval. */
2958                 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2959                                 scrub->os_bad_oimap_time +
2960                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2961                         scrub->os_bad_oimap_count >>= 1;
2962                         scrub->os_bad_oimap_time +=
2963                                 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2964                 }
2965
2966                 scrub->os_bad_oimap_time = now;
2967                 if (++scrub->os_bad_oimap_count >
2968                     dev->od_full_scrub_threshold_rate)
2969                         scrub->os_full_scrub = 1;
2970         }
2971
2972         spin_lock(&scrub->os_lock);
2973         if (unlikely(!thread_is_running(thread))) {
2974                 spin_unlock(&scrub->os_lock);
2975                 OBD_FREE_PTR(oii);
2976                 RETURN(-EAGAIN);
2977         }
2978
2979         if (list_empty(&scrub->os_inconsistent_items))
2980                 wakeup = 1;
2981         list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2982         spin_unlock(&scrub->os_lock);
2983
2984         if (wakeup != 0)
2985                 wake_up_all(&thread->t_ctl_waitq);
2986
2987         RETURN(0);
2988 }
2989
2990 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2991                    struct osd_inode_id *id)
2992 {
2993         struct osd_scrub             *scrub = &dev->od_scrub;
2994         struct osd_inconsistent_item *oii;
2995         ENTRY;
2996
2997         spin_lock(&scrub->os_lock);
2998         list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2999                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3000                         *id = oii->oii_cache.oic_lid;
3001                         spin_unlock(&scrub->os_lock);
3002                         RETURN(0);
3003                 }
3004         }
3005         spin_unlock(&scrub->os_lock);
3006
3007         RETURN(-ENOENT);
3008 }
3009
3010 /* OI scrub dump */
3011
3012 static const char *scrub_status_names[] = {
3013         "init",
3014         "scanning",
3015         "completed",
3016         "failed",
3017         "stopped",
3018         "paused",
3019         "crashed",
3020         NULL
3021 };
3022
3023 static const char *scrub_flags_names[] = {
3024         "recreated",
3025         "inconsistent",
3026         "auto",
3027         "upgrade",
3028         NULL
3029 };
3030
3031 static const char *scrub_param_names[] = {
3032         "failout",
3033         "dryrun",
3034         NULL
3035 };
3036
3037 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3038                            const char *prefix)
3039 {
3040         int flag;
3041         int rc;
3042         int i;
3043
3044         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3045         if (rc < 0)
3046                 return rc;
3047
3048         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3049                 if (flag & bits) {
3050                         bits &= ~flag;
3051                         rc = seq_printf(m, "%s%c", names[i],
3052                                         bits != 0 ? ',' : '\n');
3053                         if (rc < 0)
3054                                 return rc;
3055                 }
3056         }
3057         return 0;
3058 }
3059
3060 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3061 {
3062         int rc;
3063
3064         if (time != 0)
3065                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
3066                               cfs_time_current_sec() - time);
3067         else
3068                 rc = seq_printf(m, "%s: N/A\n", prefix);
3069         return rc;
3070 }
3071
3072 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3073 {
3074         int rc;
3075
3076         if (pos != 0)
3077                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
3078         else
3079                 rc = seq_printf(m, "%s: N/A\n", prefix);
3080         return rc;
3081 }
3082
3083 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3084 {
3085         struct osd_scrub  *scrub   = &dev->od_scrub;
3086         struct scrub_file *sf      = &scrub->os_file;
3087         __u64              checked;
3088         __u64              speed;
3089         int                rc;
3090
3091         down_read(&scrub->os_rwsem);
3092         rc = seq_printf(m, "name: OI_scrub\n"
3093                         "magic: 0x%x\n"
3094                         "oi_files: %d\n"
3095                         "status: %s\n",
3096                         sf->sf_magic, (int)sf->sf_oi_count,
3097                         scrub_status_names[sf->sf_status]);
3098         if (rc < 0)
3099                 goto out;
3100
3101         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
3102                              "flags");
3103         if (rc < 0)
3104                 goto out;
3105
3106         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,