Whamcloud - gitweb
LU-1346 libcfs: replace libcfs wrappers with kernel API
[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 Whamcloud, Inc.
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 #ifndef EXPORT_SYMTAB
39 # define EXPORT_SYMTAB
40 #endif
41 #define DEBUG_SUBSYSTEM S_MDS
42
43 #include <lustre/lustre_idl.h>
44 #include <lustre_disk.h>
45 #include <dt_object.h>
46
47 #include "osd_internal.h"
48 #include "osd_oi.h"
49 #include "osd_scrub.h"
50
51 #define HALF_SEC        (CFS_HZ >> 1)
52
53 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
54 {
55         return container_of0(scrub, struct osd_device, od_scrub);
56 }
57
58 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
59 {
60         return osd_sb(osd_scrub2dev(scrub));
61 }
62
63 static void osd_scrub_file_to_cpu(struct scrub_file *des,
64                                   struct scrub_file *src)
65 {
66         memcpy(des->sf_uuid, src->sf_uuid, 16);
67         des->sf_flags   = le64_to_cpu(src->sf_flags);
68         des->sf_magic   = le32_to_cpu(src->sf_magic);
69         des->sf_status  = le16_to_cpu(src->sf_status);
70         des->sf_param   = le16_to_cpu(src->sf_param);
71         des->sf_time_last_complete      =
72                                 le64_to_cpu(src->sf_time_last_complete);
73         des->sf_time_latest_start       =
74                                 le64_to_cpu(src->sf_time_latest_start);
75         des->sf_time_last_checkpoint    =
76                                 le64_to_cpu(src->sf_time_last_checkpoint);
77         des->sf_pos_latest_start        =
78                                 le64_to_cpu(src->sf_pos_latest_start);
79         des->sf_pos_last_checkpoint     =
80                                 le64_to_cpu(src->sf_pos_last_checkpoint);
81         des->sf_pos_first_inconsistent  =
82                                 le64_to_cpu(src->sf_pos_first_inconsistent);
83         des->sf_items_checked           =
84                                 le64_to_cpu(src->sf_items_checked);
85         des->sf_items_updated           =
86                                 le64_to_cpu(src->sf_items_updated);
87         des->sf_items_failed            =
88                                 le64_to_cpu(src->sf_items_failed);
89         des->sf_items_updated_prior     =
90                                 le64_to_cpu(src->sf_items_updated_prior);
91         des->sf_run_time        = le32_to_cpu(src->sf_run_time);
92         des->sf_success_count   = le32_to_cpu(src->sf_success_count);
93         des->sf_oi_count        = le16_to_cpu(src->sf_oi_count);
94         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
95 }
96
97 static void osd_scrub_file_to_le(struct scrub_file *des,
98                                  struct scrub_file *src)
99 {
100         memcpy(des->sf_uuid, src->sf_uuid, 16);
101         des->sf_flags   = cpu_to_le64(src->sf_flags);
102         des->sf_magic   = cpu_to_le32(src->sf_magic);
103         des->sf_status  = cpu_to_le16(src->sf_status);
104         des->sf_param   = cpu_to_le16(src->sf_param);
105         des->sf_time_last_complete      =
106                                 cpu_to_le64(src->sf_time_last_complete);
107         des->sf_time_latest_start       =
108                                 cpu_to_le64(src->sf_time_latest_start);
109         des->sf_time_last_checkpoint    =
110                                 cpu_to_le64(src->sf_time_last_checkpoint);
111         des->sf_pos_latest_start        =
112                                 cpu_to_le64(src->sf_pos_latest_start);
113         des->sf_pos_last_checkpoint     =
114                                 cpu_to_le64(src->sf_pos_last_checkpoint);
115         des->sf_pos_first_inconsistent  =
116                                 cpu_to_le64(src->sf_pos_first_inconsistent);
117         des->sf_items_checked           =
118                                 cpu_to_le64(src->sf_items_checked);
119         des->sf_items_updated           =
120                                 cpu_to_le64(src->sf_items_updated);
121         des->sf_items_failed            =
122                                 cpu_to_le64(src->sf_items_failed);
123         des->sf_items_updated_prior     =
124                                 cpu_to_le64(src->sf_items_updated_prior);
125         des->sf_run_time        = cpu_to_le32(src->sf_run_time);
126         des->sf_success_count   = cpu_to_le32(src->sf_success_count);
127         des->sf_oi_count        = cpu_to_le16(src->sf_oi_count);
128         memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
129 }
130
131 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
132 {
133         struct scrub_file *sf = &scrub->os_file;
134
135         memset(sf, 0, sizeof(*sf));
136         memcpy(sf->sf_uuid, uuid, 16);
137         sf->sf_magic = SCRUB_MAGIC_V1;
138         sf->sf_status = SS_INIT;
139 }
140
141 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
142 {
143         struct scrub_file *sf = &scrub->os_file;
144
145         CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
146         memcpy(sf->sf_uuid, uuid, 16);
147         sf->sf_status = SS_INIT;
148         sf->sf_flags |= flags;
149         sf->sf_param = 0;
150         sf->sf_run_time = 0;
151         sf->sf_time_latest_start = 0;
152         sf->sf_time_last_checkpoint = 0;
153         sf->sf_pos_latest_start = 0;
154         sf->sf_pos_last_checkpoint = 0;
155         sf->sf_pos_first_inconsistent = 0;
156         sf->sf_items_checked = 0;
157         sf->sf_items_updated = 0;
158         sf->sf_items_failed = 0;
159         sf->sf_items_updated_prior = 0;
160         sf->sf_items_noscrub = 0;
161         sf->sf_items_igif = 0;
162 }
163
164 static int osd_scrub_file_load(struct osd_scrub *scrub)
165 {
166         loff_t  pos  = 0;
167         char   *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
168         int     len  = sizeof(scrub->os_file_disk);
169         int     rc;
170
171         rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
172         if (rc == len) {
173                 struct scrub_file *sf = &scrub->os_file;
174
175                 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
176                 if (sf->sf_magic != SCRUB_MAGIC_V1) {
177                         CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
178                               name, sf->sf_magic, SCRUB_MAGIC_V1);
179                         /* Process it as new scrub file. */
180                         rc = -ENOENT;
181                 } else {
182                         rc = 0;
183                 }
184         } else if (rc != 0) {
185                 CERROR("%.16s: fail to load scrub file, expected = %d, "
186                        "rc = %d\n", name, len, rc);
187                 if (rc > 0)
188                         rc = -EFAULT;
189         } else {
190                 /* return -ENOENT for empty scrub file case. */
191                 rc = -ENOENT;
192         }
193
194         return rc;
195 }
196
197 int osd_scrub_file_store(struct osd_scrub *scrub)
198 {
199         struct osd_device *dev;
200         handle_t          *jh;
201         loff_t             pos     = 0;
202         int                len     = sizeof(scrub->os_file_disk);
203         int                credits;
204         int                rc;
205
206         dev = container_of0(scrub, struct osd_device, od_scrub);
207         credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
208                   osd_dto_credits_noquota[DTO_WRITE_BLOCK];
209         jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
210         if (IS_ERR(jh)) {
211                 rc = PTR_ERR(jh);
212                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
213                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
214                 return rc;
215         }
216
217         osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
218         rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
219                                       len, 0, &pos, jh);
220         ldiskfs_journal_stop(jh);
221         if (rc != 0)
222                 CERROR("%.16s: fail to store scrub file, expected = %d, "
223                        "rc = %d\n",
224                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
225                        len, rc);
226         scrub->os_time_last_checkpoint = cfs_time_current();
227         scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
228                                 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
229         return rc;
230 }
231
232 static int osd_scrub_prep(struct osd_device *dev)
233 {
234         struct osd_scrub     *scrub  = &dev->od_scrub;
235         struct ptlrpc_thread *thread = &scrub->os_thread;
236         struct scrub_file    *sf     = &scrub->os_file;
237         __u32                 flags  = scrub->os_start_flags;
238         int                   rc;
239         ENTRY;
240
241         down_write(&scrub->os_rwsem);
242         if (flags & SS_SET_FAILOUT)
243                 sf->sf_param |= SP_FAILOUT;
244
245         if (flags & SS_CLEAR_FAILOUT)
246                 sf->sf_param &= ~SP_FAILOUT;
247
248         if (flags & SS_RESET)
249                 osd_scrub_file_reset(scrub,
250                         LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, sf->sf_flags);
251
252         if (flags & SS_AUTO) {
253                 scrub->os_full_speed = 1;
254                 sf->sf_flags |= SF_AUTO;
255         } else {
256                 scrub->os_full_speed = 0;
257         }
258
259         if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT))
260                 scrub->os_full_speed = 1;
261
262         scrub->os_in_prior = 0;
263         scrub->os_waiting = 0;
264         scrub->os_new_checked = 0;
265         if (sf->sf_pos_last_checkpoint != 0)
266                 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
267         else
268                 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev));
269
270         scrub->os_pos_current = sf->sf_pos_latest_start;
271         sf->sf_status = SS_SCANNING;
272         sf->sf_time_latest_start = cfs_time_current_sec();
273         sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
274         rc = osd_scrub_file_store(scrub);
275         if (rc == 0) {
276                 spin_lock(&scrub->os_lock);
277                 thread_set_flags(thread, SVC_RUNNING);
278                 spin_unlock(&scrub->os_lock);
279                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
280         }
281         up_write(&scrub->os_rwsem);
282
283         RETURN(rc);
284 }
285
286 static int
287 osd_scrub_error(struct osd_device *dev, struct osd_inode_id *lid, int rc)
288 {
289         struct osd_scrub  *scrub = &dev->od_scrub;
290         struct scrub_file *sf    = &scrub->os_file;
291
292         down_write(&scrub->os_rwsem);
293         scrub->os_new_checked++;
294         sf->sf_items_failed++;
295         if (sf->sf_pos_first_inconsistent == 0 ||
296             sf->sf_pos_first_inconsistent > lid->oii_ino)
297                 sf->sf_pos_first_inconsistent = lid->oii_ino;
298         up_write(&scrub->os_rwsem);
299         return sf->sf_param & SP_FAILOUT ? rc : 0;
300 }
301
302 static int
303 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
304                        struct osd_idmap_cache *oic)
305 {
306         struct osd_scrub             *scrub  = &dev->od_scrub;
307         struct scrub_file            *sf     = &scrub->os_file;
308         struct osd_inode_id          *lid2   = &info->oti_id;
309         struct lu_fid                *oi_fid = &info->oti_fid;
310         struct osd_inode_id          *oi_id  = &info->oti_id2;
311         handle_t                     *jh     = NULL;
312         struct osd_inconsistent_item *oii    = NULL;
313         struct inode                 *inode  = NULL;
314         struct lu_fid                *fid    = &oic->oic_fid;
315         struct osd_inode_id          *lid    = &oic->oic_lid;
316         struct iam_container         *bag;
317         struct iam_path_descr        *ipd;
318         int                           ops    = DTO_INDEX_UPDATE;
319         int                           idx;
320         int                           rc;
321         ENTRY;
322
323         if (scrub->os_in_prior)
324                 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
325                                      oii_cache);
326
327         down_write(&scrub->os_rwsem);
328         scrub->os_new_checked++;
329         if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
330                 GOTO(out, rc = 0);
331
332         if (oii != NULL && oii->oii_insert)
333                 goto iget;
334
335         rc = osd_oi_lookup(info, dev, fid, lid2);
336         if (rc != 0) {
337                 if (rc != -ENOENT)
338                         GOTO(out, rc);
339
340 iget:
341                 inode = osd_iget(info, dev, lid);
342                 if (IS_ERR(inode)) {
343                         rc = PTR_ERR(inode);
344                         /* Someone removed the inode. */
345                         if (rc == -ENOENT || rc == -ESTALE)
346                                 rc = 0;
347                         GOTO(out, rc);
348                 }
349
350                 /* Prevent the inode to be unlinked during OI scrub. */
351                 mutex_lock(&inode->i_mutex);
352                 if (unlikely(inode->i_nlink == 0)) {
353                         mutex_unlock(&inode->i_mutex);
354                         iput(inode);
355                         GOTO(out, rc = 0);
356                 }
357
358                 ops = DTO_INDEX_INSERT;
359                 idx = osd_oi_fid2idx(dev, fid);
360                 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
361                         ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
362                 sf->sf_flags |= SF_RECREATED;
363         } else if (osd_id_eq(lid, lid2)) {
364                         GOTO(out, rc = 0);
365         }
366
367         sf->sf_flags |= SF_INCONSISTENT;
368         fid_cpu_to_be(oi_fid, fid);
369         osd_id_pack(oi_id, &oic->oic_lid);
370         jh = ldiskfs_journal_start_sb(osd_sb(dev),
371                                 osd_dto_credits_noquota[ops]);
372         if (IS_ERR(jh)) {
373                 rc = PTR_ERR(jh);
374                 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
375                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
376                 GOTO(out, rc);
377         }
378
379         bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
380         ipd = osd_idx_ipd_get(info->oti_env, bag);
381         if (unlikely(ipd == NULL)) {
382                 ldiskfs_journal_stop(jh);
383                 CERROR("%.16s: fail to get ipd for scrub store\n",
384                         LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
385                 GOTO(out, rc = -ENOMEM);
386         }
387
388         if (ops == DTO_INDEX_UPDATE) {
389                 rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
390                                 (struct iam_rec *)oi_id, ipd);
391         } else {
392                 rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
393                                 (struct iam_rec *)oi_id, ipd);
394                 if (rc == -EEXIST)
395                         rc = 1;
396         }
397         osd_ipd_put(info->oti_env, bag, ipd);
398         ldiskfs_journal_stop(jh);
399         if (rc == 0) {
400                 if (scrub->os_in_prior)
401                         sf->sf_items_updated_prior++;
402                 else
403                         sf->sf_items_updated++;
404         }
405
406         GOTO(out, rc);
407
408 out:
409         if (rc < 0) {
410                 sf->sf_items_failed++;
411                 if (sf->sf_pos_first_inconsistent == 0 ||
412                     sf->sf_pos_first_inconsistent > lid->oii_ino)
413                         sf->sf_pos_first_inconsistent = lid->oii_ino;
414         } else {
415                 rc = 0;
416         }
417
418         if (ops == DTO_INDEX_INSERT) {
419                 mutex_unlock(&inode->i_mutex);
420                 iput(inode);
421         }
422         up_write(&scrub->os_rwsem);
423
424         if (oii != NULL) {
425                 LASSERT(!cfs_list_empty(&oii->oii_list));
426
427                 spin_lock(&scrub->os_lock);
428                 cfs_list_del_init(&oii->oii_list);
429                 spin_unlock(&scrub->os_lock);
430                 OBD_FREE_PTR(oii);
431         }
432         RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
433 }
434
435 static int do_osd_scrub_checkpoint(struct osd_scrub *scrub)
436 {
437         struct scrub_file *sf = &scrub->os_file;
438         int                rc;
439         ENTRY;
440
441         down_write(&scrub->os_rwsem);
442         sf->sf_items_checked += scrub->os_new_checked;
443         scrub->os_new_checked = 0;
444         sf->sf_pos_last_checkpoint = scrub->os_pos_current;
445         sf->sf_time_last_checkpoint = cfs_time_current_sec();
446         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
447                                             scrub->os_time_last_checkpoint);
448         rc = osd_scrub_file_store(scrub);
449         up_write(&scrub->os_rwsem);
450
451         RETURN(rc);
452 }
453
454 static inline int osd_scrub_checkpoint(struct osd_scrub *scrub)
455 {
456         if (unlikely(cfs_time_beforeq(scrub->os_time_next_checkpoint,
457                                       cfs_time_current()) &&
458                      scrub->os_new_checked > 0))
459                 return do_osd_scrub_checkpoint(scrub);
460         return 0;
461 }
462
463 static void osd_scrub_post(struct osd_scrub *scrub, int result)
464 {
465         struct scrub_file *sf = &scrub->os_file;
466         ENTRY;
467
468         down_write(&scrub->os_rwsem);
469         spin_lock(&scrub->os_lock);
470         thread_set_flags(&scrub->os_thread, SVC_STOPPING);
471         spin_unlock(&scrub->os_lock);
472         if (scrub->os_new_checked > 0) {
473                 sf->sf_items_checked += scrub->os_new_checked;
474                 scrub->os_new_checked = 0;
475                 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
476         }
477         sf->sf_time_last_checkpoint = cfs_time_current_sec();
478         if (result > 0) {
479                 sf->sf_status = SS_COMPLETED;
480                 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
481                 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT | SF_AUTO);
482                 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
483                 sf->sf_success_count++;
484         } else if (result == 0) {
485                 if (scrub->os_paused)
486                         sf->sf_status = SS_PAUSED;
487                 else
488                         sf->sf_status = SS_STOPPED;
489         } else {
490                 sf->sf_status = SS_FAILED;
491         }
492         sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
493                                             scrub->os_time_last_checkpoint);
494         result = osd_scrub_file_store(scrub);
495         if (result < 0)
496                 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
497                        LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
498                        result);
499         up_write(&scrub->os_rwsem);
500
501         EXIT;
502 }
503
504 #define SCRUB_NEXT_BREAK        1 /* exit current loop and process next group */
505 #define SCRUB_NEXT_CONTINUE     2 /* skip current object and process next bit */
506 #define SCRUB_NEXT_EXIT         3 /* exit all the loops */
507 #define SCRUB_NEXT_WAIT         4 /* wait for free cache slot */
508 #define SCRUB_NEXT_CRASH        5 /* simulate system crash during OI scrub */
509 #define SCRUB_NEXT_FATAL        6 /* simulate failure during OI scrub */
510 #define SCRUB_NEXT_NOSCRUB      7 /* new created object, no scrub on it */
511 #define SCRUB_NEXT_IGIF         8 /* IGIF object */
512
513 struct osd_iit_param {
514         struct super_block *sb;
515         struct buffer_head *bitmap;
516         ldiskfs_group_t bg;
517         __u32 gbase;
518         __u32 offset;
519 };
520
521 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
522                                    struct osd_device *dev,
523                                    struct osd_iit_param *param,
524                                    struct osd_idmap_cache **oic,
525                                    int noslot);
526
527 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
528                                    struct osd_device *dev,
529                                    struct osd_iit_param *param,
530                                    struct osd_idmap_cache *oic,
531                                    int *noslot, int rc);
532
533 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
534                                        struct osd_otable_cache *ooc)
535 {
536         return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
537 }
538
539 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
540 {
541         param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
542                         LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
543         if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
544                 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
545                 return SCRUB_NEXT_BREAK;
546         } else {
547                 *pos = param->gbase + param->offset;
548                 return 0;
549         }
550 }
551
552 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
553                         struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
554                         struct super_block *sb, struct inode **pinode)
555 {
556         struct inode *inode;
557         int           rc;
558
559         osd_id_gen(lid, pos, OSD_OII_NOGEN);
560         inode = osd_iget_fid(info, dev, lid, fid);
561         if (IS_ERR(inode)) {
562                 rc = PTR_ERR(inode);
563                 /* The inode may be removed after bitmap searching, or the
564                  * file is new created without inode initialized yet. */
565                 if (rc == -ENOENT || rc == -ESTALE)
566                         return SCRUB_NEXT_CONTINUE;
567
568                 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
569                        LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
570                 return rc;
571         }
572
573         *pinode = inode;
574         return 0;
575 }
576
577 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
578                           struct osd_iit_param *param,
579                           struct osd_idmap_cache **oic, int noslot)
580 {
581         struct osd_scrub     *scrub  = &dev->od_scrub;
582         struct ptlrpc_thread *thread = &scrub->os_thread;
583         struct lu_fid        *fid;
584         struct osd_inode_id  *lid;
585         struct inode         *inode;
586         int                   rc;
587
588         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
589                 struct l_wait_info lwi;
590
591                 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
592                 l_wait_event(thread->t_ctl_waitq,
593                              !cfs_list_empty(&scrub->os_inconsistent_items) ||
594                              !thread_is_running(thread),
595                              &lwi);
596         }
597
598         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
599                 spin_lock(&scrub->os_lock);
600                 thread_set_flags(thread, SVC_STOPPING);
601                 spin_unlock(&scrub->os_lock);
602                 return SCRUB_NEXT_CRASH;
603         }
604
605         if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
606                 return SCRUB_NEXT_FATAL;
607
608         if (unlikely(!thread_is_running(thread)))
609                 return SCRUB_NEXT_EXIT;
610
611         if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
612                 struct osd_inconsistent_item *oii;
613
614                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
615                                      struct osd_inconsistent_item, oii_list);
616                 *oic = &oii->oii_cache;
617                 scrub->os_in_prior = 1;
618                 return 0;
619         }
620
621         if (noslot != 0)
622                 return SCRUB_NEXT_WAIT;
623
624         rc = osd_iit_next(param, &scrub->os_pos_current);
625         if (rc != 0)
626                 return rc;
627
628         *oic = &scrub->os_oic;
629         fid = &(*oic)->oic_fid;
630         lid = &(*oic)->oic_lid;
631         rc = osd_iit_iget(info, dev, fid, lid,
632                           scrub->os_pos_current, param->sb, &inode);
633         if (rc != 0)
634                 return rc;
635
636         if (inode->i_state & I_LUSTRE_NOSCRUB) {
637                 /* Only skip it for the first OI scrub accessing. */
638                 inode->i_state &= ~I_LUSTRE_NOSCRUB;
639                 rc = SCRUB_NEXT_NOSCRUB;
640         } else if (!fid_is_norm(fid)) {
641                 rc = SCRUB_NEXT_IGIF;
642         }
643
644         iput(inode);
645         return rc;
646 }
647
648 static int osd_preload_next(struct osd_thread_info *info,
649                             struct osd_device *dev, struct osd_iit_param *param,
650                             struct osd_idmap_cache **oic, int noslot)
651 {
652         struct osd_otable_cache *ooc    = &dev->od_otable_it->ooi_cache;
653         struct osd_scrub        *scrub;
654         struct ptlrpc_thread    *thread;
655         struct inode            *inode;
656         int                      rc;
657
658         rc = osd_iit_next(param, &ooc->ooc_pos_preload);
659         if (rc != 0)
660                 return rc;
661
662         scrub = &dev->od_scrub;
663         thread = &scrub->os_thread;
664         if (thread_is_running(thread) &&
665             ooc->ooc_pos_preload >= scrub->os_pos_current)
666                 return SCRUB_NEXT_EXIT;
667
668         rc = osd_iit_iget(info, dev,
669                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
670                           &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
671                           ooc->ooc_pos_preload, param->sb, &inode);
672         /* If succeed, it needs to move forward; otherwise up layer LFSCK may
673          * ignore the failure, so it still need to skip the inode next time. */
674         ooc->ooc_pos_preload = param->gbase + ++(param->offset);
675         if (rc == 0)
676                 iput(inode);
677         return rc;
678 }
679
680 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
681                           struct osd_iit_param *param,
682                           struct osd_idmap_cache *oic, int *noslot, int rc)
683 {
684         struct l_wait_info       lwi    = { 0 };
685         struct osd_scrub        *scrub  = &dev->od_scrub;
686         struct scrub_file       *sf     = &scrub->os_file;
687         __u64                   *items  = NULL;
688         struct ptlrpc_thread    *thread = &scrub->os_thread;
689         struct osd_otable_it    *it     = dev->od_otable_it;
690         struct osd_otable_cache *ooc    = it ? &it->ooi_cache : NULL;
691
692         switch (rc) {
693         case SCRUB_NEXT_CONTINUE:
694                 goto next;
695         case SCRUB_NEXT_WAIT:
696                 goto wait;
697         case SCRUB_NEXT_NOSCRUB:
698                 items = &sf->sf_items_noscrub;
699                 break;
700         case SCRUB_NEXT_IGIF:
701                 items = &sf->sf_items_igif;
702                 break;
703         }
704
705         if (items != NULL) {
706                 down_write(&scrub->os_rwsem);
707                 scrub->os_new_checked++;
708                 (*items)++;
709                 up_write(&scrub->os_rwsem);
710                 goto next;
711         }
712
713         LASSERTF(rc <= 0, "unexpected rc = %d\n", rc);
714
715         if (rc != 0)
716                 rc = osd_scrub_error(dev, &oic->oic_lid, rc);
717         else
718                 rc = osd_scrub_check_update(info, dev, oic);
719         if (rc != 0)
720                 return rc;
721
722         rc = osd_scrub_checkpoint(scrub);
723         if (rc != 0) {
724                 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
725                        LDISKFS_SB(param->sb)->s_es->s_volume_name,
726                        scrub->os_pos_current, rc);
727                 /* Continue, as long as the scrub itself can go ahead. */
728         }
729
730         if (scrub->os_in_prior) {
731                 scrub->os_in_prior = 0;
732                 return 0;
733         }
734
735 next:
736         scrub->os_pos_current = param->gbase + ++(param->offset);
737         if (it != NULL && it->ooi_waiting &&
738             ooc->ooc_pos_preload < scrub->os_pos_current) {
739                 it->ooi_waiting = 0;
740                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
741         }
742
743         if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
744                 return 0;
745
746 wait:
747         if (osd_scrub_has_window(scrub, ooc)) {
748                 *noslot = 0;
749                 return 0;
750         }
751
752         scrub->os_waiting = 1;
753         l_wait_event(thread->t_ctl_waitq,
754                      osd_scrub_has_window(scrub, ooc) ||
755                      !cfs_list_empty(&scrub->os_inconsistent_items) ||
756                      !thread_is_running(thread),
757                      &lwi);
758         scrub->os_waiting = 0;
759
760         if (osd_scrub_has_window(scrub, ooc))
761                 *noslot = 0;
762         else
763                 *noslot = 1;
764         return 0;
765 }
766
767 static int osd_preload_exec(struct osd_thread_info *info,
768                             struct osd_device *dev, struct osd_iit_param *param,
769                             struct osd_idmap_cache *oic, int *noslot, int rc)
770 {
771         struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
772
773         if (rc == 0) {
774                 ooc->ooc_cached_items++;
775                 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
776                                         ~OSD_OTABLE_IT_CACHE_MASK;
777         }
778         return rc > 0 ? 0 : rc;
779 }
780
781 #define SCRUB_IT_ALL    1
782 #define SCRUB_IT_CRASH  2
783
784 static int osd_inode_iteration(struct osd_thread_info *info,
785                                struct osd_device *dev, __u32 max, int preload)
786 {
787         osd_iit_next_policy   next;
788         osd_iit_exec_policy   exec;
789         __u32                *pos;
790         __u32                *count;
791         struct osd_iit_param  param;
792         __u32                 limit;
793         int                   noslot = 0;
794         int                   rc;
795         ENTRY;
796
797         if (preload == 0) {
798                 struct osd_scrub *scrub = &dev->od_scrub;
799
800                 next = osd_scrub_next;
801                 exec = osd_scrub_exec;
802                 pos = &scrub->os_pos_current;
803                 count = &scrub->os_new_checked;
804         } else {
805                 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
806
807                 next = osd_preload_next;
808                 exec = osd_preload_exec;
809                 pos = &ooc->ooc_pos_preload;
810                 count = &ooc->ooc_cached_items;
811         }
812         param.sb = osd_sb(dev);
813         limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
814
815         while (*pos <= limit && *count < max) {
816                 struct osd_idmap_cache *oic = NULL;
817
818                 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
819                 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
820                 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
821                 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
822                 if (param.bitmap == NULL) {
823                         CERROR("%.16s: fail to read bitmap for %u, "
824                                "scrub will stop, urgent mode\n",
825                                LDISKFS_SB(param.sb)->s_es->s_volume_name,
826                                (__u32)param.bg);
827                         RETURN(-EIO);
828                 }
829
830                 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
831                        *count < max) {
832                         rc = next(info, dev, &param, &oic, noslot);
833                         switch (rc) {
834                         case SCRUB_NEXT_BREAK:
835                                 goto next_group;
836                         case SCRUB_NEXT_EXIT:
837                                 brelse(param.bitmap);
838                                 RETURN(0);
839                         case SCRUB_NEXT_CRASH:
840                                 brelse(param.bitmap);
841                                 RETURN(SCRUB_IT_CRASH);
842                         case SCRUB_NEXT_FATAL:
843                                 brelse(param.bitmap);
844                                 RETURN(-EINVAL);
845                         }
846
847                         rc = exec(info, dev, &param, oic, &noslot, rc);
848                         if (rc != 0) {
849                                 brelse(param.bitmap);
850                                 RETURN(rc);
851                         }
852                 }
853
854 next_group:
855                 brelse(param.bitmap);
856         }
857
858         if (*pos > limit)
859                 RETURN(SCRUB_IT_ALL);
860         RETURN(0);
861 }
862
863 static int osd_scrub_main(void *args)
864 {
865         struct lu_env         env;
866         struct osd_device    *dev    = (struct osd_device *)args;
867         struct osd_scrub     *scrub  = &dev->od_scrub;
868         struct ptlrpc_thread *thread = &scrub->os_thread;
869         struct super_block   *sb     = osd_sb(dev);
870         int                   rc;
871         ENTRY;
872
873         cfs_daemonize("OI_scrub");
874         rc = lu_env_init(&env, LCT_DT_THREAD);
875         if (rc != 0) {
876                 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
877                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
878                 GOTO(noenv, rc);
879         }
880
881         rc = osd_scrub_prep(dev);
882         if (rc != 0) {
883                 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
884                        LDISKFS_SB(sb)->s_es->s_volume_name, rc);
885                 GOTO(out, rc);
886         }
887
888         if (!scrub->os_full_speed) {
889                 struct l_wait_info lwi = { 0 };
890                 struct osd_otable_it *it = dev->od_otable_it;
891                 struct osd_otable_cache *ooc = &it->ooi_cache;
892
893                 l_wait_event(thread->t_ctl_waitq,
894                              it->ooi_user_ready || !thread_is_running(thread),
895                              &lwi);
896                 if (unlikely(!thread_is_running(thread)))
897                         GOTO(post, rc = 0);
898
899                 LASSERT(scrub->os_pos_current >= ooc->ooc_pos_preload);
900                 scrub->os_pos_current = ooc->ooc_pos_preload;
901         }
902
903         CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
904                scrub->os_start_flags, scrub->os_pos_current);
905
906         rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, 0);
907         if (unlikely(rc == SCRUB_IT_CRASH))
908                 GOTO(out, rc = -EINVAL);
909         GOTO(post, rc);
910
911 post:
912         osd_scrub_post(scrub, rc);
913         CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
914                rc, scrub->os_pos_current);
915
916 out:
917         while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
918                 struct osd_inconsistent_item *oii;
919
920                 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
921                                      struct osd_inconsistent_item, oii_list);
922                 cfs_list_del_init(&oii->oii_list);
923                 OBD_FREE_PTR(oii);
924         }
925         lu_env_fini(&env);
926
927 noenv:
928         spin_lock(&scrub->os_lock);
929         thread_set_flags(thread, SVC_STOPPED);
930         cfs_waitq_broadcast(&thread->t_ctl_waitq);
931         spin_unlock(&scrub->os_lock);
932         return rc;
933 }
934
935 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
936 {
937         struct osd_scrub     *scrub  = &dev->od_scrub;
938         struct ptlrpc_thread *thread = &scrub->os_thread;
939         struct l_wait_info    lwi    = { 0 };
940         int                   rc;
941         ENTRY;
942
943 again:
944         /* os_lock: sync status between stop and scrub thread */
945         spin_lock(&scrub->os_lock);
946         if (thread_is_running(thread)) {
947                 spin_unlock(&scrub->os_lock);
948                 RETURN(-EALREADY);
949         } else if (unlikely(thread_is_stopping(thread))) {
950                 spin_unlock(&scrub->os_lock);
951                 l_wait_event(thread->t_ctl_waitq,
952                              thread_is_stopped(thread),
953                              &lwi);
954                 goto again;
955         }
956         spin_unlock(&scrub->os_lock);
957
958         if (scrub->os_file.sf_status == SS_COMPLETED)
959                 flags |= SS_RESET;
960
961         scrub->os_start_flags = flags;
962         thread_set_flags(thread, 0);
963         rc = cfs_create_thread(osd_scrub_main, dev, 0);
964         if (rc < 0) {
965                 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
966                        LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
967                 RETURN(rc);
968         }
969
970         l_wait_event(thread->t_ctl_waitq,
971                      thread_is_running(thread) || thread_is_stopped(thread),
972                      &lwi);
973
974         RETURN(0);
975 }
976
977 int osd_scrub_start(struct osd_device *dev)
978 {
979         int rc;
980         ENTRY;
981
982         /* od_otable_mutex: prevent curcurrent start/stop */
983         mutex_lock(&dev->od_otable_mutex);
984         rc = do_osd_scrub_start(dev, SS_AUTO);
985         mutex_unlock(&dev->od_otable_mutex);
986
987         RETURN(rc == -EALREADY ? 0 : rc);
988 }
989
990 static void do_osd_scrub_stop(struct osd_scrub *scrub)
991 {
992         struct ptlrpc_thread *thread = &scrub->os_thread;
993         struct l_wait_info    lwi    = { 0 };
994
995         /* os_lock: sync status between stop and scrub thread */
996         spin_lock(&scrub->os_lock);
997         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
998                 thread_set_flags(thread, SVC_STOPPING);
999                 spin_unlock(&scrub->os_lock);
1000                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1001                 l_wait_event(thread->t_ctl_waitq,
1002                              thread_is_stopped(thread),
1003                              &lwi);
1004                 /* Do not skip the last lock/unlock, which can guarantee that
1005                  * the caller cannot return until the OI scrub thread exit. */
1006                 spin_lock(&scrub->os_lock);
1007         }
1008         spin_unlock(&scrub->os_lock);
1009 }
1010
1011 static void osd_scrub_stop(struct osd_device *dev)
1012 {
1013         /* od_otable_mutex: prevent curcurrent start/stop */
1014         mutex_lock(&dev->od_otable_mutex);
1015         dev->od_scrub.os_paused = 1;
1016         do_osd_scrub_stop(&dev->od_scrub);
1017         mutex_unlock(&dev->od_otable_mutex);
1018 }
1019
1020 static const char osd_scrub_name[] = "OI_scrub";
1021
1022 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1023 {
1024         struct osd_thread_info     *info   = osd_oti_get(env);
1025         struct osd_scrub           *scrub  = &dev->od_scrub;
1026         struct lvfs_run_ctxt       *ctxt   = &scrub->os_ctxt;
1027         struct scrub_file          *sf     = &scrub->os_file;
1028         struct osd_inode_id        *id     = &scrub->os_oic.oic_lid;
1029         struct super_block         *sb     = osd_sb(dev);
1030         struct ldiskfs_super_block *es     = LDISKFS_SB(sb)->s_es;
1031         struct inode               *inode;
1032         struct lvfs_run_ctxt        saved;
1033         struct file                *filp;
1034         int                         dirty  = 0;
1035         int                         init   = 0;
1036         int                         rc     = 0;
1037         ENTRY;
1038
1039         memset(scrub, 0, sizeof(*scrub));
1040         OBD_SET_CTXT_MAGIC(ctxt);
1041         ctxt->pwdmnt = dev->od_mnt;
1042         ctxt->pwd = dev->od_mnt->mnt_root;
1043         ctxt->fs = get_ds();
1044
1045         cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1046         init_rwsem(&scrub->os_rwsem);
1047         spin_lock_init(&scrub->os_lock);
1048         CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1049
1050         push_ctxt(&saved, ctxt, NULL);
1051         filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1052         if (IS_ERR(filp))
1053                 RETURN(PTR_ERR(filp));
1054
1055         scrub->os_inode = igrab(filp->f_dentry->d_inode);
1056         filp_close(filp, 0);
1057         pop_ctxt(&saved, ctxt, NULL);
1058
1059         rc = osd_scrub_file_load(scrub);
1060         if (rc == -ENOENT) {
1061                 osd_scrub_file_init(scrub, es->s_uuid);
1062                 dirty = 1;
1063                 init = 1;
1064         } else if (rc != 0) {
1065                 RETURN(rc);
1066         } else {
1067                 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1068                         osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1069                         dirty = 1;
1070                 } else if (sf->sf_status == SS_SCANNING) {
1071                         sf->sf_status = SS_CRASHED;
1072                         dirty = 1;
1073                 }
1074         }
1075
1076         if (sf->sf_pos_last_checkpoint != 0)
1077                 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1078         else
1079                 scrub->os_pos_current = LDISKFS_FIRST_INO(sb);
1080
1081         if (dirty != 0) {
1082                 rc = osd_scrub_file_store(scrub);
1083                 if (rc != 0)
1084                         RETURN(rc);
1085         }
1086
1087         /* Initialize OI files. */
1088         rc = osd_oi_init(info, dev);
1089         if (rc < 0)
1090                 RETURN(rc);
1091
1092         if (init != 0) {
1093                 rc = __osd_oi_lookup(info, dev, &LU_DOT_LUSTRE_FID, id);
1094                 if (rc == 0) {
1095                         inode = osd_iget(info, dev, id);
1096                         if (IS_ERR(inode)) {
1097                                 rc = PTR_ERR(inode);
1098                                 /* It is restored from old 2.x backup. */
1099                                 if (rc == -ENOENT || rc == -ESTALE) {
1100                                         osd_scrub_file_reset(scrub, es->s_uuid,
1101                                                              SF_INCONSISTENT);
1102                                         rc = osd_scrub_file_store(scrub);
1103                                 }
1104                         } else {
1105                                 iput(inode);
1106                         }
1107                 } else if (rc == -ENOENT) {
1108                         rc = 0;
1109                 }
1110         }
1111
1112         if (rc == 0 && !dev->od_noscrub &&
1113             ((sf->sf_status == SS_PAUSED) ||
1114              (sf->sf_status == SS_CRASHED &&
1115               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_AUTO)) ||
1116              (sf->sf_status == SS_INIT &&
1117               sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT))))
1118                 rc = osd_scrub_start(dev);
1119
1120         RETURN(rc);
1121 }
1122
1123 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1124 {
1125         struct osd_scrub *scrub = &dev->od_scrub;
1126
1127         LASSERT(dev->od_otable_it == NULL);
1128
1129         if (scrub->os_inode != NULL) {
1130                 osd_scrub_stop(dev);
1131                 iput(scrub->os_inode);
1132                 scrub->os_inode = NULL;
1133         }
1134         if (dev->od_oi_table != NULL)
1135                 osd_oi_fini(osd_oti_get(env), dev);
1136 }
1137
1138 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1139                                        struct dt_object *dt, __u32 attr,
1140                                        struct lustre_capa *capa)
1141 {
1142         enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1143         enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1144         struct osd_device      *dev   = osd_dev(dt->do_lu.lo_dev);
1145         struct osd_scrub       *scrub = &dev->od_scrub;
1146         struct osd_otable_it   *it;
1147         __u32                   start = 0;
1148         int                     rc;
1149         ENTRY;
1150
1151         /* od_otable_mutex: prevent curcurrent init/fini */
1152         mutex_lock(&dev->od_otable_mutex);
1153         if (dev->od_otable_it != NULL)
1154                 GOTO(out, it = ERR_PTR(-EALREADY));
1155
1156         OBD_ALLOC_PTR(it);
1157         if (it == NULL)
1158                 GOTO(out, it = ERR_PTR(-ENOMEM));
1159
1160         dev->od_otable_it = it;
1161         it->ooi_dev = dev;
1162         it->ooi_cache.ooc_consumer_idx = -1;
1163         if (flags & DOIF_OUTUSED)
1164                 it->ooi_used_outside = 1;
1165
1166         if (flags & DOIF_RESET)
1167                 start |= SS_RESET;
1168
1169         if (valid & DOIV_ERROR_HANDLE) {
1170                 if (flags & DOIF_FAILOUT)
1171                         start |= SS_SET_FAILOUT;
1172                 else
1173                         start |= SS_CLEAR_FAILOUT;
1174         }
1175
1176         rc = do_osd_scrub_start(dev, start);
1177         if (rc == -EALREADY) {
1178                 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current - 1;
1179         } else if (rc < 0) {
1180                 dev->od_otable_it = NULL;
1181                 OBD_FREE_PTR(it);
1182                 GOTO(out, it = ERR_PTR(-EALREADY));
1183         } else {
1184                 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
1185         }
1186
1187         GOTO(out, it);
1188
1189 out:
1190         mutex_unlock(&dev->od_otable_mutex);
1191         return (struct dt_it *)it;
1192 }
1193
1194 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1195 {
1196         struct osd_otable_it *it  = (struct osd_otable_it *)di;
1197         struct osd_device    *dev = it->ooi_dev;
1198
1199         /* od_otable_mutex: prevent curcurrent init/fini */
1200         mutex_lock(&dev->od_otable_mutex);
1201         do_osd_scrub_stop(&dev->od_scrub);
1202         LASSERT(dev->od_otable_it == it);
1203
1204         dev->od_otable_it = NULL;
1205         mutex_unlock(&dev->od_otable_mutex);
1206         OBD_FREE_PTR(it);
1207 }
1208
1209 /**
1210  * XXX: Temporary used to notify otable iteration to be paused.
1211  */
1212 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1213 {
1214         struct osd_device *dev = ((struct osd_otable_it *)di)->ooi_dev;
1215
1216         /* od_otable_mutex: prevent curcurrent init/fini */
1217         mutex_lock(&dev->od_otable_mutex);
1218         dev->od_scrub.os_paused = 1;
1219         mutex_unlock(&dev->od_otable_mutex);
1220 }
1221
1222 /**
1223  * Set the OSD layer iteration start position as the specified key.
1224  *
1225  * The LFSCK out of OSD layer does not know the detail of the key, so if there
1226  * are several keys, they cannot be compared out of OSD, so call "::get()" for
1227  * each key, and OSD will select the smallest one by itself.
1228  */
1229 static int osd_otable_it_get(const struct lu_env *env,
1230                              struct dt_it *di, const struct dt_key *key)
1231 {
1232         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1233         struct osd_otable_cache *ooc = &it->ooi_cache;
1234         const char              *str = (const char *)key;
1235         __u32                    ino;
1236         ENTRY;
1237
1238         /* Forbid to set iteration position after iteration started. */
1239         if (it->ooi_user_ready)
1240                 RETURN(-EPERM);
1241
1242         if (str[0] == '\0')
1243                 RETURN(-EINVAL);
1244
1245         if (sscanf(str, "%u", &ino) <= 0)
1246                 RETURN(-EINVAL);
1247
1248         /* Skip the one that has been processed last time. */
1249         if (ooc->ooc_pos_preload > ++ino)
1250                 ooc->ooc_pos_preload = ino;
1251
1252         RETURN(0);
1253 }
1254
1255 static int osd_otable_it_preload(const struct lu_env *env,
1256                                  struct osd_otable_it *it)
1257 {
1258         struct osd_device       *dev   = it->ooi_dev;
1259         struct osd_scrub        *scrub = &dev->od_scrub;
1260         struct osd_otable_cache *ooc   = &it->ooi_cache;
1261         int                      rc;
1262         ENTRY;
1263
1264         rc = osd_inode_iteration(osd_oti_get(env), dev,
1265                                  OSD_OTABLE_IT_CACHE_SIZE, 1);
1266         if (rc == SCRUB_IT_ALL)
1267                 it->ooi_all_cached = 1;
1268
1269         CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1270                le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1271                ooc->ooc_pos_preload, rc);
1272
1273         if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1274                 scrub->os_waiting = 0;
1275                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1276         }
1277
1278         RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1279 }
1280
1281 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1282 {
1283         struct osd_otable_it    *it     = (struct osd_otable_it *)di;
1284         struct osd_device       *dev    = it->ooi_dev;
1285         struct osd_scrub        *scrub  = &dev->od_scrub;
1286         struct osd_otable_cache *ooc    = &it->ooi_cache;
1287         struct ptlrpc_thread    *thread = &scrub->os_thread;
1288         struct l_wait_info       lwi    = { 0 };
1289         int                      rc;
1290         ENTRY;
1291
1292         LASSERT(it->ooi_user_ready);
1293
1294 again:
1295         if (!thread_is_running(thread) && !it->ooi_used_outside)
1296                 RETURN(1);
1297
1298         if (ooc->ooc_cached_items > 0) {
1299                 ooc->ooc_cached_items--;
1300                 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
1301                                         ~OSD_OTABLE_IT_CACHE_MASK;
1302                 RETURN(0);
1303         }
1304
1305         if (it->ooi_all_cached) {
1306                 l_wait_event(thread->t_ctl_waitq,
1307                              !thread_is_running(thread),
1308                              &lwi);
1309                 RETURN(1);
1310         }
1311
1312         it->ooi_waiting = 1;
1313         l_wait_event(thread->t_ctl_waitq,
1314                      ooc->ooc_pos_preload < scrub->os_pos_current ||
1315                      !thread_is_running(thread),
1316                      &lwi);
1317         it->ooi_waiting = 0;
1318
1319         if (!thread_is_running(thread) && !it->ooi_used_outside)
1320                 RETURN(1);
1321
1322         rc = osd_otable_it_preload(env, it);
1323         if (rc >= 0)
1324                 goto again;
1325
1326         RETURN(rc);
1327 }
1328
1329 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1330                                         const struct dt_it *di)
1331 {
1332         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1333         struct osd_otable_cache *ooc = &it->ooi_cache;
1334
1335         sprintf(it->ooi_key, "%u",
1336                 ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino);
1337         return (struct dt_key *)it->ooi_key;
1338 }
1339
1340 static int osd_otable_it_key_size(const struct lu_env *env,
1341                                   const struct dt_it *di)
1342 {
1343         return sizeof(((struct osd_otable_it *)di)->ooi_key);
1344 }
1345
1346 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1347                              struct dt_rec *rec, __u32 attr)
1348 {
1349         struct osd_otable_it    *it  = (struct osd_otable_it *)di;
1350         struct osd_otable_cache *ooc = &it->ooi_cache;
1351
1352         *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
1353         return 0;
1354 }
1355
1356 static int osd_otable_it_load(const struct lu_env *env,
1357                               const struct dt_it *di, __u64 hash)
1358 {
1359         struct osd_otable_it    *it    = (struct osd_otable_it *)di;
1360         struct osd_device       *dev   = it->ooi_dev;
1361         struct osd_otable_cache *ooc   = &it->ooi_cache;
1362         struct osd_scrub        *scrub = &dev->od_scrub;
1363
1364         if (it->ooi_user_ready)
1365                 return 0;
1366
1367         if (ooc->ooc_pos_preload < LDISKFS_FIRST_INO(osd_sb(dev)))
1368                 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev));
1369         it->ooi_user_ready = 1;
1370         if (!scrub->os_full_speed)
1371                 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1372
1373         /* Unplug OSD layer iteration by the first next() call. */
1374         return osd_otable_it_next(env, (struct dt_it *)it);
1375 }
1376
1377 const struct dt_index_operations osd_otable_ops = {
1378         .dio_it = {
1379                 .init     = osd_otable_it_init,
1380                 .fini     = osd_otable_it_fini,
1381                 .put      = osd_otable_it_put,
1382                 .get      = osd_otable_it_get,
1383                 .next     = osd_otable_it_next,
1384                 .key      = osd_otable_it_key,
1385                 .key_size = osd_otable_it_key_size,
1386                 .rec      = osd_otable_it_rec,
1387                 .load     = osd_otable_it_load,
1388         }
1389 };
1390
1391 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
1392                    int insert)
1393 {
1394         struct osd_inconsistent_item *oii;
1395         struct osd_scrub             *scrub  = &dev->od_scrub;
1396         struct ptlrpc_thread         *thread = &scrub->os_thread;
1397         int                           wakeup = 0;
1398         ENTRY;
1399
1400         OBD_ALLOC_PTR(oii);
1401         if (unlikely(oii == NULL))
1402                 RETURN(-ENOMEM);
1403
1404         CFS_INIT_LIST_HEAD(&oii->oii_list);
1405         oii->oii_cache = *oic;
1406         oii->oii_insert = insert;
1407
1408         spin_lock(&scrub->os_lock);
1409         if (unlikely(!thread_is_running(thread))) {
1410                 spin_unlock(&scrub->os_lock);
1411                 OBD_FREE_PTR(oii);
1412                 RETURN(-EAGAIN);
1413         }
1414
1415         if (cfs_list_empty(&scrub->os_inconsistent_items))
1416                 wakeup = 1;
1417         cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
1418         spin_unlock(&scrub->os_lock);
1419
1420         if (wakeup != 0)
1421                 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1422
1423         RETURN(0);
1424 }
1425
1426 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
1427                    struct osd_inode_id *id)
1428 {
1429         struct osd_scrub             *scrub = &dev->od_scrub;
1430         struct osd_inconsistent_item *oii;
1431         ENTRY;
1432
1433         spin_lock(&scrub->os_lock);
1434         cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
1435                 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
1436                         *id = oii->oii_cache.oic_lid;
1437                         spin_unlock(&scrub->os_lock);
1438                         RETURN(0);
1439                 }
1440         }
1441         spin_unlock(&scrub->os_lock);
1442
1443         RETURN(-ENOENT);
1444 }
1445
1446 static const char *scrub_status_names[] = {
1447         "init",
1448         "scanning",
1449         "completed",
1450         "failed",
1451         "stopped",
1452         "paused",
1453         "crashed",
1454         NULL
1455 };
1456
1457 static const char *scrub_flags_names[] = {
1458         "recreated",
1459         "inconsistent",
1460         "auto",
1461         NULL
1462 };
1463
1464 static const char *scrub_param_names[] = {
1465         "failout",
1466         NULL
1467 };
1468
1469 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
1470                            const char *prefix)
1471 {
1472         int save = *len;
1473         int flag;
1474         int rc;
1475         int i;
1476
1477         rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1478         if (rc <= 0)
1479                 return -ENOSPC;
1480
1481         *buf += rc;
1482         *len -= rc;
1483         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
1484                 if (flag & bits) {
1485                         bits &= ~flag;
1486                         rc = snprintf(*buf, *len, "%s%c", names[i],
1487                                       bits != 0 ? ',' : '\n');
1488                         if (rc <= 0)
1489                                 return -ENOSPC;
1490
1491                         *buf += rc;
1492                         *len -= rc;
1493                 }
1494         }
1495         return save - *len;
1496 }
1497
1498 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
1499 {
1500         int rc;
1501
1502         if (time != 0)
1503                 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
1504                               cfs_time_current_sec() - time);
1505         else
1506                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
1507         if (rc <= 0)
1508                 return -ENOSPC;
1509
1510         *buf += rc;
1511         *len -= rc;
1512         return rc;
1513 }
1514
1515 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
1516 {
1517         int rc;
1518
1519         if (pos != 0)
1520                 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
1521         else
1522                 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
1523         if (rc <= 0)
1524                 return -ENOSPC;
1525
1526         *buf += rc;
1527         *len -= rc;
1528         return rc;
1529 }
1530
1531 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
1532 {
1533         struct osd_scrub  *scrub   = &dev->od_scrub;
1534         struct scrub_file *sf      = &scrub->os_file;
1535         __u64              checked;
1536         __u64              speed;
1537         int                save    = len;
1538         int                ret     = -ENOSPC;
1539         int                rc;
1540
1541         down_read(&scrub->os_rwsem);
1542         rc = snprintf(buf, len,
1543                       "name: OI scrub\n"
1544                       "magic: 0x%x\n"
1545                       "oi_files: %d\n"
1546                       "status: %s\n",
1547                       sf->sf_magic, (int)sf->sf_oi_count,
1548                       scrub_status_names[sf->sf_status]);
1549         if (rc <= 0)
1550                 goto out;
1551
1552         buf += rc;
1553         len -= rc;
1554         rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
1555                              "flags");
1556         if (rc < 0)
1557                 goto out;
1558
1559         rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
1560                              "param");
1561         if (rc < 0)
1562                 goto out;
1563
1564         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
1565                              "time_since_last_completed");
1566         if (rc < 0)
1567                 goto out;
1568
1569         rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
1570                              "time_since_latest_start");
1571         if (rc < 0)
1572                 goto out;
1573
1574         rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
1575                              "time_since_last_checkpoint");
1576         if (rc < 0)
1577                 goto out;
1578
1579         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
1580                             "latest_start_position");
1581         if (rc < 0)
1582                 goto out;
1583
1584         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
1585                             "last_checkpoint_position");
1586         if (rc < 0)
1587                 goto out;
1588
1589         rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
1590                             "first_failure_position");
1591         if (rc < 0)
1592                 goto out;
1593
1594         checked = sf->sf_items_checked + scrub->os_new_checked;
1595         rc = snprintf(buf, len,
1596                       "checked: "LPU64"\n"
1597                       "updated: "LPU64"\n"
1598                       "failed: "LPU64"\n"
1599                       "prior_updated: "LPU64"\n"
1600                       "noscrub: "LPU64"\n"
1601                       "igif: "LPU64"\n"
1602                       "success_count: %u\n",
1603                       checked, sf->sf_items_updated, sf->sf_items_failed,
1604                       sf->sf_items_updated_prior, sf->sf_items_noscrub,
1605                       sf->sf_items_igif, sf->sf_success_count);
1606         if (rc <= 0)
1607                 goto out;
1608
1609         buf += rc;
1610         len -= rc;
1611         speed = checked;
1612         if (thread_is_running(&scrub->os_thread)) {
1613                 cfs_duration_t duration = cfs_time_current() -
1614                                           scrub->os_time_last_checkpoint;
1615                 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
1616                 __u32 rtime = sf->sf_run_time +
1617                               cfs_duration_sec(duration + HALF_SEC);
1618
1619                 if (duration != 0)
1620                         do_div(new_checked, duration);
1621                 if (rtime != 0)
1622                         do_div(speed, rtime);
1623                 rc = snprintf(buf, len,
1624                               "run_time: %u seconds\n"
1625                               "average_speed: "LPU64" objects/sec\n"
1626                               "real-time_speed: "LPU64" objects/sec\n"
1627                               "current_position: %u\n",
1628                               rtime, speed, new_checked, scrub->os_pos_current);
1629         } else {
1630                 if (sf->sf_run_time != 0)
1631                         do_div(speed, sf->sf_run_time);
1632                 rc = snprintf(buf, len,
1633                               "run_time: %u seconds\n"
1634                               "average_speed: "LPU64" objects/sec\n"
1635                               "real-time_speed: N/A\n"
1636                               "current_position: N/A\n",
1637                               sf->sf_run_time, speed);
1638         }
1639         if (rc <= 0)
1640                 goto out;
1641
1642         buf += rc;
1643         len -= rc;
1644         ret = save - len;
1645
1646 out:
1647         up_read(&scrub->os_rwsem);
1648         return ret;
1649 }