4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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.
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
23 * Copyright (c) 2012, 2013, Intel Corporation.
26 * lustre/osd-ldiskfs/osd_scrub.c
28 * Top-level entry points into osd module
30 * The OI scrub is used for rebuilding Object Index files when restores MDT from
33 * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
35 * Author: Fan Yong <yong.fan@whamcloud.com>
39 # define EXPORT_SYMTAB
41 #define DEBUG_SUBSYSTEM S_MDS
43 #include <lustre/lustre_idl.h>
44 #include <lustre_disk.h>
45 #include <dt_object.h>
47 #include "osd_internal.h"
49 #include "osd_scrub.h"
51 #define HALF_SEC (CFS_HZ >> 1)
53 #define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL
55 #define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */
56 #define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */
57 #define SCRUB_NEXT_EXIT 3 /* exit all the loops */
58 #define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */
59 #define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */
60 #define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */
61 #define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */
62 #define SCRUB_NEXT_NOLMA 8 /* the inode has no FID-in-LMA */
66 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
68 return container_of0(scrub, struct osd_device, od_scrub);
71 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
73 return osd_sb(osd_scrub2dev(scrub));
76 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
77 struct osd_otable_cache *ooc)
79 return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
83 * update/insert/delete the specified OI mapping (@fid @id) according to the ops
85 * \retval 1, changed nothing
86 * \retval 0, changed successfully
87 * \retval -ve, on error
89 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
90 struct osd_device *dev,
91 const struct lu_fid *fid,
92 const struct osd_inode_id *id, int ops)
94 struct lu_fid *oi_fid = &info->oti_fid2;
95 struct osd_inode_id *oi_id = &info->oti_id2;
96 struct iam_container *bag;
97 struct iam_path_descr *ipd;
102 fid_cpu_to_be(oi_fid, fid);
103 osd_id_pack(oi_id, id);
104 jh = ldiskfs_journal_start_sb(osd_sb(dev),
105 osd_dto_credits_noquota[ops]);
108 CERROR("%.16s: fail to start trans for scrub store: rc = %d\n",
109 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
113 bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
114 ipd = osd_idx_ipd_get(info->oti_env, bag);
115 if (unlikely(ipd == NULL)) {
116 ldiskfs_journal_stop(jh);
117 CERROR("%.16s: fail to get ipd for scrub store\n",
118 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
123 case DTO_INDEX_UPDATE:
124 rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
125 (struct iam_rec *)oi_id, ipd);
126 if (unlikely(rc == -ENOENT)) {
127 /* Some unlink thread may removed the OI mapping. */
131 case DTO_INDEX_INSERT:
132 rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
133 (struct iam_rec *)oi_id, ipd);
134 if (unlikely(rc == -EEXIST)) {
136 /* XXX: There are trouble things when adding OI
137 * mapping for IGIF object, which may cause
138 * multiple objects to be mapped to the same
139 * IGIF formatted FID. Consider the following
142 * 1) The MDT is upgrading from 1.8 device.
143 * The OI scrub generates IGIF FID1 for the
144 * OBJ1 and adds the OI mapping.
146 * 2) For some reason, the OI scrub does not
147 * process all the IGIF objects completely.
149 * 3) The MDT is backuped and restored against
152 * 4) When the MDT mounts up, the OI scrub will
153 * try to rebuild the OI files. For some IGIF
154 * object, OBJ2, which was not processed by the
155 * OI scrub before the backup/restore, and the
156 * new generated IGIF formatted FID may be just
157 * the FID1, the same as OBJ1.
159 * Under such case, the OI scrub cannot know how
160 * to generate new FID for the OBJ2.
162 * Currently, we do nothing for that. One possible
163 * solution is to generate new normal FID for the
166 * Anyway, it is rare, only exists in theory. */
169 case DTO_INDEX_DELETE:
170 rc = iam_delete(jh, bag, (const struct iam_key *)oi_fid, ipd);
172 /* It is normal that the unlink thread has removed the
173 * OI mapping already. */
178 LASSERTF(0, "Unexpected ops %d\n", ops);
181 osd_ipd_put(info->oti_env, bag, ipd);
182 ldiskfs_journal_stop(jh);
186 /* OI_scrub file ops */
188 static void osd_scrub_file_to_cpu(struct scrub_file *des,
189 struct scrub_file *src)
191 memcpy(des->sf_uuid, src->sf_uuid, 16);
192 des->sf_flags = le64_to_cpu(src->sf_flags);
193 des->sf_magic = le32_to_cpu(src->sf_magic);
194 des->sf_status = le16_to_cpu(src->sf_status);
195 des->sf_param = le16_to_cpu(src->sf_param);
196 des->sf_time_last_complete =
197 le64_to_cpu(src->sf_time_last_complete);
198 des->sf_time_latest_start =
199 le64_to_cpu(src->sf_time_latest_start);
200 des->sf_time_last_checkpoint =
201 le64_to_cpu(src->sf_time_last_checkpoint);
202 des->sf_pos_latest_start =
203 le64_to_cpu(src->sf_pos_latest_start);
204 des->sf_pos_last_checkpoint =
205 le64_to_cpu(src->sf_pos_last_checkpoint);
206 des->sf_pos_first_inconsistent =
207 le64_to_cpu(src->sf_pos_first_inconsistent);
208 des->sf_items_checked =
209 le64_to_cpu(src->sf_items_checked);
210 des->sf_items_updated =
211 le64_to_cpu(src->sf_items_updated);
212 des->sf_items_failed =
213 le64_to_cpu(src->sf_items_failed);
214 des->sf_items_updated_prior =
215 le64_to_cpu(src->sf_items_updated_prior);
216 des->sf_run_time = le32_to_cpu(src->sf_run_time);
217 des->sf_success_count = le32_to_cpu(src->sf_success_count);
218 des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
219 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
222 static void osd_scrub_file_to_le(struct scrub_file *des,
223 struct scrub_file *src)
225 memcpy(des->sf_uuid, src->sf_uuid, 16);
226 des->sf_flags = cpu_to_le64(src->sf_flags);
227 des->sf_magic = cpu_to_le32(src->sf_magic);
228 des->sf_status = cpu_to_le16(src->sf_status);
229 des->sf_param = cpu_to_le16(src->sf_param);
230 des->sf_time_last_complete =
231 cpu_to_le64(src->sf_time_last_complete);
232 des->sf_time_latest_start =
233 cpu_to_le64(src->sf_time_latest_start);
234 des->sf_time_last_checkpoint =
235 cpu_to_le64(src->sf_time_last_checkpoint);
236 des->sf_pos_latest_start =
237 cpu_to_le64(src->sf_pos_latest_start);
238 des->sf_pos_last_checkpoint =
239 cpu_to_le64(src->sf_pos_last_checkpoint);
240 des->sf_pos_first_inconsistent =
241 cpu_to_le64(src->sf_pos_first_inconsistent);
242 des->sf_items_checked =
243 cpu_to_le64(src->sf_items_checked);
244 des->sf_items_updated =
245 cpu_to_le64(src->sf_items_updated);
246 des->sf_items_failed =
247 cpu_to_le64(src->sf_items_failed);
248 des->sf_items_updated_prior =
249 cpu_to_le64(src->sf_items_updated_prior);
250 des->sf_run_time = cpu_to_le32(src->sf_run_time);
251 des->sf_success_count = cpu_to_le32(src->sf_success_count);
252 des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
253 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
256 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
258 struct scrub_file *sf = &scrub->os_file;
260 memset(sf, 0, sizeof(*sf));
261 memcpy(sf->sf_uuid, uuid, 16);
262 sf->sf_magic = SCRUB_MAGIC_V1;
263 sf->sf_status = SS_INIT;
266 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
268 struct scrub_file *sf = &scrub->os_file;
270 CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
271 memcpy(sf->sf_uuid, uuid, 16);
272 sf->sf_status = SS_INIT;
273 sf->sf_flags |= flags;
276 sf->sf_time_latest_start = 0;
277 sf->sf_time_last_checkpoint = 0;
278 sf->sf_pos_latest_start = 0;
279 sf->sf_pos_last_checkpoint = 0;
280 sf->sf_pos_first_inconsistent = 0;
281 sf->sf_items_checked = 0;
282 sf->sf_items_updated = 0;
283 sf->sf_items_failed = 0;
284 sf->sf_items_updated_prior = 0;
285 sf->sf_items_noscrub = 0;
286 sf->sf_items_igif = 0;
289 static int osd_scrub_file_load(struct osd_scrub *scrub)
292 char *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
293 int len = sizeof(scrub->os_file_disk);
296 rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
298 struct scrub_file *sf = &scrub->os_file;
300 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
301 if (sf->sf_magic != SCRUB_MAGIC_V1) {
302 CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
303 name, sf->sf_magic, SCRUB_MAGIC_V1);
304 /* Process it as new scrub file. */
309 } else if (rc != 0) {
310 CERROR("%.16s: fail to load scrub file, expected = %d, "
311 "rc = %d\n", name, len, rc);
315 /* return -ENOENT for empty scrub file case. */
322 int osd_scrub_file_store(struct osd_scrub *scrub)
324 struct osd_device *dev;
327 int len = sizeof(scrub->os_file_disk);
331 dev = container_of0(scrub, struct osd_device, od_scrub);
332 credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
333 osd_dto_credits_noquota[DTO_WRITE_BLOCK];
334 jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
337 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
338 LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
342 osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
343 rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
345 ldiskfs_journal_stop(jh);
347 CERROR("%.16s: fail to store scrub file, expected = %d, "
349 LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
351 scrub->os_time_last_checkpoint = cfs_time_current();
352 scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
353 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
359 static int osd_scrub_prep(struct osd_device *dev)
361 struct osd_scrub *scrub = &dev->od_scrub;
362 struct ptlrpc_thread *thread = &scrub->os_thread;
363 struct scrub_file *sf = &scrub->os_file;
364 __u32 flags = scrub->os_start_flags;
368 down_write(&scrub->os_rwsem);
369 if (flags & SS_SET_FAILOUT)
370 sf->sf_param |= SP_FAILOUT;
372 if (flags & SS_CLEAR_FAILOUT)
373 sf->sf_param &= ~SP_FAILOUT;
375 if (flags & SS_RESET)
376 osd_scrub_file_reset(scrub,
377 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
379 if (flags & SS_AUTO) {
380 scrub->os_full_speed = 1;
381 sf->sf_flags |= SF_AUTO;
383 scrub->os_full_speed = 0;
386 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
387 scrub->os_full_speed = 1;
389 scrub->os_in_prior = 0;
390 scrub->os_waiting = 0;
391 scrub->os_paused = 0;
392 scrub->os_new_checked = 0;
393 if (sf->sf_pos_last_checkpoint != 0)
394 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
396 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
398 scrub->os_pos_current = sf->sf_pos_latest_start;
399 sf->sf_status = SS_SCANNING;
400 sf->sf_time_latest_start = cfs_time_current_sec();
401 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
402 rc = osd_scrub_file_store(scrub);
404 spin_lock(&scrub->os_lock);
405 thread_set_flags(thread, SVC_RUNNING);
406 spin_unlock(&scrub->os_lock);
407 cfs_waitq_broadcast(&thread->t_ctl_waitq);
409 up_write(&scrub->os_rwsem);
415 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
416 struct osd_idmap_cache *oic, int val)
418 struct osd_scrub *scrub = &dev->od_scrub;
419 struct scrub_file *sf = &scrub->os_file;
420 struct lu_fid *fid = &oic->oic_fid;
421 struct osd_inode_id *lid = &oic->oic_lid;
422 struct osd_inode_id *lid2 = &info->oti_id;
423 struct osd_inconsistent_item *oii = NULL;
424 struct inode *inode = NULL;
425 int ops = DTO_INDEX_UPDATE;
430 down_write(&scrub->os_rwsem);
431 scrub->os_new_checked++;
435 if (scrub->os_in_prior)
436 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
439 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
442 if (fid_is_igif(fid))
445 if ((val == SCRUB_NEXT_NOLMA) &&
446 (!dev->od_handle_nolma || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
449 if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
452 /* XXX: Currently, no FID-in-LMA for OST object, so osd_oi_lookup()
453 * without checking FLD is enough.
455 * It should be updated if FID-in-LMA for OSD object introduced
457 rc = osd_oi_lookup(info, dev, fid, lid2, false);
463 inode = osd_iget(info, dev, lid);
466 /* Someone removed the inode. */
467 if (rc == -ENOENT || rc == -ESTALE)
472 /* Check whether the inode to be unlinked during OI scrub. */
473 if (unlikely(inode->i_nlink == 0)) {
478 ops = DTO_INDEX_INSERT;
479 idx = osd_oi_fid2idx(dev, fid);
480 if (val == SCRUB_NEXT_NOLMA) {
481 sf->sf_flags |= SF_UPGRADE;
482 scrub->os_full_speed = 1;
483 rc = osd_ea_fid_set(info, inode, fid, 0);
487 if (!(sf->sf_flags & SF_INCONSISTENT))
488 dev->od_igif_inoi = 0;
490 sf->sf_flags |= SF_RECREATED;
491 scrub->os_full_speed = 1;
492 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
493 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
495 } else if (osd_id_eq(lid, lid2)) {
498 sf->sf_flags |= SF_INCONSISTENT;
499 scrub->os_full_speed = 1;
501 /* XXX: If the device is restored from file-level backup, then
502 * some IGIFs may have been already in OI files, and some
503 * may be not yet. Means upgrading from 1.8 may be partly
504 * processed, but some clients may hold some immobilized
505 * IGIFs, and use them to access related objects. Under
506 * such case, OSD does not know whether an given IGIF has
507 * been processed or to be processed, and it also cannot
508 * generate local ino#/gen# directly from the immobilized
509 * IGIF because of the backup/restore. Then force OSD to
510 * lookup the given IGIF in OI files, and if no entry,
511 * then ask the client to retry after upgrading completed.
512 * No better choice. */
513 dev->od_igif_inoi = 1;
516 rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops);
518 if (scrub->os_in_prior)
519 sf->sf_items_updated_prior++;
521 sf->sf_items_updated++;
528 sf->sf_items_failed++;
529 if (sf->sf_pos_first_inconsistent == 0 ||
530 sf->sf_pos_first_inconsistent > lid->oii_ino)
531 sf->sf_pos_first_inconsistent = lid->oii_ino;
536 if (ops == DTO_INDEX_INSERT) {
537 /* There may be conflict unlink during the OI scrub,
538 * if happend, then remove the new added OI mapping. */
539 if (unlikely(inode->i_nlink == 0))
540 osd_scrub_refresh_mapping(info, dev, fid, lid,
544 up_write(&scrub->os_rwsem);
547 LASSERT(!cfs_list_empty(&oii->oii_list));
549 spin_lock(&scrub->os_lock);
550 cfs_list_del_init(&oii->oii_list);
551 spin_unlock(&scrub->os_lock);
554 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
557 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
559 struct scrub_file *sf = &scrub->os_file;
562 if (likely(cfs_time_before(cfs_time_current(),
563 scrub->os_time_next_checkpoint) ||
564 scrub->os_new_checked == 0))
567 down_write(&scrub->os_rwsem);
568 sf->sf_items_checked += scrub->os_new_checked;
569 scrub->os_new_checked = 0;
570 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
571 sf->sf_time_last_checkpoint = cfs_time_current_sec();
572 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
573 scrub->os_time_last_checkpoint);
574 rc = osd_scrub_file_store(scrub);
575 up_write(&scrub->os_rwsem);
580 static void osd_scrub_post(struct osd_scrub *scrub, int result)
582 struct scrub_file *sf = &scrub->os_file;
585 down_write(&scrub->os_rwsem);
586 spin_lock(&scrub->os_lock);
587 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
588 spin_unlock(&scrub->os_lock);
589 if (scrub->os_new_checked > 0) {
590 sf->sf_items_checked += scrub->os_new_checked;
591 scrub->os_new_checked = 0;
592 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
594 sf->sf_time_last_checkpoint = cfs_time_current_sec();
596 struct osd_device *dev =
597 container_of0(scrub, struct osd_device, od_scrub);
599 dev->od_igif_inoi = 1;
600 sf->sf_status = SS_COMPLETED;
601 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
602 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
603 SF_UPGRADE | SF_AUTO);
604 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
605 sf->sf_success_count++;
606 } else if (result == 0) {
607 if (scrub->os_paused)
608 sf->sf_status = SS_PAUSED;
610 sf->sf_status = SS_STOPPED;
612 sf->sf_status = SS_FAILED;
614 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
615 scrub->os_time_last_checkpoint);
616 result = osd_scrub_file_store(scrub);
618 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
619 LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
621 up_write(&scrub->os_rwsem);
626 /* iteration engine */
628 struct osd_iit_param {
629 struct super_block *sb;
630 struct buffer_head *bitmap;
636 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
637 struct osd_device *dev,
638 struct osd_iit_param *param,
639 struct osd_idmap_cache **oic,
642 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
643 struct osd_device *dev,
644 struct osd_iit_param *param,
645 struct osd_idmap_cache *oic,
646 int *noslot, int rc);
648 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
650 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
651 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
652 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
653 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
654 return SCRUB_NEXT_BREAK;
656 *pos = param->gbase + param->offset;
661 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
662 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
663 struct super_block *sb, bool scrub)
665 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
669 osd_id_gen(lid, pos, OSD_OII_NOGEN);
670 inode = osd_iget(info, dev, lid);
673 /* The inode may be removed after bitmap searching, or the
674 * file is new created without inode initialized yet. */
675 if (rc == -ENOENT || rc == -ESTALE)
676 return SCRUB_NEXT_CONTINUE;
678 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
679 LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
683 /* If the inode has no OI mapping, then it is special locally used,
684 * should be invisible to OI scrub or up layer LFSCK. */
685 if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI)) {
687 return SCRUB_NEXT_CONTINUE;
691 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
692 /* Only skip it for the first OI scrub accessing. */
693 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
695 return SCRUB_NEXT_NOSCRUB;
698 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
700 if (fid_is_llog(&lma->lma_self_fid) ||
701 (!scrub && fid_is_internal(&lma->lma_self_fid)))
702 rc = SCRUB_NEXT_CONTINUE;
704 *fid = lma->lma_self_fid;
705 } else if (rc == -ENODATA) {
706 lu_igif_build(fid, inode->i_ino, inode->i_generation);
708 rc = SCRUB_NEXT_NOLMA;
716 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
717 struct osd_iit_param *param,
718 struct osd_idmap_cache **oic, int noslot)
720 struct osd_scrub *scrub = &dev->od_scrub;
721 struct ptlrpc_thread *thread = &scrub->os_thread;
723 struct osd_inode_id *lid;
726 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
727 struct l_wait_info lwi;
729 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
730 l_wait_event(thread->t_ctl_waitq,
731 !cfs_list_empty(&scrub->os_inconsistent_items) ||
732 !thread_is_running(thread),
736 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
737 spin_lock(&scrub->os_lock);
738 thread_set_flags(thread, SVC_STOPPING);
739 spin_unlock(&scrub->os_lock);
740 return SCRUB_NEXT_CRASH;
743 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
744 return SCRUB_NEXT_FATAL;
746 if (unlikely(!thread_is_running(thread)))
747 return SCRUB_NEXT_EXIT;
749 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
750 struct osd_inconsistent_item *oii;
752 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
753 struct osd_inconsistent_item, oii_list);
754 *oic = &oii->oii_cache;
755 scrub->os_in_prior = 1;
760 return SCRUB_NEXT_WAIT;
762 rc = osd_iit_next(param, &scrub->os_pos_current);
766 *oic = &scrub->os_oic;
767 fid = &(*oic)->oic_fid;
768 lid = &(*oic)->oic_lid;
769 rc = osd_iit_iget(info, dev, fid, lid,
770 scrub->os_pos_current, param->sb, true);
774 static int osd_preload_next(struct osd_thread_info *info,
775 struct osd_device *dev, struct osd_iit_param *param,
776 struct osd_idmap_cache **oic, int noslot)
778 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
779 struct osd_scrub *scrub;
780 struct ptlrpc_thread *thread;
783 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
787 scrub = &dev->od_scrub;
788 thread = &scrub->os_thread;
789 if (thread_is_running(thread) &&
790 ooc->ooc_pos_preload >= scrub->os_pos_current)
791 return SCRUB_NEXT_EXIT;
793 rc = osd_iit_iget(info, dev,
794 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
795 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
796 ooc->ooc_pos_preload, param->sb, false);
797 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
798 * ignore the failure, so it still need to skip the inode next time. */
799 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
804 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
806 spin_lock(&scrub->os_lock);
807 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
808 !cfs_list_empty(&scrub->os_inconsistent_items) ||
809 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
810 scrub->os_waiting = 0;
812 scrub->os_waiting = 1;
813 spin_unlock(&scrub->os_lock);
815 return !scrub->os_waiting;
818 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
819 struct osd_iit_param *param,
820 struct osd_idmap_cache *oic, int *noslot, int rc)
822 struct l_wait_info lwi = { 0 };
823 struct osd_scrub *scrub = &dev->od_scrub;
824 struct scrub_file *sf = &scrub->os_file;
825 struct ptlrpc_thread *thread = &scrub->os_thread;
826 struct osd_otable_it *it = dev->od_otable_it;
827 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
830 case SCRUB_NEXT_CONTINUE:
832 case SCRUB_NEXT_WAIT:
834 case SCRUB_NEXT_NOSCRUB:
835 down_write(&scrub->os_rwsem);
836 scrub->os_new_checked++;
837 sf->sf_items_noscrub++;
838 up_write(&scrub->os_rwsem);
842 rc = osd_scrub_check_update(info, dev, oic, rc);
846 rc = osd_scrub_checkpoint(scrub);
848 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
849 LDISKFS_SB(param->sb)->s_es->s_volume_name,
850 scrub->os_pos_current, rc);
851 /* Continue, as long as the scrub itself can go ahead. */
854 if (scrub->os_in_prior) {
855 scrub->os_in_prior = 0;
860 scrub->os_pos_current = param->gbase + ++(param->offset);
863 if (it != NULL && it->ooi_waiting &&
864 ooc->ooc_pos_preload < scrub->os_pos_current) {
865 spin_lock(&scrub->os_lock);
867 cfs_waitq_broadcast(&thread->t_ctl_waitq);
868 spin_unlock(&scrub->os_lock);
871 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
874 if (osd_scrub_has_window(scrub, ooc)) {
879 l_wait_event(thread->t_ctl_waitq,
880 osd_scrub_wakeup(scrub, it),
883 if (osd_scrub_has_window(scrub, ooc))
890 static int osd_preload_exec(struct osd_thread_info *info,
891 struct osd_device *dev, struct osd_iit_param *param,
892 struct osd_idmap_cache *oic, int *noslot, int rc)
894 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
897 ooc->ooc_cached_items++;
898 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
899 ~OSD_OTABLE_IT_CACHE_MASK;
901 return rc > 0 ? 0 : rc;
904 #define SCRUB_IT_ALL 1
905 #define SCRUB_IT_CRASH 2
907 static int osd_inode_iteration(struct osd_thread_info *info,
908 struct osd_device *dev, __u32 max, bool preload)
910 osd_iit_next_policy next;
911 osd_iit_exec_policy exec;
914 struct osd_iit_param param;
921 struct osd_scrub *scrub = &dev->od_scrub;
923 next = osd_scrub_next;
924 exec = osd_scrub_exec;
925 pos = &scrub->os_pos_current;
926 count = &scrub->os_new_checked;
928 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
930 next = osd_preload_next;
931 exec = osd_preload_exec;
932 pos = &ooc->ooc_pos_preload;
933 count = &ooc->ooc_cached_items;
935 param.sb = osd_sb(dev);
936 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
938 while (*pos <= limit && *count < max) {
939 struct osd_idmap_cache *oic = NULL;
941 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
942 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
943 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
944 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
945 if (param.bitmap == NULL) {
946 CERROR("%.16s: fail to read bitmap for %u, "
947 "scrub will stop, urgent mode\n",
948 LDISKFS_SB(param.sb)->s_es->s_volume_name,
953 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
955 rc = next(info, dev, ¶m, &oic, noslot);
957 case SCRUB_NEXT_BREAK:
959 case SCRUB_NEXT_EXIT:
960 brelse(param.bitmap);
962 case SCRUB_NEXT_CRASH:
963 brelse(param.bitmap);
964 RETURN(SCRUB_IT_CRASH);
965 case SCRUB_NEXT_FATAL:
966 brelse(param.bitmap);
970 rc = exec(info, dev, ¶m, oic, &noslot, rc);
972 brelse(param.bitmap);
976 if (preload && dev->od_otable_it->ooi_stopping) {
977 brelse(param.bitmap);
983 brelse(param.bitmap);
985 if (preload && dev->od_otable_it->ooi_stopping)
990 RETURN(SCRUB_IT_ALL);
994 static int osd_otable_it_preload(const struct lu_env *env,
995 struct osd_otable_it *it)
997 struct osd_device *dev = it->ooi_dev;
998 struct osd_scrub *scrub = &dev->od_scrub;
999 struct osd_otable_cache *ooc = &it->ooi_cache;
1003 rc = osd_inode_iteration(osd_oti_get(env), dev,
1004 OSD_OTABLE_IT_CACHE_SIZE, true);
1005 if (rc == SCRUB_IT_ALL)
1006 it->ooi_all_cached = 1;
1008 CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1009 le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1010 ooc->ooc_pos_preload, rc);
1012 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1013 scrub->os_waiting = 0;
1014 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1017 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1020 static int osd_scrub_main(void *args)
1023 struct osd_device *dev = (struct osd_device *)args;
1024 struct osd_scrub *scrub = &dev->od_scrub;
1025 struct ptlrpc_thread *thread = &scrub->os_thread;
1026 struct super_block *sb = osd_sb(dev);
1030 cfs_daemonize("OI_scrub");
1031 rc = lu_env_init(&env, LCT_DT_THREAD);
1033 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
1034 LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1038 rc = osd_scrub_prep(dev);
1040 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
1041 LDISKFS_SB(sb)->s_es->s_volume_name, rc);
1045 if (!scrub->os_full_speed) {
1046 struct l_wait_info lwi = { 0 };
1047 struct osd_otable_it *it = dev->od_otable_it;
1048 struct osd_otable_cache *ooc = &it->ooi_cache;
1050 l_wait_event(thread->t_ctl_waitq,
1051 it->ooi_user_ready || !thread_is_running(thread),
1053 if (unlikely(!thread_is_running(thread)))
1056 scrub->os_pos_current = ooc->ooc_pos_preload;
1059 CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
1060 scrub->os_start_flags, scrub->os_pos_current);
1062 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1063 if (unlikely(rc == SCRUB_IT_CRASH))
1064 GOTO(out, rc = -EINVAL);
1068 osd_scrub_post(scrub, rc);
1069 CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
1070 rc, scrub->os_pos_current);
1073 while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
1074 struct osd_inconsistent_item *oii;
1076 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
1077 struct osd_inconsistent_item, oii_list);
1078 cfs_list_del_init(&oii->oii_list);
1084 spin_lock(&scrub->os_lock);
1085 thread_set_flags(thread, SVC_STOPPED);
1086 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1087 spin_unlock(&scrub->os_lock);
1091 /* initial OI scrub */
1093 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1094 struct dentry *, filldir_t filldir);
1096 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1097 loff_t offset, __u64 ino, unsigned d_type);
1100 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1101 struct dentry *dentry, filldir_t filldir);
1103 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1104 struct dentry *dentry, filldir_t filldir);
1107 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1108 struct dentry *dentry, filldir_t filldir);
1111 OLF_SCAN_SUBITEMS = 0x0001,
1112 OLF_HIDE_FID = 0x0002,
1113 OLF_SHOW_NAME = 0x0004,
1118 struct lu_fid olm_fid;
1120 scandir_t olm_scandir;
1121 filldir_t olm_filldir;
1124 /* Add the new introduced local files in the list in the future. */
1125 static const struct osd_lf_map osd_lf_maps[] = {
1127 { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1131 { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1132 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1133 osd_ios_varfid_fill },
1135 /* NIDTBL_VERSIONS */
1136 { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1137 osd_ios_general_scan, osd_ios_varfid_fill },
1140 { "PENDING", { FID_SEQ_LOCAL_FILE, MDD_ORPHAN_OID, 0 }, 0, NULL, NULL },
1143 { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1144 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1146 /* changelog_catalog */
1147 { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1149 /* changelog_users */
1150 { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1153 { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1157 { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1160 /* lfsck_bookmark */
1161 { "lfsck_bookmark", { FID_SEQ_LOCAL_FILE, LFSCK_BOOKMARK_OID, 0 }, 0,
1165 { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1169 { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1170 OLF_SHOW_NAME, NULL, NULL },
1173 { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1174 osd_ios_general_scan, osd_ios_varfid_fill },
1177 { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1178 osd_ios_general_scan, osd_ios_varfid_fill },
1181 { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1182 OLF_SHOW_NAME, NULL, NULL },
1185 { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1186 OLF_SHOW_NAME, NULL, NULL },
1189 { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1190 OLF_SHOW_NAME, NULL, NULL },
1193 { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1194 OLF_SHOW_NAME, NULL, NULL },
1196 /* lfsck_namespace */
1197 { "lfsck_namespace", { FID_SEQ_LOCAL_FILE, LFSCK_BOOKMARK_OID, 0 }, 0,
1200 /* OBJECTS, upgrade from old device */
1201 { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1203 /* lquota_v2.user, upgrade from old device */
1204 { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1206 /* lquota_v2.group, upgrade from old device */
1207 { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1209 { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1212 struct osd_ios_item {
1213 cfs_list_t oii_list;
1214 struct dentry *oii_dentry;
1215 scandir_t oii_scandir;
1216 filldir_t oii_filldir;
1219 struct osd_ios_filldir_buf {
1220 struct osd_thread_info *oifb_info;
1221 struct osd_device *oifb_dev;
1222 struct dentry *oifb_dentry;
1225 static inline struct dentry *
1226 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1228 struct dentry *dentry;
1230 dentry = ll_lookup_one_len(name, parent, namelen);
1231 if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1233 return ERR_PTR(-ENOENT);
1240 osd_ios_llogname2fid(struct lu_fid *fid, const char *name, int namelen)
1245 fid->f_seq = FID_SEQ_LLOG;
1247 id = id * 10 + name[i++] - '0';
1249 fid->f_oid = id & 0x00000000ffffffffULL;
1250 fid->f_ver = id >> 32;
1254 osd_ios_Oname2fid(struct lu_fid *fid, const char *name, int namelen)
1260 seq = seq * 10 + name[i++] - '0';
1262 lu_last_id_fid(fid, seq);
1266 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1267 scandir_t scandir, filldir_t filldir)
1269 struct osd_ios_item *item;
1271 OBD_ALLOC_PTR(item);
1275 CFS_INIT_LIST_HEAD(&item->oii_list);
1276 item->oii_dentry = dget(dentry);
1277 item->oii_scandir = scandir;
1278 item->oii_filldir = filldir;
1279 cfs_list_add_tail(&item->oii_list, &dev->od_ios_list);
1284 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1286 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1287 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1288 * reference the inode, or fixed if it is missing or references another inode.
1291 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1292 struct inode *inode, const struct lu_fid *fid, int flags)
1294 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1295 struct osd_inode_id *id = &info->oti_id;
1296 struct osd_inode_id *id2 = &info->oti_id2;
1297 struct osd_scrub *scrub = &dev->od_scrub;
1298 struct scrub_file *sf = &scrub->os_file;
1303 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1304 if (rc != 0 && rc != -ENODATA)
1307 osd_id_gen(id, inode->i_ino, inode->i_generation);
1308 if (rc == -ENODATA) {
1309 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1310 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1313 rc = osd_ea_fid_set(info, inode, &tfid, 0);
1317 tfid = lma->lma_self_fid;
1320 rc = __osd_oi_lookup(info, dev, &tfid, id2);
1325 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1330 if (osd_id_eq_strict(id, id2))
1333 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1334 osd_scrub_file_reset(scrub,
1335 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1337 rc = osd_scrub_file_store(scrub);
1342 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, DTO_INDEX_UPDATE);
1347 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1348 loff_t offset, __u64 ino, unsigned d_type)
1350 struct osd_ios_filldir_buf *fill_buf = buf;
1351 struct osd_device *dev = fill_buf->oifb_dev;
1352 struct dentry *child;
1356 /* skip any '.' started names */
1360 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1362 RETURN(PTR_ERR(child));
1364 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1366 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1367 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1368 osd_ios_varfid_fill);
1374 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1375 loff_t offset, __u64 ino, unsigned d_type)
1377 struct osd_ios_filldir_buf *fill_buf = buf;
1378 struct osd_device *dev = fill_buf->oifb_dev;
1379 const struct osd_lf_map *map;
1380 struct dentry *child;
1384 /* skip any '.' started names */
1388 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1389 if (strlen(map->olm_name) != namelen)
1392 if (strncmp(map->olm_name, name, namelen) == 0)
1396 if (map->olm_name == NULL)
1399 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1401 RETURN(PTR_ERR(child));
1403 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1404 &map->olm_fid, map->olm_flags);
1405 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1406 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1414 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1415 struct dentry *dentry, filldir_t filldir)
1417 struct osd_ios_filldir_buf buf = { info, dev, dentry };
1418 struct file *filp = &info->oti_it_ea.oie_file;
1419 struct inode *inode = dentry->d_inode;
1420 const struct file_operations *fops = inode->i_fop;
1424 LASSERT(filldir != NULL);
1427 filp->f_dentry = dentry;
1428 filp->f_mode = FMODE_64BITHASH;
1429 filp->f_mapping = inode->i_mapping;
1431 filp->private_data = NULL;
1433 rc = fops->readdir(filp, &buf, filldir);
1434 fops->release(inode, filp);
1440 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1441 struct dentry *dentry, filldir_t filldir)
1443 struct osd_scrub *scrub = &dev->od_scrub;
1444 struct scrub_file *sf = &scrub->os_file;
1445 struct dentry *child;
1449 /* It is existing MDT device. */
1450 dev->od_handle_nolma = 1;
1451 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1452 strlen(dot_lustre_name));
1453 if (IS_ERR(child)) {
1454 rc = PTR_ERR(child);
1455 if (rc == -ENOENT) {
1456 /* It is 1.8 MDT device. */
1457 if (!(sf->sf_flags & SF_UPGRADE)) {
1458 osd_scrub_file_reset(scrub,
1459 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1461 rc = osd_scrub_file_store(scrub);
1467 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1475 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1476 struct dentry *dentry, filldir_t filldir)
1478 struct dentry *child;
1482 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1483 if (!IS_ERR(child)) {
1484 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1487 rc = PTR_ERR(child);
1490 if (rc != 0 && rc != -ENOENT)
1493 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1494 if (!IS_ERR(child)) {
1495 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1498 rc = PTR_ERR(child);
1507 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1508 struct osd_device *dev)
1510 struct osd_ios_item *item = NULL;
1511 scandir_t scandir = osd_ios_general_scan;
1512 filldir_t filldir = osd_ios_root_fill;
1513 struct dentry *dentry = osd_sb(dev)->s_root;
1518 rc = scandir(info, dev, dentry, filldir);
1520 dput(item->oii_dentry);
1527 if (cfs_list_empty(&dev->od_ios_list))
1530 item = cfs_list_entry(dev->od_ios_list.next,
1531 struct osd_ios_item, oii_list);
1532 cfs_list_del_init(&item->oii_list);
1534 LASSERT(item->oii_scandir != NULL);
1535 scandir = item->oii_scandir;
1536 filldir = item->oii_filldir;
1537 dentry = item->oii_dentry;
1540 while (!cfs_list_empty(&dev->od_ios_list)) {
1541 item = cfs_list_entry(dev->od_ios_list.next,
1542 struct osd_ios_item, oii_list);
1543 cfs_list_del_init(&item->oii_list);
1544 dput(item->oii_dentry);
1551 char *osd_lf_fid2name(const struct lu_fid *fid)
1553 const struct osd_lf_map *map = osd_lf_maps;
1555 while (map->olm_name != NULL) {
1556 if (!lu_fid_eq(fid, &map->olm_fid)) {
1561 if (map->olm_flags & OLF_SHOW_NAME)
1562 return map->olm_name;
1570 /* OI scrub start/stop */
1572 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
1574 struct osd_scrub *scrub = &dev->od_scrub;
1575 struct ptlrpc_thread *thread = &scrub->os_thread;
1576 struct l_wait_info lwi = { 0 };
1581 /* os_lock: sync status between stop and scrub thread */
1582 spin_lock(&scrub->os_lock);
1583 if (thread_is_running(thread)) {
1584 spin_unlock(&scrub->os_lock);
1586 } else if (unlikely(thread_is_stopping(thread))) {
1587 spin_unlock(&scrub->os_lock);
1588 l_wait_event(thread->t_ctl_waitq,
1589 thread_is_stopped(thread),
1593 spin_unlock(&scrub->os_lock);
1595 if (scrub->os_file.sf_status == SS_COMPLETED)
1598 scrub->os_start_flags = flags;
1599 thread_set_flags(thread, 0);
1600 rc = cfs_create_thread(osd_scrub_main, dev, 0);
1602 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
1603 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
1607 l_wait_event(thread->t_ctl_waitq,
1608 thread_is_running(thread) || thread_is_stopped(thread),
1614 int osd_scrub_start(struct osd_device *dev)
1619 /* od_otable_mutex: prevent curcurrent start/stop */
1620 mutex_lock(&dev->od_otable_mutex);
1621 rc = do_osd_scrub_start(dev, SS_AUTO);
1622 mutex_unlock(&dev->od_otable_mutex);
1624 RETURN(rc == -EALREADY ? 0 : rc);
1627 static void do_osd_scrub_stop(struct osd_scrub *scrub)
1629 struct ptlrpc_thread *thread = &scrub->os_thread;
1630 struct l_wait_info lwi = { 0 };
1632 /* os_lock: sync status between stop and scrub thread */
1633 spin_lock(&scrub->os_lock);
1634 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
1635 thread_set_flags(thread, SVC_STOPPING);
1636 spin_unlock(&scrub->os_lock);
1637 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1638 l_wait_event(thread->t_ctl_waitq,
1639 thread_is_stopped(thread),
1641 /* Do not skip the last lock/unlock, which can guarantee that
1642 * the caller cannot return until the OI scrub thread exit. */
1643 spin_lock(&scrub->os_lock);
1645 spin_unlock(&scrub->os_lock);
1648 static void osd_scrub_stop(struct osd_device *dev)
1650 /* od_otable_mutex: prevent curcurrent start/stop */
1651 mutex_lock(&dev->od_otable_mutex);
1652 dev->od_scrub.os_paused = 1;
1653 do_osd_scrub_stop(&dev->od_scrub);
1654 mutex_unlock(&dev->od_otable_mutex);
1657 /* OI scrub setup/cleanup */
1659 static const char osd_scrub_name[] = "OI_scrub";
1661 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1663 struct osd_thread_info *info = osd_oti_get(env);
1664 struct osd_scrub *scrub = &dev->od_scrub;
1665 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
1666 struct scrub_file *sf = &scrub->os_file;
1667 struct super_block *sb = osd_sb(dev);
1668 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
1669 struct lvfs_run_ctxt saved;
1675 memset(scrub, 0, sizeof(*scrub));
1676 OBD_SET_CTXT_MAGIC(ctxt);
1677 ctxt->pwdmnt = dev->od_mnt;
1678 ctxt->pwd = dev->od_mnt->mnt_root;
1679 ctxt->fs = get_ds();
1681 cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1682 init_rwsem(&scrub->os_rwsem);
1683 spin_lock_init(&scrub->os_lock);
1684 CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1686 push_ctxt(&saved, ctxt, NULL);
1687 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1689 RETURN(PTR_ERR(filp));
1691 scrub->os_inode = igrab(filp->f_dentry->d_inode);
1692 filp_close(filp, 0);
1693 pop_ctxt(&saved, ctxt, NULL);
1694 ldiskfs_set_inode_state(scrub->os_inode,
1695 LDISKFS_STATE_LUSTRE_NO_OI);
1697 rc = osd_scrub_file_load(scrub);
1698 if (rc == -ENOENT) {
1699 osd_scrub_file_init(scrub, es->s_uuid);
1701 } else if (rc != 0) {
1704 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1705 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1707 } else if (sf->sf_status == SS_SCANNING) {
1708 sf->sf_status = SS_CRASHED;
1713 if (sf->sf_pos_last_checkpoint != 0)
1714 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1716 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
1719 rc = osd_scrub_file_store(scrub);
1724 /* Initialize OI files. */
1725 rc = osd_oi_init(info, dev);
1729 rc = osd_initial_OI_scrub(info, dev);
1731 if ((sf->sf_flags & SF_UPGRADE) &&
1732 !(sf->sf_flags & SF_INCONSISTENT))
1733 /* The 'od_igif_inoi' will be set after the
1734 * upgrading completed, needs NOT remount. */
1735 dev->od_igif_inoi = 0;
1737 /* The 'od_igif_inoi' will be set under the
1739 * 1) new created system, or
1740 * 2) restored from file-level backup, or
1741 * 3) the upgrading completed.
1743 * The 'od_igif_inoi' may be cleared by OI scrub
1744 * later if found that the system is upgrading. */
1745 dev->od_igif_inoi = 1;
1747 if (!dev->od_noscrub &&
1748 ((sf->sf_status == SS_PAUSED) ||
1749 (sf->sf_status == SS_CRASHED &&
1750 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1751 SF_UPGRADE | SF_AUTO)) ||
1752 (sf->sf_status == SS_INIT &&
1753 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
1755 rc = osd_scrub_start(dev);
1761 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1763 struct osd_scrub *scrub = &dev->od_scrub;
1765 LASSERT(dev->od_otable_it == NULL);
1767 if (scrub->os_inode != NULL) {
1768 osd_scrub_stop(dev);
1769 iput(scrub->os_inode);
1770 scrub->os_inode = NULL;
1772 if (dev->od_oi_table != NULL)
1773 osd_oi_fini(osd_oti_get(env), dev);
1776 /* object table based iteration APIs */
1778 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1779 struct dt_object *dt, __u32 attr,
1780 struct lustre_capa *capa)
1782 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1783 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1784 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1785 struct osd_scrub *scrub = &dev->od_scrub;
1786 struct osd_otable_it *it;
1791 /* od_otable_mutex: prevent curcurrent init/fini */
1792 mutex_lock(&dev->od_otable_mutex);
1793 if (dev->od_otable_it != NULL)
1794 GOTO(out, it = ERR_PTR(-EALREADY));
1798 GOTO(out, it = ERR_PTR(-ENOMEM));
1800 dev->od_otable_it = it;
1802 it->ooi_pid = cfs_curproc_pid();
1803 it->ooi_cache.ooc_consumer_idx = -1;
1804 if (flags & DOIF_OUTUSED)
1805 it->ooi_used_outside = 1;
1807 if (flags & DOIF_RESET)
1810 if (valid & DOIV_ERROR_HANDLE) {
1811 if (flags & DOIF_FAILOUT)
1812 start |= SS_SET_FAILOUT;
1814 start |= SS_CLEAR_FAILOUT;
1817 rc = do_osd_scrub_start(dev, start);
1818 if (rc < 0 && rc != -EALREADY) {
1819 dev->od_otable_it = NULL;
1821 GOTO(out, it = ERR_PTR(rc));
1824 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
1829 mutex_unlock(&dev->od_otable_mutex);
1830 return (struct dt_it *)it;
1833 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1835 struct osd_otable_it *it = (struct osd_otable_it *)di;
1836 struct osd_device *dev = it->ooi_dev;
1838 /* od_otable_mutex: prevent curcurrent init/fini */
1839 mutex_lock(&dev->od_otable_mutex);
1840 do_osd_scrub_stop(&dev->od_scrub);
1841 LASSERT(dev->od_otable_it == it);
1843 dev->od_otable_it = NULL;
1844 mutex_unlock(&dev->od_otable_mutex);
1848 static int osd_otable_it_get(const struct lu_env *env,
1849 struct dt_it *di, const struct dt_key *key)
1857 * Sometimes the otable-based iteration driver (LFSCK) may be blocked in OSD
1858 * layer when someone wants to stop/pause the iteration. Under such case, we
1859 * need some mechanism to notify the event and wakeup the blocker.
1861 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1863 struct osd_otable_it *it = (struct osd_otable_it *)di;
1864 struct osd_device *dev = it->ooi_dev;
1866 /* od_otable_mutex: prevent curcurrent init/fini */
1867 mutex_lock(&dev->od_otable_mutex);
1868 if (it->ooi_pid == cfs_curproc_pid()) {
1869 dev->od_scrub.os_paused = 1;
1871 struct ptlrpc_thread *thread = &dev->od_scrub.os_thread;
1873 it->ooi_stopping = 1;
1874 if (it->ooi_waiting)
1875 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1877 mutex_unlock(&dev->od_otable_mutex);
1881 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1883 spin_lock(&scrub->os_lock);
1884 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
1885 scrub->os_waiting || it->ooi_stopping ||
1886 !thread_is_running(&scrub->os_thread))
1887 it->ooi_waiting = 0;
1889 it->ooi_waiting = 1;
1890 spin_unlock(&scrub->os_lock);
1892 return !it->ooi_waiting;
1895 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1897 struct osd_otable_it *it = (struct osd_otable_it *)di;
1898 struct osd_device *dev = it->ooi_dev;
1899 struct osd_scrub *scrub = &dev->od_scrub;
1900 struct osd_otable_cache *ooc = &it->ooi_cache;
1901 struct ptlrpc_thread *thread = &scrub->os_thread;
1902 struct l_wait_info lwi = { 0 };
1906 LASSERT(it->ooi_user_ready);
1909 if (!thread_is_running(thread) && !it->ooi_used_outside)
1912 if (ooc->ooc_cached_items > 0) {
1913 ooc->ooc_cached_items--;
1914 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
1915 ~OSD_OTABLE_IT_CACHE_MASK;
1919 if (it->ooi_all_cached) {
1920 l_wait_event(thread->t_ctl_waitq,
1921 !thread_is_running(thread),
1926 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1927 spin_lock(&scrub->os_lock);
1928 scrub->os_waiting = 0;
1929 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1930 spin_unlock(&scrub->os_lock);
1933 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
1934 l_wait_event(thread->t_ctl_waitq,
1935 osd_otable_it_wakeup(scrub, it),
1938 if (!thread_is_running(thread) && !it->ooi_used_outside)
1941 if (it->ooi_stopping)
1944 rc = osd_otable_it_preload(env, it);
1951 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1952 const struct dt_it *di)
1957 static int osd_otable_it_key_size(const struct lu_env *env,
1958 const struct dt_it *di)
1960 return sizeof(__u64);
1963 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1964 struct dt_rec *rec, __u32 attr)
1966 struct osd_otable_it *it = (struct osd_otable_it *)di;
1967 struct osd_otable_cache *ooc = &it->ooi_cache;
1969 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
1971 /* Filter out Invald FID already. */
1972 LASSERTF(fid_is_sane((struct lu_fid *)rec),
1973 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
1974 PFID((struct lu_fid *)rec),
1975 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
1980 static __u64 osd_otable_it_store(const struct lu_env *env,
1981 const struct dt_it *di)
1983 struct osd_otable_it *it = (struct osd_otable_it *)di;
1984 struct osd_otable_cache *ooc = &it->ooi_cache;
1987 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
1988 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
1990 hash = ooc->ooc_pos_preload;
1995 * Set the OSD layer iteration start position as the specified hash.
1997 static int osd_otable_it_load(const struct lu_env *env,
1998 const struct dt_it *di, __u64 hash)
2000 struct osd_otable_it *it = (struct osd_otable_it *)di;
2001 struct osd_device *dev = it->ooi_dev;
2002 struct osd_otable_cache *ooc = &it->ooi_cache;
2003 struct osd_scrub *scrub = &dev->od_scrub;
2007 /* Forbid to set iteration position after iteration started. */
2008 if (it->ooi_user_ready)
2011 if (hash > OSD_OTABLE_MAX_HASH)
2012 hash = OSD_OTABLE_MAX_HASH;
2014 ooc->ooc_pos_preload = hash;
2015 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2016 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2018 it->ooi_user_ready = 1;
2019 if (!scrub->os_full_speed)
2020 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
2022 /* Unplug OSD layer iteration by the first next() call. */
2023 rc = osd_otable_it_next(env, (struct dt_it *)it);
2028 static int osd_otable_it_key_rec(const struct lu_env *env,
2029 const struct dt_it *di, void *key_rec)
2034 const struct dt_index_operations osd_otable_ops = {
2036 .init = osd_otable_it_init,
2037 .fini = osd_otable_it_fini,
2038 .get = osd_otable_it_get,
2039 .put = osd_otable_it_put,
2040 .next = osd_otable_it_next,
2041 .key = osd_otable_it_key,
2042 .key_size = osd_otable_it_key_size,
2043 .rec = osd_otable_it_rec,
2044 .store = osd_otable_it_store,
2045 .load = osd_otable_it_load,
2046 .key_rec = osd_otable_it_key_rec,
2050 /* high priority inconsistent items list APIs */
2052 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2055 struct osd_inconsistent_item *oii;
2056 struct osd_scrub *scrub = &dev->od_scrub;
2057 struct ptlrpc_thread *thread = &scrub->os_thread;
2062 if (unlikely(oii == NULL))
2065 CFS_INIT_LIST_HEAD(&oii->oii_list);
2066 oii->oii_cache = *oic;
2067 oii->oii_insert = insert;
2069 spin_lock(&scrub->os_lock);
2070 if (unlikely(!thread_is_running(thread))) {
2071 spin_unlock(&scrub->os_lock);
2076 if (cfs_list_empty(&scrub->os_inconsistent_items))
2078 cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2079 spin_unlock(&scrub->os_lock);
2082 cfs_waitq_broadcast(&thread->t_ctl_waitq);
2087 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2088 struct osd_inode_id *id)
2090 struct osd_scrub *scrub = &dev->od_scrub;
2091 struct osd_inconsistent_item *oii;
2094 spin_lock(&scrub->os_lock);
2095 cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2096 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2097 *id = oii->oii_cache.oic_lid;
2098 spin_unlock(&scrub->os_lock);
2102 spin_unlock(&scrub->os_lock);
2109 static const char *scrub_status_names[] = {
2120 static const char *scrub_flags_names[] = {
2128 static const char *scrub_param_names[] = {
2133 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
2141 rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2147 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2150 rc = snprintf(*buf, *len, "%s%c", names[i],
2151 bits != 0 ? ',' : '\n');
2162 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
2167 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
2168 cfs_time_current_sec() - time);
2170 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2179 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
2184 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
2186 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2195 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
2197 struct osd_scrub *scrub = &dev->od_scrub;
2198 struct scrub_file *sf = &scrub->os_file;
2205 down_read(&scrub->os_rwsem);
2206 rc = snprintf(buf, len,
2211 sf->sf_magic, (int)sf->sf_oi_count,
2212 scrub_status_names[sf->sf_status]);
2218 rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
2223 rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
2228 rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
2229 "time_since_last_completed");
2233 rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
2234 "time_since_latest_start");
2238 rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
2239 "time_since_last_checkpoint");
2243 rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
2244 "latest_start_position");
2248 rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
2249 "last_checkpoint_position");
2253 rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
2254 "first_failure_position");
2258 checked = sf->sf_items_checked + scrub->os_new_checked;
2259 rc = snprintf(buf, len,
2260 "checked: "LPU64"\n"
2261 "updated: "LPU64"\n"
2263 "prior_updated: "LPU64"\n"
2264 "noscrub: "LPU64"\n"
2266 "success_count: %u\n",
2267 checked, sf->sf_items_updated, sf->sf_items_failed,
2268 sf->sf_items_updated_prior, sf->sf_items_noscrub,
2269 sf->sf_items_igif, sf->sf_success_count);
2276 if (thread_is_running(&scrub->os_thread)) {
2277 cfs_duration_t duration = cfs_time_current() -
2278 scrub->os_time_last_checkpoint;
2279 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
2280 __u32 rtime = sf->sf_run_time +
2281 cfs_duration_sec(duration + HALF_SEC);
2284 do_div(new_checked, duration);
2286 do_div(speed, rtime);
2287 rc = snprintf(buf, len,
2288 "run_time: %u seconds\n"
2289 "average_speed: "LPU64" objects/sec\n"
2290 "real-time_speed: "LPU64" objects/sec\n"
2291 "current_position: %u\n",
2292 rtime, speed, new_checked, scrub->os_pos_current);
2294 if (sf->sf_run_time != 0)
2295 do_div(speed, sf->sf_run_time);
2296 rc = snprintf(buf, len,
2297 "run_time: %u seconds\n"
2298 "average_speed: "LPU64" objects/sec\n"
2299 "real-time_speed: N/A\n"
2300 "current_position: N/A\n",
2301 sf->sf_run_time, speed);
2311 up_read(&scrub->os_rwsem);