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, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2012, 2017, 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>
38 #define DEBUG_SUBSYSTEM S_LFSCK
40 #include <linux/kthread.h>
41 #include <uapi/linux/lustre/lustre_idl.h>
42 #include <lustre_disk.h>
43 #include <dt_object.h>
44 #include <linux/xattr.h>
45 #include <lustre_scrub.h>
46 #include <lustre_nodemap.h>
48 #include "osd_internal.h"
50 #include "osd_scrub.h"
52 #define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL
54 static inline int osd_scrub_has_window(struct lustre_scrub *scrub,
55 struct osd_otable_cache *ooc)
57 return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
61 * update/insert/delete the specified OI mapping (@fid @id) according to the ops
63 * \retval 1, changed nothing
64 * \retval 0, changed successfully
65 * \retval -ve, on error
67 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
68 struct osd_device *dev,
69 const struct lu_fid *fid,
70 const struct osd_inode_id *id,
72 enum oi_check_flags flags, bool *exist)
78 if (dev->od_scrub.os_scrub.os_file.sf_param & SP_DRYRUN && !force)
81 /* DTO_INDEX_INSERT is enough for other two ops:
82 * delete/update, but save stack. */
83 th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
84 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
87 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
88 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
89 PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
95 case DTO_INDEX_UPDATE:
96 rc = osd_oi_update(info, dev, fid, id, th, flags);
97 if (unlikely(rc == -ENOENT)) {
98 /* Some unlink thread may removed the OI mapping. */
102 case DTO_INDEX_INSERT:
103 rc = osd_oi_insert(info, dev, fid, id, th, flags, exist);
104 if (unlikely(rc == -EEXIST)) {
106 /* XXX: There are trouble things when adding OI
107 * mapping for IGIF object, which may cause
108 * multiple objects to be mapped to the same
109 * IGIF formatted FID. Consider the following
112 * 1) The MDT is upgrading from 1.8 device.
113 * The OI scrub generates IGIF FID1 for the
114 * OBJ1 and adds the OI mapping.
116 * 2) For some reason, the OI scrub does not
117 * process all the IGIF objects completely.
119 * 3) The MDT is backuped and restored against
122 * 4) When the MDT mounts up, the OI scrub will
123 * try to rebuild the OI files. For some IGIF
124 * object, OBJ2, which was not processed by the
125 * OI scrub before the backup/restore, and the
126 * new generated IGIF formatted FID may be just
127 * the FID1, the same as OBJ1.
129 * Under such case, the OI scrub cannot know how
130 * to generate new FID for the OBJ2.
132 * Currently, we do nothing for that. One possible
133 * solution is to generate new normal FID for the
136 * Anyway, it is rare, only exists in theory. */
139 case DTO_INDEX_DELETE:
140 rc = osd_oi_delete(info, dev, fid, th, flags);
142 /* It is normal that the unlink thread has removed the
143 * OI mapping already. */
148 LASSERTF(0, "Unexpected ops %d\n", ops);
152 ldiskfs_journal_stop(th);
154 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
155 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
156 PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
163 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
164 struct inode *inode, const struct lu_fid *fid)
166 struct filter_fid_18_23 *ff = &info->oti_ff_old;
167 struct dentry *dentry = &info->oti_obj_dentry;
168 struct lu_fid *tfid = &info->oti_fid;
169 bool fid_18_23 = false;
175 if (dev->od_scrub.os_scrub.os_file.sf_param & SP_DRYRUN)
178 if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
179 struct ost_id *oi = &info->oti_ostid;
181 fid_to_ostid(fid, oi);
182 ostid_to_fid(tfid, oi, 0);
187 /* We want the LMA to fit into the 256-byte OST inode, so operate
189 * 1) read old XATTR_NAME_FID and save the parent FID;
190 * 2) delete the old XATTR_NAME_FID;
191 * 3) make new LMA and add it;
192 * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
194 * Making the LMA to fit into the 256-byte OST inode can save time for
195 * normal osd_check_lma() and for other OI scrub scanning in future.
196 * So it is worth to make some slow conversion here. */
197 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
198 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
201 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
202 DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
206 /* 1) read old XATTR_NAME_FID and save the parent FID */
207 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
208 if (rc == sizeof(*ff)) {
209 /* 2) delete the old XATTR_NAME_FID */
210 dquot_initialize(inode);
211 rc = osd_removexattr(dentry, inode, XATTR_NAME_FID);
216 } else if (rc != -ENODATA && rc < (int)sizeof(struct filter_fid_24_29)) {
217 GOTO(stop, rc = -EINVAL);
220 /* 3) make new LMA and add it */
221 rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
224 /* If failed, we should try to add the old back. */
227 /* The new PFID EA will only contains ::ff_parent */
228 size = sizeof(ff->ff_parent);
231 /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
235 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
237 if (rc1 != 0 && rc == 0)
244 ldiskfs_journal_stop(jh);
246 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
247 osd_name(dev), PFID(tfid), rc);
252 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
253 struct osd_idmap_cache *oic, int val)
255 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
256 struct scrub_file *sf = &scrub->os_file;
257 struct lu_fid *fid = &oic->oic_fid;
258 struct osd_inode_id *lid = &oic->oic_lid;
259 struct osd_inode_id *lid2 = &info->oti_id;
260 struct osd_inconsistent_item *oii = NULL;
261 struct inode *inode = NULL;
262 int ops = DTO_INDEX_UPDATE;
264 bool converted = false;
268 down_write(&scrub->os_rwsem);
269 scrub->os_new_checked++;
273 if (scrub->os_in_prior)
274 oii = list_entry(oic, struct osd_inconsistent_item,
277 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
280 if (fid_is_igif(fid))
283 if (val == SCRUB_NEXT_OSTOBJ_OLD) {
284 inode = osd_iget(info, dev, lid);
287 /* Someone removed the inode. */
288 if (rc == -ENOENT || rc == -ESTALE)
293 /* The inode has been reused as EA inode, ignore it. */
294 if (unlikely(osd_is_ea_inode(inode)))
297 sf->sf_flags |= SF_UPGRADE;
298 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
299 dev->od_check_ff = 1;
300 rc = osd_scrub_convert_ff(info, dev, inode, fid);
307 if ((val == SCRUB_NEXT_NOLMA) &&
308 (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
311 if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
312 ops = DTO_INDEX_INSERT;
317 rc = osd_oi_lookup(info, dev, fid, lid2,
318 (val == SCRUB_NEXT_OSTOBJ ||
319 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
322 ops = DTO_INDEX_INSERT;
323 else if (rc != -ESTALE)
328 inode = osd_iget(info, dev, lid);
331 /* Someone removed the inode. */
332 if (rc == -ENOENT || rc == -ESTALE)
337 /* The inode has been reused as EA inode, ignore it. */
338 if (unlikely(osd_is_ea_inode(inode)))
342 if (!scrub->os_partial_scan)
343 scrub->os_full_speed = 1;
346 case SCRUB_NEXT_NOLMA:
347 sf->sf_flags |= SF_UPGRADE;
348 if (!(sf->sf_param & SP_DRYRUN)) {
349 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
354 if (!(sf->sf_flags & SF_INCONSISTENT))
355 dev->od_igif_inoi = 0;
357 case SCRUB_NEXT_OSTOBJ:
358 sf->sf_flags |= SF_INCONSISTENT;
359 case SCRUB_NEXT_OSTOBJ_OLD:
364 } else if (osd_id_eq(lid, lid2)) {
366 sf->sf_items_updated++;
370 if (!scrub->os_partial_scan)
371 scrub->os_full_speed = 1;
373 sf->sf_flags |= SF_INCONSISTENT;
375 /* XXX: If the device is restored from file-level backup, then
376 * some IGIFs may have been already in OI files, and some
377 * may be not yet. Means upgrading from 1.8 may be partly
378 * processed, but some clients may hold some immobilized
379 * IGIFs, and use them to access related objects. Under
380 * such case, OSD does not know whether an given IGIF has
381 * been processed or to be processed, and it also cannot
382 * generate local ino#/gen# directly from the immobilized
383 * IGIF because of the backup/restore. Then force OSD to
384 * lookup the given IGIF in OI files, and if no entry,
385 * then ask the client to retry after upgrading completed.
386 * No better choice. */
387 dev->od_igif_inoi = 1;
390 rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
391 (val == SCRUB_NEXT_OSTOBJ ||
392 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0,
395 if (scrub->os_in_prior)
396 sf->sf_items_updated_prior++;
398 sf->sf_items_updated++;
400 if (ops == DTO_INDEX_INSERT && val == 0 && !exist) {
401 int idx = osd_oi_fid2idx(dev, fid);
403 sf->sf_flags |= SF_RECREATED;
404 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
405 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
413 sf->sf_items_failed++;
414 if (sf->sf_pos_first_inconsistent == 0 ||
415 sf->sf_pos_first_inconsistent > lid->oii_ino)
416 sf->sf_pos_first_inconsistent = lid->oii_ino;
421 /* There may be conflict unlink during the OI scrub,
422 * if happend, then remove the new added OI mapping. */
423 if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
424 unlikely(ldiskfs_test_inode_state(inode,
425 LDISKFS_STATE_LUSTRE_DESTROY)))
426 osd_scrub_refresh_mapping(info, dev, fid, lid,
427 DTO_INDEX_DELETE, false,
428 (val == SCRUB_NEXT_OSTOBJ ||
429 val == SCRUB_NEXT_OSTOBJ_OLD) ?
430 OI_KNOWN_ON_OST : 0, NULL);
431 up_write(&scrub->os_rwsem);
433 if (inode != NULL && !IS_ERR(inode))
437 spin_lock(&scrub->os_lock);
438 if (likely(!list_empty(&oii->oii_list)))
439 list_del(&oii->oii_list);
440 spin_unlock(&scrub->os_lock);
445 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
448 static int osd_scrub_prep(const struct lu_env *env, struct osd_device *dev)
450 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
451 struct ptlrpc_thread *thread = &scrub->os_thread;
452 struct scrub_file *sf = &scrub->os_file;
453 __u32 flags = scrub->os_start_flags;
455 bool drop_dryrun = false;
458 CDEBUG(D_LFSCK, "%s: OI scrub prep, flags = 0x%x\n",
459 osd_scrub2name(scrub), flags);
461 down_write(&scrub->os_rwsem);
462 if (flags & SS_SET_FAILOUT)
463 sf->sf_param |= SP_FAILOUT;
464 else if (flags & SS_CLEAR_FAILOUT)
465 sf->sf_param &= ~SP_FAILOUT;
467 if (flags & SS_SET_DRYRUN) {
468 sf->sf_param |= SP_DRYRUN;
469 } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
470 sf->sf_param &= ~SP_DRYRUN;
474 if (flags & SS_RESET)
475 scrub_file_reset(scrub, dev->od_uuid, 0);
477 if (flags & SS_AUTO_FULL) {
478 scrub->os_full_speed = 1;
479 scrub->os_partial_scan = 0;
480 sf->sf_flags |= SF_AUTO;
481 } else if (flags & SS_AUTO_PARTIAL) {
482 scrub->os_full_speed = 0;
483 scrub->os_partial_scan = 1;
484 sf->sf_flags |= SF_AUTO;
485 } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
487 scrub->os_full_speed = 1;
488 scrub->os_partial_scan = 0;
490 scrub->os_full_speed = 0;
491 scrub->os_partial_scan = 0;
494 spin_lock(&scrub->os_lock);
495 scrub->os_in_prior = 0;
496 scrub->os_waiting = 0;
497 scrub->os_paused = 0;
498 scrub->os_in_join = 0;
499 scrub->os_full_scrub = 0;
500 spin_unlock(&scrub->os_lock);
501 scrub->os_new_checked = 0;
502 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
503 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
504 else if (sf->sf_pos_last_checkpoint != 0)
505 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
507 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
509 scrub->os_pos_current = sf->sf_pos_latest_start;
510 sf->sf_status = SS_SCANNING;
511 sf->sf_time_latest_start = ktime_get_real_seconds();
512 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
513 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
514 rc = scrub_file_store(env, scrub);
516 spin_lock(&scrub->os_lock);
517 thread_set_flags(thread, SVC_RUNNING);
518 spin_unlock(&scrub->os_lock);
519 wake_up_all(&thread->t_ctl_waitq);
521 up_write(&scrub->os_rwsem);
526 static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev,
529 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
530 struct scrub_file *sf = &scrub->os_file;
534 CDEBUG(D_LFSCK, "%s: OI scrub post with result = %d\n",
535 osd_scrub2name(scrub), result);
537 down_write(&scrub->os_rwsem);
538 spin_lock(&scrub->os_lock);
539 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
540 spin_unlock(&scrub->os_lock);
541 if (scrub->os_new_checked > 0) {
542 sf->sf_items_checked += scrub->os_new_checked;
543 scrub->os_new_checked = 0;
544 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
546 sf->sf_time_last_checkpoint = ktime_get_real_seconds();
548 dev->od_igif_inoi = 1;
549 dev->od_check_ff = 0;
550 sf->sf_status = SS_COMPLETED;
551 if (!(sf->sf_param & SP_DRYRUN)) {
552 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
553 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
554 SF_UPGRADE | SF_AUTO);
556 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
557 sf->sf_success_count++;
558 } else if (result == 0) {
559 if (scrub->os_paused)
560 sf->sf_status = SS_PAUSED;
562 sf->sf_status = SS_STOPPED;
564 sf->sf_status = SS_FAILED;
566 sf->sf_run_time += ktime_get_seconds() -
567 scrub->os_time_last_checkpoint;
569 rc = scrub_file_store(env, scrub);
570 up_write(&scrub->os_rwsem);
572 RETURN(rc < 0 ? rc : result);
575 /* iteration engine */
577 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
578 struct osd_device *dev,
579 struct osd_iit_param *param,
580 struct osd_idmap_cache **oic,
583 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
584 struct osd_device *dev,
585 struct osd_iit_param *param,
586 struct osd_idmap_cache *oic,
587 bool *noslot, int rc);
589 static int osd_iit_next(struct osd_iit_param *param, __u64 *pos)
594 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
595 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
596 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
597 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
598 return SCRUB_NEXT_BREAK;
601 offset = param->offset++;
602 if (unlikely(*pos == param->gbase + offset && *pos != param->start)) {
603 /* We should NOT find the same object more than once. */
604 CERROR("%s: scan the same object multiple times at the pos: "
605 "group = %u, base = %u, offset = %u, start = %u\n",
606 param->sb->s_id, (__u32)param->bg, param->gbase,
607 offset, param->start);
611 *pos = param->gbase + offset;
616 * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
617 * \retval 0: FID-on-MDT
619 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
620 struct osd_device *dev,
623 /* XXX: The initial OI scrub will scan the top level /O to generate
624 * a small local FLDB according to the <seq>. If the given FID
625 * is in the local FLDB, then it is FID-on-OST; otherwise it's
626 * quite possible for FID-on-MDT. */
628 return SCRUB_NEXT_OSTOBJ_OLD;
633 static int osd_scrub_get_fid(struct osd_thread_info *info,
634 struct osd_device *dev, struct inode *inode,
635 struct lu_fid *fid, bool scrub)
637 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
638 bool has_lma = false;
641 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
642 &info->oti_ost_attrs);
645 if (lma->lma_compat & LMAC_NOT_IN_OI ||
646 lma->lma_incompat & LMAI_AGENT)
647 return SCRUB_NEXT_CONTINUE;
649 *fid = lma->lma_self_fid;
653 if (lma->lma_compat & LMAC_FID_ON_OST)
654 return SCRUB_NEXT_OSTOBJ;
656 if (fid_is_idif(fid))
657 return SCRUB_NEXT_OSTOBJ_OLD;
659 /* For local object. */
660 if (fid_is_internal(fid))
663 /* For external visible MDT-object with non-normal FID. */
664 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
667 /* For the object with normal FID, it may be MDT-object,
668 * or may be 2.4 OST-object, need further distinguish.
669 * Fall through to next section. */
672 if (rc == -ENODATA || rc == 0) {
673 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
676 /* It is 2.3 or older OST-object. */
677 rc = SCRUB_NEXT_OSTOBJ_OLD;
683 /* It is FID-on-OST, but we do not know how
684 * to generate its FID, ignore it directly. */
685 rc = SCRUB_NEXT_CONTINUE;
687 /* It is 2.4 or newer OST-object. */
688 rc = SCRUB_NEXT_OSTOBJ_OLD;
696 if (dev->od_scrub.os_scrub.os_convert_igif) {
697 lu_igif_build(fid, inode->i_ino,
698 inode->i_generation);
700 rc = SCRUB_NEXT_NOLMA;
704 /* It may be FID-on-OST, or may be FID for
705 * non-MDT0, anyway, we do not know how to
706 * generate its FID, ignore it directly. */
707 rc = SCRUB_NEXT_CONTINUE;
712 /* For OI scrub case only: the object has LMA but has no ff
713 * (or ff crashed). It may be MDT-object, may be OST-object
714 * with crashed ff. The last check is local FLDB. */
715 rc = osd_scrub_check_local_fldb(info, dev, fid);
721 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
722 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
723 struct super_block *sb, bool scrub)
729 /* Not handle the backend root object and agent parent object.
730 * They are neither visible to namespace nor have OI mappings. */
731 if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
732 is_remote_parent_ino(dev, pos)))
733 RETURN(SCRUB_NEXT_CONTINUE);
735 /* Skip project quota inode since it is greater than s_first_ino. */
736 #ifdef HAVE_PROJECT_QUOTA
737 if (ldiskfs_has_feature_project(sb) &&
738 pos == le32_to_cpu(LDISKFS_SB(sb)->s_es->s_prj_quota_inum))
739 RETURN(SCRUB_NEXT_CONTINUE);
742 osd_id_gen(lid, pos, OSD_OII_NOGEN);
743 inode = osd_iget(info, dev, lid);
746 /* The inode may be removed after bitmap searching, or the
747 * file is new created without inode initialized yet. */
748 if (rc == -ENOENT || rc == -ESTALE)
749 RETURN(SCRUB_NEXT_CONTINUE);
751 CDEBUG(D_LFSCK, "%s: fail to read inode, ino# = %u: "
752 "rc = %d\n", osd_dev2name(dev), pos, rc);
756 /* It is an EA inode, no OI mapping for it, skip it. */
757 if (osd_is_ea_inode(inode))
758 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
761 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
762 /* Only skip it for the first OI scrub accessing. */
763 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
764 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
767 rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
776 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
777 struct osd_iit_param *param,
778 struct osd_idmap_cache **oic, const bool noslot)
780 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
781 struct ptlrpc_thread *thread = &scrub->os_thread;
783 struct osd_inode_id *lid;
786 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
787 struct l_wait_info lwi;
789 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
790 if (likely(lwi.lwi_timeout > 0))
791 l_wait_event(thread->t_ctl_waitq,
792 !list_empty(&scrub->os_inconsistent_items) ||
793 !thread_is_running(thread),
797 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
798 spin_lock(&scrub->os_lock);
799 thread_set_flags(thread, SVC_STOPPING);
800 spin_unlock(&scrub->os_lock);
801 return SCRUB_NEXT_CRASH;
804 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
805 return SCRUB_NEXT_FATAL;
807 if (unlikely(!thread_is_running(thread)))
808 return SCRUB_NEXT_EXIT;
810 if (!list_empty(&scrub->os_inconsistent_items)) {
811 spin_lock(&scrub->os_lock);
812 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
813 struct osd_inconsistent_item *oii;
815 oii = list_entry(scrub->os_inconsistent_items.next,
816 struct osd_inconsistent_item, oii_list);
817 spin_unlock(&scrub->os_lock);
819 *oic = &oii->oii_cache;
820 scrub->os_in_prior = 1;
824 spin_unlock(&scrub->os_lock);
828 return SCRUB_NEXT_WAIT;
830 rc = osd_iit_next(param, &scrub->os_pos_current);
834 *oic = &dev->od_scrub.os_oic;
835 fid = &(*oic)->oic_fid;
836 lid = &(*oic)->oic_lid;
837 rc = osd_iit_iget(info, dev, fid, lid,
838 scrub->os_pos_current, param->sb, true);
842 static int osd_preload_next(struct osd_thread_info *info,
843 struct osd_device *dev, struct osd_iit_param *param,
844 struct osd_idmap_cache **oic, const bool noslot)
846 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
847 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
848 struct ptlrpc_thread *thread = &scrub->os_thread;
851 if (thread_is_running(thread) &&
852 ooc->ooc_pos_preload >= scrub->os_pos_current)
853 return SCRUB_NEXT_EXIT;
855 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
859 rc = osd_iit_iget(info, dev,
860 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
861 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
862 ooc->ooc_pos_preload, param->sb, false);
867 osd_scrub_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
869 spin_lock(&scrub->os_lock);
870 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
871 !list_empty(&scrub->os_inconsistent_items) ||
872 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
873 scrub->os_waiting = 0;
875 scrub->os_waiting = 1;
876 spin_unlock(&scrub->os_lock);
878 return !scrub->os_waiting;
881 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
882 struct osd_iit_param *param,
883 struct osd_idmap_cache *oic, bool *noslot, int rc)
885 struct l_wait_info lwi = { 0 };
886 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
887 struct scrub_file *sf = &scrub->os_file;
888 struct ptlrpc_thread *thread = &scrub->os_thread;
889 struct osd_otable_it *it = dev->od_otable_it;
890 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
893 case SCRUB_NEXT_NOSCRUB:
894 down_write(&scrub->os_rwsem);
895 scrub->os_new_checked++;
896 sf->sf_items_noscrub++;
897 up_write(&scrub->os_rwsem);
898 case SCRUB_NEXT_CONTINUE:
899 case SCRUB_NEXT_WAIT:
903 rc = osd_scrub_check_update(info, dev, oic, rc);
905 scrub->os_in_prior = 0;
909 rc = scrub_checkpoint(info->oti_env, scrub);
911 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %llu: "
912 "rc = %d\n", osd_scrub2name(scrub),
913 scrub->os_pos_current, rc);
914 /* Continue, as long as the scrub itself can go ahead. */
917 if (scrub->os_in_prior) {
918 scrub->os_in_prior = 0;
923 if (it != NULL && it->ooi_waiting && ooc != NULL &&
924 ooc->ooc_pos_preload < scrub->os_pos_current) {
925 spin_lock(&scrub->os_lock);
927 wake_up_all(&thread->t_ctl_waitq);
928 spin_unlock(&scrub->os_lock);
931 if (rc == SCRUB_NEXT_CONTINUE)
934 if (scrub->os_full_speed || !ooc || osd_scrub_has_window(scrub, ooc)) {
940 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
943 if (!ooc || osd_scrub_has_window(scrub, ooc))
950 static int osd_preload_exec(struct osd_thread_info *info,
951 struct osd_device *dev, struct osd_iit_param *param,
952 struct osd_idmap_cache *oic, bool *noslot, int rc)
954 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
957 ooc->ooc_cached_items++;
958 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
959 ~OSD_OTABLE_IT_CACHE_MASK;
961 return rc > 0 ? 0 : rc;
964 #define SCRUB_IT_ALL 1
965 #define SCRUB_IT_CRASH 2
967 static void osd_scrub_join(const struct lu_env *env, struct osd_device *dev,
968 __u32 flags, bool inconsistent)
970 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
971 struct ptlrpc_thread *thread = &scrub->os_thread;
972 struct scrub_file *sf = &scrub->os_file;
976 LASSERT(!(flags & SS_AUTO_PARTIAL));
978 down_write(&scrub->os_rwsem);
979 scrub->os_in_join = 1;
980 if (flags & SS_SET_FAILOUT)
981 sf->sf_param |= SP_FAILOUT;
982 else if (flags & SS_CLEAR_FAILOUT)
983 sf->sf_param &= ~SP_FAILOUT;
985 if (flags & SS_SET_DRYRUN)
986 sf->sf_param |= SP_DRYRUN;
987 else if (flags & SS_CLEAR_DRYRUN)
988 sf->sf_param &= ~SP_DRYRUN;
990 if (flags & SS_RESET) {
991 scrub_file_reset(scrub, dev->od_uuid,
992 inconsistent ? SF_INCONSISTENT : 0);
993 sf->sf_status = SS_SCANNING;
996 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
997 scrub->os_full_speed = 1;
999 scrub->os_full_speed = 0;
1001 if (flags & SS_AUTO_FULL) {
1002 sf->sf_flags |= SF_AUTO;
1003 scrub->os_full_speed = 1;
1006 scrub->os_new_checked = 0;
1007 if (sf->sf_pos_last_checkpoint != 0)
1008 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1010 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1012 scrub->os_pos_current = sf->sf_pos_latest_start;
1013 sf->sf_time_latest_start = ktime_get_real_seconds();
1014 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1015 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1016 rc = scrub_file_store(env, scrub);
1018 spin_lock(&scrub->os_lock);
1019 scrub->os_waiting = 0;
1020 scrub->os_paused = 0;
1021 scrub->os_partial_scan = 0;
1022 scrub->os_in_join = 0;
1023 scrub->os_full_scrub = 0;
1024 spin_unlock(&scrub->os_lock);
1025 wake_up_all(&thread->t_ctl_waitq);
1026 up_write(&scrub->os_rwsem);
1028 CDEBUG(D_LFSCK, "%s: joined in the OI scrub with flag %u: rc = %d\n",
1029 osd_scrub2name(scrub), flags, rc);
1034 static int osd_inode_iteration(struct osd_thread_info *info,
1035 struct osd_device *dev, __u32 max, bool preload)
1037 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1038 struct ptlrpc_thread *thread = &scrub->os_thread;
1039 struct scrub_file *sf = &scrub->os_file;
1040 osd_iit_next_policy next;
1041 osd_iit_exec_policy exec;
1044 struct osd_iit_param *param;
1045 struct l_wait_info lwi = { 0 };
1054 param = &dev->od_scrub.os_iit_param;
1055 memset(param, 0, sizeof(*param));
1056 param->sb = osd_sb(dev);
1058 while (scrub->os_partial_scan && !scrub->os_in_join) {
1059 struct osd_idmap_cache *oic = NULL;
1061 rc = osd_scrub_next(info, dev, param, &oic, noslot);
1063 case SCRUB_NEXT_EXIT:
1065 case SCRUB_NEXT_CRASH:
1066 RETURN(SCRUB_IT_CRASH);
1067 case SCRUB_NEXT_FATAL:
1069 case SCRUB_NEXT_WAIT: {
1070 struct kstatfs *ksfs = &info->oti_ksfs;
1073 if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1074 unlikely(sf->sf_items_updated_prior == 0))
1077 if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1078 scrub->os_full_scrub) {
1079 osd_scrub_join(info->oti_env, dev,
1080 SS_AUTO_FULL | SS_RESET, true);
1084 rc = param->sb->s_op->statfs(param->sb->s_root, ksfs);
1086 __u64 used = ksfs->f_files - ksfs->f_ffree;
1088 do_div(used, sf->sf_items_updated_prior);
1089 /* If we hit too much inconsistent OI
1090 * mappings during the partial scan,
1091 * then scan the device completely. */
1092 if (used < dev->od_full_scrub_ratio) {
1093 osd_scrub_join(info->oti_env, dev,
1094 SS_AUTO_FULL | SS_RESET, true);
1100 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1104 saved_flags = sf->sf_flags;
1105 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1106 SF_UPGRADE | SF_AUTO);
1107 sf->sf_status = SS_COMPLETED;
1108 l_wait_event(thread->t_ctl_waitq,
1109 !thread_is_running(thread) ||
1110 !scrub->os_partial_scan ||
1111 scrub->os_in_join ||
1112 !list_empty(&scrub->os_inconsistent_items),
1114 sf->sf_flags = saved_flags;
1115 sf->sf_status = SS_SCANNING;
1117 if (unlikely(!thread_is_running(thread)))
1120 if (!scrub->os_partial_scan || scrub->os_in_join)
1126 LASSERTF(rc == 0, "rc = %d\n", rc);
1128 osd_scrub_exec(info, dev, param, oic, &noslot, rc);
1135 l_wait_event(thread->t_ctl_waitq,
1136 !thread_is_running(thread) || !scrub->os_in_join,
1139 if (unlikely(!thread_is_running(thread)))
1145 next = osd_scrub_next;
1146 exec = osd_scrub_exec;
1147 pos = &scrub->os_pos_current;
1148 count = &scrub->os_new_checked;
1149 param->start = *pos;
1150 param->bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param->sb);
1152 (*pos - 1) % LDISKFS_INODES_PER_GROUP(param->sb);
1154 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1156 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1158 next = osd_preload_next;
1159 exec = osd_preload_exec;
1160 pos = &ooc->ooc_pos_preload;
1161 count = &ooc->ooc_cached_items;
1162 param = &dev->od_otable_it->ooi_iit_param;
1166 limit = le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count);
1167 while (*pos <= limit && *count < max) {
1168 struct ldiskfs_group_desc *desc;
1169 bool next_group = false;
1171 desc = ldiskfs_get_group_desc(param->sb, param->bg, NULL);
1175 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1180 param->bitmap = ldiskfs_read_inode_bitmap(param->sb, param->bg);
1181 if (!param->bitmap) {
1182 CERROR("%s: fail to read bitmap for %u, "
1183 "scrub will stop, urgent mode\n",
1184 osd_scrub2name(scrub), (__u32)param->bg);
1189 struct osd_idmap_cache *oic = NULL;
1192 ldiskfs_itable_unused_count(param->sb, desc) >=
1193 LDISKFS_INODES_PER_GROUP(param->sb)) {
1198 rc = next(info, dev, param, &oic, noslot);
1200 case SCRUB_NEXT_BREAK:
1203 case SCRUB_NEXT_EXIT:
1204 brelse(param->bitmap);
1206 case SCRUB_NEXT_CRASH:
1207 brelse(param->bitmap);
1208 RETURN(SCRUB_IT_CRASH);
1209 case SCRUB_NEXT_FATAL:
1210 brelse(param->bitmap);
1214 rc = exec(info, dev, param, oic, &noslot, rc);
1215 } while (!rc && *pos <= limit && *count < max);
1218 if (param->bitmap) {
1219 brelse(param->bitmap);
1220 param->bitmap = NULL;
1230 param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1231 *pos = param->gbase;
1232 param->start = *pos;
1237 RETURN(SCRUB_IT_ALL);
1243 static int osd_otable_it_preload(const struct lu_env *env,
1244 struct osd_otable_it *it)
1246 struct osd_device *dev = it->ooi_dev;
1247 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1248 struct osd_otable_cache *ooc = &it->ooi_cache;
1252 rc = osd_inode_iteration(osd_oti_get(env), dev,
1253 OSD_OTABLE_IT_CACHE_SIZE, true);
1254 if (rc == SCRUB_IT_ALL)
1255 it->ooi_all_cached = 1;
1257 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1258 spin_lock(&scrub->os_lock);
1259 scrub->os_waiting = 0;
1260 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1261 spin_unlock(&scrub->os_lock);
1264 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1267 static int osd_scrub_main(void *args)
1270 struct osd_device *dev = (struct osd_device *)args;
1271 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1272 struct ptlrpc_thread *thread = &scrub->os_thread;
1276 rc = lu_env_init(&env, LCT_LOCAL | LCT_DT_THREAD);
1278 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
1279 osd_scrub2name(scrub), rc);
1283 rc = osd_scrub_prep(&env, dev);
1285 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
1286 osd_scrub2name(scrub), rc);
1290 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1291 struct l_wait_info lwi = { 0 };
1292 struct osd_otable_it *it = dev->od_otable_it;
1293 struct osd_otable_cache *ooc = &it->ooi_cache;
1295 l_wait_event(thread->t_ctl_waitq,
1296 it->ooi_user_ready || !thread_is_running(thread),
1298 if (unlikely(!thread_is_running(thread)))
1301 scrub->os_pos_current = ooc->ooc_pos_preload;
1304 CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %llu\n",
1305 osd_scrub2name(scrub), scrub->os_start_flags,
1306 scrub->os_pos_current);
1308 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1309 if (unlikely(rc == SCRUB_IT_CRASH)) {
1310 spin_lock(&scrub->os_lock);
1311 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
1312 spin_unlock(&scrub->os_lock);
1313 GOTO(out, rc = -EINVAL);
1319 rc = osd_scrub_post(&env, dev, rc);
1320 CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %llu: rc = %d\n",
1321 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1324 while (!list_empty(&scrub->os_inconsistent_items)) {
1325 struct osd_inconsistent_item *oii;
1327 oii = list_entry(scrub->os_inconsistent_items.next,
1328 struct osd_inconsistent_item, oii_list);
1329 list_del_init(&oii->oii_list);
1335 spin_lock(&scrub->os_lock);
1336 thread_set_flags(thread, SVC_STOPPED);
1337 wake_up_all(&thread->t_ctl_waitq);
1338 spin_unlock(&scrub->os_lock);
1342 /* initial OI scrub */
1344 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1345 struct dentry *, filldir_t filldir);
1347 #ifdef HAVE_FILLDIR_USE_CTX
1348 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1349 int namelen, loff_t offset, __u64 ino,
1351 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1352 int namelen, loff_t offset, __u64 ino,
1354 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1355 int namelen, loff_t offset, __u64 ino,
1357 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1358 int namelen, loff_t offset, __u64 ino,
1361 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1362 loff_t offset, __u64 ino, unsigned d_type);
1363 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1364 loff_t offset, __u64 ino, unsigned d_type);
1365 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1366 loff_t offset, __u64 ino, unsigned d_type);
1367 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1368 loff_t offset, __u64 ino, unsigned d_type);
1372 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1373 struct dentry *dentry, filldir_t filldir);
1375 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1376 struct dentry *dentry, filldir_t filldir);
1379 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1380 struct dentry *dentry, filldir_t filldir);
1384 struct lu_fid olm_fid;
1387 scandir_t olm_scandir;
1388 filldir_t olm_filldir;
1391 /* Add the new introduced local files in the list in the future. */
1392 static const struct osd_lf_map osd_lf_maps[] = {
1395 .olm_name = CATLIST,
1397 .f_seq = FID_SEQ_LOCAL_FILE,
1398 .f_oid = LLOG_CATALOGS_OID,
1400 .olm_flags = OLF_SHOW_NAME,
1401 .olm_namelen = sizeof(CATLIST) - 1,
1406 .olm_name = MOUNT_CONFIGS_DIR,
1408 .f_seq = FID_SEQ_LOCAL_FILE,
1409 .f_oid = MGS_CONFIGS_OID,
1411 .olm_flags = OLF_SCAN_SUBITEMS,
1412 .olm_namelen = sizeof(MOUNT_CONFIGS_DIR) - 1,
1413 .olm_scandir = osd_ios_general_scan,
1414 .olm_filldir = osd_ios_varfid_fill,
1417 /* NIDTBL_VERSIONS */
1419 .olm_name = MGS_NIDTBL_DIR,
1420 .olm_flags = OLF_SCAN_SUBITEMS,
1421 .olm_namelen = sizeof(MGS_NIDTBL_DIR) - 1,
1422 .olm_scandir = osd_ios_general_scan,
1423 .olm_filldir = osd_ios_varfid_fill,
1428 .olm_name = MDT_ORPHAN_DIR,
1429 .olm_namelen = sizeof(MDT_ORPHAN_DIR) - 1,
1436 .f_seq = FID_SEQ_ROOT,
1437 .f_oid = FID_OID_ROOT,
1439 .olm_flags = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1440 .olm_namelen = sizeof("ROOT") - 1,
1441 .olm_scandir = osd_ios_ROOT_scan,
1444 /* changelog_catalog */
1446 .olm_name = CHANGELOG_CATALOG,
1447 .olm_namelen = sizeof(CHANGELOG_CATALOG) - 1,
1450 /* changelog_users */
1452 .olm_name = CHANGELOG_USERS,
1453 .olm_namelen = sizeof(CHANGELOG_USERS) - 1,
1460 .f_seq = FID_SEQ_LOCAL_FILE,
1461 .f_oid = FLD_INDEX_OID,
1463 .olm_flags = OLF_SHOW_NAME,
1464 .olm_namelen = sizeof("fld") - 1,
1469 .olm_name = LAST_RCVD,
1471 .f_seq = FID_SEQ_LOCAL_FILE,
1472 .f_oid = LAST_RECV_OID,
1474 .olm_flags = OLF_SHOW_NAME,
1475 .olm_namelen = sizeof(LAST_RCVD) - 1,
1480 .olm_name = REPLY_DATA,
1482 .f_seq = FID_SEQ_LOCAL_FILE,
1483 .f_oid = REPLY_DATA_OID,
1485 .olm_flags = OLF_SHOW_NAME,
1486 .olm_namelen = sizeof(REPLY_DATA) - 1,
1491 .olm_name = LOV_OBJID,
1493 .f_seq = FID_SEQ_LOCAL_FILE,
1494 .f_oid = MDD_LOV_OBJ_OID,
1496 .olm_flags = OLF_SHOW_NAME,
1497 .olm_namelen = sizeof(LOV_OBJID) - 1,
1502 .olm_name = LOV_OBJSEQ,
1504 .f_seq = FID_SEQ_LOCAL_FILE,
1505 .f_oid = MDD_LOV_OBJ_OSEQ,
1507 .olm_flags = OLF_SHOW_NAME,
1508 .olm_namelen = sizeof(LOV_OBJSEQ) - 1,
1513 .olm_name = QMT_DIR,
1514 .olm_flags = OLF_SCAN_SUBITEMS,
1515 .olm_namelen = sizeof(QMT_DIR) - 1,
1516 .olm_scandir = osd_ios_general_scan,
1517 .olm_filldir = osd_ios_varfid_fill,
1522 .olm_name = QSD_DIR,
1523 .olm_flags = OLF_SCAN_SUBITEMS,
1524 .olm_namelen = sizeof(QSD_DIR) - 1,
1525 .olm_scandir = osd_ios_general_scan,
1526 .olm_filldir = osd_ios_varfid_fill,
1531 .olm_name = "seq_ctl",
1533 .f_seq = FID_SEQ_LOCAL_FILE,
1534 .f_oid = FID_SEQ_CTL_OID,
1536 .olm_flags = OLF_SHOW_NAME,
1537 .olm_namelen = sizeof("seq_ctl") - 1,
1542 .olm_name = "seq_srv",
1544 .f_seq = FID_SEQ_LOCAL_FILE,
1545 .f_oid = FID_SEQ_SRV_OID,
1547 .olm_flags = OLF_SHOW_NAME,
1548 .olm_namelen = sizeof("seq_srv") - 1,
1553 .olm_name = HEALTH_CHECK,
1555 .f_seq = FID_SEQ_LOCAL_FILE,
1556 .f_oid = OFD_HEALTH_CHECK_OID,
1558 .olm_flags = OLF_SHOW_NAME,
1559 .olm_namelen = sizeof(HEALTH_CHECK) - 1,
1564 .olm_name = LFSCK_DIR,
1565 .olm_flags = OLF_SCAN_SUBITEMS,
1566 .olm_namelen = sizeof(LFSCK_DIR) - 1,
1567 .olm_scandir = osd_ios_general_scan,
1568 .olm_filldir = osd_ios_varfid_fill,
1571 /* lfsck_bookmark */
1573 .olm_name = LFSCK_BOOKMARK,
1574 .olm_namelen = sizeof(LFSCK_BOOKMARK) - 1,
1579 .olm_name = LFSCK_LAYOUT,
1580 .olm_namelen = sizeof(LFSCK_LAYOUT) - 1,
1583 /* lfsck_namespace */
1585 .olm_name = LFSCK_NAMESPACE,
1586 .olm_namelen = sizeof(LFSCK_NAMESPACE) - 1,
1589 /* OBJECTS, upgrade from old device */
1591 .olm_name = OBJECTS,
1592 .olm_flags = OLF_SCAN_SUBITEMS,
1593 .olm_namelen = sizeof(OBJECTS) - 1,
1594 .olm_scandir = osd_ios_OBJECTS_scan,
1597 /* lquota_v2.user, upgrade from old device */
1599 .olm_name = "lquota_v2.user",
1600 .olm_namelen = sizeof("lquota_v2.user") - 1,
1603 /* lquota_v2.group, upgrade from old device */
1605 .olm_name = "lquota_v2.group",
1606 .olm_namelen = sizeof("lquota_v2.group") - 1,
1609 /* LAST_GROUP, upgrade from old device */
1611 .olm_name = "LAST_GROUP",
1613 .f_seq = FID_SEQ_LOCAL_FILE,
1614 .f_oid = OFD_LAST_GROUP_OID,
1616 .olm_flags = OLF_SHOW_NAME,
1617 .olm_namelen = sizeof("LAST_GROUP") - 1,
1620 /* committed batchid for cross-MDT operation */
1622 .olm_name = "BATCHID",
1624 .f_seq = FID_SEQ_LOCAL_FILE,
1625 .f_oid = BATCHID_COMMITTED_OID,
1627 .olm_flags = OLF_SHOW_NAME,
1628 .olm_namelen = sizeof("BATCHID") - 1,
1631 /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1632 * and f_oid = index for their log files. See lu_update_log{_dir}_fid()
1633 * for more details. */
1637 .olm_name = "update_log",
1639 .f_seq = FID_SEQ_UPDATE_LOG,
1641 .olm_flags = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1642 .olm_namelen = sizeof("update_log") - 1,
1645 /* update_log_dir */
1647 .olm_name = "update_log_dir",
1649 .f_seq = FID_SEQ_UPDATE_LOG_DIR,
1651 .olm_flags = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1653 .olm_namelen = sizeof("update_log_dir") - 1,
1654 .olm_scandir = osd_ios_general_scan,
1655 .olm_filldir = osd_ios_uld_fill,
1660 .olm_name = "lost+found",
1662 .f_seq = FID_SEQ_LOCAL_FILE,
1663 .f_oid = OSD_LPF_OID,
1665 .olm_flags = OLF_SCAN_SUBITEMS,
1666 .olm_namelen = sizeof("lost+found") - 1,
1667 .olm_scandir = osd_ios_general_scan,
1668 .olm_filldir = osd_ios_lf_fill,
1673 .olm_name = HSM_ACTIONS,
1678 .olm_name = LUSTRE_NODEMAP_NAME,
1683 .olm_name = INDEX_BACKUP_DIR,
1685 .f_seq = FID_SEQ_LOCAL_FILE,
1686 .f_oid = INDEX_BACKUP_OID,
1688 .olm_flags = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP,
1689 .olm_namelen = sizeof(INDEX_BACKUP_DIR) - 1,
1690 .olm_scandir = osd_ios_general_scan,
1691 .olm_filldir = osd_ios_varfid_fill,
1699 /* Add the new introduced files under .lustre/ in the list in the future. */
1700 static const struct osd_lf_map osd_dl_maps[] = {
1705 .f_seq = FID_SEQ_DOT_LUSTRE,
1706 .f_oid = FID_OID_DOT_LUSTRE_OBF,
1708 .olm_namelen = sizeof("fid") - 1,
1711 /* .lustre/lost+found */
1713 .olm_name = "lost+found",
1715 .f_seq = FID_SEQ_DOT_LUSTRE,
1716 .f_oid = FID_OID_DOT_LUSTRE_LPF,
1718 .olm_namelen = sizeof("lost+found") - 1,
1726 struct osd_ios_item {
1727 struct list_head oii_list;
1728 struct dentry *oii_dentry;
1729 scandir_t oii_scandir;
1730 filldir_t oii_filldir;
1733 struct osd_ios_filldir_buf {
1734 #ifdef HAVE_DIR_CONTEXT
1735 /* please keep it as first member */
1736 struct dir_context ctx;
1738 struct osd_thread_info *oifb_info;
1739 struct osd_device *oifb_dev;
1740 struct dentry *oifb_dentry;
1745 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1746 scandir_t scandir, filldir_t filldir)
1748 struct osd_ios_item *item;
1751 OBD_ALLOC_PTR(item);
1755 INIT_LIST_HEAD(&item->oii_list);
1756 item->oii_dentry = dget(dentry);
1757 item->oii_scandir = scandir;
1758 item->oii_filldir = filldir;
1759 list_add_tail(&item->oii_list, &dev->od_ios_list);
1764 static bool osd_index_need_recreate(const struct lu_env *env,
1765 struct osd_device *dev, struct inode *inode)
1767 struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
1768 struct iam_container *bag = &iam->od_container;
1772 rc = iam_container_init(bag, &iam->od_descr, inode);
1776 rc = iam_container_setup(bag);
1777 iam_container_fini(bag);
1784 static void osd_ios_index_register(const struct lu_env *env,
1785 struct osd_device *osd,
1786 const struct lu_fid *fid,
1787 struct inode *inode)
1789 struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
1790 struct iam_container *bag = &iam->od_container;
1791 struct super_block *sb = osd_sb(osd);
1792 struct iam_descr *descr;
1798 /* Index must be a regular file. */
1799 if (!S_ISREG(inode->i_mode))
1802 /* Index's size must be block aligned. */
1803 if (inode->i_size < sb->s_blocksize ||
1804 (inode->i_size & (sb->s_blocksize - 1)) != 0)
1807 iam_container_init(bag, &iam->od_descr, inode);
1808 rc = iam_container_setup(bag);
1812 descr = bag->ic_descr;
1813 /* May be regular file with IAM_LFIX_ROOT_MAGIC matched
1814 * coincidentally, or corrupted index object, skip it. */
1815 if (descr->id_ptr_size != 4)
1818 keysize = descr->id_key_size;
1819 recsize = descr->id_rec_size;
1820 rc = osd_index_register(osd, fid, keysize, recsize);
1825 iam_container_fini(bag);
1827 CDEBUG(D_LFSCK, "%s: index object "DFID" (%u/%u) registered\n",
1828 osd_name(osd), PFID(fid), keysize, recsize);
1831 static void osd_index_restore(const struct lu_env *env, struct osd_device *dev,
1832 struct lustre_index_restore_unit *liru,
1833 void *buf, int bufsize)
1835 struct osd_thread_info *info = osd_oti_get(env);
1836 struct osd_inode_id *id = &info->oti_id;
1837 struct lu_fid *tgt_fid = &liru->liru_cfid;
1838 struct inode *bak_inode = NULL;
1839 struct ldiskfs_dir_entry_2 *de = NULL;
1840 struct buffer_head *bh = NULL;
1841 struct dentry *dentry;
1843 struct lu_fid bak_fid;
1847 lustre_fid2lbx(name, tgt_fid, bufsize);
1848 dentry = osd_child_dentry_by_inode(env, dev->od_index_backup_inode,
1849 name, strlen(name));
1850 bh = osd_ldiskfs_find_entry(dev->od_index_backup_inode,
1851 &dentry->d_name, &de, NULL, NULL);
1853 GOTO(log, rc = PTR_ERR(bh));
1855 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
1857 bak_inode = osd_iget_fid(info, dev, id, &bak_fid);
1858 if (IS_ERR(bak_inode))
1859 GOTO(log, rc = PTR_ERR(bak_inode));
1862 /* The OI mapping for index may be invalid, since it will be
1863 * re-created, not update the OI mapping, just cache it in RAM. */
1864 osd_id_gen(id, liru->liru_clid, OSD_OII_NOGEN);
1865 osd_add_oi_cache(info, dev, id, tgt_fid);
1866 rc = lustre_index_restore(env, &dev->od_dt_dev, &liru->liru_pfid,
1867 tgt_fid, &bak_fid, liru->liru_name,
1868 &dev->od_index_backup_list, &dev->od_lock,
1873 CDEBUG(D_WARNING, "%s: restore index '%s' with "DFID": rc = %d\n",
1874 osd_name(dev), liru->liru_name, PFID(tgt_fid), rc);
1878 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1880 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1881 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1882 * reference the inode, or fixed if it is missing or references another inode.
1885 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1886 struct inode *parent, struct inode *inode,
1887 const struct lu_fid *fid, const char *name,
1888 int namelen, int flags)
1890 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1891 struct osd_inode_id *id = &info->oti_id;
1892 struct osd_inode_id *id2 = &info->oti_id2;
1893 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1894 struct scrub_file *sf = &scrub->os_file;
1900 CDEBUG(D_INODE, "%s: child '%.*s' lacks inode: rc = -2\n",
1901 osd_name(dev), namelen, name);
1905 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
1906 &info->oti_ost_attrs);
1907 if (rc != 0 && rc != -ENODATA) {
1908 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1909 "rc = %d\n", osd_name(dev), rc);
1914 osd_id_gen(id, inode->i_ino, inode->i_generation);
1915 if (rc == -ENODATA) {
1916 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1917 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1920 if (flags & OLF_IDX_IN_FID) {
1921 LASSERT(dev->od_index >= 0);
1923 tfid.f_oid = dev->od_index;
1926 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1928 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1929 "scrub: rc = %d\n", osd_name(dev), rc);
1934 if (lma->lma_compat & LMAC_NOT_IN_OI)
1937 tfid = lma->lma_self_fid;
1938 if (lma->lma_compat & LMAC_IDX_BACKUP &&
1939 osd_index_need_recreate(info->oti_env, dev, inode)) {
1940 struct lu_fid *pfid = &info->oti_fid3;
1942 if (parent == osd_sb(dev)->s_root->d_inode) {
1943 lu_local_obj_fid(pfid, OSD_FS_ROOT_OID);
1945 rc = osd_scrub_get_fid(info, dev, parent, pfid,
1951 rc = lustre_liru_new(&dev->od_index_restore_list, pfid,
1952 &tfid, inode->i_ino, name, namelen);
1957 if (!(flags & OLF_NOT_BACKUP))
1958 osd_ios_index_register(info->oti_env, dev, &tfid,
1962 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1967 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1968 DTO_INDEX_INSERT, true, 0, NULL);
1975 if (osd_id_eq_strict(id, id2))
1978 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1979 scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
1980 rc = scrub_file_store(info->oti_env, scrub);
1985 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1986 DTO_INDEX_UPDATE, true, 0, NULL);
1994 * It scans the /lost+found, and for the OST-object (with filter_fid
1995 * or filter_fid_18_23), move them back to its proper /O/<seq>/d<x>.
1997 #ifdef HAVE_FILLDIR_USE_CTX
1998 static int osd_ios_lf_fill(struct dir_context *buf,
2000 static int osd_ios_lf_fill(void *buf,
2002 const char *name, int namelen,
2003 loff_t offset, __u64 ino, unsigned d_type)
2005 struct osd_ios_filldir_buf *fill_buf =
2006 (struct osd_ios_filldir_buf *)buf;
2007 struct osd_thread_info *info = fill_buf->oifb_info;
2008 struct osd_device *dev = fill_buf->oifb_dev;
2009 struct lu_fid *fid = &info->oti_fid;
2010 struct osd_scrub *scrub = &dev->od_scrub;
2011 struct dentry *parent = fill_buf->oifb_dentry;
2012 struct dentry *child;
2013 struct inode *dir = parent->d_inode;
2014 struct inode *inode;
2018 fill_buf->oifb_items++;
2020 /* skip any '.' started names */
2024 scrub->os_lf_scanned++;
2025 child = osd_ios_lookup_one_len(name, parent, namelen);
2026 if (IS_ERR(child)) {
2027 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2028 osd_name(dev), namelen, name, (int)PTR_ERR(child));
2030 } else if (!child->d_inode) {
2032 CDEBUG(D_INODE, "%s: child '%.*s' lacks inode\n",
2033 osd_name(dev), namelen, name);
2037 inode = child->d_inode;
2038 if (S_ISDIR(inode->i_mode)) {
2039 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2042 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2043 "rc = %d\n", osd_name(dev), namelen, name, rc);
2047 if (!S_ISREG(inode->i_mode))
2050 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2051 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2052 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2054 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2055 "/lost+found.\n", namelen, name, PFID(fid));
2056 scrub->os_lf_repaired++;
2058 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2060 osd_name(dev), namelen, name, PFID(fid), rc);
2064 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2065 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2066 * can process them in furtuer. */
2072 scrub->os_lf_failed++;
2074 /* skip the failure to make the scanning to continue. */
2078 #ifdef HAVE_FILLDIR_USE_CTX
2079 static int osd_ios_varfid_fill(struct dir_context *buf,
2081 static int osd_ios_varfid_fill(void *buf,
2083 const char *name, int namelen,
2084 loff_t offset, __u64 ino, unsigned d_type)
2086 struct osd_ios_filldir_buf *fill_buf =
2087 (struct osd_ios_filldir_buf *)buf;
2088 struct osd_device *dev = fill_buf->oifb_dev;
2089 struct dentry *child;
2093 fill_buf->oifb_items++;
2095 /* skip any '.' started names */
2099 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2101 RETURN(PTR_ERR(child));
2103 rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2104 fill_buf->oifb_dentry->d_inode, child->d_inode,
2105 NULL, name, namelen, 0);
2106 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2107 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2108 osd_ios_varfid_fill);
2114 #ifdef HAVE_FILLDIR_USE_CTX
2115 static int osd_ios_dl_fill(struct dir_context *buf,
2117 static int osd_ios_dl_fill(void *buf,
2119 const char *name, int namelen,
2120 loff_t offset, __u64 ino, unsigned d_type)
2122 struct osd_ios_filldir_buf *fill_buf =
2123 (struct osd_ios_filldir_buf *)buf;
2124 struct osd_device *dev = fill_buf->oifb_dev;
2125 const struct osd_lf_map *map;
2126 struct dentry *child;
2130 fill_buf->oifb_items++;
2132 /* skip any '.' started names */
2136 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2137 if (map->olm_namelen != namelen)
2140 if (strncmp(map->olm_name, name, namelen) == 0)
2144 if (map->olm_name == NULL)
2147 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2149 RETURN(PTR_ERR(child));
2151 rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2152 fill_buf->oifb_dentry->d_inode, child->d_inode,
2153 &map->olm_fid, name, namelen, map->olm_flags);
2159 #ifdef HAVE_FILLDIR_USE_CTX
2160 static int osd_ios_uld_fill(struct dir_context *buf,
2162 static int osd_ios_uld_fill(void *buf,
2164 const char *name, int namelen,
2165 loff_t offset, __u64 ino, unsigned d_type)
2167 struct osd_ios_filldir_buf *fill_buf =
2168 (struct osd_ios_filldir_buf *)buf;
2169 struct dentry *child;
2174 fill_buf->oifb_items++;
2176 /* skip any non-DFID format name */
2180 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2182 RETURN(PTR_ERR(child));
2184 /* skip the start '[' */
2185 sscanf(&name[1], SFID, RFID(&tfid));
2186 if (fid_is_sane(&tfid))
2187 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2188 fill_buf->oifb_dentry->d_inode,
2189 child->d_inode, &tfid, name, namelen, 0);
2197 #ifdef HAVE_FILLDIR_USE_CTX
2198 static int osd_ios_root_fill(struct dir_context *buf,
2200 static int osd_ios_root_fill(void *buf,
2202 const char *name, int namelen,
2203 loff_t offset, __u64 ino, unsigned d_type)
2205 struct osd_ios_filldir_buf *fill_buf =
2206 (struct osd_ios_filldir_buf *)buf;
2207 struct osd_device *dev = fill_buf->oifb_dev;
2208 const struct osd_lf_map *map;
2209 struct dentry *child;
2213 fill_buf->oifb_items++;
2215 /* skip any '.' started names */
2219 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2220 if (map->olm_namelen != namelen)
2223 if (strncmp(map->olm_name, name, namelen) == 0)
2227 if (map->olm_name == NULL)
2230 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2232 RETURN(PTR_ERR(child));
2233 else if (!child->d_inode)
2234 GOTO(out_put, rc = -ENOENT);
2236 if (!(map->olm_flags & OLF_NO_OI))
2237 rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2238 fill_buf->oifb_dentry->d_inode, child->d_inode,
2239 &map->olm_fid, name, namelen, map->olm_flags);
2240 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2241 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2250 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2251 struct dentry *dentry, filldir_t filldir)
2253 struct osd_ios_filldir_buf buf = {
2254 #ifdef HAVE_DIR_CONTEXT
2255 .ctx.actor = filldir,
2259 .oifb_dentry = dentry };
2260 struct file *filp = &info->oti_file;
2261 struct inode *inode = dentry->d_inode;
2262 const struct file_operations *fops = inode->i_fop;
2266 LASSERT(filldir != NULL);
2269 filp->f_path.dentry = dentry;
2270 filp->f_mode = FMODE_64BITHASH;
2271 filp->f_mapping = inode->i_mapping;
2273 filp->private_data = NULL;
2274 set_file_inode(filp, inode);
2278 #ifdef HAVE_DIR_CONTEXT
2279 buf.ctx.pos = filp->f_pos;
2280 #ifdef HAVE_ITERATE_SHARED
2281 rc = fops->iterate_shared(filp, &buf.ctx);
2283 rc = fops->iterate(filp, &buf.ctx);
2285 filp->f_pos = buf.ctx.pos;
2287 rc = fops->readdir(filp, &buf, filldir);
2289 } while (rc >= 0 && buf.oifb_items > 0 &&
2290 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
2291 fops->release(inode, filp);
2297 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2298 struct dentry *dentry, filldir_t filldir)
2300 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2301 struct scrub_file *sf = &scrub->os_file;
2302 struct dentry *child;
2306 /* It is existing MDT0 device. We only allow the case of object without
2307 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2308 * can generate IGIF mode FID for the object and related OI mapping. If
2309 * it is on other MDTs, then becuase file-level backup/restore, related
2310 * OI mapping may be invalid already, we do not know which is the right
2311 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2313 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2314 * "fid" xattr, then something crashed. We cannot re-generate the
2315 * FID directly, instead, the OI scrub will scan the OI structure
2316 * and try to re-generate the LMA from the OI mapping. But if the
2317 * OI mapping crashed or lost also, then we have to give up under
2318 * double failure cases. */
2319 scrub->os_convert_igif = 1;
2320 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2321 strlen(dot_lustre_name));
2322 if (IS_ERR(child) && PTR_ERR(child) != -ENOENT) {
2323 rc = PTR_ERR(child);
2324 } else if (IS_ERR(child) || !child->d_inode) {
2325 /* It is 1.8 MDT device. */
2326 if (!(sf->sf_flags & SF_UPGRADE)) {
2327 scrub_file_reset(scrub, dev->od_uuid,
2329 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2330 rc = scrub_file_store(info->oti_env, scrub);
2337 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2338 * so the client will get IGIF for the ".lustre" object when
2341 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2342 * it does not know whether there are some old clients cached
2343 * the ".lustre" IGIF during the upgrading. Two choices:
2345 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2346 * It will allow the old connected clients to access the
2347 * ".lustre" with cached IGIF. But it will cause others
2348 * on the MDT failed to check "fid_is_dot_lustre()".
2350 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2351 * for ".lustre" in spite of whether there are some clients
2352 * cached the ".lustre" IGIF or not. It enables the check
2353 * "fid_is_dot_lustre()" on the MDT, although it will cause
2354 * that the old connected clients cannot access the ".lustre"
2355 * with the cached IGIF.
2357 * Usually, it is rare case for the old connected clients
2358 * to access the ".lustre" with cached IGIF. So we prefer
2359 * to the solution 2). */
2360 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2361 child->d_inode, &LU_DOT_LUSTRE_FID,
2363 strlen(dot_lustre_name), 0);
2365 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2374 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2375 struct dentry *dentry, filldir_t filldir)
2377 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2378 struct scrub_file *sf = &scrub->os_file;
2379 struct dentry *child;
2383 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2384 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2385 rc = scrub_file_store(info->oti_env, scrub);
2390 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2391 if (IS_ERR(child)) {
2392 rc = PTR_ERR(child);
2394 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2395 child->d_inode, NULL, ADMIN_USR,
2396 strlen(ADMIN_USR), 0);
2400 if (rc != 0 && rc != -ENOENT)
2403 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2405 GOTO(out, rc = PTR_ERR(child));
2407 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2408 child->d_inode, NULL, ADMIN_GRP,
2409 strlen(ADMIN_GRP), 0);
2412 RETURN(rc == -ENOENT ? 0 : rc);
2415 static void osd_initial_OI_scrub(struct osd_thread_info *info,
2416 struct osd_device *dev)
2418 struct osd_ios_item *item = NULL;
2419 scandir_t scandir = osd_ios_general_scan;
2420 filldir_t filldir = osd_ios_root_fill;
2421 struct dentry *dentry = osd_sb(dev)->s_root;
2422 const struct osd_lf_map *map = osd_lf_maps;
2425 /* Lookup IGIF in OI by force for initial OI scrub. */
2426 dev->od_igif_inoi = 1;
2429 scandir(info, dev, dentry, filldir);
2431 dput(item->oii_dentry);
2435 if (list_empty(&dev->od_ios_list))
2438 item = list_entry(dev->od_ios_list.next,
2439 struct osd_ios_item, oii_list);
2440 list_del_init(&item->oii_list);
2442 LASSERT(item->oii_scandir != NULL);
2443 scandir = item->oii_scandir;
2444 filldir = item->oii_filldir;
2445 dentry = item->oii_dentry;
2448 /* There maybe the case that the object has been removed, but its OI
2449 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2450 * file-level backup/restore. So here cleanup the stale OI mappings. */
2451 while (map->olm_name != NULL) {
2452 struct dentry *child;
2454 if (fid_is_zero(&map->olm_fid)) {
2459 child = osd_ios_lookup_one_len(map->olm_name,
2460 osd_sb(dev)->s_root,
2462 if (PTR_ERR(child) == -ENOENT ||
2463 (!IS_ERR(child) && !child->d_inode))
2464 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2465 NULL, DTO_INDEX_DELETE,
2472 if (!list_empty(&dev->od_index_restore_list)) {
2475 OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
2477 CERROR("%s: not enough RAM for rebuild index\n",
2480 while (!list_empty(&dev->od_index_restore_list)) {
2481 struct lustre_index_restore_unit *liru;
2483 liru = list_entry(dev->od_index_restore_list.next,
2484 struct lustre_index_restore_unit,
2486 list_del(&liru->liru_link);
2488 osd_index_restore(info->oti_env, dev, liru,
2489 buf, INDEX_BACKUP_BUFSIZE);
2490 OBD_FREE(liru, liru->liru_len);
2494 OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
2500 char *osd_lf_fid2name(const struct lu_fid *fid)
2502 const struct osd_lf_map *map = osd_lf_maps;
2504 while (map->olm_name != NULL) {
2505 if (!lu_fid_eq(fid, &map->olm_fid)) {
2510 if (map->olm_flags & OLF_SHOW_NAME)
2511 return map->olm_name;
2519 /* OI scrub start/stop */
2521 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
2524 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2528 if (dev->od_dt_dev.dd_rdonly)
2531 /* od_otable_mutex: prevent curcurrent start/stop */
2532 mutex_lock(&dev->od_otable_mutex);
2533 rc = scrub_start(osd_scrub_main, scrub, dev, flags);
2534 if (rc == -EALREADY) {
2536 if ((scrub->os_file.sf_flags & SF_AUTO ||
2537 scrub->os_partial_scan) &&
2538 !(flags & SS_AUTO_PARTIAL))
2539 osd_scrub_join(env, dev, flags, false);
2541 mutex_unlock(&dev->od_otable_mutex);
2546 void osd_scrub_stop(struct osd_device *dev)
2548 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2550 /* od_otable_mutex: prevent curcurrent start/stop */
2551 mutex_lock(&dev->od_otable_mutex);
2552 scrub->os_paused = 1;
2554 mutex_unlock(&dev->od_otable_mutex);
2557 /* OI scrub setup/cleanup */
2559 static const char osd_scrub_name[] = "OI_scrub";
2561 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2563 struct osd_thread_info *info = osd_oti_get(env);
2564 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2565 struct lvfs_run_ctxt *ctxt = &dev->od_scrub.os_ctxt;
2566 struct scrub_file *sf = &scrub->os_file;
2567 struct super_block *sb = osd_sb(dev);
2568 struct lvfs_run_ctxt saved;
2570 struct inode *inode;
2571 struct lu_fid *fid = &info->oti_fid;
2572 struct osd_inode_id *id = &info->oti_id;
2573 struct dt_object *obj;
2575 bool restored = false;
2579 memset(&dev->od_scrub, 0, sizeof(struct osd_scrub));
2580 OBD_SET_CTXT_MAGIC(ctxt);
2581 ctxt->pwdmnt = dev->od_mnt;
2582 ctxt->pwd = dev->od_mnt->mnt_root;
2583 ctxt->fs = KERNEL_DS;
2585 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2586 init_rwsem(&scrub->os_rwsem);
2587 spin_lock_init(&scrub->os_lock);
2588 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2589 scrub->os_name = osd_name(dev);
2591 push_ctxt(&saved, ctxt);
2592 filp = filp_open(osd_scrub_name, O_RDWR |
2593 (dev->od_dt_dev.dd_rdonly ? 0 : O_CREAT), 0644);
2595 pop_ctxt(&saved, ctxt);
2596 RETURN(PTR_ERR(filp));
2599 inode = file_inode(filp);
2600 if (!dev->od_dt_dev.dd_rdonly) {
2601 /* 'What the @fid is' is not imporatant, because the object
2602 * has no OI mapping, and only is visible inside the OSD.*/
2603 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2604 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2606 filp_close(filp, NULL);
2607 pop_ctxt(&saved, ctxt);
2612 osd_id_gen(id, inode->i_ino, inode->i_generation);
2613 osd_add_oi_cache(info, dev, id, fid);
2614 filp_close(filp, NULL);
2615 pop_ctxt(&saved, ctxt);
2617 obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, NULL));
2618 if (IS_ERR_OR_NULL(obj))
2619 RETURN(obj ? PTR_ERR(obj) : -ENOENT);
2621 #ifndef HAVE_S_UUID_AS_UUID_T
2622 memcpy(dev->od_uuid.b, sb->s_uuid, UUID_SIZE);
2624 uuid_copy(&dev->od_uuid, &sb->s_uuid);
2626 scrub->os_obj = obj;
2627 rc = scrub_file_load(env, scrub);
2628 if (rc == -ENOENT || rc == -EFAULT) {
2629 scrub_file_init(scrub, dev->od_uuid);
2630 /* If the "/O" dir does not exist when mount (indicated by
2631 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2632 * then it is quite probably that the device is a new one,
2633 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2635 * For the rare case that "/O" and "OI_scrub" both lost on
2636 * an old device, it can be found and cleared later.
2638 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2639 * need to check "filter_fid_18_23" and to convert it to
2640 * "filter_fid" for each object, and all the IGIF should
2641 * have their FID mapping in OI files already. */
2642 if (dev->od_maybe_new && rc == -ENOENT)
2643 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2645 } else if (rc < 0) {
2646 GOTO(cleanup_obj, rc);
2648 if (!uuid_equal(&sf->sf_uuid, &dev->od_uuid)) {
2650 "%s: UUID has been changed from %pU to %pU\n",
2651 osd_dev2name(dev), &sf->sf_uuid, &dev->od_uuid);
2652 scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
2655 } else if (sf->sf_status == SS_SCANNING) {
2656 sf->sf_status = SS_CRASHED;
2660 if ((sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
2661 LCONSOLE_WARN("%s: invalid oi count %d, set it to %d\n",
2662 osd_dev2name(dev), sf->sf_oi_count,
2664 sf->sf_oi_count = osd_oi_count;
2669 if (sf->sf_pos_last_checkpoint != 0)
2670 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2672 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2675 rc = scrub_file_store(env, scrub);
2677 GOTO(cleanup_obj, rc);
2680 /* Initialize OI files. */
2681 rc = osd_oi_init(info, dev, restored);
2683 GOTO(cleanup_obj, rc);
2685 if (!dev->od_dt_dev.dd_rdonly)
2686 osd_initial_OI_scrub(info, dev);
2688 if (sf->sf_flags & SF_UPGRADE ||
2689 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2690 sf->sf_success_count > 0)) {
2691 dev->od_igif_inoi = 0;
2692 dev->od_check_ff = dev->od_is_ost;
2694 dev->od_igif_inoi = 1;
2695 dev->od_check_ff = 0;
2698 if (sf->sf_flags & SF_INCONSISTENT)
2699 /* The 'od_igif_inoi' will be set under the
2701 * 1) new created system, or
2702 * 2) restored from file-level backup, or
2703 * 3) the upgrading completed.
2705 * The 'od_igif_inoi' may be cleared by OI scrub
2706 * later if found that the system is upgrading. */
2707 dev->od_igif_inoi = 1;
2709 if (!dev->od_dt_dev.dd_rdonly &&
2710 dev->od_auto_scrub_interval != AS_NEVER &&
2711 ((sf->sf_status == SS_PAUSED) ||
2712 (sf->sf_status == SS_CRASHED &&
2713 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2714 SF_UPGRADE | SF_AUTO)) ||
2715 (sf->sf_status == SS_INIT &&
2716 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2718 rc = osd_scrub_start(env, dev, SS_AUTO_FULL);
2721 GOTO(cleanup_oi, rc);
2723 /* it is possible that dcache entries may keep objects after they are
2724 * deleted by OSD. While it looks safe this can cause object data to
2725 * stay until umount causing failures in tests calculating free space,
2726 * e.g. replay-ost-single. Since those dcache entries are not used
2727 * anymore let's just free them after use here */
2728 shrink_dcache_sb(sb);
2732 osd_oi_fini(info, dev);
2734 dt_object_put_nocache(env, scrub->os_obj);
2735 scrub->os_obj = NULL;
2740 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2742 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2744 LASSERT(dev->od_otable_it == NULL);
2746 if (scrub->os_obj != NULL) {
2747 osd_scrub_stop(dev);
2748 dt_object_put_nocache(env, scrub->os_obj);
2749 scrub->os_obj = NULL;
2751 if (dev->od_oi_table != NULL)
2752 osd_oi_fini(osd_oti_get(env), dev);
2755 /* object table based iteration APIs */
2757 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2758 struct dt_object *dt, __u32 attr)
2760 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2761 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2762 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2763 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2764 struct osd_otable_it *it;
2769 /* od_otable_mutex: prevent curcurrent init/fini */
2770 mutex_lock(&dev->od_otable_mutex);
2771 if (dev->od_otable_it != NULL)
2772 GOTO(out, it = ERR_PTR(-EALREADY));
2776 GOTO(out, it = ERR_PTR(-ENOMEM));
2778 dev->od_otable_it = it;
2780 it->ooi_cache.ooc_consumer_idx = -1;
2781 if (flags & DOIF_OUTUSED)
2782 it->ooi_used_outside = 1;
2784 if (flags & DOIF_RESET)
2787 if (valid & DOIV_ERROR_HANDLE) {
2788 if (flags & DOIF_FAILOUT)
2789 start |= SS_SET_FAILOUT;
2791 start |= SS_CLEAR_FAILOUT;
2794 if (valid & DOIV_DRYRUN) {
2795 if (flags & DOIF_DRYRUN)
2796 start |= SS_SET_DRYRUN;
2798 start |= SS_CLEAR_DRYRUN;
2801 rc = scrub_start(osd_scrub_main, scrub, dev, start & ~SS_AUTO_PARTIAL);
2802 if (rc == -EALREADY) {
2803 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2804 } else if (rc < 0) {
2805 dev->od_otable_it = NULL;
2809 /* We have to start from the begining. */
2810 it->ooi_cache.ooc_pos_preload =
2811 LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2817 mutex_unlock(&dev->od_otable_mutex);
2818 return (struct dt_it *)it;
2821 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2823 struct osd_otable_it *it = (struct osd_otable_it *)di;
2824 struct osd_device *dev = it->ooi_dev;
2826 /* od_otable_mutex: prevent curcurrent init/fini */
2827 mutex_lock(&dev->od_otable_mutex);
2828 scrub_stop(&dev->od_scrub.os_scrub);
2829 LASSERT(dev->od_otable_it == it);
2831 dev->od_otable_it = NULL;
2832 mutex_unlock(&dev->od_otable_mutex);
2836 static int osd_otable_it_get(const struct lu_env *env,
2837 struct dt_it *di, const struct dt_key *key)
2842 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2847 osd_otable_it_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
2849 spin_lock(&scrub->os_lock);
2850 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2851 scrub->os_waiting ||
2852 !thread_is_running(&scrub->os_thread))
2853 it->ooi_waiting = 0;
2855 it->ooi_waiting = 1;
2856 spin_unlock(&scrub->os_lock);
2858 return !it->ooi_waiting;
2861 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2863 struct osd_otable_it *it = (struct osd_otable_it *)di;
2864 struct osd_device *dev = it->ooi_dev;
2865 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2866 struct osd_otable_cache *ooc = &it->ooi_cache;
2867 struct ptlrpc_thread *thread = &scrub->os_thread;
2868 struct l_wait_info lwi = { 0 };
2872 LASSERT(it->ooi_user_ready);
2875 if (!thread_is_running(thread) && !it->ooi_used_outside)
2878 if (ooc->ooc_cached_items > 0) {
2879 ooc->ooc_cached_items--;
2880 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2881 ~OSD_OTABLE_IT_CACHE_MASK;
2885 if (it->ooi_all_cached) {
2886 l_wait_event(thread->t_ctl_waitq,
2887 !thread_is_running(thread),
2892 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2893 spin_lock(&scrub->os_lock);
2894 scrub->os_waiting = 0;
2895 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2896 spin_unlock(&scrub->os_lock);
2899 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2900 l_wait_event(thread->t_ctl_waitq,
2901 osd_otable_it_wakeup(scrub, it),
2904 if (!thread_is_running(thread) && !it->ooi_used_outside)
2907 rc = osd_otable_it_preload(env, it);
2914 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2915 const struct dt_it *di)
2920 static int osd_otable_it_key_size(const struct lu_env *env,
2921 const struct dt_it *di)
2923 return sizeof(__u64);
2926 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2927 struct dt_rec *rec, __u32 attr)
2929 struct osd_otable_it *it = (struct osd_otable_it *)di;
2930 struct osd_otable_cache *ooc = &it->ooi_cache;
2932 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2934 /* Filter out Invald FID already. */
2935 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2936 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2937 PFID((struct lu_fid *)rec),
2938 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2943 static __u64 osd_otable_it_store(const struct lu_env *env,
2944 const struct dt_it *di)
2946 struct osd_otable_it *it = (struct osd_otable_it *)di;
2947 struct osd_otable_cache *ooc = &it->ooi_cache;
2950 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2951 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2953 hash = ooc->ooc_pos_preload;
2958 * Set the OSD layer iteration start position as the specified hash.
2960 static int osd_otable_it_load(const struct lu_env *env,
2961 const struct dt_it *di, __u64 hash)
2963 struct osd_otable_it *it = (struct osd_otable_it *)di;
2964 struct osd_device *dev = it->ooi_dev;
2965 struct osd_otable_cache *ooc = &it->ooi_cache;
2966 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2967 struct osd_iit_param *param = &it->ooi_iit_param;
2971 /* Forbid to set iteration position after iteration started. */
2972 if (it->ooi_user_ready)
2975 LASSERT(!scrub->os_partial_scan);
2977 if (hash > OSD_OTABLE_MAX_HASH)
2978 hash = OSD_OTABLE_MAX_HASH;
2980 /* The hash is the last checkpoint position,
2981 * we will start from the next one. */
2982 ooc->ooc_pos_preload = hash + 1;
2983 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2984 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2986 it->ooi_user_ready = 1;
2987 if (!scrub->os_full_speed)
2988 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2990 memset(param, 0, sizeof(*param));
2991 param->sb = osd_sb(dev);
2992 param->start = ooc->ooc_pos_preload;
2993 param->bg = (ooc->ooc_pos_preload - 1) /
2994 LDISKFS_INODES_PER_GROUP(param->sb);
2995 param->offset = (ooc->ooc_pos_preload - 1) %
2996 LDISKFS_INODES_PER_GROUP(param->sb);
2997 param->gbase = 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
2999 /* Unplug OSD layer iteration by the first next() call. */
3000 rc = osd_otable_it_next(env, (struct dt_it *)it);
3005 static int osd_otable_it_key_rec(const struct lu_env *env,
3006 const struct dt_it *di, void *key_rec)
3011 const struct dt_index_operations osd_otable_ops = {
3013 .init = osd_otable_it_init,
3014 .fini = osd_otable_it_fini,
3015 .get = osd_otable_it_get,
3016 .put = osd_otable_it_put,
3017 .next = osd_otable_it_next,
3018 .key = osd_otable_it_key,
3019 .key_size = osd_otable_it_key_size,
3020 .rec = osd_otable_it_rec,
3021 .store = osd_otable_it_store,
3022 .load = osd_otable_it_load,
3023 .key_rec = osd_otable_it_key_rec,
3027 /* high priority inconsistent items list APIs */
3029 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
3031 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3034 struct osd_inconsistent_item *oii;
3035 struct osd_scrub *oscrub = &dev->od_scrub;
3036 struct lustre_scrub *lscrub = &oscrub->os_scrub;
3037 struct ptlrpc_thread *thread = &lscrub->os_thread;
3042 if (unlikely(oii == NULL))
3045 INIT_LIST_HEAD(&oii->oii_list);
3046 oii->oii_cache = *oic;
3047 oii->oii_insert = insert;
3049 if (lscrub->os_partial_scan) {
3050 __u64 now = ktime_get_real_seconds();
3052 /* If there haven't been errors in a long time,
3053 * decay old count until either the errors are
3054 * gone or we reach the current interval. */
3055 while (unlikely(oscrub->os_bad_oimap_count > 0 &&
3056 oscrub->os_bad_oimap_time +
3057 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3058 oscrub->os_bad_oimap_count >>= 1;
3059 oscrub->os_bad_oimap_time +=
3060 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3063 oscrub->os_bad_oimap_time = now;
3064 if (++oscrub->os_bad_oimap_count >
3065 dev->od_full_scrub_threshold_rate)
3066 lscrub->os_full_scrub = 1;
3069 spin_lock(&lscrub->os_lock);
3070 if (unlikely(!thread_is_running(thread))) {
3071 spin_unlock(&lscrub->os_lock);
3076 if (list_empty(&lscrub->os_inconsistent_items))
3078 list_add_tail(&oii->oii_list, &lscrub->os_inconsistent_items);
3079 spin_unlock(&lscrub->os_lock);
3082 wake_up_all(&thread->t_ctl_waitq);
3087 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3088 struct osd_inode_id *id)
3090 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
3091 struct osd_inconsistent_item *oii;
3094 spin_lock(&scrub->os_lock);
3095 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3096 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3097 *id = oii->oii_cache.oic_lid;
3098 spin_unlock(&scrub->os_lock);
3102 spin_unlock(&scrub->os_lock);
3107 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3109 struct osd_scrub *scrub = &dev->od_scrub;
3111 scrub_dump(m, &scrub->os_scrub);
3112 seq_printf(m, "lf_scanned: %llu\n"
3114 "lf_failed: %llu\n",
3115 scrub->os_lf_scanned,
3116 scrub->os_scrub.os_file.sf_param & SP_DRYRUN ?
3117 "inconsistent" : "repaired",
3118 scrub->os_lf_repaired,
3119 scrub->os_lf_failed);