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 lustre_scrub *scrub = &dev->od_scrub.os_scrub;
886 struct scrub_file *sf = &scrub->os_file;
887 struct ptlrpc_thread *thread = &scrub->os_thread;
888 struct osd_otable_it *it = dev->od_otable_it;
889 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
892 case SCRUB_NEXT_NOSCRUB:
893 down_write(&scrub->os_rwsem);
894 scrub->os_new_checked++;
895 sf->sf_items_noscrub++;
896 up_write(&scrub->os_rwsem);
897 case SCRUB_NEXT_CONTINUE:
898 case SCRUB_NEXT_WAIT:
902 rc = osd_scrub_check_update(info, dev, oic, rc);
904 scrub->os_in_prior = 0;
908 rc = scrub_checkpoint(info->oti_env, scrub);
910 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %llu: "
911 "rc = %d\n", osd_scrub2name(scrub),
912 scrub->os_pos_current, rc);
913 /* Continue, as long as the scrub itself can go ahead. */
916 if (scrub->os_in_prior) {
917 scrub->os_in_prior = 0;
922 if (it != NULL && it->ooi_waiting && ooc != NULL &&
923 ooc->ooc_pos_preload < scrub->os_pos_current) {
924 spin_lock(&scrub->os_lock);
926 wake_up_all(&thread->t_ctl_waitq);
927 spin_unlock(&scrub->os_lock);
930 if (rc == SCRUB_NEXT_CONTINUE)
933 if (scrub->os_full_speed || !ooc || osd_scrub_has_window(scrub, ooc)) {
939 wait_event_idle(thread->t_ctl_waitq,
940 osd_scrub_wakeup(scrub, it));
942 if (!ooc || osd_scrub_has_window(scrub, ooc))
949 static int osd_preload_exec(struct osd_thread_info *info,
950 struct osd_device *dev, struct osd_iit_param *param,
951 struct osd_idmap_cache *oic, bool *noslot, int rc)
953 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
956 ooc->ooc_cached_items++;
957 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
958 ~OSD_OTABLE_IT_CACHE_MASK;
960 return rc > 0 ? 0 : rc;
963 #define SCRUB_IT_ALL 1
964 #define SCRUB_IT_CRASH 2
966 static void osd_scrub_join(const struct lu_env *env, struct osd_device *dev,
967 __u32 flags, bool inconsistent)
969 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
970 struct ptlrpc_thread *thread = &scrub->os_thread;
971 struct scrub_file *sf = &scrub->os_file;
975 LASSERT(!(flags & SS_AUTO_PARTIAL));
977 down_write(&scrub->os_rwsem);
978 scrub->os_in_join = 1;
979 if (flags & SS_SET_FAILOUT)
980 sf->sf_param |= SP_FAILOUT;
981 else if (flags & SS_CLEAR_FAILOUT)
982 sf->sf_param &= ~SP_FAILOUT;
984 if (flags & SS_SET_DRYRUN)
985 sf->sf_param |= SP_DRYRUN;
986 else if (flags & SS_CLEAR_DRYRUN)
987 sf->sf_param &= ~SP_DRYRUN;
989 if (flags & SS_RESET) {
990 scrub_file_reset(scrub, dev->od_uuid,
991 inconsistent ? SF_INCONSISTENT : 0);
992 sf->sf_status = SS_SCANNING;
995 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
996 scrub->os_full_speed = 1;
998 scrub->os_full_speed = 0;
1000 if (flags & SS_AUTO_FULL) {
1001 sf->sf_flags |= SF_AUTO;
1002 scrub->os_full_speed = 1;
1005 scrub->os_new_checked = 0;
1006 if (sf->sf_pos_last_checkpoint != 0)
1007 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1009 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1011 scrub->os_pos_current = sf->sf_pos_latest_start;
1012 sf->sf_time_latest_start = ktime_get_real_seconds();
1013 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1014 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1015 rc = scrub_file_store(env, scrub);
1017 spin_lock(&scrub->os_lock);
1018 scrub->os_waiting = 0;
1019 scrub->os_paused = 0;
1020 scrub->os_partial_scan = 0;
1021 scrub->os_in_join = 0;
1022 scrub->os_full_scrub = 0;
1023 spin_unlock(&scrub->os_lock);
1024 wake_up_all(&thread->t_ctl_waitq);
1025 up_write(&scrub->os_rwsem);
1027 CDEBUG(D_LFSCK, "%s: joined in the OI scrub with flag %u: rc = %d\n",
1028 osd_scrub2name(scrub), flags, rc);
1033 static int osd_inode_iteration(struct osd_thread_info *info,
1034 struct osd_device *dev, __u32 max, bool preload)
1036 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1037 struct ptlrpc_thread *thread = &scrub->os_thread;
1038 struct scrub_file *sf = &scrub->os_file;
1039 osd_iit_next_policy next;
1040 osd_iit_exec_policy exec;
1043 struct osd_iit_param *param;
1052 param = &dev->od_scrub.os_iit_param;
1053 memset(param, 0, sizeof(*param));
1054 param->sb = osd_sb(dev);
1056 while (scrub->os_partial_scan && !scrub->os_in_join) {
1057 struct osd_idmap_cache *oic = NULL;
1059 rc = osd_scrub_next(info, dev, param, &oic, noslot);
1061 case SCRUB_NEXT_EXIT:
1063 case SCRUB_NEXT_CRASH:
1064 RETURN(SCRUB_IT_CRASH);
1065 case SCRUB_NEXT_FATAL:
1067 case SCRUB_NEXT_WAIT: {
1068 struct kstatfs *ksfs = &info->oti_ksfs;
1071 if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1072 unlikely(sf->sf_items_updated_prior == 0))
1075 if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1076 scrub->os_full_scrub) {
1077 osd_scrub_join(info->oti_env, dev,
1078 SS_AUTO_FULL | SS_RESET, true);
1082 rc = param->sb->s_op->statfs(param->sb->s_root, ksfs);
1084 __u64 used = ksfs->f_files - ksfs->f_ffree;
1086 do_div(used, sf->sf_items_updated_prior);
1087 /* If we hit too much inconsistent OI
1088 * mappings during the partial scan,
1089 * then scan the device completely. */
1090 if (used < dev->od_full_scrub_ratio) {
1091 osd_scrub_join(info->oti_env, dev,
1092 SS_AUTO_FULL | SS_RESET, true);
1098 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1102 saved_flags = sf->sf_flags;
1103 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1104 SF_UPGRADE | SF_AUTO);
1105 sf->sf_status = SS_COMPLETED;
1107 thread->t_ctl_waitq,
1108 !thread_is_running(thread) ||
1109 !scrub->os_partial_scan ||
1110 scrub->os_in_join ||
1111 !list_empty(&scrub->os_inconsistent_items));
1112 sf->sf_flags = saved_flags;
1113 sf->sf_status = SS_SCANNING;
1115 if (unlikely(!thread_is_running(thread)))
1118 if (!scrub->os_partial_scan || scrub->os_in_join)
1124 LASSERTF(rc == 0, "rc = %d\n", rc);
1126 osd_scrub_exec(info, dev, param, oic, &noslot, rc);
1133 wait_event_idle(thread->t_ctl_waitq,
1134 !thread_is_running(thread) ||
1135 !scrub->os_in_join);
1137 if (unlikely(!thread_is_running(thread)))
1143 next = osd_scrub_next;
1144 exec = osd_scrub_exec;
1145 pos = &scrub->os_pos_current;
1146 count = &scrub->os_new_checked;
1147 param->start = *pos;
1148 param->bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param->sb);
1150 (*pos - 1) % LDISKFS_INODES_PER_GROUP(param->sb);
1152 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1154 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1156 next = osd_preload_next;
1157 exec = osd_preload_exec;
1158 pos = &ooc->ooc_pos_preload;
1159 count = &ooc->ooc_cached_items;
1160 param = &dev->od_otable_it->ooi_iit_param;
1164 limit = le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count);
1165 while (*pos <= limit && *count < max) {
1166 struct ldiskfs_group_desc *desc;
1167 bool next_group = false;
1169 desc = ldiskfs_get_group_desc(param->sb, param->bg, NULL);
1173 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1178 param->bitmap = ldiskfs_read_inode_bitmap(param->sb, param->bg);
1179 if (!param->bitmap) {
1180 CERROR("%s: fail to read bitmap for %u, "
1181 "scrub will stop, urgent mode\n",
1182 osd_scrub2name(scrub), (__u32)param->bg);
1187 struct osd_idmap_cache *oic = NULL;
1190 ldiskfs_itable_unused_count(param->sb, desc) >=
1191 LDISKFS_INODES_PER_GROUP(param->sb)) {
1196 rc = next(info, dev, param, &oic, noslot);
1198 case SCRUB_NEXT_BREAK:
1201 case SCRUB_NEXT_EXIT:
1202 brelse(param->bitmap);
1204 case SCRUB_NEXT_CRASH:
1205 brelse(param->bitmap);
1206 RETURN(SCRUB_IT_CRASH);
1207 case SCRUB_NEXT_FATAL:
1208 brelse(param->bitmap);
1212 rc = exec(info, dev, param, oic, &noslot, rc);
1213 } while (!rc && *pos <= limit && *count < max);
1216 if (param->bitmap) {
1217 brelse(param->bitmap);
1218 param->bitmap = NULL;
1228 param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1229 *pos = param->gbase;
1230 param->start = *pos;
1235 RETURN(SCRUB_IT_ALL);
1241 static int osd_otable_it_preload(const struct lu_env *env,
1242 struct osd_otable_it *it)
1244 struct osd_device *dev = it->ooi_dev;
1245 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1246 struct osd_otable_cache *ooc = &it->ooi_cache;
1250 rc = osd_inode_iteration(osd_oti_get(env), dev,
1251 OSD_OTABLE_IT_CACHE_SIZE, true);
1252 if (rc == SCRUB_IT_ALL)
1253 it->ooi_all_cached = 1;
1255 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1256 spin_lock(&scrub->os_lock);
1257 scrub->os_waiting = 0;
1258 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1259 spin_unlock(&scrub->os_lock);
1262 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1265 static int osd_scrub_main(void *args)
1268 struct osd_device *dev = (struct osd_device *)args;
1269 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1270 struct ptlrpc_thread *thread = &scrub->os_thread;
1274 rc = lu_env_init(&env, LCT_LOCAL | LCT_DT_THREAD);
1276 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
1277 osd_scrub2name(scrub), rc);
1281 rc = osd_scrub_prep(&env, dev);
1283 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
1284 osd_scrub2name(scrub), rc);
1288 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1289 struct osd_otable_it *it = dev->od_otable_it;
1290 struct osd_otable_cache *ooc = &it->ooi_cache;
1292 wait_event_idle(thread->t_ctl_waitq,
1293 it->ooi_user_ready ||
1294 !thread_is_running(thread));
1295 if (unlikely(!thread_is_running(thread)))
1298 scrub->os_pos_current = ooc->ooc_pos_preload;
1301 CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %llu\n",
1302 osd_scrub2name(scrub), scrub->os_start_flags,
1303 scrub->os_pos_current);
1305 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1306 if (unlikely(rc == SCRUB_IT_CRASH)) {
1307 spin_lock(&scrub->os_lock);
1308 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
1309 spin_unlock(&scrub->os_lock);
1310 GOTO(out, rc = -EINVAL);
1316 rc = osd_scrub_post(&env, dev, rc);
1317 CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %llu: rc = %d\n",
1318 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1321 while (!list_empty(&scrub->os_inconsistent_items)) {
1322 struct osd_inconsistent_item *oii;
1324 oii = list_entry(scrub->os_inconsistent_items.next,
1325 struct osd_inconsistent_item, oii_list);
1326 list_del_init(&oii->oii_list);
1332 spin_lock(&scrub->os_lock);
1333 thread_set_flags(thread, SVC_STOPPED);
1334 wake_up_all(&thread->t_ctl_waitq);
1335 spin_unlock(&scrub->os_lock);
1339 /* initial OI scrub */
1341 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1342 struct dentry *, filldir_t filldir);
1344 #ifdef HAVE_FILLDIR_USE_CTX
1345 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1346 int namelen, loff_t offset, __u64 ino,
1348 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1349 int namelen, loff_t offset, __u64 ino,
1351 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1352 int namelen, loff_t offset, __u64 ino,
1354 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1355 int namelen, loff_t offset, __u64 ino,
1358 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1359 loff_t offset, __u64 ino, unsigned d_type);
1360 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1361 loff_t offset, __u64 ino, unsigned d_type);
1362 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1363 loff_t offset, __u64 ino, unsigned d_type);
1364 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1365 loff_t offset, __u64 ino, unsigned d_type);
1369 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1370 struct dentry *dentry, filldir_t filldir);
1372 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1373 struct dentry *dentry, filldir_t filldir);
1376 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1377 struct dentry *dentry, filldir_t filldir);
1381 struct lu_fid olm_fid;
1384 scandir_t olm_scandir;
1385 filldir_t olm_filldir;
1388 /* Add the new introduced local files in the list in the future. */
1389 static const struct osd_lf_map osd_lf_maps[] = {
1392 .olm_name = CATLIST,
1394 .f_seq = FID_SEQ_LOCAL_FILE,
1395 .f_oid = LLOG_CATALOGS_OID,
1397 .olm_flags = OLF_SHOW_NAME,
1398 .olm_namelen = sizeof(CATLIST) - 1,
1403 .olm_name = MOUNT_CONFIGS_DIR,
1405 .f_seq = FID_SEQ_LOCAL_FILE,
1406 .f_oid = MGS_CONFIGS_OID,
1408 .olm_flags = OLF_SCAN_SUBITEMS,
1409 .olm_namelen = sizeof(MOUNT_CONFIGS_DIR) - 1,
1410 .olm_scandir = osd_ios_general_scan,
1411 .olm_filldir = osd_ios_varfid_fill,
1414 /* NIDTBL_VERSIONS */
1416 .olm_name = MGS_NIDTBL_DIR,
1417 .olm_flags = OLF_SCAN_SUBITEMS,
1418 .olm_namelen = sizeof(MGS_NIDTBL_DIR) - 1,
1419 .olm_scandir = osd_ios_general_scan,
1420 .olm_filldir = osd_ios_varfid_fill,
1425 .olm_name = MDT_ORPHAN_DIR,
1426 .olm_namelen = sizeof(MDT_ORPHAN_DIR) - 1,
1433 .f_seq = FID_SEQ_ROOT,
1434 .f_oid = FID_OID_ROOT,
1436 .olm_flags = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1437 .olm_namelen = sizeof("ROOT") - 1,
1438 .olm_scandir = osd_ios_ROOT_scan,
1441 /* changelog_catalog */
1443 .olm_name = CHANGELOG_CATALOG,
1444 .olm_namelen = sizeof(CHANGELOG_CATALOG) - 1,
1447 /* changelog_users */
1449 .olm_name = CHANGELOG_USERS,
1450 .olm_namelen = sizeof(CHANGELOG_USERS) - 1,
1457 .f_seq = FID_SEQ_LOCAL_FILE,
1458 .f_oid = FLD_INDEX_OID,
1460 .olm_flags = OLF_SHOW_NAME,
1461 .olm_namelen = sizeof("fld") - 1,
1466 .olm_name = LAST_RCVD,
1468 .f_seq = FID_SEQ_LOCAL_FILE,
1469 .f_oid = LAST_RECV_OID,
1471 .olm_flags = OLF_SHOW_NAME,
1472 .olm_namelen = sizeof(LAST_RCVD) - 1,
1477 .olm_name = REPLY_DATA,
1479 .f_seq = FID_SEQ_LOCAL_FILE,
1480 .f_oid = REPLY_DATA_OID,
1482 .olm_flags = OLF_SHOW_NAME,
1483 .olm_namelen = sizeof(REPLY_DATA) - 1,
1488 .olm_name = LOV_OBJID,
1490 .f_seq = FID_SEQ_LOCAL_FILE,
1491 .f_oid = MDD_LOV_OBJ_OID,
1493 .olm_flags = OLF_SHOW_NAME,
1494 .olm_namelen = sizeof(LOV_OBJID) - 1,
1499 .olm_name = LOV_OBJSEQ,
1501 .f_seq = FID_SEQ_LOCAL_FILE,
1502 .f_oid = MDD_LOV_OBJ_OSEQ,
1504 .olm_flags = OLF_SHOW_NAME,
1505 .olm_namelen = sizeof(LOV_OBJSEQ) - 1,
1510 .olm_name = QMT_DIR,
1511 .olm_flags = OLF_SCAN_SUBITEMS,
1512 .olm_namelen = sizeof(QMT_DIR) - 1,
1513 .olm_scandir = osd_ios_general_scan,
1514 .olm_filldir = osd_ios_varfid_fill,
1519 .olm_name = QSD_DIR,
1520 .olm_flags = OLF_SCAN_SUBITEMS,
1521 .olm_namelen = sizeof(QSD_DIR) - 1,
1522 .olm_scandir = osd_ios_general_scan,
1523 .olm_filldir = osd_ios_varfid_fill,
1528 .olm_name = "seq_ctl",
1530 .f_seq = FID_SEQ_LOCAL_FILE,
1531 .f_oid = FID_SEQ_CTL_OID,
1533 .olm_flags = OLF_SHOW_NAME,
1534 .olm_namelen = sizeof("seq_ctl") - 1,
1539 .olm_name = "seq_srv",
1541 .f_seq = FID_SEQ_LOCAL_FILE,
1542 .f_oid = FID_SEQ_SRV_OID,
1544 .olm_flags = OLF_SHOW_NAME,
1545 .olm_namelen = sizeof("seq_srv") - 1,
1550 .olm_name = HEALTH_CHECK,
1552 .f_seq = FID_SEQ_LOCAL_FILE,
1553 .f_oid = OFD_HEALTH_CHECK_OID,
1555 .olm_flags = OLF_SHOW_NAME,
1556 .olm_namelen = sizeof(HEALTH_CHECK) - 1,
1561 .olm_name = LFSCK_DIR,
1562 .olm_flags = OLF_SCAN_SUBITEMS,
1563 .olm_namelen = sizeof(LFSCK_DIR) - 1,
1564 .olm_scandir = osd_ios_general_scan,
1565 .olm_filldir = osd_ios_varfid_fill,
1568 /* lfsck_bookmark */
1570 .olm_name = LFSCK_BOOKMARK,
1571 .olm_namelen = sizeof(LFSCK_BOOKMARK) - 1,
1576 .olm_name = LFSCK_LAYOUT,
1577 .olm_namelen = sizeof(LFSCK_LAYOUT) - 1,
1580 /* lfsck_namespace */
1582 .olm_name = LFSCK_NAMESPACE,
1583 .olm_namelen = sizeof(LFSCK_NAMESPACE) - 1,
1586 /* OBJECTS, upgrade from old device */
1588 .olm_name = OBJECTS,
1589 .olm_flags = OLF_SCAN_SUBITEMS,
1590 .olm_namelen = sizeof(OBJECTS) - 1,
1591 .olm_scandir = osd_ios_OBJECTS_scan,
1594 /* lquota_v2.user, upgrade from old device */
1596 .olm_name = "lquota_v2.user",
1597 .olm_namelen = sizeof("lquota_v2.user") - 1,
1600 /* lquota_v2.group, upgrade from old device */
1602 .olm_name = "lquota_v2.group",
1603 .olm_namelen = sizeof("lquota_v2.group") - 1,
1606 /* LAST_GROUP, upgrade from old device */
1608 .olm_name = "LAST_GROUP",
1610 .f_seq = FID_SEQ_LOCAL_FILE,
1611 .f_oid = OFD_LAST_GROUP_OID,
1613 .olm_flags = OLF_SHOW_NAME,
1614 .olm_namelen = sizeof("LAST_GROUP") - 1,
1617 /* committed batchid for cross-MDT operation */
1619 .olm_name = "BATCHID",
1621 .f_seq = FID_SEQ_LOCAL_FILE,
1622 .f_oid = BATCHID_COMMITTED_OID,
1624 .olm_flags = OLF_SHOW_NAME,
1625 .olm_namelen = sizeof("BATCHID") - 1,
1628 /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1629 * and f_oid = index for their log files. See lu_update_log{_dir}_fid()
1630 * for more details. */
1634 .olm_name = "update_log",
1636 .f_seq = FID_SEQ_UPDATE_LOG,
1638 .olm_flags = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1639 .olm_namelen = sizeof("update_log") - 1,
1642 /* update_log_dir */
1644 .olm_name = "update_log_dir",
1646 .f_seq = FID_SEQ_UPDATE_LOG_DIR,
1648 .olm_flags = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1650 .olm_namelen = sizeof("update_log_dir") - 1,
1651 .olm_scandir = osd_ios_general_scan,
1652 .olm_filldir = osd_ios_uld_fill,
1657 .olm_name = "lost+found",
1659 .f_seq = FID_SEQ_LOCAL_FILE,
1660 .f_oid = OSD_LPF_OID,
1662 .olm_flags = OLF_SCAN_SUBITEMS,
1663 .olm_namelen = sizeof("lost+found") - 1,
1664 .olm_scandir = osd_ios_general_scan,
1665 .olm_filldir = osd_ios_lf_fill,
1670 .olm_name = HSM_ACTIONS,
1675 .olm_name = LUSTRE_NODEMAP_NAME,
1680 .olm_name = INDEX_BACKUP_DIR,
1682 .f_seq = FID_SEQ_LOCAL_FILE,
1683 .f_oid = INDEX_BACKUP_OID,
1685 .olm_flags = OLF_SCAN_SUBITEMS | OLF_NOT_BACKUP,
1686 .olm_namelen = sizeof(INDEX_BACKUP_DIR) - 1,
1687 .olm_scandir = osd_ios_general_scan,
1688 .olm_filldir = osd_ios_varfid_fill,
1696 /* Add the new introduced files under .lustre/ in the list in the future. */
1697 static const struct osd_lf_map osd_dl_maps[] = {
1702 .f_seq = FID_SEQ_DOT_LUSTRE,
1703 .f_oid = FID_OID_DOT_LUSTRE_OBF,
1705 .olm_namelen = sizeof("fid") - 1,
1708 /* .lustre/lost+found */
1710 .olm_name = "lost+found",
1712 .f_seq = FID_SEQ_DOT_LUSTRE,
1713 .f_oid = FID_OID_DOT_LUSTRE_LPF,
1715 .olm_namelen = sizeof("lost+found") - 1,
1723 struct osd_ios_item {
1724 struct list_head oii_list;
1725 struct dentry *oii_dentry;
1726 scandir_t oii_scandir;
1727 filldir_t oii_filldir;
1730 struct osd_ios_filldir_buf {
1731 /* please keep it as first member */
1732 struct dir_context ctx;
1733 struct osd_thread_info *oifb_info;
1734 struct osd_device *oifb_dev;
1735 struct dentry *oifb_dentry;
1740 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1741 scandir_t scandir, filldir_t filldir)
1743 struct osd_ios_item *item;
1746 OBD_ALLOC_PTR(item);
1750 INIT_LIST_HEAD(&item->oii_list);
1751 item->oii_dentry = dget(dentry);
1752 item->oii_scandir = scandir;
1753 item->oii_filldir = filldir;
1754 list_add_tail(&item->oii_list, &dev->od_ios_list);
1759 static bool osd_index_need_recreate(const struct lu_env *env,
1760 struct osd_device *dev, struct inode *inode)
1762 struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
1763 struct iam_container *bag = &iam->od_container;
1767 rc = iam_container_init(bag, &iam->od_descr, inode);
1771 rc = iam_container_setup(bag);
1772 iam_container_fini(bag);
1779 static void osd_ios_index_register(const struct lu_env *env,
1780 struct osd_device *osd,
1781 const struct lu_fid *fid,
1782 struct inode *inode)
1784 struct osd_directory *iam = &osd_oti_get(env)->oti_iam;
1785 struct iam_container *bag = &iam->od_container;
1786 struct super_block *sb = osd_sb(osd);
1787 struct iam_descr *descr;
1793 /* Index must be a regular file. */
1794 if (!S_ISREG(inode->i_mode))
1797 /* Index's size must be block aligned. */
1798 if (inode->i_size < sb->s_blocksize ||
1799 (inode->i_size & (sb->s_blocksize - 1)) != 0)
1802 iam_container_init(bag, &iam->od_descr, inode);
1803 rc = iam_container_setup(bag);
1807 descr = bag->ic_descr;
1808 /* May be regular file with IAM_LFIX_ROOT_MAGIC matched
1809 * coincidentally, or corrupted index object, skip it. */
1810 if (descr->id_ptr_size != 4)
1813 keysize = descr->id_key_size;
1814 recsize = descr->id_rec_size;
1815 rc = osd_index_register(osd, fid, keysize, recsize);
1820 iam_container_fini(bag);
1822 CDEBUG(D_LFSCK, "%s: index object "DFID" (%u/%u) registered\n",
1823 osd_name(osd), PFID(fid), keysize, recsize);
1826 static void osd_index_restore(const struct lu_env *env, struct osd_device *dev,
1827 struct lustre_index_restore_unit *liru,
1828 void *buf, int bufsize)
1830 struct osd_thread_info *info = osd_oti_get(env);
1831 struct osd_inode_id *id = &info->oti_id;
1832 struct lu_fid *tgt_fid = &liru->liru_cfid;
1833 struct inode *bak_inode = NULL;
1834 struct ldiskfs_dir_entry_2 *de = NULL;
1835 struct buffer_head *bh = NULL;
1836 struct dentry *dentry;
1838 struct lu_fid bak_fid;
1842 lustre_fid2lbx(name, tgt_fid, bufsize);
1843 dentry = osd_child_dentry_by_inode(env, dev->od_index_backup_inode,
1844 name, strlen(name));
1845 bh = osd_ldiskfs_find_entry(dev->od_index_backup_inode,
1846 &dentry->d_name, &de, NULL, NULL);
1848 GOTO(log, rc = PTR_ERR(bh));
1850 osd_id_gen(id, le32_to_cpu(de->inode), OSD_OII_NOGEN);
1852 bak_inode = osd_iget_fid(info, dev, id, &bak_fid);
1853 if (IS_ERR(bak_inode))
1854 GOTO(log, rc = PTR_ERR(bak_inode));
1857 /* The OI mapping for index may be invalid, since it will be
1858 * re-created, not update the OI mapping, just cache it in RAM. */
1859 osd_id_gen(id, liru->liru_clid, OSD_OII_NOGEN);
1860 osd_add_oi_cache(info, dev, id, tgt_fid);
1861 rc = lustre_index_restore(env, &dev->od_dt_dev, &liru->liru_pfid,
1862 tgt_fid, &bak_fid, liru->liru_name,
1863 &dev->od_index_backup_list, &dev->od_lock,
1868 CDEBUG(D_WARNING, "%s: restore index '%s' with "DFID": rc = %d\n",
1869 osd_name(dev), liru->liru_name, PFID(tgt_fid), rc);
1873 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1875 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1876 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1877 * reference the inode, or fixed if it is missing or references another inode.
1880 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1881 struct inode *parent, struct inode *inode,
1882 const struct lu_fid *fid, const char *name,
1883 int namelen, int flags)
1885 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1886 struct osd_inode_id *id = &info->oti_id;
1887 struct osd_inode_id *id2 = &info->oti_id2;
1888 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1889 struct scrub_file *sf = &scrub->os_file;
1895 CDEBUG(D_INODE, "%s: child '%.*s' lacks inode: rc = -2\n",
1896 osd_name(dev), namelen, name);
1900 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
1901 &info->oti_ost_attrs);
1902 if (rc != 0 && rc != -ENODATA) {
1903 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1904 "rc = %d\n", osd_name(dev), rc);
1909 osd_id_gen(id, inode->i_ino, inode->i_generation);
1910 if (rc == -ENODATA) {
1911 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1912 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1915 if (flags & OLF_IDX_IN_FID) {
1916 LASSERT(dev->od_index >= 0);
1918 tfid.f_oid = dev->od_index;
1921 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1923 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1924 "scrub: rc = %d\n", osd_name(dev), rc);
1929 if (lma->lma_compat & LMAC_NOT_IN_OI)
1932 tfid = lma->lma_self_fid;
1933 if (lma->lma_compat & LMAC_IDX_BACKUP &&
1934 osd_index_need_recreate(info->oti_env, dev, inode)) {
1935 struct lu_fid *pfid = &info->oti_fid3;
1937 if (parent == osd_sb(dev)->s_root->d_inode) {
1938 lu_local_obj_fid(pfid, OSD_FS_ROOT_OID);
1940 rc = osd_scrub_get_fid(info, dev, parent, pfid,
1946 rc = lustre_liru_new(&dev->od_index_restore_list, pfid,
1947 &tfid, inode->i_ino, name, namelen);
1952 if (!(flags & OLF_NOT_BACKUP))
1953 osd_ios_index_register(info->oti_env, dev, &tfid,
1957 /* Since this called from iterate_dir() the inode lock will be taken */
1958 rc = osd_oi_lookup(info, dev, &tfid, id2, OI_LOCKED);
1963 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1964 DTO_INDEX_INSERT, true, 0, NULL);
1971 if (osd_id_eq_strict(id, id2))
1974 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1975 scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
1976 rc = scrub_file_store(info->oti_env, scrub);
1981 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1982 DTO_INDEX_UPDATE, true, 0, NULL);
1990 * It scans the /lost+found, and for the OST-object (with filter_fid
1991 * or filter_fid_18_23), move them back to its proper /O/<seq>/d<x>.
1993 #ifdef HAVE_FILLDIR_USE_CTX
1994 static int osd_ios_lf_fill(struct dir_context *buf,
1996 static int osd_ios_lf_fill(void *buf,
1998 const char *name, int namelen,
1999 loff_t offset, __u64 ino, unsigned d_type)
2001 struct osd_ios_filldir_buf *fill_buf =
2002 (struct osd_ios_filldir_buf *)buf;
2003 struct osd_thread_info *info = fill_buf->oifb_info;
2004 struct osd_device *dev = fill_buf->oifb_dev;
2005 struct lu_fid *fid = &info->oti_fid;
2006 struct osd_scrub *scrub = &dev->od_scrub;
2007 struct dentry *parent = fill_buf->oifb_dentry;
2008 struct dentry *child;
2009 struct inode *dir = parent->d_inode;
2010 struct inode *inode;
2014 fill_buf->oifb_items++;
2016 /* skip any '.' started names */
2020 scrub->os_lf_scanned++;
2021 child = osd_lookup_one_len(dev, name, parent, namelen);
2022 if (IS_ERR(child)) {
2023 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2024 osd_name(dev), namelen, name, (int)PTR_ERR(child));
2026 } else if (!child->d_inode) {
2028 CDEBUG(D_INODE, "%s: child '%.*s' lacks inode\n",
2029 osd_name(dev), namelen, name);
2033 inode = child->d_inode;
2034 if (S_ISDIR(inode->i_mode)) {
2035 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2038 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2039 "rc = %d\n", osd_name(dev), namelen, name, rc);
2043 if (!S_ISREG(inode->i_mode))
2046 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2047 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2048 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2050 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2051 "/lost+found.\n", namelen, name, PFID(fid));
2052 scrub->os_lf_repaired++;
2054 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2056 osd_name(dev), namelen, name, PFID(fid), rc);
2060 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2061 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2062 * can process them in furtuer. */
2068 scrub->os_lf_failed++;
2070 /* skip the failure to make the scanning to continue. */
2074 #ifdef HAVE_FILLDIR_USE_CTX
2075 static int osd_ios_varfid_fill(struct dir_context *buf,
2077 static int osd_ios_varfid_fill(void *buf,
2079 const char *name, int namelen,
2080 loff_t offset, __u64 ino, unsigned d_type)
2082 struct osd_ios_filldir_buf *fill_buf =
2083 (struct osd_ios_filldir_buf *)buf;
2084 struct osd_device *dev = fill_buf->oifb_dev;
2085 struct dentry *child;
2089 fill_buf->oifb_items++;
2091 /* skip any '.' started names */
2095 child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
2097 RETURN(PTR_ERR(child));
2099 rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2100 fill_buf->oifb_dentry->d_inode, child->d_inode,
2101 NULL, name, namelen, 0);
2102 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2103 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2104 osd_ios_varfid_fill);
2110 #ifdef HAVE_FILLDIR_USE_CTX
2111 static int osd_ios_dl_fill(struct dir_context *buf,
2113 static int osd_ios_dl_fill(void *buf,
2115 const char *name, int namelen,
2116 loff_t offset, __u64 ino, unsigned d_type)
2118 struct osd_ios_filldir_buf *fill_buf =
2119 (struct osd_ios_filldir_buf *)buf;
2120 struct osd_device *dev = fill_buf->oifb_dev;
2121 const struct osd_lf_map *map;
2122 struct dentry *child;
2126 fill_buf->oifb_items++;
2128 /* skip any '.' started names */
2132 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2133 if (map->olm_namelen != namelen)
2136 if (strncmp(map->olm_name, name, namelen) == 0)
2140 if (map->olm_name == NULL)
2143 child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
2145 RETURN(PTR_ERR(child));
2147 rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2148 fill_buf->oifb_dentry->d_inode, child->d_inode,
2149 &map->olm_fid, name, namelen, map->olm_flags);
2155 #ifdef HAVE_FILLDIR_USE_CTX
2156 static int osd_ios_uld_fill(struct dir_context *buf,
2158 static int osd_ios_uld_fill(void *buf,
2160 const char *name, int namelen,
2161 loff_t offset, __u64 ino, unsigned d_type)
2163 struct osd_ios_filldir_buf *fill_buf =
2164 (struct osd_ios_filldir_buf *)buf;
2165 struct osd_device *dev = fill_buf->oifb_dev;
2166 struct dentry *child;
2171 fill_buf->oifb_items++;
2173 /* skip any non-DFID format name */
2177 child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
2179 RETURN(PTR_ERR(child));
2181 /* skip the start '[' */
2182 sscanf(&name[1], SFID, RFID(&tfid));
2183 if (fid_is_sane(&tfid))
2184 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2185 fill_buf->oifb_dentry->d_inode,
2186 child->d_inode, &tfid, name, namelen, 0);
2194 #ifdef HAVE_FILLDIR_USE_CTX
2195 static int osd_ios_root_fill(struct dir_context *buf,
2197 static int osd_ios_root_fill(void *buf,
2199 const char *name, int namelen,
2200 loff_t offset, __u64 ino, unsigned d_type)
2202 struct osd_ios_filldir_buf *fill_buf =
2203 (struct osd_ios_filldir_buf *)buf;
2204 struct osd_device *dev = fill_buf->oifb_dev;
2205 const struct osd_lf_map *map;
2206 struct dentry *child;
2210 fill_buf->oifb_items++;
2212 /* skip any '.' started names */
2216 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2217 if (map->olm_namelen != namelen)
2220 if (strncmp(map->olm_name, name, namelen) == 0)
2224 if (map->olm_name == NULL)
2227 child = osd_lookup_one_len(dev, name, fill_buf->oifb_dentry, namelen);
2229 RETURN(PTR_ERR(child));
2230 else if (!child->d_inode)
2231 GOTO(out_put, rc = -ENOENT);
2233 if (!(map->olm_flags & OLF_NO_OI))
2234 rc = osd_ios_scan_one(fill_buf->oifb_info, dev,
2235 fill_buf->oifb_dentry->d_inode, child->d_inode,
2236 &map->olm_fid, name, namelen, map->olm_flags);
2237 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2238 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2247 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2248 struct dentry *dentry, filldir_t filldir)
2250 struct osd_ios_filldir_buf buf = {
2251 .ctx.actor = filldir,
2254 .oifb_dentry = dentry };
2255 struct file *filp = &info->oti_file;
2256 struct inode *inode = dentry->d_inode;
2257 const struct file_operations *fops = inode->i_fop;
2261 LASSERT(filldir != NULL);
2264 filp->f_path.dentry = dentry;
2265 filp->f_flags |= O_NOATIME;
2266 filp->f_mode = FMODE_64BITHASH | FMODE_NONOTIFY;
2267 filp->f_mapping = inode->i_mapping;
2269 filp->private_data = NULL;
2270 set_file_inode(filp, inode);
2271 rc = osd_security_file_alloc(filp);
2277 rc = iterate_dir(filp, &buf.ctx);
2278 } while (rc >= 0 && buf.oifb_items > 0 &&
2279 filp->f_pos != LDISKFS_HTREE_EOF_64BIT);
2280 fops->release(inode, filp);
2286 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2287 struct dentry *dentry, filldir_t filldir)
2289 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2290 struct scrub_file *sf = &scrub->os_file;
2291 struct dentry *child;
2295 /* It is existing MDT0 device. We only allow the case of object without
2296 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2297 * can generate IGIF mode FID for the object and related OI mapping. If
2298 * it is on other MDTs, then becuase file-level backup/restore, related
2299 * OI mapping may be invalid already, we do not know which is the right
2300 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2302 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2303 * "fid" xattr, then something crashed. We cannot re-generate the
2304 * FID directly, instead, the OI scrub will scan the OI structure
2305 * and try to re-generate the LMA from the OI mapping. But if the
2306 * OI mapping crashed or lost also, then we have to give up under
2307 * double failure cases. */
2308 scrub->os_convert_igif = 1;
2309 child = osd_lookup_one_len_unlocked(dev, dot_lustre_name, dentry,
2310 strlen(dot_lustre_name));
2311 if (IS_ERR(child)) {
2312 if (PTR_ERR(child) != -ENOENT)
2313 RETURN(PTR_ERR(child));
2317 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2318 * so the client will get IGIF for the ".lustre" object when
2321 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2322 * it does not know whether there are some old clients cached
2323 * the ".lustre" IGIF during the upgrading. Two choices:
2325 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2326 * It will allow the old connected clients to access the
2327 * ".lustre" with cached IGIF. But it will cause others
2328 * on the MDT failed to check "fid_is_dot_lustre()".
2330 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2331 * for ".lustre" in spite of whether there are some clients
2332 * cached the ".lustre" IGIF or not. It enables the check
2333 * "fid_is_dot_lustre()" on the MDT, although it will cause
2334 * that the old connected clients cannot access the ".lustre"
2335 * with the cached IGIF.
2337 * Usually, it is rare case for the old connected clients
2338 * to access the ".lustre" with cached IGIF. So we prefer
2339 * to the solution 2).
2341 inode_lock(dentry->d_inode);
2342 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2343 child->d_inode, &LU_DOT_LUSTRE_FID,
2345 strlen(dot_lustre_name), 0);
2346 inode_unlock(dentry->d_inode);
2347 if (rc == -ENOENT) {
2349 /* It is 1.8 MDT device. */
2350 if (!(sf->sf_flags & SF_UPGRADE)) {
2351 scrub_file_reset(scrub, dev->od_uuid,
2353 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2354 rc = scrub_file_store(info->oti_env, scrub);
2358 } else if (rc == 0) {
2359 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2368 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2369 struct dentry *dentry, filldir_t filldir)
2371 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2372 struct scrub_file *sf = &scrub->os_file;
2373 struct dentry *child;
2377 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2378 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2379 rc = scrub_file_store(info->oti_env, scrub);
2384 child = osd_lookup_one_len_unlocked(dev, ADMIN_USR, dentry,
2386 if (IS_ERR(child)) {
2387 rc = PTR_ERR(child);
2389 inode_lock(dentry->d_inode);
2390 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2391 child->d_inode, NULL, ADMIN_USR,
2392 strlen(ADMIN_USR), 0);
2393 inode_unlock(dentry->d_inode);
2397 if (rc != 0 && rc != -ENOENT)
2400 child = osd_lookup_one_len_unlocked(dev, ADMIN_GRP, dentry,
2403 GOTO(out, rc = PTR_ERR(child));
2405 inode_lock(dentry->d_inode);
2406 rc = osd_ios_scan_one(info, dev, dentry->d_inode,
2407 child->d_inode, NULL, ADMIN_GRP,
2408 strlen(ADMIN_GRP), 0);
2409 inode_unlock(dentry->d_inode);
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 /* Don't take inode_lock here since scandir() callbacks
2430 * can call VFS functions which may manully take the
2431 * inode lock itself like iterate_dir(). Since this
2432 * is the case it is best to leave the scandir()
2433 * callbacks to managing the inode lock.
2435 scandir(info, dev, dentry, filldir);
2437 dput(item->oii_dentry);
2441 if (list_empty(&dev->od_ios_list))
2444 item = list_entry(dev->od_ios_list.next,
2445 struct osd_ios_item, oii_list);
2446 list_del_init(&item->oii_list);
2448 LASSERT(item->oii_scandir != NULL);
2449 scandir = item->oii_scandir;
2450 filldir = item->oii_filldir;
2451 dentry = item->oii_dentry;
2454 /* There maybe the case that the object has been removed, but its OI
2455 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2456 * file-level backup/restore. So here cleanup the stale OI mappings. */
2457 while (map->olm_name != NULL) {
2458 struct dentry *child;
2460 if (fid_is_zero(&map->olm_fid)) {
2465 child = osd_lookup_one_len_unlocked(dev, map->olm_name,
2466 osd_sb(dev)->s_root,
2468 if (PTR_ERR(child) == -ENOENT ||
2469 (!IS_ERR(child) && !child->d_inode))
2470 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2471 NULL, DTO_INDEX_DELETE,
2478 if (!list_empty(&dev->od_index_restore_list)) {
2481 OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
2483 CERROR("%s: not enough RAM for rebuild index\n",
2486 while (!list_empty(&dev->od_index_restore_list)) {
2487 struct lustre_index_restore_unit *liru;
2489 liru = list_entry(dev->od_index_restore_list.next,
2490 struct lustre_index_restore_unit,
2492 list_del(&liru->liru_link);
2494 osd_index_restore(info->oti_env, dev, liru,
2495 buf, INDEX_BACKUP_BUFSIZE);
2496 OBD_FREE(liru, liru->liru_len);
2500 OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
2506 char *osd_lf_fid2name(const struct lu_fid *fid)
2508 const struct osd_lf_map *map = osd_lf_maps;
2510 while (map->olm_name != NULL) {
2511 if (!lu_fid_eq(fid, &map->olm_fid)) {
2516 if (map->olm_flags & OLF_SHOW_NAME)
2517 return map->olm_name;
2525 /* OI scrub start/stop */
2527 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
2530 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2534 if (dev->od_dt_dev.dd_rdonly)
2537 /* od_otable_mutex: prevent curcurrent start/stop */
2538 mutex_lock(&dev->od_otable_mutex);
2539 rc = scrub_start(osd_scrub_main, scrub, dev, flags);
2540 if (rc == -EALREADY) {
2542 if ((scrub->os_file.sf_flags & SF_AUTO ||
2543 scrub->os_partial_scan) &&
2544 !(flags & SS_AUTO_PARTIAL))
2545 osd_scrub_join(env, dev, flags, false);
2547 mutex_unlock(&dev->od_otable_mutex);
2552 void osd_scrub_stop(struct osd_device *dev)
2554 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2556 /* od_otable_mutex: prevent curcurrent start/stop */
2557 mutex_lock(&dev->od_otable_mutex);
2558 scrub->os_paused = 1;
2560 mutex_unlock(&dev->od_otable_mutex);
2563 /* OI scrub setup/cleanup */
2565 static const char osd_scrub_name[] = "OI_scrub";
2567 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2569 struct osd_thread_info *info = osd_oti_get(env);
2570 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2571 struct lvfs_run_ctxt *ctxt = &dev->od_scrub.os_ctxt;
2572 struct scrub_file *sf = &scrub->os_file;
2573 struct super_block *sb = osd_sb(dev);
2574 struct lvfs_run_ctxt saved;
2576 struct inode *inode;
2577 struct lu_fid *fid = &info->oti_fid;
2578 struct osd_inode_id *id = &info->oti_id;
2579 struct dt_object *obj;
2581 bool restored = false;
2585 memset(&dev->od_scrub, 0, sizeof(struct osd_scrub));
2586 OBD_SET_CTXT_MAGIC(ctxt);
2587 ctxt->pwdmnt = dev->od_mnt;
2588 ctxt->pwd = dev->od_mnt->mnt_root;
2589 ctxt->fs = KERNEL_DS;
2591 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2592 init_rwsem(&scrub->os_rwsem);
2593 spin_lock_init(&scrub->os_lock);
2594 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2595 scrub->os_name = osd_name(dev);
2597 push_ctxt(&saved, ctxt);
2598 filp = filp_open(osd_scrub_name, O_RDWR |
2599 (dev->od_dt_dev.dd_rdonly ? 0 : O_CREAT), 0644);
2601 pop_ctxt(&saved, ctxt);
2602 RETURN(PTR_ERR(filp));
2605 inode = file_inode(filp);
2606 if (!dev->od_dt_dev.dd_rdonly) {
2607 /* 'What the @fid is' is not imporatant, because the object
2608 * has no OI mapping, and only is visible inside the OSD.*/
2609 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2610 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2612 filp_close(filp, NULL);
2613 pop_ctxt(&saved, ctxt);
2618 osd_id_gen(id, inode->i_ino, inode->i_generation);
2619 osd_add_oi_cache(info, dev, id, fid);
2620 filp_close(filp, NULL);
2621 pop_ctxt(&saved, ctxt);
2623 obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, NULL));
2624 if (IS_ERR_OR_NULL(obj))
2625 RETURN(obj ? PTR_ERR(obj) : -ENOENT);
2627 #ifndef HAVE_S_UUID_AS_UUID_T
2628 memcpy(dev->od_uuid.b, sb->s_uuid, UUID_SIZE);
2630 uuid_copy(&dev->od_uuid, &sb->s_uuid);
2632 scrub->os_obj = obj;
2633 rc = scrub_file_load(env, scrub);
2634 if (rc == -ENOENT || rc == -EFAULT) {
2635 scrub_file_init(scrub, dev->od_uuid);
2636 /* If the "/O" dir does not exist when mount (indicated by
2637 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2638 * then it is quite probably that the device is a new one,
2639 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2641 * For the rare case that "/O" and "OI_scrub" both lost on
2642 * an old device, it can be found and cleared later.
2644 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2645 * need to check "filter_fid_18_23" and to convert it to
2646 * "filter_fid" for each object, and all the IGIF should
2647 * have their FID mapping in OI files already. */
2648 if (dev->od_maybe_new && rc == -ENOENT)
2649 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2651 } else if (rc < 0) {
2652 GOTO(cleanup_obj, rc);
2654 if (!uuid_equal(&sf->sf_uuid, &dev->od_uuid)) {
2656 "%s: UUID has been changed from %pU to %pU\n",
2657 osd_dev2name(dev), &sf->sf_uuid, &dev->od_uuid);
2658 scrub_file_reset(scrub, dev->od_uuid, SF_INCONSISTENT);
2661 } else if (sf->sf_status == SS_SCANNING) {
2662 sf->sf_status = SS_CRASHED;
2666 if ((sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
2667 LCONSOLE_WARN("%s: invalid oi count %d, set it to %d\n",
2668 osd_dev2name(dev), sf->sf_oi_count,
2670 sf->sf_oi_count = osd_oi_count;
2675 if (sf->sf_pos_last_checkpoint != 0)
2676 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2678 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2681 rc = scrub_file_store(env, scrub);
2683 GOTO(cleanup_obj, rc);
2686 /* Initialize OI files. */
2687 rc = osd_oi_init(info, dev, restored);
2689 GOTO(cleanup_obj, rc);
2691 if (!dev->od_dt_dev.dd_rdonly)
2692 osd_initial_OI_scrub(info, dev);
2694 if (sf->sf_flags & SF_UPGRADE ||
2695 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2696 sf->sf_success_count > 0)) {
2697 dev->od_igif_inoi = 0;
2698 dev->od_check_ff = dev->od_is_ost;
2700 dev->od_igif_inoi = 1;
2701 dev->od_check_ff = 0;
2704 if (sf->sf_flags & SF_INCONSISTENT)
2705 /* The 'od_igif_inoi' will be set under the
2707 * 1) new created system, or
2708 * 2) restored from file-level backup, or
2709 * 3) the upgrading completed.
2711 * The 'od_igif_inoi' may be cleared by OI scrub
2712 * later if found that the system is upgrading. */
2713 dev->od_igif_inoi = 1;
2715 if (!dev->od_dt_dev.dd_rdonly &&
2716 dev->od_auto_scrub_interval != AS_NEVER &&
2717 ((sf->sf_status == SS_PAUSED) ||
2718 (sf->sf_status == SS_CRASHED &&
2719 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2720 SF_UPGRADE | SF_AUTO)) ||
2721 (sf->sf_status == SS_INIT &&
2722 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2724 rc = osd_scrub_start(env, dev, SS_AUTO_FULL);
2727 GOTO(cleanup_oi, rc);
2729 /* it is possible that dcache entries may keep objects after they are
2730 * deleted by OSD. While it looks safe this can cause object data to
2731 * stay until umount causing failures in tests calculating free space,
2732 * e.g. replay-ost-single. Since those dcache entries are not used
2733 * anymore let's just free them after use here */
2734 shrink_dcache_sb(sb);
2738 osd_oi_fini(info, dev);
2740 dt_object_put_nocache(env, scrub->os_obj);
2741 scrub->os_obj = NULL;
2746 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2748 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2750 LASSERT(dev->od_otable_it == NULL);
2752 if (scrub->os_obj != NULL) {
2753 osd_scrub_stop(dev);
2754 dt_object_put_nocache(env, scrub->os_obj);
2755 scrub->os_obj = NULL;
2757 if (dev->od_oi_table != NULL)
2758 osd_oi_fini(osd_oti_get(env), dev);
2761 /* object table based iteration APIs */
2763 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2764 struct dt_object *dt, __u32 attr)
2766 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2767 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2768 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2769 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2770 struct osd_otable_it *it;
2775 /* od_otable_mutex: prevent curcurrent init/fini */
2776 mutex_lock(&dev->od_otable_mutex);
2777 if (dev->od_otable_it != NULL)
2778 GOTO(out, it = ERR_PTR(-EALREADY));
2782 GOTO(out, it = ERR_PTR(-ENOMEM));
2784 dev->od_otable_it = it;
2786 it->ooi_cache.ooc_consumer_idx = -1;
2787 if (flags & DOIF_OUTUSED)
2788 it->ooi_used_outside = 1;
2790 if (flags & DOIF_RESET)
2793 if (valid & DOIV_ERROR_HANDLE) {
2794 if (flags & DOIF_FAILOUT)
2795 start |= SS_SET_FAILOUT;
2797 start |= SS_CLEAR_FAILOUT;
2800 if (valid & DOIV_DRYRUN) {
2801 if (flags & DOIF_DRYRUN)
2802 start |= SS_SET_DRYRUN;
2804 start |= SS_CLEAR_DRYRUN;
2807 rc = scrub_start(osd_scrub_main, scrub, dev, start & ~SS_AUTO_PARTIAL);
2808 if (rc == -EALREADY) {
2809 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2810 } else if (rc < 0) {
2811 dev->od_otable_it = NULL;
2815 /* We have to start from the begining. */
2816 it->ooi_cache.ooc_pos_preload =
2817 LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2823 mutex_unlock(&dev->od_otable_mutex);
2824 return (struct dt_it *)it;
2827 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2829 struct osd_otable_it *it = (struct osd_otable_it *)di;
2830 struct osd_device *dev = it->ooi_dev;
2832 /* od_otable_mutex: prevent curcurrent init/fini */
2833 mutex_lock(&dev->od_otable_mutex);
2834 scrub_stop(&dev->od_scrub.os_scrub);
2835 LASSERT(dev->od_otable_it == it);
2837 dev->od_otable_it = NULL;
2838 mutex_unlock(&dev->od_otable_mutex);
2842 static int osd_otable_it_get(const struct lu_env *env,
2843 struct dt_it *di, const struct dt_key *key)
2848 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2853 osd_otable_it_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
2855 spin_lock(&scrub->os_lock);
2856 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2857 scrub->os_waiting ||
2858 !thread_is_running(&scrub->os_thread))
2859 it->ooi_waiting = 0;
2861 it->ooi_waiting = 1;
2862 spin_unlock(&scrub->os_lock);
2864 return !it->ooi_waiting;
2867 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2869 struct osd_otable_it *it = (struct osd_otable_it *)di;
2870 struct osd_device *dev = it->ooi_dev;
2871 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2872 struct osd_otable_cache *ooc = &it->ooi_cache;
2873 struct ptlrpc_thread *thread = &scrub->os_thread;
2877 LASSERT(it->ooi_user_ready);
2880 if (!thread_is_running(thread) && !it->ooi_used_outside)
2883 if (ooc->ooc_cached_items > 0) {
2884 ooc->ooc_cached_items--;
2885 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2886 ~OSD_OTABLE_IT_CACHE_MASK;
2890 if (it->ooi_all_cached) {
2891 wait_event_idle(thread->t_ctl_waitq,
2892 !thread_is_running(thread));
2896 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2897 spin_lock(&scrub->os_lock);
2898 scrub->os_waiting = 0;
2899 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2900 spin_unlock(&scrub->os_lock);
2903 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2904 wait_event_idle(thread->t_ctl_waitq,
2905 osd_otable_it_wakeup(scrub, it));
2907 if (!thread_is_running(thread) && !it->ooi_used_outside)
2910 rc = osd_otable_it_preload(env, it);
2917 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2918 const struct dt_it *di)
2923 static int osd_otable_it_key_size(const struct lu_env *env,
2924 const struct dt_it *di)
2926 return sizeof(__u64);
2929 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2930 struct dt_rec *rec, __u32 attr)
2932 struct osd_otable_it *it = (struct osd_otable_it *)di;
2933 struct osd_otable_cache *ooc = &it->ooi_cache;
2935 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2937 /* Filter out Invald FID already. */
2938 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2939 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2940 PFID((struct lu_fid *)rec),
2941 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2946 static __u64 osd_otable_it_store(const struct lu_env *env,
2947 const struct dt_it *di)
2949 struct osd_otable_it *it = (struct osd_otable_it *)di;
2950 struct osd_otable_cache *ooc = &it->ooi_cache;
2953 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2954 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2956 hash = ooc->ooc_pos_preload;
2961 * Set the OSD layer iteration start position as the specified hash.
2963 static int osd_otable_it_load(const struct lu_env *env,
2964 const struct dt_it *di, __u64 hash)
2966 struct osd_otable_it *it = (struct osd_otable_it *)di;
2967 struct osd_device *dev = it->ooi_dev;
2968 struct osd_otable_cache *ooc = &it->ooi_cache;
2969 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2970 struct osd_iit_param *param = &it->ooi_iit_param;
2974 /* Forbid to set iteration position after iteration started. */
2975 if (it->ooi_user_ready)
2978 LASSERT(!scrub->os_partial_scan);
2980 if (hash > OSD_OTABLE_MAX_HASH)
2981 hash = OSD_OTABLE_MAX_HASH;
2983 /* The hash is the last checkpoint position,
2984 * we will start from the next one. */
2985 ooc->ooc_pos_preload = hash + 1;
2986 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2987 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2989 it->ooi_user_ready = 1;
2990 if (!scrub->os_full_speed)
2991 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2993 memset(param, 0, sizeof(*param));
2994 param->sb = osd_sb(dev);
2995 param->start = ooc->ooc_pos_preload;
2996 param->bg = (ooc->ooc_pos_preload - 1) /
2997 LDISKFS_INODES_PER_GROUP(param->sb);
2998 param->offset = (ooc->ooc_pos_preload - 1) %
2999 LDISKFS_INODES_PER_GROUP(param->sb);
3000 param->gbase = 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
3002 /* Unplug OSD layer iteration by the first next() call. */
3003 rc = osd_otable_it_next(env, (struct dt_it *)it);
3008 static int osd_otable_it_key_rec(const struct lu_env *env,
3009 const struct dt_it *di, void *key_rec)
3014 const struct dt_index_operations osd_otable_ops = {
3016 .init = osd_otable_it_init,
3017 .fini = osd_otable_it_fini,
3018 .get = osd_otable_it_get,
3019 .put = osd_otable_it_put,
3020 .next = osd_otable_it_next,
3021 .key = osd_otable_it_key,
3022 .key_size = osd_otable_it_key_size,
3023 .rec = osd_otable_it_rec,
3024 .store = osd_otable_it_store,
3025 .load = osd_otable_it_load,
3026 .key_rec = osd_otable_it_key_rec,
3030 /* high priority inconsistent items list APIs */
3032 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
3034 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3037 struct osd_inconsistent_item *oii;
3038 struct osd_scrub *oscrub = &dev->od_scrub;
3039 struct lustre_scrub *lscrub = &oscrub->os_scrub;
3040 struct ptlrpc_thread *thread = &lscrub->os_thread;
3045 if (unlikely(oii == NULL))
3048 INIT_LIST_HEAD(&oii->oii_list);
3049 oii->oii_cache = *oic;
3050 oii->oii_insert = insert;
3052 if (lscrub->os_partial_scan) {
3053 __u64 now = ktime_get_real_seconds();
3055 /* If there haven't been errors in a long time,
3056 * decay old count until either the errors are
3057 * gone or we reach the current interval. */
3058 while (unlikely(oscrub->os_bad_oimap_count > 0 &&
3059 oscrub->os_bad_oimap_time +
3060 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3061 oscrub->os_bad_oimap_count >>= 1;
3062 oscrub->os_bad_oimap_time +=
3063 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3066 oscrub->os_bad_oimap_time = now;
3067 if (++oscrub->os_bad_oimap_count >
3068 dev->od_full_scrub_threshold_rate)
3069 lscrub->os_full_scrub = 1;
3072 spin_lock(&lscrub->os_lock);
3073 if (unlikely(!thread_is_running(thread))) {
3074 spin_unlock(&lscrub->os_lock);
3079 if (list_empty(&lscrub->os_inconsistent_items))
3081 list_add_tail(&oii->oii_list, &lscrub->os_inconsistent_items);
3082 spin_unlock(&lscrub->os_lock);
3085 wake_up_all(&thread->t_ctl_waitq);
3090 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3091 struct osd_inode_id *id)
3093 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
3094 struct osd_inconsistent_item *oii;
3097 spin_lock(&scrub->os_lock);
3098 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3099 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3100 *id = oii->oii_cache.oic_lid;
3101 spin_unlock(&scrub->os_lock);
3105 spin_unlock(&scrub->os_lock);
3110 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3112 struct osd_scrub *scrub = &dev->od_scrub;
3114 scrub_dump(m, &scrub->os_scrub);
3115 seq_printf(m, "lf_scanned: %llu\n"
3117 "lf_failed: %llu\n",
3118 scrub->os_lf_scanned,
3119 scrub->os_scrub.os_file.sf_param & SP_DRYRUN ?
3120 "inconsistent" : "repaired",
3121 scrub->os_lf_repaired,
3122 scrub->os_lf_failed);