Whamcloud - gitweb
LU-4927 scrub: rebuild stale OST OI mapping
[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, 2013, 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_MDS
39
40 #include <lustre/lustre_idl.h>
41 #include <lustre_disk.h>
42 #include <dt_object.h>
43 #include <linux/xattr.h>
44
45 #include "osd_internal.h"
46 #include "osd_oi.h"
47 #include "osd_scrub.h"
48
49 #define HALF_SEC        (HZ >> 1)
50
51 #define OSD_OTABLE_MAX_HASH             0x00000000ffffffffULL
52
53 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
54 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
55 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
56 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
57 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
58 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
59 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
60 #define SCRUB_NEXT_NOLMA        8 /* the inode has no FID-in-LMA */
61 #define SCRUB_NEXT_OSTOBJ       9 /* for OST-object */
62 #define SCRUB_NEXT_OSTOBJ_OLD   10 /* old OST-object, no LMA or no FID-on-OST
63                                     * flags in LMA */
64
65 /* misc functions */
66
67 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
68 {
69         return container_of0(scrub, struct osd_device, od_scrub);
70 }
71
72 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
73 {
74         return osd_sb(osd_scrub2dev(scrub));
75 }
76
77 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
78                                        struct osd_otable_cache *ooc)
79 {
80         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
81 }
82
83 /**
84  * update/insert/delete the specified OI mapping (@fid @id) according to the ops
85  *
86  * \retval   1, changed nothing
87  * \retval   0, changed successfully
88  * \retval -ve, on error
89  */
90 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
91                                      struct osd_device *dev,
92                                      const struct lu_fid *fid,
93                                      const struct osd_inode_id *id,
94                                      int ops, bool force,
95                                      enum oi_check_flags flags)
96 {
97         handle_t *th;
98         int       rc;
99         ENTRY;
100
101         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
102                 RETURN(0);
103
104         /* DTO_INDEX_INSERT is enough for other two ops:
105          * delete/update, but save stack. */
106         th = ldiskfs_journal_start_sb(osd_sb(dev),
107                                 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
108         if (IS_ERR(th)) {
109                 rc = PTR_ERR(th);
110                 CERROR("%s: fail to start trans for scrub %d: rc = %d\n",
111                        osd_name(dev), ops, rc);
112                 RETURN(rc);
113         }
114
115         switch (ops) {
116         case DTO_INDEX_UPDATE:
117                 rc = osd_oi_update(info, dev, fid, id, th, flags);
118                 if (unlikely(rc == -ENOENT)) {
119                         /* Some unlink thread may removed the OI mapping. */
120                         rc = 1;
121                 }
122                 break;
123         case DTO_INDEX_INSERT:
124                 rc = osd_oi_insert(info, dev, fid, id, th, flags);
125                 if (unlikely(rc == -EEXIST)) {
126                         rc = 1;
127                         /* XXX: There are trouble things when adding OI
128                          *      mapping for IGIF object, which may cause
129                          *      multiple objects to be mapped to the same
130                          *      IGIF formatted FID. Consider the following
131                          *      situations:
132                          *
133                          *      1) The MDT is upgrading from 1.8 device.
134                          *      The OI scrub generates IGIF FID1 for the
135                          *      OBJ1 and adds the OI mapping.
136                          *
137                          *      2) For some reason, the OI scrub does not
138                          *      process all the IGIF objects completely.
139                          *
140                          *      3) The MDT is backuped and restored against
141                          *      this device.
142                          *
143                          *      4) When the MDT mounts up, the OI scrub will
144                          *      try to rebuild the OI files. For some IGIF
145                          *      object, OBJ2, which was not processed by the
146                          *      OI scrub before the backup/restore, and the
147                          *      new generated IGIF formatted FID may be just
148                          *      the FID1, the same as OBJ1.
149                          *
150                          *      Under such case, the OI scrub cannot know how
151                          *      to generate new FID for the OBJ2.
152                          *
153                          *      Currently, we do nothing for that. One possible
154                          *      solution is to generate new normal FID for the
155                          *      conflict object.
156                          *
157                          *      Anyway, it is rare, only exists in theory. */
158                 }
159                 break;
160         case DTO_INDEX_DELETE:
161                 rc = osd_oi_delete(info, dev, fid, th, flags);
162                 if (rc == -ENOENT) {
163                         /* It is normal that the unlink thread has removed the
164                          * OI mapping already. */
165                         rc = 1;
166                 }
167                 break;
168         default:
169                 LASSERTF(0, "Unexpected ops %d\n", ops);
170                 break;
171         }
172
173         ldiskfs_journal_stop(th);
174         return rc;
175 }
176
177 /* OI_scrub file ops */
178
179 static void osd_scrub_file_to_cpu(struct scrub_file *des,
180                                   struct scrub_file *src)
181 {
182         memcpy(des->sf_uuid, src->sf_uuid, 16);
183         des->sf_flags   = le64_to_cpu(src->sf_flags);
184         des->sf_magic   = le32_to_cpu(src->sf_magic);
185         des->sf_status  = le16_to_cpu(src->sf_status);
186         des->sf_param   = le16_to_cpu(src->sf_param);
187         des->sf_time_last_complete      =
188                                 le64_to_cpu(src->sf_time_last_complete);
189         des->sf_time_latest_start       =
190                                 le64_to_cpu(src->sf_time_latest_start);
191         des->sf_time_last_checkpoint    =
192                                 le64_to_cpu(src->sf_time_last_checkpoint);
193         des->sf_pos_latest_start        =
194                                 le64_to_cpu(src->sf_pos_latest_start);
195         des->sf_pos_last_checkpoint     =
196                                 le64_to_cpu(src->sf_pos_last_checkpoint);
197         des->sf_pos_first_inconsistent  =
198                                 le64_to_cpu(src->sf_pos_first_inconsistent);
199         des->sf_items_checked           =
200                                 le64_to_cpu(src->sf_items_checked);
201         des->sf_items_updated           =
202                                 le64_to_cpu(src->sf_items_updated);
203         des->sf_items_failed            =
204                                 le64_to_cpu(src->sf_items_failed);
205         des->sf_items_updated_prior     =
206                                 le64_to_cpu(src->sf_items_updated_prior);
207         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
208         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
209         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
210         des->sf_internal_flags  = le16_to_cpu(src->sf_internal_flags);
211         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
212 }
213
214 static void osd_scrub_file_to_le(struct scrub_file *des,
215                                  struct scrub_file *src)
216 {
217         memcpy(des->sf_uuid, src->sf_uuid, 16);
218         des->sf_flags   = cpu_to_le64(src->sf_flags);
219         des->sf_magic   = cpu_to_le32(src->sf_magic);
220         des->sf_status  = cpu_to_le16(src->sf_status);
221         des->sf_param   = cpu_to_le16(src->sf_param);
222         des->sf_time_last_complete      =
223                                 cpu_to_le64(src->sf_time_last_complete);
224         des->sf_time_latest_start       =
225                                 cpu_to_le64(src->sf_time_latest_start);
226         des->sf_time_last_checkpoint    =
227                                 cpu_to_le64(src->sf_time_last_checkpoint);
228         des->sf_pos_latest_start        =
229                                 cpu_to_le64(src->sf_pos_latest_start);
230         des->sf_pos_last_checkpoint     =
231                                 cpu_to_le64(src->sf_pos_last_checkpoint);
232         des->sf_pos_first_inconsistent  =
233                                 cpu_to_le64(src->sf_pos_first_inconsistent);
234         des->sf_items_checked           =
235                                 cpu_to_le64(src->sf_items_checked);
236         des->sf_items_updated           =
237                                 cpu_to_le64(src->sf_items_updated);
238         des->sf_items_failed            =
239                                 cpu_to_le64(src->sf_items_failed);
240         des->sf_items_updated_prior     =
241                                 cpu_to_le64(src->sf_items_updated_prior);
242         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
243         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
244         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
245         des->sf_internal_flags  = cpu_to_le16(src->sf_internal_flags);
246         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
247 }
248
249 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
250 {
251         struct scrub_file *sf = &scrub->os_file;
252
253         memset(sf, 0, sizeof(*sf));
254         memcpy(sf->sf_uuid, uuid, 16);
255         sf->sf_magic = SCRUB_MAGIC_V1;
256         sf->sf_status = SS_INIT;
257 }
258
259 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
260 {
261         struct scrub_file *sf = &scrub->os_file;
262
263         CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
264         memcpy(sf->sf_uuid, uuid, 16);
265         sf->sf_status = SS_INIT;
266         sf->sf_flags |= flags;
267         sf->sf_run_time = 0;
268         sf->sf_time_latest_start = 0;
269         sf->sf_time_last_checkpoint = 0;
270         sf->sf_pos_latest_start = 0;
271         sf->sf_pos_last_checkpoint = 0;
272         sf->sf_pos_first_inconsistent = 0;
273         sf->sf_items_checked = 0;
274         sf->sf_items_updated = 0;
275         sf->sf_items_failed = 0;
276         sf->sf_items_updated_prior = 0;
277         sf->sf_items_noscrub = 0;
278         sf->sf_items_igif = 0;
279 }
280
281 static int osd_scrub_file_load(struct osd_scrub *scrub)
282 {
283         loff_t  pos  = 0;
284         char   *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
285         int     len  = sizeof(scrub->os_file_disk);
286         int     rc;
287
288         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
289         if (rc == len) {
290                 struct scrub_file *sf = &scrub->os_file;
291
292                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
293                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
294                         CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
295                               name, sf->sf_magic, SCRUB_MAGIC_V1);
296                         /* Process it as new scrub file. */
297                         rc = -ENOENT;
298                 } else {
299                         rc = 0;
300                 }
301         } else if (rc != 0) {
302                 CERROR("%.16s: fail to load scrub file, expected = %d, "
303                        "rc = %d\n", name, len, rc);
304                 if (rc > 0)
305                         rc = -EFAULT;
306         } else {
307                 /* return -ENOENT for empty scrub file case. */
308                 rc = -ENOENT;
309         }
310
311         return rc;
312 }
313
314 int osd_scrub_file_store(struct osd_scrub *scrub)
315 {
316         struct osd_device *dev;
317         handle_t          *jh;
318         loff_t             pos     = 0;
319         int                len     = sizeof(scrub->os_file_disk);
320         int                credits;
321         int                rc;
322
323         dev = container_of0(scrub, struct osd_device, od_scrub);
324         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
325                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
326         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
327         if (IS_ERR(jh)) {
328                 rc = PTR_ERR(jh);
329                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
330                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
331                 return rc;
332         }
333
334         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
335         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
336                                       len, 0, &pos, jh);
337         ldiskfs_journal_stop(jh);
338         if (rc != 0)
339                 CERROR("%.16s: fail to store scrub file, expected = %d, "
340                        "rc = %d\n",
341                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
342                        len, rc);
343         scrub->os_time_last_checkpoint = cfs_time_current();
344         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
345                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
346         return rc;
347 }
348
349 static int
350 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
351                      struct inode *inode, const struct lu_fid *fid)
352 {
353         struct filter_fid_old   *ff      = &info->oti_ff;
354         struct dentry           *dentry  = &info->oti_obj_dentry;
355         handle_t                *jh;
356         int                      size    = 0;
357         int                      rc;
358         bool                     removed = false;
359         bool                     reset   = true;
360         ENTRY;
361
362         if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
363                 RETURN(0);
364
365         /* We want the LMA to fit into the 256-byte OST inode, so operate
366          * as following:
367          * 1) read old XATTR_NAME_FID and save the parent FID;
368          * 2) delete the old XATTR_NAME_FID;
369          * 3) make new LMA and add it;
370          * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
371          *
372          * Making the LMA to fit into the 256-byte OST inode can save time for
373          * normal osd_check_lma() and for other OI scrub scanning in future.
374          * So it is worth to make some slow conversion here. */
375         jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
376                                 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
377         if (IS_ERR(jh)) {
378                 rc = PTR_ERR(jh);
379                 CERROR("%s: fail to start trans for convert ff: "DFID
380                        ": rc = %d\n",
381                        osd_name(dev), PFID(fid), rc);
382                 RETURN(rc);
383         }
384
385         /* 1) read old XATTR_NAME_FID and save the parent FID */
386         rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
387         if (rc == sizeof(*ff)) {
388                 /* 2) delete the old XATTR_NAME_FID */
389                 ll_vfs_dq_init(inode);
390                 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
391                 if (rc != 0)
392                         GOTO(stop, rc);
393
394                 removed = true;
395         } else if (unlikely(rc == -ENODATA)) {
396                 reset = false;
397         } else if (rc != sizeof(struct filter_fid)) {
398                 GOTO(stop, rc = -EINVAL);
399         }
400
401         /* 3) make new LMA and add it */
402         rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0);
403         if (rc == 0 && reset)
404                 size = sizeof(struct filter_fid);
405         else if (rc != 0 && removed)
406                 /* If failed, we should try to add the old back. */
407                 size = sizeof(struct filter_fid_old);
408
409         /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
410         if (size > 0) {
411                 int rc1;
412
413                 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
414                                       XATTR_CREATE);
415                 if (rc1 != 0 && rc == 0)
416                         rc = rc1;
417         }
418
419         GOTO(stop, rc);
420
421 stop:
422         ldiskfs_journal_stop(jh);
423         return rc;
424 }
425
426 static int
427 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
428                        struct osd_idmap_cache *oic, int val)
429 {
430         struct osd_scrub             *scrub  = &dev->od_scrub;
431         struct scrub_file            *sf     = &scrub->os_file;
432         struct lu_fid                *fid    = &oic->oic_fid;
433         struct osd_inode_id          *lid    = &oic->oic_lid;
434         struct osd_inode_id          *lid2   = &info->oti_id;
435         struct osd_inconsistent_item *oii    = NULL;
436         struct inode                 *inode  = NULL;
437         int                           ops    = DTO_INDEX_UPDATE;
438         int                           idx;
439         int                           rc;
440         bool                          converted = false;
441         ENTRY;
442
443         down_write(&scrub->os_rwsem);
444         scrub->os_new_checked++;
445         if (val < 0)
446                 GOTO(out, rc = val);
447
448         if (scrub->os_in_prior)
449                 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
450                                      oii_cache);
451
452         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
453                 GOTO(out, rc = 0);
454
455         if (fid_is_igif(fid))
456                 sf->sf_items_igif++;
457
458         if (val == SCRUB_NEXT_OSTOBJ_OLD) {
459                 inode = osd_iget(info, dev, lid);
460                 if (IS_ERR(inode)) {
461                         rc = PTR_ERR(inode);
462                         /* Someone removed the inode. */
463                         if (rc == -ENOENT || rc == -ESTALE)
464                                 rc = 0;
465                         GOTO(out, rc);
466                 }
467
468                 sf->sf_flags |= SF_UPGRADE;
469                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
470                 dev->od_check_ff = 1;
471                 rc = osd_scrub_convert_ff(info, dev, inode, fid);
472                 if (rc != 0)
473                         GOTO(out, rc);
474
475                 converted = true;
476         }
477
478         if ((val == SCRUB_NEXT_NOLMA) &&
479             (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
480                 GOTO(out, rc = 0);
481
482         if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
483                 goto iget;
484
485         rc = osd_oi_lookup(info, dev, fid, lid2,
486                 (val == SCRUB_NEXT_OSTOBJ ||
487                  val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
488         if (rc != 0) {
489                 if (rc != -ENOENT && rc != -ESTALE)
490                         GOTO(out, rc);
491
492 iget:
493                 if (inode == NULL) {
494                         inode = osd_iget(info, dev, lid);
495                         if (IS_ERR(inode)) {
496                                 rc = PTR_ERR(inode);
497                                 /* Someone removed the inode. */
498                                 if (rc == -ENOENT || rc == -ESTALE)
499                                         rc = 0;
500                                 GOTO(out, rc);
501                         }
502                 }
503
504                 scrub->os_full_speed = 1;
505                 ops = DTO_INDEX_INSERT;
506                 idx = osd_oi_fid2idx(dev, fid);
507                 switch (val) {
508                 case SCRUB_NEXT_NOLMA:
509                         sf->sf_flags |= SF_UPGRADE;
510                         if (!(sf->sf_param & SP_DRYRUN)) {
511                                 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
512                                 if (rc != 0)
513                                         GOTO(out, rc);
514                         }
515
516                         if (!(sf->sf_flags & SF_INCONSISTENT))
517                                 dev->od_igif_inoi = 0;
518                         break;
519                 case SCRUB_NEXT_OSTOBJ:
520                         sf->sf_flags |= SF_INCONSISTENT;
521                 case SCRUB_NEXT_OSTOBJ_OLD:
522                         break;
523                 default:
524                         sf->sf_flags |= SF_RECREATED;
525                         if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
526                                 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
527                         break;
528                 }
529         } else if (osd_id_eq(lid, lid2)) {
530                 if (converted)
531                         sf->sf_items_updated++;
532
533                 GOTO(out, rc = 0);
534         } else {
535                 scrub->os_full_speed = 1;
536                 sf->sf_flags |= SF_INCONSISTENT;
537
538                 /* XXX: If the device is restored from file-level backup, then
539                  *      some IGIFs may have been already in OI files, and some
540                  *      may be not yet. Means upgrading from 1.8 may be partly
541                  *      processed, but some clients may hold some immobilized
542                  *      IGIFs, and use them to access related objects. Under
543                  *      such case, OSD does not know whether an given IGIF has
544                  *      been processed or to be processed, and it also cannot
545                  *      generate local ino#/gen# directly from the immobilized
546                  *      IGIF because of the backup/restore. Then force OSD to
547                  *      lookup the given IGIF in OI files, and if no entry,
548                  *      then ask the client to retry after upgrading completed.
549                  *      No better choice. */
550                 dev->od_igif_inoi = 1;
551         }
552
553         rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
554                         (val == SCRUB_NEXT_OSTOBJ ||
555                          val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
556         if (rc == 0) {
557                 if (scrub->os_in_prior)
558                         sf->sf_items_updated_prior++;
559                 else
560                         sf->sf_items_updated++;
561
562                 /* The target has been changed, need to be re-loaded. */
563                 lu_object_purge(info->oti_env, osd2lu_dev(dev), fid);
564         }
565
566         GOTO(out, rc);
567
568 out:
569         if (rc < 0) {
570                 sf->sf_items_failed++;
571                 if (sf->sf_pos_first_inconsistent == 0 ||
572                     sf->sf_pos_first_inconsistent > lid->oii_ino)
573                         sf->sf_pos_first_inconsistent = lid->oii_ino;
574         } else {
575                 rc = 0;
576         }
577
578         /* There may be conflict unlink during the OI scrub,
579          * if happend, then remove the new added OI mapping. */
580         if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
581             unlikely(inode->i_nlink == 0))
582                 osd_scrub_refresh_mapping(info, dev, fid, lid,
583                                 DTO_INDEX_DELETE, false,
584                                 (val == SCRUB_NEXT_OSTOBJ ||
585                                  val == SCRUB_NEXT_OSTOBJ_OLD) ?
586                                 OI_KNOWN_ON_OST : 0);
587         up_write(&scrub->os_rwsem);
588
589         if (inode != NULL && !IS_ERR(inode))
590                 iput(inode);
591
592         if (oii != NULL) {
593                 LASSERT(!cfs_list_empty(&oii->oii_list));
594
595                 spin_lock(&scrub->os_lock);
596                 cfs_list_del_init(&oii->oii_list);
597                 spin_unlock(&scrub->os_lock);
598                 OBD_FREE_PTR(oii);
599         }
600         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
601 }
602
603 /* OI scrub APIs */
604
605 static int osd_scrub_prep(struct osd_device *dev)
606 {
607         struct osd_scrub     *scrub  = &dev->od_scrub;
608         struct ptlrpc_thread *thread = &scrub->os_thread;
609         struct scrub_file    *sf     = &scrub->os_file;
610         __u32                 flags  = scrub->os_start_flags;
611         int                   rc;
612         bool                  drop_dryrun = false;
613         ENTRY;
614
615         down_write(&scrub->os_rwsem);
616         if (flags & SS_SET_FAILOUT)
617                 sf->sf_param |= SP_FAILOUT;
618
619         if (flags & SS_CLEAR_FAILOUT)
620                 sf->sf_param &= ~SP_FAILOUT;
621
622         if (flags & SS_SET_DRYRUN)
623                 sf->sf_param |= SP_DRYRUN;
624
625         if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
626                 sf->sf_param &= ~SP_DRYRUN;
627                 drop_dryrun = true;
628         }
629
630         if (flags & SS_RESET)
631                 osd_scrub_file_reset(scrub,
632                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
633
634         if (flags & SS_AUTO) {
635                 scrub->os_full_speed = 1;
636                 sf->sf_flags |= SF_AUTO;
637                 /* For the case of OI scrub auto triggered, NOT dryrun. */
638                 sf->sf_param &= ~SP_FAILOUT;
639         } else {
640                 scrub->os_full_speed = 0;
641         }
642
643         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
644                 scrub->os_full_speed = 1;
645
646         scrub->os_in_prior = 0;
647         spin_lock(&scrub->os_lock);
648         scrub->os_waiting = 0;
649         scrub->os_paused = 0;
650         spin_unlock(&scrub->os_lock);
651         scrub->os_new_checked = 0;
652         if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
653                 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
654         else if (sf->sf_pos_last_checkpoint != 0)
655                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
656         else
657                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
658
659         scrub->os_pos_current = sf->sf_pos_latest_start;
660         sf->sf_status = SS_SCANNING;
661         sf->sf_time_latest_start = cfs_time_current_sec();
662         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
663         rc = osd_scrub_file_store(scrub);
664         if (rc == 0) {
665                 spin_lock(&scrub->os_lock);
666                 thread_set_flags(thread, SVC_RUNNING);
667                 spin_unlock(&scrub->os_lock);
668                 wake_up_all(&thread->t_ctl_waitq);
669         }
670         up_write(&scrub->os_rwsem);
671
672         RETURN(rc);
673 }
674
675 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
676 {
677         struct scrub_file *sf = &scrub->os_file;
678         int                rc;
679
680         if (likely(cfs_time_before(cfs_time_current(),
681                                    scrub->os_time_next_checkpoint) ||
682                    scrub->os_new_checked == 0))
683                 return 0;
684
685         down_write(&scrub->os_rwsem);
686         sf->sf_items_checked += scrub->os_new_checked;
687         scrub->os_new_checked = 0;
688         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
689         sf->sf_time_last_checkpoint = cfs_time_current_sec();
690         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
691                                             scrub->os_time_last_checkpoint);
692         rc = osd_scrub_file_store(scrub);
693         up_write(&scrub->os_rwsem);
694
695         return rc;
696 }
697
698 static void osd_scrub_post(struct osd_scrub *scrub, int result)
699 {
700         struct scrub_file *sf = &scrub->os_file;
701         ENTRY;
702
703         down_write(&scrub->os_rwsem);
704         spin_lock(&scrub->os_lock);
705         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
706         spin_unlock(&scrub->os_lock);
707         if (scrub->os_new_checked > 0) {
708                 sf->sf_items_checked += scrub->os_new_checked;
709                 scrub->os_new_checked = 0;
710                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
711         }
712         sf->sf_time_last_checkpoint = cfs_time_current_sec();
713         if (result > 0) {
714                 struct osd_device *dev =
715                         container_of0(scrub, struct osd_device, od_scrub);
716
717                 dev->od_igif_inoi = 1;
718                 dev->od_check_ff = 0;
719                 sf->sf_status = SS_COMPLETED;
720                 if (!(sf->sf_param & SP_DRYRUN)) {
721                         memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
722                         sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
723                                           SF_UPGRADE | SF_AUTO);
724                 }
725                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
726                 sf->sf_success_count++;
727         } else if (result == 0) {
728                 if (scrub->os_paused)
729                         sf->sf_status = SS_PAUSED;
730                 else
731                         sf->sf_status = SS_STOPPED;
732         } else {
733                 sf->sf_status = SS_FAILED;
734         }
735         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
736                                             scrub->os_time_last_checkpoint);
737         result = osd_scrub_file_store(scrub);
738         if (result < 0)
739                 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
740                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
741                        result);
742         up_write(&scrub->os_rwsem);
743
744         EXIT;
745 }
746
747 /* iteration engine */
748
749 struct osd_iit_param {
750         struct super_block *sb;
751         struct buffer_head *bitmap;
752         ldiskfs_group_t bg;
753         __u32 gbase;
754         __u32 offset;
755 };
756
757 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
758                                    struct osd_device *dev,
759                                    struct osd_iit_param *param,
760                                    struct osd_idmap_cache **oic,
761                                    int noslot);
762
763 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
764                                    struct osd_device *dev,
765                                    struct osd_iit_param *param,
766                                    struct osd_idmap_cache *oic,
767                                    int *noslot, int rc);
768
769 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
770 {
771         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
772                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
773         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
774                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
775                 return SCRUB_NEXT_BREAK;
776         } else {
777                 *pos = param->gbase + param->offset;
778                 return 0;
779         }
780 }
781
782 /**
783  * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
784  * \retval 0: FID-on-MDT
785  */
786 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
787                                       struct osd_device *dev,
788                                       struct lu_fid *fid)
789 {
790         /* XXX: The initial OI scrub will scan the top level /O to generate
791          *      a small local FLDB according to the <seq>. If the given FID
792          *      is in the local FLDB, then it is FID-on-OST; otherwise it's
793          *      quite possible for FID-on-MDT. */
794         if (dev->od_is_ost)
795                 return SCRUB_NEXT_OSTOBJ_OLD;
796         else
797                 return 0;
798 }
799
800 static int osd_scrub_get_fid(struct osd_thread_info *info,
801                              struct osd_device *dev, struct inode *inode,
802                              struct lu_fid *fid, bool scrub)
803 {
804         struct lustre_mdt_attrs *lma     = &info->oti_mdt_attrs;
805         int                      rc;
806         bool                     has_lma = false;
807
808         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
809         if (rc == 0) {
810                 has_lma = true;
811                 if (lma->lma_compat & LMAC_NOT_IN_OI ||
812                     lma->lma_incompat & LMAI_AGENT)
813                         return SCRUB_NEXT_CONTINUE;
814
815                 *fid = lma->lma_self_fid;
816                 if (!scrub)
817                         return 0;
818
819                 if (lma->lma_compat & LMAC_FID_ON_OST)
820                         return SCRUB_NEXT_OSTOBJ;
821
822                 if (fid_is_idif(fid))
823                         return SCRUB_NEXT_OSTOBJ_OLD;
824
825                 /* For local object. */
826                 if (fid_is_internal(fid))
827                         return 0;
828
829                 /* For external visible MDT-object with non-normal FID. */
830                 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
831                         return 0;
832
833                 /* For the object with normal FID, it may be MDT-object,
834                  * or may be 2.4 OST-object, need further distinguish.
835                  * Fall through to next section. */
836         }
837
838         if (rc == -ENODATA || rc == 0) {
839                 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
840                 if (rc == 0) {
841                         if (scrub)
842                                 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
843                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
844                         return rc;
845                 }
846
847                 if (rc > 0) {
848                         if (!has_lma)
849                                 /* It is FID-on-OST, but we do not know how
850                                  * to generate its FID, ignore it directly. */
851                                 rc = SCRUB_NEXT_CONTINUE;
852                         else
853                                 /* It is 2.4 OST-object. */
854                                 rc = SCRUB_NEXT_OSTOBJ_OLD;
855                         return rc;
856                 }
857
858                 if (rc != -ENODATA)
859                         return rc;
860
861                 if (!has_lma) {
862                         if (dev->od_scrub.os_convert_igif) {
863                                 lu_igif_build(fid, inode->i_ino,
864                                               inode->i_generation);
865                                 if (scrub)
866                                         rc = SCRUB_NEXT_NOLMA;
867                                 else
868                                         rc = 0;
869                         } else {
870                                 /* It may be FID-on-OST, or may be FID for
871                                  * non-MDT0, anyway, we do not know how to
872                                  * generate its FID, ignore it directly. */
873                                 rc = SCRUB_NEXT_CONTINUE;
874                         }
875                         return rc;
876                 }
877
878                 /* For OI scrub case only: the object has LMA but has no ff
879                  * (or ff crashed). It may be MDT-object, may be OST-object
880                  * with crashed ff. The last check is local FLDB. */
881                 rc = osd_scrub_check_local_fldb(info, dev, fid);
882         }
883
884         return rc;
885 }
886
887 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
888                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
889                         struct super_block *sb, bool scrub)
890 {
891         struct inode *inode;
892         int           rc;
893         ENTRY;
894
895         osd_id_gen(lid, pos, OSD_OII_NOGEN);
896         inode = osd_iget(info, dev, lid);
897         if (IS_ERR(inode)) {
898                 rc = PTR_ERR(inode);
899                 /* The inode may be removed after bitmap searching, or the
900                  * file is new created without inode initialized yet. */
901                 if (rc == -ENOENT || rc == -ESTALE)
902                         RETURN(SCRUB_NEXT_CONTINUE);
903
904                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
905                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
906                 RETURN(rc);
907         }
908
909         if (scrub &&
910             ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
911                 /* Only skip it for the first OI scrub accessing. */
912                 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
913                 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
914         }
915
916         rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
917
918         GOTO(put, rc);
919
920 put:
921         iput(inode);
922         return rc;
923 }
924
925 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
926                           struct osd_iit_param *param,
927                           struct osd_idmap_cache **oic, int noslot)
928 {
929         struct osd_scrub     *scrub  = &dev->od_scrub;
930         struct ptlrpc_thread *thread = &scrub->os_thread;
931         struct lu_fid        *fid;
932         struct osd_inode_id  *lid;
933         int                   rc;
934
935         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
936                 struct l_wait_info lwi;
937
938                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
939                 l_wait_event(thread->t_ctl_waitq,
940                              !cfs_list_empty(&scrub->os_inconsistent_items) ||
941                              !thread_is_running(thread),
942                              &lwi);
943         }
944
945         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
946                 spin_lock(&scrub->os_lock);
947                 thread_set_flags(thread, SVC_STOPPING);
948                 spin_unlock(&scrub->os_lock);
949                 return SCRUB_NEXT_CRASH;
950         }
951
952         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
953                 return SCRUB_NEXT_FATAL;
954
955         if (unlikely(!thread_is_running(thread)))
956                 return SCRUB_NEXT_EXIT;
957
958         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
959                 struct osd_inconsistent_item *oii;
960
961                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
962                                      struct osd_inconsistent_item, oii_list);
963                 *oic = &oii->oii_cache;
964                 scrub->os_in_prior = 1;
965                 return 0;
966         }
967
968         if (noslot != 0)
969                 return SCRUB_NEXT_WAIT;
970
971         rc = osd_iit_next(param, &scrub->os_pos_current);
972         if (rc != 0)
973                 return rc;
974
975         *oic = &scrub->os_oic;
976         fid = &(*oic)->oic_fid;
977         lid = &(*oic)->oic_lid;
978         rc = osd_iit_iget(info, dev, fid, lid,
979                           scrub->os_pos_current, param->sb, true);
980         return rc;
981 }
982
983 static int osd_preload_next(struct osd_thread_info *info,
984                             struct osd_device *dev, struct osd_iit_param *param,
985                             struct osd_idmap_cache **oic, int noslot)
986 {
987         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
988         struct osd_scrub        *scrub;
989         struct ptlrpc_thread    *thread;
990         int                      rc;
991
992         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
993         if (rc != 0)
994                 return rc;
995
996         scrub = &dev->od_scrub;
997         thread = &scrub->os_thread;
998         if (thread_is_running(thread) &&
999             ooc->ooc_pos_preload >= scrub->os_pos_current)
1000                 return SCRUB_NEXT_EXIT;
1001
1002         rc = osd_iit_iget(info, dev,
1003                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1004                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1005                           ooc->ooc_pos_preload, param->sb, false);
1006         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1007          * ignore the failure, so it still need to skip the inode next time. */
1008         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1009         return rc;
1010 }
1011
1012 static inline int
1013 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1014 {
1015         spin_lock(&scrub->os_lock);
1016         if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1017             !cfs_list_empty(&scrub->os_inconsistent_items) ||
1018             it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1019                 scrub->os_waiting = 0;
1020         else
1021                 scrub->os_waiting = 1;
1022         spin_unlock(&scrub->os_lock);
1023
1024         return !scrub->os_waiting;
1025 }
1026
1027 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1028                           struct osd_iit_param *param,
1029                           struct osd_idmap_cache *oic, int *noslot, int rc)
1030 {
1031         struct l_wait_info       lwi    = { 0 };
1032         struct osd_scrub        *scrub  = &dev->od_scrub;
1033         struct scrub_file       *sf     = &scrub->os_file;
1034         struct ptlrpc_thread    *thread = &scrub->os_thread;
1035         struct osd_otable_it    *it     = dev->od_otable_it;
1036         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
1037
1038         switch (rc) {
1039         case SCRUB_NEXT_CONTINUE:
1040                 goto next;
1041         case SCRUB_NEXT_WAIT:
1042                 goto wait;
1043         case SCRUB_NEXT_NOSCRUB:
1044                 down_write(&scrub->os_rwsem);
1045                 scrub->os_new_checked++;
1046                 sf->sf_items_noscrub++;
1047                 up_write(&scrub->os_rwsem);
1048                 goto next;
1049         }
1050
1051         rc = osd_scrub_check_update(info, dev, oic, rc);
1052         if (rc != 0)
1053                 return rc;
1054
1055         rc = osd_scrub_checkpoint(scrub);
1056         if (rc != 0) {
1057                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
1058                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
1059                        scrub->os_pos_current, rc);
1060                 /* Continue, as long as the scrub itself can go ahead. */
1061         }
1062
1063         if (scrub->os_in_prior) {
1064                 scrub->os_in_prior = 0;
1065                 return 0;
1066         }
1067
1068 next:
1069         scrub->os_pos_current = param->gbase + ++(param->offset);
1070
1071 wait:
1072         if (it != NULL && it->ooi_waiting && ooc != NULL &&
1073             ooc->ooc_pos_preload < scrub->os_pos_current) {
1074                 spin_lock(&scrub->os_lock);
1075                 it->ooi_waiting = 0;
1076                 wake_up_all(&thread->t_ctl_waitq);
1077                 spin_unlock(&scrub->os_lock);
1078         }
1079
1080         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1081                 return 0;
1082
1083         if (osd_scrub_has_window(scrub, ooc)) {
1084                 *noslot = 0;
1085                 return 0;
1086         }
1087
1088         l_wait_event(thread->t_ctl_waitq,
1089                      osd_scrub_wakeup(scrub, it),
1090                      &lwi);
1091
1092         if (osd_scrub_has_window(scrub, ooc))
1093                 *noslot = 0;
1094         else
1095                 *noslot = 1;
1096         return 0;
1097 }
1098
1099 static int osd_preload_exec(struct osd_thread_info *info,
1100                             struct osd_device *dev, struct osd_iit_param *param,
1101                             struct osd_idmap_cache *oic, int *noslot, int rc)
1102 {
1103         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1104
1105         if (rc == 0) {
1106                 ooc->ooc_cached_items++;
1107                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1108                                         ~OSD_OTABLE_IT_CACHE_MASK;
1109         }
1110         return rc > 0 ? 0 : rc;
1111 }
1112
1113 #define SCRUB_IT_ALL    1
1114 #define SCRUB_IT_CRASH  2
1115
1116 static int osd_inode_iteration(struct osd_thread_info *info,
1117                                struct osd_device *dev, __u32 max, bool preload)
1118 {
1119         osd_iit_next_policy   next;
1120         osd_iit_exec_policy   exec;
1121         __u32                *pos;
1122         __u32                *count;
1123         struct osd_iit_param  param;
1124         __u32                 limit;
1125         int                   noslot = 0;
1126         int                   rc;
1127         ENTRY;
1128
1129         if (!preload) {
1130                 struct osd_scrub *scrub = &dev->od_scrub;
1131
1132                 next = osd_scrub_next;
1133                 exec = osd_scrub_exec;
1134                 pos = &scrub->os_pos_current;
1135                 count = &scrub->os_new_checked;
1136         } else {
1137                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1138
1139                 next = osd_preload_next;
1140                 exec = osd_preload_exec;
1141                 pos = &ooc->ooc_pos_preload;
1142                 count = &ooc->ooc_cached_items;
1143         }
1144         param.sb = osd_sb(dev);
1145         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1146
1147         while (*pos <= limit && *count < max) {
1148                 struct osd_idmap_cache *oic = NULL;
1149
1150                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1151                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1152                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1153                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1154                 if (param.bitmap == NULL) {
1155                         CERROR("%.16s: fail to read bitmap for %u, "
1156                                "scrub will stop, urgent mode\n",
1157                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
1158                                (__u32)param.bg);
1159                         RETURN(-EIO);
1160                 }
1161
1162                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1163                        *count < max) {
1164                         rc = next(info, dev, &param, &oic, noslot);
1165                         switch (rc) {
1166                         case SCRUB_NEXT_BREAK:
1167                                 goto next_group;
1168                         case SCRUB_NEXT_EXIT:
1169                                 brelse(param.bitmap);
1170                                 RETURN(0);
1171                         case SCRUB_NEXT_CRASH:
1172                                 brelse(param.bitmap);
1173                                 RETURN(SCRUB_IT_CRASH);
1174                         case SCRUB_NEXT_FATAL:
1175                                 brelse(param.bitmap);
1176                                 RETURN(-EINVAL);
1177                         }
1178
1179                         rc = exec(info, dev, &param, oic, &noslot, rc);
1180                         if (rc != 0) {
1181                                 brelse(param.bitmap);
1182                                 RETURN(rc);
1183                         }
1184                 }
1185
1186 next_group:
1187                 brelse(param.bitmap);
1188         }
1189
1190         if (*pos > limit)
1191                 RETURN(SCRUB_IT_ALL);
1192         RETURN(0);
1193 }
1194
1195 static int osd_otable_it_preload(const struct lu_env *env,
1196                                  struct osd_otable_it *it)
1197 {
1198         struct osd_device       *dev   = it->ooi_dev;
1199         struct osd_scrub        *scrub = &dev->od_scrub;
1200         struct osd_otable_cache *ooc   = &it->ooi_cache;
1201         int                      rc;
1202         ENTRY;
1203
1204         rc = osd_inode_iteration(osd_oti_get(env), dev,
1205                                  OSD_OTABLE_IT_CACHE_SIZE, true);
1206         if (rc == SCRUB_IT_ALL)
1207                 it->ooi_all_cached = 1;
1208
1209         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1210                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1211                ooc->ooc_pos_preload, rc);
1212
1213         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1214                 spin_lock(&scrub->os_lock);
1215                 scrub->os_waiting = 0;
1216                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1217                 spin_unlock(&scrub->os_lock);
1218         }
1219
1220         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1221 }
1222
1223 static int osd_scrub_main(void *args)
1224 {
1225         struct lu_env         env;
1226         struct osd_device    *dev    = (struct osd_device *)args;
1227         struct osd_scrub     *scrub  = &dev->od_scrub;
1228         struct ptlrpc_thread *thread = &scrub->os_thread;
1229         struct super_block   *sb     = osd_sb(dev);
1230         int                   rc;
1231         ENTRY;
1232
1233         rc = lu_env_init(&env, LCT_LOCAL);
1234         if (rc != 0) {
1235                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
1236                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1237                 GOTO(noenv, rc);
1238         }
1239
1240         rc = osd_scrub_prep(dev);
1241         if (rc != 0) {
1242                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
1243                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1244                 GOTO(out, rc);
1245         }
1246
1247         if (!scrub->os_full_speed) {
1248                 struct l_wait_info lwi = { 0 };
1249                 struct osd_otable_it *it = dev->od_otable_it;
1250                 struct osd_otable_cache *ooc = &it->ooi_cache;
1251
1252                 l_wait_event(thread->t_ctl_waitq,
1253                              it->ooi_user_ready || !thread_is_running(thread),
1254                              &lwi);
1255                 if (unlikely(!thread_is_running(thread)))
1256                         GOTO(post, rc = 0);
1257
1258                 scrub->os_pos_current = ooc->ooc_pos_preload;
1259         }
1260
1261         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
1262                scrub->os_start_flags, scrub->os_pos_current);
1263
1264         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1265         if (unlikely(rc == SCRUB_IT_CRASH))
1266                 GOTO(out, rc = -EINVAL);
1267         GOTO(post, rc);
1268
1269 post:
1270         osd_scrub_post(scrub, rc);
1271         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
1272                rc, scrub->os_pos_current);
1273
1274 out:
1275         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
1276                 struct osd_inconsistent_item *oii;
1277
1278                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
1279                                      struct osd_inconsistent_item, oii_list);
1280                 cfs_list_del_init(&oii->oii_list);
1281                 OBD_FREE_PTR(oii);
1282         }
1283         lu_env_fini(&env);
1284
1285 noenv:
1286         spin_lock(&scrub->os_lock);
1287         thread_set_flags(thread, SVC_STOPPED);
1288         wake_up_all(&thread->t_ctl_waitq);
1289         spin_unlock(&scrub->os_lock);
1290         return rc;
1291 }
1292
1293 /* initial OI scrub */
1294
1295 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1296                          struct dentry *, filldir_t filldir);
1297
1298 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1299                                loff_t offset, __u64 ino, unsigned d_type);
1300 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1301                            loff_t offset, __u64 ino, unsigned d_type);
1302 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1303                            loff_t offset, __u64 ino, unsigned d_type);
1304
1305 static int
1306 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1307                      struct dentry *dentry, filldir_t filldir);
1308 static int
1309 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1310                   struct dentry *dentry, filldir_t filldir);
1311
1312 static int
1313 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1314                      struct dentry *dentry, filldir_t filldir);
1315
1316 enum osd_lf_flags {
1317         OLF_SCAN_SUBITEMS       = 0x0001,
1318         OLF_HIDE_FID            = 0x0002,
1319         OLF_SHOW_NAME           = 0x0004,
1320         OLF_NO_OI               = 0x0008,
1321 };
1322
1323 struct osd_lf_map {
1324         char            *olm_name;
1325         struct lu_fid    olm_fid;
1326         __u16            olm_flags;
1327         scandir_t        olm_scandir;
1328         filldir_t        olm_filldir;
1329 };
1330
1331 /* Add the new introduced local files in the list in the future. */
1332 static const struct osd_lf_map osd_lf_maps[] = {
1333         /* CATALOGS */
1334         { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1335                 NULL, NULL },
1336
1337         /* CONFIGS */
1338         { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1339                 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1340                 osd_ios_varfid_fill },
1341
1342         /* NIDTBL_VERSIONS */
1343         { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1344                 osd_ios_general_scan, osd_ios_varfid_fill },
1345
1346         /* PENDING */
1347         { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
1348
1349         /* ROOT */
1350         { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1351                 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1352
1353         /* changelog_catalog */
1354         { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1355
1356         /* changelog_users */
1357         { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1358
1359         /* fld */
1360         { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1361                 NULL, NULL },
1362
1363         /* last_rcvd */
1364         { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1365                 NULL, NULL },
1366
1367         /* lfsck_bookmark */
1368         { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
1369
1370         /* lov_objid */
1371         { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1372                 NULL, NULL },
1373
1374         /* lov_objseq */
1375         { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1376                 OLF_SHOW_NAME, NULL, NULL },
1377
1378         /* quota_master */
1379         { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1380                 osd_ios_general_scan, osd_ios_varfid_fill },
1381
1382         /* quota_slave */
1383         { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1384                 osd_ios_general_scan, osd_ios_varfid_fill },
1385
1386         /* seq_ctl */
1387         { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1388                 OLF_SHOW_NAME, NULL, NULL },
1389
1390         /* seq_srv */
1391         { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1392                 OLF_SHOW_NAME, NULL, NULL },
1393
1394         /* health_check */
1395         { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1396                 OLF_SHOW_NAME, NULL, NULL },
1397
1398         /* lfsck_namespace */
1399         { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
1400
1401         /* OBJECTS, upgrade from old device */
1402         { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1403
1404         /* lquota_v2.user, upgrade from old device */
1405         { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1406
1407         /* lquota_v2.group, upgrade from old device */
1408         { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1409
1410         /* LAST_GROUP, upgrade from old device */
1411         { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1412                 OLF_SHOW_NAME, NULL, NULL },
1413
1414         /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1415         { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1416                OLF_SHOW_NAME, NULL, NULL },
1417
1418         /* lost+found */
1419         { "lost+found", { 0, 0, 0 }, OLF_SCAN_SUBITEMS | OLF_NO_OI,
1420                 osd_ios_general_scan, osd_ios_lf_fill },
1421
1422         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1423 };
1424
1425 /* Add the new introduced files under .lustre/ in the list in the future. */
1426 static const struct osd_lf_map osd_dl_maps[] = {
1427         /* .lustre/fid */
1428         { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1429                 NULL, NULL },
1430
1431         /* .lustre/lost+found */
1432         { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1433                 NULL, NULL },
1434
1435         { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1436 };
1437
1438 struct osd_ios_item {
1439         cfs_list_t       oii_list;
1440         struct dentry   *oii_dentry;
1441         scandir_t        oii_scandir;
1442         filldir_t        oii_filldir;
1443 };
1444
1445 struct osd_ios_filldir_buf {
1446 #ifdef HAVE_DIR_CONTEXT
1447         /* please keep it as first member */
1448         struct dir_context       ctx;
1449 #endif
1450         struct osd_thread_info  *oifb_info;
1451         struct osd_device       *oifb_dev;
1452         struct dentry           *oifb_dentry;
1453 };
1454
1455 static inline struct dentry *
1456 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1457 {
1458         struct dentry *dentry;
1459
1460         CDEBUG(D_LFSCK, "init lookup one: parent = %.*s, name = %.*s\n",
1461                parent->d_name.len, parent->d_name.name, namelen, name);
1462
1463         dentry = ll_lookup_one_len(name, parent, namelen);
1464         if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1465                 dput(dentry);
1466                 return ERR_PTR(-ENOENT);
1467         }
1468
1469         return dentry;
1470 }
1471
1472 static int
1473 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1474                  scandir_t scandir, filldir_t filldir)
1475 {
1476         struct osd_ios_item *item;
1477         ENTRY;
1478
1479         OBD_ALLOC_PTR(item);
1480         if (item == NULL)
1481                 RETURN(-ENOMEM);
1482
1483         CFS_INIT_LIST_HEAD(&item->oii_list);
1484         item->oii_dentry = dget(dentry);
1485         item->oii_scandir = scandir;
1486         item->oii_filldir = filldir;
1487         cfs_list_add_tail(&item->oii_list, &dev->od_ios_list);
1488
1489         RETURN(0);
1490 }
1491
1492 /**
1493  * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1494  *
1495  * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1496  * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1497  * reference the inode, or fixed if it is missing or references another inode.
1498  */
1499 static int
1500 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1501                  struct inode *inode, const struct lu_fid *fid, int flags)
1502 {
1503         struct lustre_mdt_attrs *lma    = &info->oti_mdt_attrs;
1504         struct osd_inode_id     *id     = &info->oti_id;
1505         struct osd_inode_id     *id2    = &info->oti_id2;
1506         struct osd_scrub        *scrub  = &dev->od_scrub;
1507         struct scrub_file       *sf     = &scrub->os_file;
1508         struct lu_fid            tfid;
1509         int                      rc;
1510         ENTRY;
1511
1512         CDEBUG(D_LFSCK, "init scan one: ino = %ld\n", inode->i_ino);
1513
1514         rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1515         if (rc != 0 && rc != -ENODATA)
1516                 RETURN(rc);
1517
1518         osd_id_gen(id, inode->i_ino, inode->i_generation);
1519         if (rc == -ENODATA) {
1520                 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1521                         lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1522                 else
1523                         tfid = *fid;
1524                 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1525                 if (rc != 0)
1526                         RETURN(rc);
1527         } else {
1528                 if (lma->lma_compat & LMAC_NOT_IN_OI)
1529                         RETURN(0);
1530
1531                 tfid = lma->lma_self_fid;
1532         }
1533
1534         rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1535         if (rc != 0) {
1536                 if (rc != -ENOENT)
1537                         RETURN(rc);
1538
1539                 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1540                                                DTO_INDEX_INSERT, true, 0);
1541                 if (rc > 0)
1542                         rc = 0;
1543
1544                 RETURN(rc);
1545         }
1546
1547         if (osd_id_eq_strict(id, id2))
1548                 RETURN(0);
1549
1550         if (!(sf->sf_flags & SF_INCONSISTENT)) {
1551                 osd_scrub_file_reset(scrub,
1552                                      LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1553                                      SF_INCONSISTENT);
1554                 rc = osd_scrub_file_store(scrub);
1555                 if (rc != 0)
1556                         RETURN(rc);
1557         }
1558
1559         rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1560                                        DTO_INDEX_UPDATE, true, 0);
1561         if (rc > 0)
1562                 rc = 0;
1563
1564         RETURN(rc);
1565 }
1566
1567 /**
1568  * It scans the /lost+found, and for the OST-object (with filter_fid
1569  * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1570  */
1571 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1572                            loff_t offset, __u64 ino, unsigned d_type)
1573 {
1574         struct osd_ios_filldir_buf *fill_buf = buf;
1575         struct osd_thread_info     *info     = fill_buf->oifb_info;
1576         struct osd_device          *dev      = fill_buf->oifb_dev;
1577         struct lu_fid              *fid      = &info->oti_fid;
1578         struct osd_scrub           *scrub    = &dev->od_scrub;
1579         struct dentry              *parent   = fill_buf->oifb_dentry;
1580         struct dentry              *child;
1581         struct inode               *dir      = parent->d_inode;
1582         struct inode               *inode;
1583         int                         rc;
1584         ENTRY;
1585
1586         /* skip any '.' started names */
1587         if (name[0] == '.')
1588                 RETURN(0);
1589
1590         scrub->os_lf_scanned++;
1591         child = osd_ios_lookup_one_len(name, parent, namelen);
1592         if (IS_ERR(child)) {
1593                 CWARN("%s: cannot lookup child '%.*s': rc = %d\n",
1594                       osd_name(dev), namelen, name, (int)PTR_ERR(child));
1595                 RETURN(0);
1596         }
1597
1598         inode = child->d_inode;
1599         if (S_ISDIR(inode->i_mode)) {
1600                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1601                                       osd_ios_lf_fill);
1602                 if (rc != 0)
1603                         CWARN("%s: cannot add child '%.*s': rc = %d\n",
1604                               osd_name(dev), namelen, name, rc);
1605                 GOTO(put, rc);
1606         }
1607
1608         if (!S_ISREG(inode->i_mode))
1609                 GOTO(put, rc = 0);
1610
1611         rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1612         if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1613                 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1614                 if (rc == 0) {
1615                         CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1616                                "/lost+found.\n", namelen, name, PFID(fid));
1617                         scrub->os_lf_repaired++;
1618                 } else {
1619                         CWARN("%s: cannot rename for '%.*s' "DFID": rc = %d\n",
1620                               osd_name(dev), namelen, name, PFID(fid), rc);
1621                 }
1622         }
1623
1624         /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1625          *      visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1626          *      can process them in furtuer. */
1627
1628         GOTO(put, rc);
1629
1630 put:
1631         if (rc < 0)
1632                 scrub->os_lf_failed++;
1633         dput(child);
1634         /* skip the failure to make the scanning to continue. */
1635         return 0;
1636 }
1637
1638 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1639                                loff_t offset, __u64 ino, unsigned d_type)
1640 {
1641         struct osd_ios_filldir_buf *fill_buf = buf;
1642         struct osd_device          *dev      = fill_buf->oifb_dev;
1643         struct dentry              *child;
1644         int                         rc;
1645         ENTRY;
1646
1647         /* skip any '.' started names */
1648         if (name[0] == '.')
1649                 RETURN(0);
1650
1651         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1652         if (IS_ERR(child))
1653                 RETURN(PTR_ERR(child));
1654
1655         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1656                               NULL, 0);
1657         if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1658                 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1659                                       osd_ios_varfid_fill);
1660         dput(child);
1661
1662         RETURN(rc);
1663 }
1664
1665 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1666                            loff_t offset, __u64 ino, unsigned d_type)
1667 {
1668         struct osd_ios_filldir_buf *fill_buf = buf;
1669         struct osd_device          *dev      = fill_buf->oifb_dev;
1670         const struct osd_lf_map    *map;
1671         struct dentry              *child;
1672         int                         rc       = 0;
1673         ENTRY;
1674
1675         /* skip any '.' started names */
1676         if (name[0] == '.')
1677                 RETURN(0);
1678
1679         for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1680                 if (strlen(map->olm_name) != namelen)
1681                         continue;
1682
1683                 if (strncmp(map->olm_name, name, namelen) == 0)
1684                         break;
1685         }
1686
1687         if (map->olm_name == NULL)
1688                 RETURN(0);
1689
1690         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1691         if (IS_ERR(child))
1692                 RETURN(PTR_ERR(child));
1693
1694         rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1695                               &map->olm_fid, map->olm_flags);
1696         dput(child);
1697
1698         RETURN(rc);
1699 }
1700
1701 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1702                              loff_t offset, __u64 ino, unsigned d_type)
1703 {
1704         struct osd_ios_filldir_buf *fill_buf = buf;
1705         struct osd_device          *dev      = fill_buf->oifb_dev;
1706         const struct osd_lf_map    *map;
1707         struct dentry              *child;
1708         int                         rc       = 0;
1709         ENTRY;
1710
1711         /* skip any '.' started names */
1712         if (name[0] == '.')
1713                 RETURN(0);
1714
1715         for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1716                 if (strlen(map->olm_name) != namelen)
1717                         continue;
1718
1719                 if (strncmp(map->olm_name, name, namelen) == 0)
1720                         break;
1721         }
1722
1723         if (map->olm_name == NULL)
1724                 RETURN(0);
1725
1726         child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1727         if (IS_ERR(child))
1728                 RETURN(PTR_ERR(child));
1729
1730         if (!(map->olm_flags & OLF_NO_OI))
1731                 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1732                                       &map->olm_fid, map->olm_flags);
1733         if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1734                 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1735                                       map->olm_filldir);
1736         dput(child);
1737
1738         RETURN(rc);
1739 }
1740
1741 static int
1742 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1743                      struct dentry *dentry, filldir_t filldir)
1744 {
1745         struct osd_ios_filldir_buf    buf   = {
1746 #ifdef HAVE_DIR_CONTEXT
1747                                                 .ctx.actor = filldir,
1748 #endif
1749                                                 .oifb_info = info,
1750                                                 .oifb_dev = dev,
1751                                                 .oifb_dentry = dentry };
1752         struct file                  *filp  = &info->oti_it_ea.oie_file;
1753         struct inode                 *inode = dentry->d_inode;
1754         const struct file_operations *fops  = inode->i_fop;
1755         int                           rc;
1756         ENTRY;
1757
1758         LASSERT(filldir != NULL);
1759
1760         filp->f_pos = 0;
1761         filp->f_dentry = dentry;
1762         filp->f_mode = FMODE_64BITHASH;
1763         filp->f_mapping = inode->i_mapping;
1764         filp->f_op = fops;
1765         filp->private_data = NULL;
1766         set_file_inode(filp, inode);
1767
1768 #ifdef HAVE_DIR_CONTEXT
1769         buf.ctx.pos = filp->f_pos;
1770         rc = fops->iterate(filp, &buf.ctx);
1771         filp->f_pos = buf.ctx.pos;
1772 #else
1773         rc = fops->readdir(filp, &buf, filldir);
1774 #endif
1775         fops->release(inode, filp);
1776
1777         RETURN(rc);
1778 }
1779
1780 static int
1781 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1782                   struct dentry *dentry, filldir_t filldir)
1783 {
1784         struct osd_scrub  *scrub  = &dev->od_scrub;
1785         struct scrub_file *sf     = &scrub->os_file;
1786         struct dentry     *child;
1787         int                rc;
1788         ENTRY;
1789
1790         /* It is existing MDT0 device. We only allow the case of object without
1791          * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
1792          * can generate IGIF mode FID for the object and related OI mapping. If
1793          * it is on other MDTs, then becuase file-level backup/restore, related
1794          * OI mapping may be invalid already, we do not know which is the right
1795          * FID for the object. We only allow IGIF objects to reside on the MDT0.
1796          *
1797          * XXX: For the case of object on non-MDT0 device with neither LMA nor
1798          *      "fid" xattr, then something crashed. We cannot re-generate the
1799          *      FID directly, instead, the OI scrub will scan the OI structure
1800          *      and try to re-generate the LMA from the OI mapping. But if the
1801          *      OI mapping crashed or lost also, then we have to give up under
1802          *      double failure cases. */
1803         scrub->os_convert_igif = 1;
1804         child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1805                                        strlen(dot_lustre_name));
1806         if (IS_ERR(child)) {
1807                 rc = PTR_ERR(child);
1808                 if (rc == -ENOENT) {
1809                         /* It is 1.8 MDT device. */
1810                         if (!(sf->sf_flags & SF_UPGRADE)) {
1811                                 osd_scrub_file_reset(scrub,
1812                                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1813                                         SF_UPGRADE);
1814                                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
1815                                 rc = osd_scrub_file_store(scrub);
1816                         } else {
1817                                 rc = 0;
1818                         }
1819                 }
1820         } else {
1821                 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
1822                  * so the client will get IGIF for the ".lustre" object when
1823                  * the MDT restart.
1824                  *
1825                  * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
1826                  * it does not know whether there are some old clients cached
1827                  * the ".lustre" IGIF during the upgrading. Two choices:
1828                  *
1829                  * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
1830                  *    It will allow the old connected clients to access the
1831                  *    ".lustre" with cached IGIF. But it will cause others
1832                  *    on the MDT failed to check "fid_is_dot_lustre()".
1833                  *
1834                  * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
1835                  *    for ".lustre" in spite of whether there are some clients
1836                  *    cached the ".lustre" IGIF or not. It enables the check
1837                  *    "fid_is_dot_lustre()" on the MDT, although it will cause
1838                  *    that the old connected clients cannot access the ".lustre"
1839                  *    with the cached IGIF.
1840                  *
1841                  * Usually, it is rare case for the old connected clients
1842                  * to access the ".lustre" with cached IGIF. So we prefer
1843                  * to the solution 2). */
1844                 rc = osd_ios_scan_one(info, dev, child->d_inode,
1845                                       &LU_DOT_LUSTRE_FID, 0);
1846                 if (rc == 0)
1847                         rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1848                                               osd_ios_dl_fill);
1849                 dput(child);
1850         }
1851
1852         RETURN(rc);
1853 }
1854
1855 static int
1856 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1857                      struct dentry *dentry, filldir_t filldir)
1858 {
1859         struct osd_scrub  *scrub  = &dev->od_scrub;
1860         struct scrub_file *sf     = &scrub->os_file;
1861         struct dentry     *child;
1862         int                rc;
1863         ENTRY;
1864
1865         if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
1866                 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
1867                 rc = osd_scrub_file_store(scrub);
1868                 if (rc != 0)
1869                         RETURN(rc);
1870         }
1871
1872         child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1873         if (!IS_ERR(child)) {
1874                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1875                 dput(child);
1876         } else {
1877                 rc = PTR_ERR(child);
1878         }
1879
1880         if (rc != 0 && rc != -ENOENT)
1881                 RETURN(rc);
1882
1883         child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1884         if (!IS_ERR(child)) {
1885                 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1886                 dput(child);
1887         } else {
1888                 rc = PTR_ERR(child);
1889         }
1890
1891         if (rc == -ENOENT)
1892                 rc = 0;
1893
1894         RETURN(rc);
1895 }
1896
1897 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1898                                 struct osd_device *dev)
1899 {
1900         struct osd_ios_item     *item    = NULL;
1901         scandir_t                scandir = osd_ios_general_scan;
1902         filldir_t                filldir = osd_ios_root_fill;
1903         struct dentry           *dentry  = osd_sb(dev)->s_root;
1904         const struct osd_lf_map *map     = osd_lf_maps;
1905         int                      rc;
1906         ENTRY;
1907
1908         /* Lookup IGIF in OI by force for initial OI scrub. */
1909         dev->od_igif_inoi = 1;
1910
1911         while (1) {
1912                 rc = scandir(info, dev, dentry, filldir);
1913                 if (item != NULL) {
1914                         dput(item->oii_dentry);
1915                         OBD_FREE_PTR(item);
1916                 }
1917
1918                 if (rc != 0)
1919                         break;
1920
1921                 if (cfs_list_empty(&dev->od_ios_list))
1922                         break;
1923
1924                 item = cfs_list_entry(dev->od_ios_list.next,
1925                                       struct osd_ios_item, oii_list);
1926                 cfs_list_del_init(&item->oii_list);
1927
1928                 LASSERT(item->oii_scandir != NULL);
1929                 scandir = item->oii_scandir;
1930                 filldir = item->oii_filldir;
1931                 dentry = item->oii_dentry;
1932         }
1933
1934         while (!cfs_list_empty(&dev->od_ios_list)) {
1935                 item = cfs_list_entry(dev->od_ios_list.next,
1936                                       struct osd_ios_item, oii_list);
1937                 cfs_list_del_init(&item->oii_list);
1938                 dput(item->oii_dentry);
1939                 OBD_FREE_PTR(item);
1940         }
1941
1942         if (rc != 0)
1943                 RETURN(rc);
1944
1945         /* There maybe the case that the object has been removed, but its OI
1946          * mapping is still in the OI file, such as the "CATALOGS" after MDT
1947          * file-level backup/restore. So here cleanup the stale OI mappings. */
1948         while (map->olm_name != NULL) {
1949                 struct dentry *child;
1950
1951                 if (fid_is_zero(&map->olm_fid)) {
1952                         map++;
1953                         continue;
1954                 }
1955
1956                 child = osd_ios_lookup_one_len(map->olm_name,
1957                                                osd_sb(dev)->s_root,
1958                                                strlen(map->olm_name));
1959                 if (!IS_ERR(child))
1960                         dput(child);
1961                 else if (PTR_ERR(child) == -ENOENT)
1962                         osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
1963                                                   NULL, DTO_INDEX_DELETE,
1964                                                   true, 0);
1965                 map++;
1966         }
1967
1968         RETURN(0);
1969 }
1970
1971 char *osd_lf_fid2name(const struct lu_fid *fid)
1972 {
1973         const struct osd_lf_map *map = osd_lf_maps;
1974
1975         while (map->olm_name != NULL) {
1976                 if (!lu_fid_eq(fid, &map->olm_fid)) {
1977                         map++;
1978                         continue;
1979                 }
1980
1981                 if (map->olm_flags & OLF_SHOW_NAME)
1982                         return map->olm_name;
1983                 else
1984                         return "";
1985         }
1986
1987         return NULL;
1988 }
1989
1990 /* OI scrub start/stop */
1991
1992 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
1993 {
1994         struct osd_scrub     *scrub  = &dev->od_scrub;
1995         struct ptlrpc_thread *thread = &scrub->os_thread;
1996         struct l_wait_info    lwi    = { 0 };
1997         struct task_struct   *task;
1998         int                   rc;
1999         ENTRY;
2000
2001 again:
2002         /* os_lock: sync status between stop and scrub thread */
2003         spin_lock(&scrub->os_lock);
2004         if (thread_is_running(thread)) {
2005                 spin_unlock(&scrub->os_lock);
2006                 RETURN(-EALREADY);
2007         } else if (unlikely(thread_is_stopping(thread))) {
2008                 spin_unlock(&scrub->os_lock);
2009                 l_wait_event(thread->t_ctl_waitq,
2010                              thread_is_stopped(thread),
2011                              &lwi);
2012                 goto again;
2013         }
2014         spin_unlock(&scrub->os_lock);
2015
2016         if (scrub->os_file.sf_status == SS_COMPLETED)
2017                 flags |= SS_RESET;
2018
2019         scrub->os_start_flags = flags;
2020         thread_set_flags(thread, 0);
2021         task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2022         if (IS_ERR(task)) {
2023                 rc = PTR_ERR(task);
2024                 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2025                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
2026                 RETURN(rc);
2027         }
2028
2029         l_wait_event(thread->t_ctl_waitq,
2030                      thread_is_running(thread) || thread_is_stopped(thread),
2031                      &lwi);
2032
2033         RETURN(0);
2034 }
2035
2036 int osd_scrub_start(struct osd_device *dev)
2037 {
2038         int rc;
2039         ENTRY;
2040
2041         /* od_otable_mutex: prevent curcurrent start/stop */
2042         mutex_lock(&dev->od_otable_mutex);
2043         rc = do_osd_scrub_start(dev, SS_AUTO);
2044         mutex_unlock(&dev->od_otable_mutex);
2045
2046         RETURN(rc == -EALREADY ? 0 : rc);
2047 }
2048
2049 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2050 {
2051         struct ptlrpc_thread *thread = &scrub->os_thread;
2052         struct l_wait_info    lwi    = { 0 };
2053
2054         /* os_lock: sync status between stop and scrub thread */
2055         spin_lock(&scrub->os_lock);
2056         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2057                 thread_set_flags(thread, SVC_STOPPING);
2058                 spin_unlock(&scrub->os_lock);
2059                 wake_up_all(&thread->t_ctl_waitq);
2060                 l_wait_event(thread->t_ctl_waitq,
2061                              thread_is_stopped(thread),
2062                              &lwi);
2063                 /* Do not skip the last lock/unlock, which can guarantee that
2064                  * the caller cannot return until the OI scrub thread exit. */
2065                 spin_lock(&scrub->os_lock);
2066         }
2067         spin_unlock(&scrub->os_lock);
2068 }
2069
2070 static void osd_scrub_stop(struct osd_device *dev)
2071 {
2072         /* od_otable_mutex: prevent curcurrent start/stop */
2073         mutex_lock(&dev->od_otable_mutex);
2074         dev->od_scrub.os_paused = 1;
2075         do_osd_scrub_stop(&dev->od_scrub);
2076         mutex_unlock(&dev->od_otable_mutex);
2077 }
2078
2079 /* OI scrub setup/cleanup */
2080
2081 static const char osd_scrub_name[] = "OI_scrub";
2082
2083 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2084 {
2085         struct osd_thread_info     *info   = osd_oti_get(env);
2086         struct osd_scrub           *scrub  = &dev->od_scrub;
2087         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
2088         struct scrub_file          *sf     = &scrub->os_file;
2089         struct super_block         *sb     = osd_sb(dev);
2090         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
2091         struct lvfs_run_ctxt        saved;
2092         struct file                *filp;
2093         struct inode               *inode;
2094         struct lu_fid              *fid    = &info->oti_fid;
2095         int                         dirty  = 0;
2096         int                         rc     = 0;
2097         ENTRY;
2098
2099         memset(scrub, 0, sizeof(*scrub));
2100         OBD_SET_CTXT_MAGIC(ctxt);
2101         ctxt->pwdmnt = dev->od_mnt;
2102         ctxt->pwd = dev->od_mnt->mnt_root;
2103         ctxt->fs = get_ds();
2104
2105         init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2106         init_rwsem(&scrub->os_rwsem);
2107         spin_lock_init(&scrub->os_lock);
2108         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2109
2110         push_ctxt(&saved, ctxt);
2111         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2112         if (IS_ERR(filp)) {
2113                 pop_ctxt(&saved, ctxt);
2114                 RETURN(PTR_ERR(filp));
2115         }
2116
2117         inode = filp->f_dentry->d_inode;
2118         /* 'What the @fid is' is not imporatant, because the object
2119          * has no OI mapping, and only is visible inside the OSD.*/
2120         lu_igif_build(fid, inode->i_ino, inode->i_generation);
2121         rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2122         if (rc != 0) {
2123                 filp_close(filp, 0);
2124                 pop_ctxt(&saved, ctxt);
2125                 RETURN(rc);
2126         }
2127
2128         scrub->os_inode = igrab(inode);
2129         filp_close(filp, 0);
2130         pop_ctxt(&saved, ctxt);
2131
2132         rc = osd_scrub_file_load(scrub);
2133         if (rc == -ENOENT) {
2134                 osd_scrub_file_init(scrub, es->s_uuid);
2135                 /* If the "/O" dir does not exist when mount (indicated by
2136                  * osd_device::od_maybe_new), neither for the "/OI_scrub",
2137                  * then it is quite probably that the device is a new one,
2138                  * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2139                  *
2140                  * For the rare case that "/O" and "OI_scrub" both lost on
2141                  * an old device, it can be found and cleared later.
2142                  *
2143                  * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2144                  * need to check "filter_fid_old" and to convert it to
2145                  * "filter_fid" for each object, and all the IGIF should
2146                  * have their FID mapping in OI files already. */
2147                 if (dev->od_maybe_new)
2148                         sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2149                 dirty = 1;
2150         } else if (rc != 0) {
2151                 RETURN(rc);
2152         } else {
2153                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2154                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2155                         dirty = 1;
2156                 } else if (sf->sf_status == SS_SCANNING) {
2157                         sf->sf_status = SS_CRASHED;
2158                         dirty = 1;
2159                 }
2160         }
2161
2162         if (sf->sf_pos_last_checkpoint != 0)
2163                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2164         else
2165                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2166
2167         if (dirty != 0) {
2168                 rc = osd_scrub_file_store(scrub);
2169                 if (rc != 0)
2170                         RETURN(rc);
2171         }
2172
2173         /* Initialize OI files. */
2174         rc = osd_oi_init(info, dev);
2175         if (rc < 0)
2176                 RETURN(rc);
2177
2178         rc = osd_initial_OI_scrub(info, dev);
2179         if (rc == 0) {
2180                 if (sf->sf_flags & SF_UPGRADE ||
2181                     !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2182                       sf->sf_success_count > 0)) {
2183                         dev->od_igif_inoi = 0;
2184                         dev->od_check_ff = dev->od_is_ost;
2185                 } else {
2186                         dev->od_igif_inoi = 1;
2187                         dev->od_check_ff = 0;
2188                 }
2189
2190                 if (sf->sf_flags & SF_INCONSISTENT)
2191                         /* The 'od_igif_inoi' will be set under the
2192                          * following cases:
2193                          * 1) new created system, or
2194                          * 2) restored from file-level backup, or
2195                          * 3) the upgrading completed.
2196                          *
2197                          * The 'od_igif_inoi' may be cleared by OI scrub
2198                          * later if found that the system is upgrading. */
2199                         dev->od_igif_inoi = 1;
2200
2201                 if (!dev->od_noscrub &&
2202                     ((sf->sf_status == SS_PAUSED) ||
2203                      (sf->sf_status == SS_CRASHED &&
2204                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2205                                       SF_UPGRADE | SF_AUTO)) ||
2206                      (sf->sf_status == SS_INIT &&
2207                       sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2208                                       SF_UPGRADE))))
2209                         rc = osd_scrub_start(dev);
2210         }
2211
2212         /* it is possible that dcache entries may keep objects after they are
2213          * deleted by OSD. While it looks safe this can cause object data to
2214          * stay until umount causing failures in tests calculating free space,
2215          * e.g. replay-ost-single. Since those dcache entries are not used
2216          * anymore let's just free them after use here */
2217         shrink_dcache_sb(sb);
2218
2219         RETURN(rc);
2220 }
2221
2222 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2223 {
2224         struct osd_scrub *scrub = &dev->od_scrub;
2225
2226         LASSERT(dev->od_otable_it == NULL);
2227
2228         if (scrub->os_inode != NULL) {
2229                 osd_scrub_stop(dev);
2230                 iput(scrub->os_inode);
2231                 scrub->os_inode = NULL;
2232         }
2233         if (dev->od_oi_table != NULL)
2234                 osd_oi_fini(osd_oti_get(env), dev);
2235 }
2236
2237 /* object table based iteration APIs */
2238
2239 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2240                                        struct dt_object *dt, __u32 attr,
2241                                        struct lustre_capa *capa)
2242 {
2243         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2244         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2245         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
2246         struct osd_scrub       *scrub = &dev->od_scrub;
2247         struct osd_otable_it   *it;
2248         __u32                   start = 0;
2249         int                     rc;
2250         ENTRY;
2251
2252         /* od_otable_mutex: prevent curcurrent init/fini */
2253         mutex_lock(&dev->od_otable_mutex);
2254         if (dev->od_otable_it != NULL)
2255                 GOTO(out, it = ERR_PTR(-EALREADY));
2256
2257         OBD_ALLOC_PTR(it);
2258         if (it == NULL)
2259                 GOTO(out, it = ERR_PTR(-ENOMEM));
2260
2261         dev->od_otable_it = it;
2262         it->ooi_dev = dev;
2263         it->ooi_cache.ooc_consumer_idx = -1;
2264         if (flags & DOIF_OUTUSED)
2265                 it->ooi_used_outside = 1;
2266
2267         if (flags & DOIF_RESET)
2268                 start |= SS_RESET;
2269
2270         if (valid & DOIV_ERROR_HANDLE) {
2271                 if (flags & DOIF_FAILOUT)
2272                         start |= SS_SET_FAILOUT;
2273                 else
2274                         start |= SS_CLEAR_FAILOUT;
2275         }
2276
2277         if (valid & DOIV_DRYRUN) {
2278                 if (flags & DOIF_DRYRUN)
2279                         start |= SS_SET_DRYRUN;
2280                 else
2281                         start |= SS_CLEAR_DRYRUN;
2282         }
2283
2284         rc = do_osd_scrub_start(dev, start);
2285         if (rc < 0 && rc != -EALREADY) {
2286                 dev->od_otable_it = NULL;
2287                 OBD_FREE_PTR(it);
2288                 GOTO(out, it = ERR_PTR(rc));
2289         }
2290
2291         it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2292
2293         GOTO(out, it);
2294
2295 out:
2296         mutex_unlock(&dev->od_otable_mutex);
2297         return (struct dt_it *)it;
2298 }
2299
2300 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2301 {
2302         struct osd_otable_it *it  = (struct osd_otable_it *)di;
2303         struct osd_device    *dev = it->ooi_dev;
2304
2305         /* od_otable_mutex: prevent curcurrent init/fini */
2306         mutex_lock(&dev->od_otable_mutex);
2307         do_osd_scrub_stop(&dev->od_scrub);
2308         LASSERT(dev->od_otable_it == it);
2309
2310         dev->od_otable_it = NULL;
2311         mutex_unlock(&dev->od_otable_mutex);
2312         OBD_FREE_PTR(it);
2313 }
2314
2315 static int osd_otable_it_get(const struct lu_env *env,
2316                              struct dt_it *di, const struct dt_key *key)
2317 {
2318         return 0;
2319 }
2320
2321 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2322 {
2323 }
2324
2325 static inline int
2326 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2327 {
2328         spin_lock(&scrub->os_lock);
2329         if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2330             scrub->os_waiting ||
2331             !thread_is_running(&scrub->os_thread))
2332                 it->ooi_waiting = 0;
2333         else
2334                 it->ooi_waiting = 1;
2335         spin_unlock(&scrub->os_lock);
2336
2337         return !it->ooi_waiting;
2338 }
2339
2340 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2341 {
2342         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
2343         struct osd_device       *dev    = it->ooi_dev;
2344         struct osd_scrub        *scrub  = &dev->od_scrub;
2345         struct osd_otable_cache *ooc    = &it->ooi_cache;
2346         struct ptlrpc_thread    *thread = &scrub->os_thread;
2347         struct l_wait_info       lwi    = { 0 };
2348         int                      rc;
2349         ENTRY;
2350
2351         LASSERT(it->ooi_user_ready);
2352
2353 again:
2354         if (!thread_is_running(thread) && !it->ooi_used_outside)
2355                 RETURN(1);
2356
2357         if (ooc->ooc_cached_items > 0) {
2358                 ooc->ooc_cached_items--;
2359                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2360                                         ~OSD_OTABLE_IT_CACHE_MASK;
2361                 RETURN(0);
2362         }
2363
2364         if (it->ooi_all_cached) {
2365                 l_wait_event(thread->t_ctl_waitq,
2366                              !thread_is_running(thread),
2367                              &lwi);
2368                 RETURN(1);
2369         }
2370
2371         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2372                 spin_lock(&scrub->os_lock);
2373                 scrub->os_waiting = 0;
2374                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2375                 spin_unlock(&scrub->os_lock);
2376         }
2377
2378         if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2379                 l_wait_event(thread->t_ctl_waitq,
2380                              osd_otable_it_wakeup(scrub, it),
2381                              &lwi);
2382
2383         if (!thread_is_running(thread) && !it->ooi_used_outside)
2384                 RETURN(1);
2385
2386         rc = osd_otable_it_preload(env, it);
2387         if (rc >= 0)
2388                 goto again;
2389
2390         RETURN(rc);
2391 }
2392
2393 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2394                                         const struct dt_it *di)
2395 {
2396         return NULL;
2397 }
2398
2399 static int osd_otable_it_key_size(const struct lu_env *env,
2400                                   const struct dt_it *di)
2401 {
2402         return sizeof(__u64);
2403 }
2404
2405 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2406                              struct dt_rec *rec, __u32 attr)
2407 {
2408         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2409         struct osd_otable_cache *ooc = &it->ooi_cache;
2410
2411         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2412
2413         /* Filter out Invald FID already. */
2414         LASSERTF(fid_is_sane((struct lu_fid *)rec),
2415                  "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2416                  PFID((struct lu_fid *)rec),
2417                  ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2418
2419         return 0;
2420 }
2421
2422 static __u64 osd_otable_it_store(const struct lu_env *env,
2423                                  const struct dt_it *di)
2424 {
2425         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
2426         struct osd_otable_cache *ooc = &it->ooi_cache;
2427         __u64                    hash;
2428
2429         if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2430                 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2431         else
2432                 hash = ooc->ooc_pos_preload;
2433         return hash;
2434 }
2435
2436 /**
2437  * Set the OSD layer iteration start position as the specified hash.
2438  */
2439 static int osd_otable_it_load(const struct lu_env *env,
2440                               const struct dt_it *di, __u64 hash)
2441 {
2442         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
2443         struct osd_device       *dev   = it->ooi_dev;
2444         struct osd_otable_cache *ooc   = &it->ooi_cache;
2445         struct osd_scrub        *scrub = &dev->od_scrub;
2446         int                      rc;
2447         ENTRY;
2448
2449         /* Forbid to set iteration position after iteration started. */
2450         if (it->ooi_user_ready)
2451                 RETURN(-EPERM);
2452
2453         if (hash > OSD_OTABLE_MAX_HASH)
2454                 hash = OSD_OTABLE_MAX_HASH;
2455
2456         ooc->ooc_pos_preload = hash;
2457         if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2458                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2459
2460         it->ooi_user_ready = 1;
2461         if (!scrub->os_full_speed)
2462                 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2463
2464         /* Unplug OSD layer iteration by the first next() call. */
2465         rc = osd_otable_it_next(env, (struct dt_it *)it);
2466
2467         RETURN(rc);
2468 }
2469
2470 static int osd_otable_it_key_rec(const struct lu_env *env,
2471                                  const struct dt_it *di, void *key_rec)
2472 {
2473         return 0;
2474 }
2475
2476 const struct dt_index_operations osd_otable_ops = {
2477         .dio_it = {
2478                 .init     = osd_otable_it_init,
2479                 .fini     = osd_otable_it_fini,
2480                 .get      = osd_otable_it_get,
2481                 .put      = osd_otable_it_put,
2482                 .next     = osd_otable_it_next,
2483                 .key      = osd_otable_it_key,
2484                 .key_size = osd_otable_it_key_size,
2485                 .rec      = osd_otable_it_rec,
2486                 .store    = osd_otable_it_store,
2487                 .load     = osd_otable_it_load,
2488                 .key_rec  = osd_otable_it_key_rec,
2489         }
2490 };
2491
2492 /* high priority inconsistent items list APIs */
2493
2494 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2495                    int insert)
2496 {
2497         struct osd_inconsistent_item *oii;
2498         struct osd_scrub             *scrub  = &dev->od_scrub;
2499         struct ptlrpc_thread         *thread = &scrub->os_thread;
2500         int                           wakeup = 0;
2501         ENTRY;
2502
2503         OBD_ALLOC_PTR(oii);
2504         if (unlikely(oii == NULL))
2505                 RETURN(-ENOMEM);
2506
2507         CFS_INIT_LIST_HEAD(&oii->oii_list);
2508         oii->oii_cache = *oic;
2509         oii->oii_insert = insert;
2510
2511         spin_lock(&scrub->os_lock);
2512         if (unlikely(!thread_is_running(thread))) {
2513                 spin_unlock(&scrub->os_lock);
2514                 OBD_FREE_PTR(oii);
2515                 RETURN(-EAGAIN);
2516         }
2517
2518         if (cfs_list_empty(&scrub->os_inconsistent_items))
2519                 wakeup = 1;
2520         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2521         spin_unlock(&scrub->os_lock);
2522
2523         if (wakeup != 0)
2524                 wake_up_all(&thread->t_ctl_waitq);
2525
2526         RETURN(0);
2527 }
2528
2529 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2530                    struct osd_inode_id *id)
2531 {
2532         struct osd_scrub             *scrub = &dev->od_scrub;
2533         struct osd_inconsistent_item *oii;
2534         ENTRY;
2535
2536         spin_lock(&scrub->os_lock);
2537         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2538                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2539                         *id = oii->oii_cache.oic_lid;
2540                         spin_unlock(&scrub->os_lock);
2541                         RETURN(0);
2542                 }
2543         }
2544         spin_unlock(&scrub->os_lock);
2545
2546         RETURN(-ENOENT);
2547 }
2548
2549 /* OI scrub dump */
2550
2551 static const char *scrub_status_names[] = {
2552         "init",
2553         "scanning",
2554         "completed",
2555         "failed",
2556         "stopped",
2557         "paused",
2558         "crashed",
2559         NULL
2560 };
2561
2562 static const char *scrub_flags_names[] = {
2563         "recreated",
2564         "inconsistent",
2565         "auto",
2566         "upgrade",
2567         NULL
2568 };
2569
2570 static const char *scrub_param_names[] = {
2571         "failout",
2572         "dryrun",
2573         NULL
2574 };
2575
2576 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2577                            const char *prefix)
2578 {
2579         int flag;
2580         int rc;
2581         int i;
2582
2583         rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2584         if (rc < 0)
2585                 return rc;
2586
2587         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2588                 if (flag & bits) {
2589                         bits &= ~flag;
2590                         rc = seq_printf(m, "%s%c", names[i],
2591                                         bits != 0 ? ',' : '\n');
2592                         if (rc < 0)
2593                                 return rc;
2594                 }
2595         }
2596         return 0;
2597 }
2598
2599 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2600 {
2601         int rc;
2602
2603         if (time != 0)
2604                 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2605                               cfs_time_current_sec() - time);
2606         else
2607                 rc = seq_printf(m, "%s: N/A\n", prefix);
2608         return rc;
2609 }
2610
2611 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2612 {
2613         int rc;
2614
2615         if (pos != 0)
2616                 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2617         else
2618                 rc = seq_printf(m, "%s: N/A\n", prefix);
2619         return rc;
2620 }
2621
2622 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2623 {
2624         struct osd_scrub  *scrub   = &dev->od_scrub;
2625         struct scrub_file *sf      = &scrub->os_file;
2626         __u64              checked;
2627         __u64              speed;
2628         int                rc;
2629
2630         down_read(&scrub->os_rwsem);
2631         rc = seq_printf(m, "name: OI_scrub\n"
2632                         "magic: 0x%x\n"
2633                         "oi_files: %d\n"
2634                         "status: %s\n",
2635                         sf->sf_magic, (int)sf->sf_oi_count,
2636                         scrub_status_names[sf->sf_status]);
2637         if (rc < 0)
2638                 goto out;
2639
2640         rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2641                              "flags");
2642         if (rc < 0)
2643                 goto out;
2644
2645         rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2646                              "param");
2647         if (rc < 0)
2648                 goto out;
2649
2650         rc = scrub_time_dump(m, sf->sf_time_last_complete,
2651                              "time_since_last_completed");
2652         if (rc < 0)
2653                 goto out;
2654
2655         rc = scrub_time_dump(m, sf->sf_time_latest_start,
2656                              "time_since_latest_start");
2657         if (rc < 0)
2658                 goto out;
2659
2660         rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2661                              "time_since_last_checkpoint");
2662         if (rc < 0)
2663                 goto out;
2664
2665         rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2666                             "latest_start_position");
2667         if (rc < 0)
2668                 goto out;
2669
2670         rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2671                             "last_checkpoint_position");
2672         if (rc < 0)
2673                 goto out;
2674
2675         rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2676                             "first_failure_position");
2677         if (rc < 0)
2678                 goto out;
2679
2680         checked = sf->sf_items_checked + scrub->os_new_checked;
2681         rc = seq_printf(m, "checked: "LPU64"\n"
2682                       "updated: "LPU64"\n"
2683                       "failed: "LPU64"\n"
2684                       "prior_updated: "LPU64"\n"
2685                       "noscrub: "LPU64"\n"
2686                       "igif: "LPU64"\n"
2687                       "success_count: %u\n",
2688                       checked, sf->sf_items_updated, sf->sf_items_failed,
2689                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
2690                       sf->sf_items_igif, sf->sf_success_count);
2691         if (rc < 0)
2692                 goto out;
2693
2694         speed = checked;
2695         if (thread_is_running(&scrub->os_thread)) {
2696                 cfs_duration_t duration = cfs_time_current() -
2697                                           scrub->os_time_last_checkpoint;
2698                 __u64 new_checked = scrub->os_new_checked * HZ;
2699                 __u32 rtime = sf->sf_run_time +
2700                               cfs_duration_sec(duration + HALF_SEC);
2701
2702                 if (duration != 0)
2703                         do_div(new_checked, duration);
2704                 if (rtime != 0)
2705                         do_div(speed, rtime);
2706                 rc = seq_printf(m, "run_time: %u seconds\n"
2707                               "average_speed: "LPU64" objects/sec\n"
2708                               "real-time_speed: "LPU64" objects/sec\n"
2709                               "current_position: %u\n"
2710                               "lf_scanned: "LPU64"\n"
2711                               "lf_reparied: "LPU64"\n"
2712                               "lf_failed: "LPU64"\n",
2713                               rtime, speed, new_checked, scrub->os_pos_current,
2714                               scrub->os_lf_scanned, scrub->os_lf_repaired,
2715                               scrub->os_lf_failed);
2716         } else {
2717                 if (sf->sf_run_time != 0)
2718                         do_div(speed, sf->sf_run_time);
2719                 rc = seq_printf(m, "run_time: %u seconds\n"
2720                               "average_speed: "LPU64" objects/sec\n"
2721                               "real-time_speed: N/A\n"
2722                               "current_position: N/A\n"
2723                               "lf_scanned: "LPU64"\n"
2724                               "lf_reparied: "LPU64"\n"
2725                               "lf_failed: "LPU64"\n",
2726                               sf->sf_run_time, speed, scrub->os_lf_scanned,
2727                               scrub->os_lf_repaired, scrub->os_lf_failed);
2728         }
2729
2730 out:
2731         up_read(&scrub->os_rwsem);
2732         return (rc < 0 ? -ENOSPC : 0);
2733 }