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_old *ff = &info->oti_ff;
167 struct dentry *dentry = &info->oti_obj_dentry;
168 struct lu_fid *tfid = &info->oti_fid;
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 ll_vfs_dq_init(inode);
211 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
216 } else if (rc != -ENODATA && rc != sizeof(struct filter_fid)) {
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,
476 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
478 if (flags & SS_AUTO_FULL) {
479 scrub->os_full_speed = 1;
480 scrub->os_partial_scan = 0;
481 sf->sf_flags |= SF_AUTO;
482 } else if (flags & SS_AUTO_PARTIAL) {
483 scrub->os_full_speed = 0;
484 scrub->os_partial_scan = 1;
485 sf->sf_flags |= SF_AUTO;
486 } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
488 scrub->os_full_speed = 1;
489 scrub->os_partial_scan = 0;
491 scrub->os_full_speed = 0;
492 scrub->os_partial_scan = 0;
495 spin_lock(&scrub->os_lock);
496 scrub->os_in_prior = 0;
497 scrub->os_waiting = 0;
498 scrub->os_paused = 0;
499 scrub->os_in_join = 0;
500 scrub->os_full_scrub = 0;
501 spin_unlock(&scrub->os_lock);
502 scrub->os_new_checked = 0;
503 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
504 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
505 else if (sf->sf_pos_last_checkpoint != 0)
506 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
508 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
510 scrub->os_pos_current = sf->sf_pos_latest_start;
511 sf->sf_status = SS_SCANNING;
512 sf->sf_time_latest_start = ktime_get_real_seconds();
513 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
514 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
515 rc = scrub_file_store(env, scrub);
517 spin_lock(&scrub->os_lock);
518 thread_set_flags(thread, SVC_RUNNING);
519 spin_unlock(&scrub->os_lock);
520 wake_up_all(&thread->t_ctl_waitq);
522 up_write(&scrub->os_rwsem);
527 static int osd_scrub_post(const struct lu_env *env, struct osd_device *dev,
530 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
531 struct scrub_file *sf = &scrub->os_file;
535 CDEBUG(D_LFSCK, "%s: OI scrub post with result = %d\n",
536 osd_scrub2name(scrub), result);
538 down_write(&scrub->os_rwsem);
539 spin_lock(&scrub->os_lock);
540 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
541 spin_unlock(&scrub->os_lock);
542 if (scrub->os_new_checked > 0) {
543 sf->sf_items_checked += scrub->os_new_checked;
544 scrub->os_new_checked = 0;
545 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
547 sf->sf_time_last_checkpoint = ktime_get_real_seconds();
549 dev->od_igif_inoi = 1;
550 dev->od_check_ff = 0;
551 sf->sf_status = SS_COMPLETED;
552 if (!(sf->sf_param & SP_DRYRUN)) {
553 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
554 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
555 SF_UPGRADE | SF_AUTO);
557 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
558 sf->sf_success_count++;
559 } else if (result == 0) {
560 if (scrub->os_paused)
561 sf->sf_status = SS_PAUSED;
563 sf->sf_status = SS_STOPPED;
565 sf->sf_status = SS_FAILED;
567 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
568 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;
639 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 old 2.x (x <= 3) or 1.8 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 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 pos == osd_remote_parent_ino(dev)))
733 RETURN(SCRUB_NEXT_CONTINUE);
735 osd_id_gen(lid, pos, OSD_OII_NOGEN);
736 inode = osd_iget(info, dev, lid);
739 /* The inode may be removed after bitmap searching, or the
740 * file is new created without inode initialized yet. */
741 if (rc == -ENOENT || rc == -ESTALE)
742 RETURN(SCRUB_NEXT_CONTINUE);
744 CDEBUG(D_LFSCK, "%s: fail to read inode, ino# = %u: "
745 "rc = %d\n", osd_dev2name(dev), pos, rc);
749 /* It is an EA inode, no OI mapping for it, skip it. */
750 if (osd_is_ea_inode(inode))
751 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
754 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
755 /* Only skip it for the first OI scrub accessing. */
756 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
757 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
760 rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
769 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
770 struct osd_iit_param *param,
771 struct osd_idmap_cache **oic, const bool noslot)
773 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
774 struct ptlrpc_thread *thread = &scrub->os_thread;
776 struct osd_inode_id *lid;
779 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
780 struct l_wait_info lwi;
782 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
783 if (likely(lwi.lwi_timeout > 0))
784 l_wait_event(thread->t_ctl_waitq,
785 !list_empty(&scrub->os_inconsistent_items) ||
786 !thread_is_running(thread),
790 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
791 spin_lock(&scrub->os_lock);
792 thread_set_flags(thread, SVC_STOPPING);
793 spin_unlock(&scrub->os_lock);
794 return SCRUB_NEXT_CRASH;
797 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
798 return SCRUB_NEXT_FATAL;
800 if (unlikely(!thread_is_running(thread)))
801 return SCRUB_NEXT_EXIT;
803 if (!list_empty(&scrub->os_inconsistent_items)) {
804 spin_lock(&scrub->os_lock);
805 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
806 struct osd_inconsistent_item *oii;
808 oii = list_entry(scrub->os_inconsistent_items.next,
809 struct osd_inconsistent_item, oii_list);
810 spin_unlock(&scrub->os_lock);
812 *oic = &oii->oii_cache;
813 scrub->os_in_prior = 1;
817 spin_unlock(&scrub->os_lock);
821 return SCRUB_NEXT_WAIT;
823 rc = osd_iit_next(param, &scrub->os_pos_current);
827 *oic = &dev->od_scrub.os_oic;
828 fid = &(*oic)->oic_fid;
829 lid = &(*oic)->oic_lid;
830 rc = osd_iit_iget(info, dev, fid, lid,
831 scrub->os_pos_current, param->sb, true);
835 static int osd_preload_next(struct osd_thread_info *info,
836 struct osd_device *dev, struct osd_iit_param *param,
837 struct osd_idmap_cache **oic, const bool noslot)
839 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
840 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
841 struct ptlrpc_thread *thread = &scrub->os_thread;
844 if (thread_is_running(thread) &&
845 ooc->ooc_pos_preload >= scrub->os_pos_current)
846 return SCRUB_NEXT_EXIT;
848 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
852 rc = osd_iit_iget(info, dev,
853 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
854 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
855 ooc->ooc_pos_preload, param->sb, false);
860 osd_scrub_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
862 spin_lock(&scrub->os_lock);
863 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
864 !list_empty(&scrub->os_inconsistent_items) ||
865 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
866 scrub->os_waiting = 0;
868 scrub->os_waiting = 1;
869 spin_unlock(&scrub->os_lock);
871 return !scrub->os_waiting;
874 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
875 struct osd_iit_param *param,
876 struct osd_idmap_cache *oic, bool *noslot, int rc)
878 struct l_wait_info lwi = { 0 };
879 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
880 struct scrub_file *sf = &scrub->os_file;
881 struct ptlrpc_thread *thread = &scrub->os_thread;
882 struct osd_otable_it *it = dev->od_otable_it;
883 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
886 case SCRUB_NEXT_NOSCRUB:
887 down_write(&scrub->os_rwsem);
888 scrub->os_new_checked++;
889 sf->sf_items_noscrub++;
890 up_write(&scrub->os_rwsem);
891 case SCRUB_NEXT_CONTINUE:
892 case SCRUB_NEXT_WAIT:
896 rc = osd_scrub_check_update(info, dev, oic, rc);
898 scrub->os_in_prior = 0;
902 rc = scrub_checkpoint(info->oti_env, scrub);
904 CDEBUG(D_LFSCK, "%s: fail to checkpoint, pos = %llu: "
905 "rc = %d\n", osd_scrub2name(scrub),
906 scrub->os_pos_current, rc);
907 /* Continue, as long as the scrub itself can go ahead. */
910 if (scrub->os_in_prior) {
911 scrub->os_in_prior = 0;
916 if (it != NULL && it->ooi_waiting && ooc != NULL &&
917 ooc->ooc_pos_preload < scrub->os_pos_current) {
918 spin_lock(&scrub->os_lock);
920 wake_up_all(&thread->t_ctl_waitq);
921 spin_unlock(&scrub->os_lock);
924 if (rc == SCRUB_NEXT_CONTINUE)
927 if (scrub->os_full_speed || !ooc || osd_scrub_has_window(scrub, ooc)) {
933 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
936 if (!ooc || osd_scrub_has_window(scrub, ooc))
943 static int osd_preload_exec(struct osd_thread_info *info,
944 struct osd_device *dev, struct osd_iit_param *param,
945 struct osd_idmap_cache *oic, bool *noslot, int rc)
947 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
950 ooc->ooc_cached_items++;
951 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
952 ~OSD_OTABLE_IT_CACHE_MASK;
954 return rc > 0 ? 0 : rc;
957 #define SCRUB_IT_ALL 1
958 #define SCRUB_IT_CRASH 2
960 static void osd_scrub_join(const struct lu_env *env, struct osd_device *dev,
961 __u32 flags, bool inconsistent)
963 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
964 struct ptlrpc_thread *thread = &scrub->os_thread;
965 struct scrub_file *sf = &scrub->os_file;
969 LASSERT(!(flags & SS_AUTO_PARTIAL));
971 down_write(&scrub->os_rwsem);
972 scrub->os_in_join = 1;
973 if (flags & SS_SET_FAILOUT)
974 sf->sf_param |= SP_FAILOUT;
975 else if (flags & SS_CLEAR_FAILOUT)
976 sf->sf_param &= ~SP_FAILOUT;
978 if (flags & SS_SET_DRYRUN)
979 sf->sf_param |= SP_DRYRUN;
980 else if (flags & SS_CLEAR_DRYRUN)
981 sf->sf_param &= ~SP_DRYRUN;
983 if (flags & SS_RESET) {
984 scrub_file_reset(scrub, LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
985 inconsistent ? SF_INCONSISTENT : 0);
986 sf->sf_status = SS_SCANNING;
989 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
990 scrub->os_full_speed = 1;
992 scrub->os_full_speed = 0;
994 if (flags & SS_AUTO_FULL) {
995 sf->sf_flags |= SF_AUTO;
996 scrub->os_full_speed = 1;
999 scrub->os_new_checked = 0;
1000 if (sf->sf_pos_last_checkpoint != 0)
1001 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1003 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1005 scrub->os_pos_current = sf->sf_pos_latest_start;
1006 sf->sf_time_latest_start = ktime_get_real_seconds();
1007 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1008 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1009 rc = scrub_file_store(env, scrub);
1011 spin_lock(&scrub->os_lock);
1012 scrub->os_waiting = 0;
1013 scrub->os_paused = 0;
1014 scrub->os_partial_scan = 0;
1015 scrub->os_in_join = 0;
1016 scrub->os_full_scrub = 0;
1017 spin_unlock(&scrub->os_lock);
1018 wake_up_all(&thread->t_ctl_waitq);
1019 up_write(&scrub->os_rwsem);
1021 CDEBUG(D_LFSCK, "%s: joined in the OI scrub with flag %u: rc = %d\n",
1022 osd_scrub2name(scrub), flags, rc);
1027 static int osd_inode_iteration(struct osd_thread_info *info,
1028 struct osd_device *dev, __u32 max, bool preload)
1030 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1031 struct ptlrpc_thread *thread = &scrub->os_thread;
1032 struct scrub_file *sf = &scrub->os_file;
1033 osd_iit_next_policy next;
1034 osd_iit_exec_policy exec;
1037 struct osd_iit_param *param;
1038 struct l_wait_info lwi = { 0 };
1047 param = &dev->od_scrub.os_iit_param;
1048 memset(param, 0, sizeof(*param));
1049 param->sb = osd_sb(dev);
1051 while (scrub->os_partial_scan && !scrub->os_in_join) {
1052 struct osd_idmap_cache *oic = NULL;
1054 rc = osd_scrub_next(info, dev, param, &oic, noslot);
1056 case SCRUB_NEXT_EXIT:
1058 case SCRUB_NEXT_CRASH:
1059 RETURN(SCRUB_IT_CRASH);
1060 case SCRUB_NEXT_FATAL:
1062 case SCRUB_NEXT_WAIT: {
1063 struct kstatfs *ksfs = &info->oti_ksfs;
1066 if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1067 unlikely(sf->sf_items_updated_prior == 0))
1070 if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1071 scrub->os_full_scrub) {
1072 osd_scrub_join(info->oti_env, dev,
1073 SS_AUTO_FULL | SS_RESET, true);
1077 rc = param->sb->s_op->statfs(param->sb->s_root, ksfs);
1079 __u64 used = ksfs->f_files - ksfs->f_ffree;
1081 do_div(used, sf->sf_items_updated_prior);
1082 /* If we hit too much inconsistent OI
1083 * mappings during the partial scan,
1084 * then scan the device completely. */
1085 if (used < dev->od_full_scrub_ratio) {
1086 osd_scrub_join(info->oti_env, dev,
1087 SS_AUTO_FULL | SS_RESET, true);
1093 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1097 saved_flags = sf->sf_flags;
1098 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1099 SF_UPGRADE | SF_AUTO);
1100 sf->sf_status = SS_COMPLETED;
1101 l_wait_event(thread->t_ctl_waitq,
1102 !thread_is_running(thread) ||
1103 !scrub->os_partial_scan ||
1104 scrub->os_in_join ||
1105 !list_empty(&scrub->os_inconsistent_items),
1107 sf->sf_flags = saved_flags;
1108 sf->sf_status = SS_SCANNING;
1110 if (unlikely(!thread_is_running(thread)))
1113 if (!scrub->os_partial_scan || scrub->os_in_join)
1119 LASSERTF(rc == 0, "rc = %d\n", rc);
1121 osd_scrub_exec(info, dev, param, oic, &noslot, rc);
1128 l_wait_event(thread->t_ctl_waitq,
1129 !thread_is_running(thread) || !scrub->os_in_join,
1132 if (unlikely(!thread_is_running(thread)))
1138 next = osd_scrub_next;
1139 exec = osd_scrub_exec;
1140 pos = &scrub->os_pos_current;
1141 count = &scrub->os_new_checked;
1142 param->start = *pos;
1143 param->bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param->sb);
1145 (*pos - 1) % LDISKFS_INODES_PER_GROUP(param->sb);
1147 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1149 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1151 next = osd_preload_next;
1152 exec = osd_preload_exec;
1153 pos = &ooc->ooc_pos_preload;
1154 count = &ooc->ooc_cached_items;
1155 param = &dev->od_otable_it->ooi_iit_param;
1159 limit = le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count);
1160 while (*pos <= limit && *count < max) {
1161 struct ldiskfs_group_desc *desc;
1162 bool next_group = false;
1164 desc = ldiskfs_get_group_desc(param->sb, param->bg, NULL);
1168 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1173 param->bitmap = ldiskfs_read_inode_bitmap(param->sb, param->bg);
1174 if (!param->bitmap) {
1175 CERROR("%s: fail to read bitmap for %u, "
1176 "scrub will stop, urgent mode\n",
1177 osd_scrub2name(scrub), (__u32)param->bg);
1182 struct osd_idmap_cache *oic = NULL;
1185 ldiskfs_itable_unused_count(param->sb, desc) >=
1186 LDISKFS_INODES_PER_GROUP(param->sb)) {
1191 rc = next(info, dev, param, &oic, noslot);
1193 case SCRUB_NEXT_BREAK:
1196 case SCRUB_NEXT_EXIT:
1197 brelse(param->bitmap);
1199 case SCRUB_NEXT_CRASH:
1200 brelse(param->bitmap);
1201 RETURN(SCRUB_IT_CRASH);
1202 case SCRUB_NEXT_FATAL:
1203 brelse(param->bitmap);
1207 rc = exec(info, dev, param, oic, &noslot, rc);
1208 } while (!rc && *pos <= limit && *count < max);
1211 if (param->bitmap) {
1212 brelse(param->bitmap);
1213 param->bitmap = NULL;
1223 param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
1224 *pos = param->gbase;
1225 param->start = *pos;
1230 RETURN(SCRUB_IT_ALL);
1236 static int osd_otable_it_preload(const struct lu_env *env,
1237 struct osd_otable_it *it)
1239 struct osd_device *dev = it->ooi_dev;
1240 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1241 struct osd_otable_cache *ooc = &it->ooi_cache;
1245 rc = osd_inode_iteration(osd_oti_get(env), dev,
1246 OSD_OTABLE_IT_CACHE_SIZE, true);
1247 if (rc == SCRUB_IT_ALL)
1248 it->ooi_all_cached = 1;
1250 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1251 spin_lock(&scrub->os_lock);
1252 scrub->os_waiting = 0;
1253 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1254 spin_unlock(&scrub->os_lock);
1257 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1260 static int osd_scrub_main(void *args)
1263 struct osd_device *dev = (struct osd_device *)args;
1264 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1265 struct ptlrpc_thread *thread = &scrub->os_thread;
1269 rc = lu_env_init(&env, LCT_LOCAL | LCT_DT_THREAD);
1271 CDEBUG(D_LFSCK, "%s: OI scrub fail to init env: rc = %d\n",
1272 osd_scrub2name(scrub), rc);
1276 rc = osd_scrub_prep(&env, dev);
1278 CDEBUG(D_LFSCK, "%s: OI scrub fail to scrub prep: rc = %d\n",
1279 osd_scrub2name(scrub), rc);
1283 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1284 struct l_wait_info lwi = { 0 };
1285 struct osd_otable_it *it = dev->od_otable_it;
1286 struct osd_otable_cache *ooc = &it->ooi_cache;
1288 l_wait_event(thread->t_ctl_waitq,
1289 it->ooi_user_ready || !thread_is_running(thread),
1291 if (unlikely(!thread_is_running(thread)))
1294 scrub->os_pos_current = ooc->ooc_pos_preload;
1297 CDEBUG(D_LFSCK, "%s: OI scrub start, flags = 0x%x, pos = %llu\n",
1298 osd_scrub2name(scrub), scrub->os_start_flags,
1299 scrub->os_pos_current);
1301 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1302 if (unlikely(rc == SCRUB_IT_CRASH)) {
1303 spin_lock(&scrub->os_lock);
1304 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
1305 spin_unlock(&scrub->os_lock);
1306 GOTO(out, rc = -EINVAL);
1312 rc = osd_scrub_post(&env, dev, rc);
1313 CDEBUG(D_LFSCK, "%s: OI scrub: stop, pos = %llu: rc = %d\n",
1314 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1317 while (!list_empty(&scrub->os_inconsistent_items)) {
1318 struct osd_inconsistent_item *oii;
1320 oii = list_entry(scrub->os_inconsistent_items.next,
1321 struct osd_inconsistent_item, oii_list);
1322 list_del_init(&oii->oii_list);
1328 spin_lock(&scrub->os_lock);
1329 thread_set_flags(thread, SVC_STOPPED);
1330 wake_up_all(&thread->t_ctl_waitq);
1331 spin_unlock(&scrub->os_lock);
1335 /* initial OI scrub */
1337 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1338 struct dentry *, filldir_t filldir);
1340 #ifdef HAVE_FILLDIR_USE_CTX
1341 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1342 int namelen, loff_t offset, __u64 ino,
1344 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1345 int namelen, loff_t offset, __u64 ino,
1347 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1348 int namelen, loff_t offset, __u64 ino,
1350 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1351 int namelen, loff_t offset, __u64 ino,
1354 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1355 loff_t offset, __u64 ino, unsigned d_type);
1356 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1357 loff_t offset, __u64 ino, unsigned d_type);
1358 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1359 loff_t offset, __u64 ino, unsigned d_type);
1360 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1361 loff_t offset, __u64 ino, unsigned d_type);
1365 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1366 struct dentry *dentry, filldir_t filldir);
1368 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1369 struct dentry *dentry, filldir_t filldir);
1372 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1373 struct dentry *dentry, filldir_t filldir);
1377 struct lu_fid olm_fid;
1380 scandir_t olm_scandir;
1381 filldir_t olm_filldir;
1384 /* Add the new introduced local files in the list in the future. */
1385 static const struct osd_lf_map osd_lf_maps[] = {
1388 .olm_name = CATLIST,
1390 .f_seq = FID_SEQ_LOCAL_FILE,
1391 .f_oid = LLOG_CATALOGS_OID,
1393 .olm_flags = OLF_SHOW_NAME,
1394 .olm_namelen = sizeof(CATLIST) - 1,
1399 .olm_name = MOUNT_CONFIGS_DIR,
1401 .f_seq = FID_SEQ_LOCAL_FILE,
1402 .f_oid = MGS_CONFIGS_OID,
1404 .olm_flags = OLF_SCAN_SUBITEMS,
1405 .olm_namelen = sizeof(MOUNT_CONFIGS_DIR) - 1,
1406 .olm_scandir = osd_ios_general_scan,
1407 .olm_filldir = osd_ios_varfid_fill,
1410 /* NIDTBL_VERSIONS */
1412 .olm_name = MGS_NIDTBL_DIR,
1413 .olm_flags = OLF_SCAN_SUBITEMS,
1414 .olm_namelen = sizeof(MGS_NIDTBL_DIR) - 1,
1415 .olm_scandir = osd_ios_general_scan,
1416 .olm_filldir = osd_ios_varfid_fill,
1421 .olm_name = "PENDING",
1422 .olm_namelen = sizeof("PENDING") - 1,
1429 .f_seq = FID_SEQ_ROOT,
1430 .f_oid = FID_OID_ROOT,
1432 .olm_flags = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1433 .olm_namelen = sizeof("ROOT") - 1,
1434 .olm_scandir = osd_ios_ROOT_scan,
1437 /* changelog_catalog */
1439 .olm_name = CHANGELOG_CATALOG,
1440 .olm_namelen = sizeof(CHANGELOG_CATALOG) - 1,
1443 /* changelog_users */
1445 .olm_name = CHANGELOG_USERS,
1446 .olm_namelen = sizeof(CHANGELOG_USERS) - 1,
1453 .f_seq = FID_SEQ_LOCAL_FILE,
1454 .f_oid = FLD_INDEX_OID,
1456 .olm_flags = OLF_SHOW_NAME,
1457 .olm_namelen = sizeof("fld") - 1,
1462 .olm_name = LAST_RCVD,
1464 .f_seq = FID_SEQ_LOCAL_FILE,
1465 .f_oid = LAST_RECV_OID,
1467 .olm_flags = OLF_SHOW_NAME,
1468 .olm_namelen = sizeof(LAST_RCVD) - 1,
1473 .olm_name = REPLY_DATA,
1475 .f_seq = FID_SEQ_LOCAL_FILE,
1476 .f_oid = REPLY_DATA_OID,
1478 .olm_flags = OLF_SHOW_NAME,
1479 .olm_namelen = sizeof(REPLY_DATA) - 1,
1484 .olm_name = LOV_OBJID,
1486 .f_seq = FID_SEQ_LOCAL_FILE,
1487 .f_oid = MDD_LOV_OBJ_OID,
1489 .olm_flags = OLF_SHOW_NAME,
1490 .olm_namelen = sizeof(LOV_OBJID) - 1,
1495 .olm_name = LOV_OBJSEQ,
1497 .f_seq = FID_SEQ_LOCAL_FILE,
1498 .f_oid = MDD_LOV_OBJ_OSEQ,
1500 .olm_flags = OLF_SHOW_NAME,
1501 .olm_namelen = sizeof(LOV_OBJSEQ) - 1,
1506 .olm_name = QMT_DIR,
1507 .olm_flags = OLF_SCAN_SUBITEMS,
1508 .olm_namelen = sizeof(QMT_DIR) - 1,
1509 .olm_scandir = osd_ios_general_scan,
1510 .olm_filldir = osd_ios_varfid_fill,
1515 .olm_name = QSD_DIR,
1516 .olm_flags = OLF_SCAN_SUBITEMS,
1517 .olm_namelen = sizeof(QSD_DIR) - 1,
1518 .olm_scandir = osd_ios_general_scan,
1519 .olm_filldir = osd_ios_varfid_fill,
1524 .olm_name = "seq_ctl",
1526 .f_seq = FID_SEQ_LOCAL_FILE,
1527 .f_oid = FID_SEQ_CTL_OID,
1529 .olm_flags = OLF_SHOW_NAME,
1530 .olm_namelen = sizeof("seq_ctl") - 1,
1535 .olm_name = "seq_srv",
1537 .f_seq = FID_SEQ_LOCAL_FILE,
1538 .f_oid = FID_SEQ_SRV_OID,
1540 .olm_flags = OLF_SHOW_NAME,
1541 .olm_namelen = sizeof("seq_srv") - 1,
1546 .olm_name = HEALTH_CHECK,
1548 .f_seq = FID_SEQ_LOCAL_FILE,
1549 .f_oid = OFD_HEALTH_CHECK_OID,
1551 .olm_flags = OLF_SHOW_NAME,
1552 .olm_namelen = sizeof(HEALTH_CHECK) - 1,
1557 .olm_name = LFSCK_DIR,
1558 .olm_flags = OLF_SCAN_SUBITEMS,
1559 .olm_namelen = sizeof(LFSCK_DIR) - 1,
1560 .olm_scandir = osd_ios_general_scan,
1561 .olm_filldir = osd_ios_varfid_fill,
1564 /* lfsck_bookmark */
1566 .olm_name = LFSCK_BOOKMARK,
1567 .olm_namelen = sizeof(LFSCK_BOOKMARK) - 1,
1572 .olm_name = LFSCK_LAYOUT,
1573 .olm_namelen = sizeof(LFSCK_LAYOUT) - 1,
1576 /* lfsck_namespace */
1578 .olm_name = LFSCK_NAMESPACE,
1579 .olm_namelen = sizeof(LFSCK_NAMESPACE) - 1,
1582 /* OBJECTS, upgrade from old device */
1584 .olm_name = OBJECTS,
1585 .olm_flags = OLF_SCAN_SUBITEMS,
1586 .olm_namelen = sizeof(OBJECTS) - 1,
1587 .olm_scandir = osd_ios_OBJECTS_scan,
1590 /* lquota_v2.user, upgrade from old device */
1592 .olm_name = "lquota_v2.user",
1593 .olm_namelen = sizeof("lquota_v2.user") - 1,
1596 /* lquota_v2.group, upgrade from old device */
1598 .olm_name = "lquota_v2.group",
1599 .olm_namelen = sizeof("lquota_v2.group") - 1,
1602 /* LAST_GROUP, upgrade from old device */
1604 .olm_name = "LAST_GROUP",
1606 .f_seq = FID_SEQ_LOCAL_FILE,
1607 .f_oid = OFD_LAST_GROUP_OID,
1609 .olm_flags = OLF_SHOW_NAME,
1610 .olm_namelen = sizeof("LAST_GROUP") - 1,
1613 /* committed batchid for cross-MDT operation */
1615 .olm_name = "BATCHID",
1617 .f_seq = FID_SEQ_LOCAL_FILE,
1618 .f_oid = BATCHID_COMMITTED_OID,
1620 .olm_flags = OLF_SHOW_NAME,
1621 .olm_namelen = sizeof("BATCHID") - 1,
1624 /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1625 * and f_oid = index for their log files. See lu_update_log{_dir}_fid()
1626 * for more details. */
1630 .olm_name = "update_log",
1632 .f_seq = FID_SEQ_UPDATE_LOG,
1634 .olm_flags = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1635 .olm_namelen = sizeof("update_log") - 1,
1638 /* update_log_dir */
1640 .olm_name = "update_log_dir",
1642 .f_seq = FID_SEQ_UPDATE_LOG_DIR,
1644 .olm_flags = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1646 .olm_namelen = sizeof("update_log_dir") - 1,
1647 .olm_scandir = osd_ios_general_scan,
1648 .olm_filldir = osd_ios_uld_fill,
1653 .olm_name = "lost+found",
1655 .f_seq = FID_SEQ_LOCAL_FILE,
1656 .f_oid = OSD_LPF_OID,
1658 .olm_flags = OLF_SCAN_SUBITEMS,
1659 .olm_namelen = sizeof("lost+found") - 1,
1660 .olm_scandir = osd_ios_general_scan,
1661 .olm_filldir = osd_ios_lf_fill,
1666 .olm_name = HSM_ACTIONS,
1671 .olm_name = LUSTRE_NODEMAP_NAME,
1679 /* Add the new introduced files under .lustre/ in the list in the future. */
1680 static const struct osd_lf_map osd_dl_maps[] = {
1685 .f_seq = FID_SEQ_DOT_LUSTRE,
1686 .f_oid = FID_OID_DOT_LUSTRE_OBF,
1688 .olm_namelen = sizeof("fid") - 1,
1691 /* .lustre/lost+found */
1693 .olm_name = "lost+found",
1695 .f_seq = FID_SEQ_DOT_LUSTRE,
1696 .f_oid = FID_OID_DOT_LUSTRE_LPF,
1698 .olm_namelen = sizeof("lost+found") - 1,
1706 struct osd_ios_item {
1707 struct list_head oii_list;
1708 struct dentry *oii_dentry;
1709 scandir_t oii_scandir;
1710 filldir_t oii_filldir;
1713 struct osd_ios_filldir_buf {
1714 #ifdef HAVE_DIR_CONTEXT
1715 /* please keep it as first member */
1716 struct dir_context ctx;
1718 struct osd_thread_info *oifb_info;
1719 struct osd_device *oifb_dev;
1720 struct dentry *oifb_dentry;
1723 static inline struct dentry *
1724 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1726 struct dentry *dentry;
1728 dentry = ll_lookup_one_len(name, parent, namelen);
1729 if (IS_ERR(dentry)) {
1730 int rc = PTR_ERR(dentry);
1733 CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1734 namelen, name, parent->d_name.len,
1735 parent->d_name.name, parent->d_inode->i_ino,
1736 parent->d_inode->i_generation, rc);
1741 if (dentry->d_inode == NULL) {
1743 return ERR_PTR(-ENOENT);
1750 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1751 scandir_t scandir, filldir_t filldir)
1753 struct osd_ios_item *item;
1756 OBD_ALLOC_PTR(item);
1760 INIT_LIST_HEAD(&item->oii_list);
1761 item->oii_dentry = dget(dentry);
1762 item->oii_scandir = scandir;
1763 item->oii_filldir = filldir;
1764 list_add_tail(&item->oii_list, &dev->od_ios_list);
1770 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1772 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1773 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1774 * reference the inode, or fixed if it is missing or references another inode.
1777 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1778 struct inode *inode, const struct lu_fid *fid, int flags)
1780 struct lustre_mdt_attrs *lma = &info->oti_ost_attrs.loa_lma;
1781 struct osd_inode_id *id = &info->oti_id;
1782 struct osd_inode_id *id2 = &info->oti_id2;
1783 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
1784 struct scrub_file *sf = &scrub->os_file;
1789 rc = osd_get_lma(info, inode, &info->oti_obj_dentry,
1790 &info->oti_ost_attrs);
1791 if (rc != 0 && rc != -ENODATA) {
1792 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1793 "rc = %d\n", osd_name(dev), rc);
1798 osd_id_gen(id, inode->i_ino, inode->i_generation);
1799 if (rc == -ENODATA) {
1800 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1801 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1804 if (flags & OLF_IDX_IN_FID) {
1805 LASSERT(dev->od_index >= 0);
1807 tfid.f_oid = dev->od_index;
1810 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1812 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1813 "scrub: rc = %d\n", osd_name(dev), rc);
1818 if (lma->lma_compat & LMAC_NOT_IN_OI)
1821 tfid = lma->lma_self_fid;
1824 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1829 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1830 DTO_INDEX_INSERT, true, 0, NULL);
1837 if (osd_id_eq_strict(id, id2))
1840 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1841 scrub_file_reset(scrub, LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1843 rc = scrub_file_store(info->oti_env, scrub);
1848 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1849 DTO_INDEX_UPDATE, true, 0, NULL);
1857 * It scans the /lost+found, and for the OST-object (with filter_fid
1858 * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1860 #ifdef HAVE_FILLDIR_USE_CTX
1861 static int osd_ios_lf_fill(struct dir_context *buf,
1863 static int osd_ios_lf_fill(void *buf,
1865 const char *name, int namelen,
1866 loff_t offset, __u64 ino, unsigned d_type)
1868 struct osd_ios_filldir_buf *fill_buf =
1869 (struct osd_ios_filldir_buf *)buf;
1870 struct osd_thread_info *info = fill_buf->oifb_info;
1871 struct osd_device *dev = fill_buf->oifb_dev;
1872 struct lu_fid *fid = &info->oti_fid;
1873 struct osd_scrub *scrub = &dev->od_scrub;
1874 struct dentry *parent = fill_buf->oifb_dentry;
1875 struct dentry *child;
1876 struct inode *dir = parent->d_inode;
1877 struct inode *inode;
1881 /* skip any '.' started names */
1885 scrub->os_lf_scanned++;
1886 child = osd_ios_lookup_one_len(name, parent, namelen);
1887 if (IS_ERR(child)) {
1888 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1889 osd_name(dev), namelen, name, (int)PTR_ERR(child));
1893 inode = child->d_inode;
1894 if (S_ISDIR(inode->i_mode)) {
1895 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1898 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1899 "rc = %d\n", osd_name(dev), namelen, name, rc);
1903 if (!S_ISREG(inode->i_mode))
1906 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1907 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1908 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1910 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1911 "/lost+found.\n", namelen, name, PFID(fid));
1912 scrub->os_lf_repaired++;
1914 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1916 osd_name(dev), namelen, name, PFID(fid), rc);
1920 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1921 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1922 * can process them in furtuer. */
1928 scrub->os_lf_failed++;
1930 /* skip the failure to make the scanning to continue. */
1934 #ifdef HAVE_FILLDIR_USE_CTX
1935 static int osd_ios_varfid_fill(struct dir_context *buf,
1937 static int osd_ios_varfid_fill(void *buf,
1939 const char *name, int namelen,
1940 loff_t offset, __u64 ino, unsigned d_type)
1942 struct osd_ios_filldir_buf *fill_buf =
1943 (struct osd_ios_filldir_buf *)buf;
1944 struct osd_device *dev = fill_buf->oifb_dev;
1945 struct dentry *child;
1949 /* skip any '.' started names */
1953 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1955 RETURN(PTR_ERR(child));
1957 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1959 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1960 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1961 osd_ios_varfid_fill);
1967 #ifdef HAVE_FILLDIR_USE_CTX
1968 static int osd_ios_dl_fill(struct dir_context *buf,
1970 static int osd_ios_dl_fill(void *buf,
1972 const char *name, int namelen,
1973 loff_t offset, __u64 ino, unsigned d_type)
1975 struct osd_ios_filldir_buf *fill_buf =
1976 (struct osd_ios_filldir_buf *)buf;
1977 struct osd_device *dev = fill_buf->oifb_dev;
1978 const struct osd_lf_map *map;
1979 struct dentry *child;
1983 /* skip any '.' started names */
1987 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1988 if (map->olm_namelen != namelen)
1991 if (strncmp(map->olm_name, name, namelen) == 0)
1995 if (map->olm_name == NULL)
1998 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2000 RETURN(PTR_ERR(child));
2002 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2003 &map->olm_fid, map->olm_flags);
2009 #ifdef HAVE_FILLDIR_USE_CTX
2010 static int osd_ios_uld_fill(struct dir_context *buf,
2012 static int osd_ios_uld_fill(void *buf,
2014 const char *name, int namelen,
2015 loff_t offset, __u64 ino, unsigned d_type)
2017 struct osd_ios_filldir_buf *fill_buf =
2018 (struct osd_ios_filldir_buf *)buf;
2019 struct dentry *child;
2024 /* skip any non-DFID format name */
2028 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2030 RETURN(PTR_ERR(child));
2032 /* skip the start '[' */
2033 sscanf(&name[1], SFID, RFID(&tfid));
2034 if (fid_is_sane(&tfid))
2035 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2036 child->d_inode, &tfid, 0);
2044 #ifdef HAVE_FILLDIR_USE_CTX
2045 static int osd_ios_root_fill(struct dir_context *buf,
2047 static int osd_ios_root_fill(void *buf,
2049 const char *name, int namelen,
2050 loff_t offset, __u64 ino, unsigned d_type)
2052 struct osd_ios_filldir_buf *fill_buf =
2053 (struct osd_ios_filldir_buf *)buf;
2054 struct osd_device *dev = fill_buf->oifb_dev;
2055 const struct osd_lf_map *map;
2056 struct dentry *child;
2060 /* skip any '.' started names */
2064 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2065 if (map->olm_namelen != namelen)
2068 if (strncmp(map->olm_name, name, namelen) == 0)
2072 if (map->olm_name == NULL)
2075 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2077 RETURN(PTR_ERR(child));
2079 if (!(map->olm_flags & OLF_NO_OI))
2080 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2081 &map->olm_fid, map->olm_flags);
2082 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2083 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2091 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2092 struct dentry *dentry, filldir_t filldir)
2094 struct osd_ios_filldir_buf buf = {
2095 #ifdef HAVE_DIR_CONTEXT
2096 .ctx.actor = filldir,
2100 .oifb_dentry = dentry };
2101 struct file *filp = &info->oti_file;
2102 struct inode *inode = dentry->d_inode;
2103 const struct file_operations *fops = inode->i_fop;
2107 LASSERT(filldir != NULL);
2110 filp->f_path.dentry = dentry;
2111 filp->f_mode = FMODE_64BITHASH;
2112 filp->f_mapping = inode->i_mapping;
2114 filp->private_data = NULL;
2115 set_file_inode(filp, inode);
2117 #ifdef HAVE_DIR_CONTEXT
2118 buf.ctx.pos = filp->f_pos;
2119 rc = fops->iterate(filp, &buf.ctx);
2120 filp->f_pos = buf.ctx.pos;
2122 rc = fops->readdir(filp, &buf, filldir);
2124 fops->release(inode, filp);
2130 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2131 struct dentry *dentry, filldir_t filldir)
2133 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2134 struct scrub_file *sf = &scrub->os_file;
2135 struct dentry *child;
2139 /* It is existing MDT0 device. We only allow the case of object without
2140 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2141 * can generate IGIF mode FID for the object and related OI mapping. If
2142 * it is on other MDTs, then becuase file-level backup/restore, related
2143 * OI mapping may be invalid already, we do not know which is the right
2144 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2146 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2147 * "fid" xattr, then something crashed. We cannot re-generate the
2148 * FID directly, instead, the OI scrub will scan the OI structure
2149 * and try to re-generate the LMA from the OI mapping. But if the
2150 * OI mapping crashed or lost also, then we have to give up under
2151 * double failure cases. */
2152 scrub->os_convert_igif = 1;
2153 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2154 strlen(dot_lustre_name));
2155 if (IS_ERR(child)) {
2156 rc = PTR_ERR(child);
2157 if (rc == -ENOENT) {
2158 /* It is 1.8 MDT device. */
2159 if (!(sf->sf_flags & SF_UPGRADE)) {
2160 scrub_file_reset(scrub,
2161 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2163 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2164 rc = scrub_file_store(info->oti_env, scrub);
2170 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2171 * so the client will get IGIF for the ".lustre" object when
2174 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2175 * it does not know whether there are some old clients cached
2176 * the ".lustre" IGIF during the upgrading. Two choices:
2178 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2179 * It will allow the old connected clients to access the
2180 * ".lustre" with cached IGIF. But it will cause others
2181 * on the MDT failed to check "fid_is_dot_lustre()".
2183 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2184 * for ".lustre" in spite of whether there are some clients
2185 * cached the ".lustre" IGIF or not. It enables the check
2186 * "fid_is_dot_lustre()" on the MDT, although it will cause
2187 * that the old connected clients cannot access the ".lustre"
2188 * with the cached IGIF.
2190 * Usually, it is rare case for the old connected clients
2191 * to access the ".lustre" with cached IGIF. So we prefer
2192 * to the solution 2). */
2193 rc = osd_ios_scan_one(info, dev, child->d_inode,
2194 &LU_DOT_LUSTRE_FID, 0);
2196 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2205 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2206 struct dentry *dentry, filldir_t filldir)
2208 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2209 struct scrub_file *sf = &scrub->os_file;
2210 struct dentry *child;
2214 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2215 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2216 rc = scrub_file_store(info->oti_env, scrub);
2221 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2222 if (!IS_ERR(child)) {
2223 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2226 rc = PTR_ERR(child);
2229 if (rc != 0 && rc != -ENOENT)
2232 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2233 if (!IS_ERR(child)) {
2234 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2237 rc = PTR_ERR(child);
2246 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2247 struct osd_device *dev)
2249 struct osd_ios_item *item = NULL;
2250 scandir_t scandir = osd_ios_general_scan;
2251 filldir_t filldir = osd_ios_root_fill;
2252 struct dentry *dentry = osd_sb(dev)->s_root;
2253 const struct osd_lf_map *map = osd_lf_maps;
2257 /* Lookup IGIF in OI by force for initial OI scrub. */
2258 dev->od_igif_inoi = 1;
2261 rc = scandir(info, dev, dentry, filldir);
2263 dput(item->oii_dentry);
2270 if (list_empty(&dev->od_ios_list))
2273 item = list_entry(dev->od_ios_list.next,
2274 struct osd_ios_item, oii_list);
2275 list_del_init(&item->oii_list);
2277 LASSERT(item->oii_scandir != NULL);
2278 scandir = item->oii_scandir;
2279 filldir = item->oii_filldir;
2280 dentry = item->oii_dentry;
2283 while (!list_empty(&dev->od_ios_list)) {
2284 item = list_entry(dev->od_ios_list.next,
2285 struct osd_ios_item, oii_list);
2286 list_del_init(&item->oii_list);
2287 dput(item->oii_dentry);
2294 /* There maybe the case that the object has been removed, but its OI
2295 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2296 * file-level backup/restore. So here cleanup the stale OI mappings. */
2297 while (map->olm_name != NULL) {
2298 struct dentry *child;
2300 if (fid_is_zero(&map->olm_fid)) {
2305 child = osd_ios_lookup_one_len(map->olm_name,
2306 osd_sb(dev)->s_root,
2310 else if (PTR_ERR(child) == -ENOENT)
2311 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2312 NULL, DTO_INDEX_DELETE,
2320 char *osd_lf_fid2name(const struct lu_fid *fid)
2322 const struct osd_lf_map *map = osd_lf_maps;
2324 while (map->olm_name != NULL) {
2325 if (!lu_fid_eq(fid, &map->olm_fid)) {
2330 if (map->olm_flags & OLF_SHOW_NAME)
2331 return map->olm_name;
2339 /* OI scrub start/stop */
2341 int osd_scrub_start(const struct lu_env *env, struct osd_device *dev,
2344 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2348 if (dev->od_dt_dev.dd_rdonly)
2351 /* od_otable_mutex: prevent curcurrent start/stop */
2352 mutex_lock(&dev->od_otable_mutex);
2353 rc = scrub_start(osd_scrub_main, scrub, dev, flags);
2354 if (rc == -EALREADY) {
2356 if ((scrub->os_file.sf_flags & SF_AUTO ||
2357 scrub->os_partial_scan) &&
2358 !(flags & SS_AUTO_PARTIAL))
2359 osd_scrub_join(env, dev, flags, false);
2361 mutex_unlock(&dev->od_otable_mutex);
2366 static void osd_scrub_stop(struct osd_device *dev)
2368 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2370 /* od_otable_mutex: prevent curcurrent start/stop */
2371 mutex_lock(&dev->od_otable_mutex);
2372 scrub->os_paused = 1;
2374 mutex_unlock(&dev->od_otable_mutex);
2377 /* OI scrub setup/cleanup */
2379 static const char osd_scrub_name[] = "OI_scrub";
2381 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2383 struct osd_thread_info *info = osd_oti_get(env);
2384 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2385 struct lvfs_run_ctxt *ctxt = &dev->od_scrub.os_ctxt;
2386 struct scrub_file *sf = &scrub->os_file;
2387 struct super_block *sb = osd_sb(dev);
2388 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2389 struct lvfs_run_ctxt saved;
2391 struct inode *inode;
2392 struct lu_fid *fid = &info->oti_fid;
2393 struct lu_object_conf conf;
2394 struct dt_object *obj;
2396 bool restored = false;
2400 memset(&dev->od_scrub, 0, sizeof(struct osd_scrub));
2401 OBD_SET_CTXT_MAGIC(ctxt);
2402 ctxt->pwdmnt = dev->od_mnt;
2403 ctxt->pwd = dev->od_mnt->mnt_root;
2404 ctxt->fs = get_ds();
2406 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2407 init_rwsem(&scrub->os_rwsem);
2408 spin_lock_init(&scrub->os_lock);
2409 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2410 scrub->os_name = osd_name(dev);
2412 push_ctxt(&saved, ctxt);
2413 filp = filp_open(osd_scrub_name, O_RDWR |
2414 (dev->od_dt_dev.dd_rdonly ? 0 : O_CREAT), 0644);
2416 pop_ctxt(&saved, ctxt);
2417 RETURN(PTR_ERR(filp));
2420 inode = file_inode(filp);
2421 if (!dev->od_dt_dev.dd_rdonly) {
2422 /* 'What the @fid is' is not imporatant, because the object
2423 * has no OI mapping, and only is visible inside the OSD.*/
2424 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2425 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2427 filp_close(filp, NULL);
2428 pop_ctxt(&saved, ctxt);
2434 filp_close(filp, NULL);
2435 pop_ctxt(&saved, ctxt);
2437 conf.loc_flags = LOC_F_NEW;
2438 obj = lu2dt(lu_object_find_slice(env, osd2lu_dev(dev), fid, &conf));
2439 if (IS_ERR_OR_NULL(obj)) {
2441 RETURN(obj == NULL ? -ENOENT : PTR_ERR(obj));
2444 osd_dt_obj(obj)->oo_inode = inode;
2445 scrub->os_obj = obj;
2446 rc = scrub_file_load(env, scrub);
2447 if (rc == -ENOENT || rc == -EFAULT) {
2448 scrub_file_init(scrub, es->s_uuid);
2449 /* If the "/O" dir does not exist when mount (indicated by
2450 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2451 * then it is quite probably that the device is a new one,
2452 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2454 * For the rare case that "/O" and "OI_scrub" both lost on
2455 * an old device, it can be found and cleared later.
2457 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2458 * need to check "filter_fid_old" and to convert it to
2459 * "filter_fid" for each object, and all the IGIF should
2460 * have their FID mapping in OI files already. */
2461 if (dev->od_maybe_new && rc == -ENOENT)
2462 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2464 } else if (rc < 0) {
2465 GOTO(cleanup_obj, rc);
2467 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2468 struct obd_uuid *old_uuid;
2469 struct obd_uuid *new_uuid;
2471 OBD_ALLOC_PTR(old_uuid);
2472 OBD_ALLOC_PTR(new_uuid);
2473 if (old_uuid == NULL || new_uuid == NULL) {
2474 CERROR("%s: UUID has been changed, but"
2475 "failed to allocate RAM for report\n",
2478 class_uuid_unparse(sf->sf_uuid, old_uuid);
2479 class_uuid_unparse(es->s_uuid, new_uuid);
2480 CDEBUG(D_LFSCK, "%s: UUID has been changed "
2481 "from %s to %s\n", osd_dev2name(dev),
2482 old_uuid->uuid, new_uuid->uuid);
2484 scrub_file_reset(scrub, es->s_uuid, SF_INCONSISTENT);
2487 if (old_uuid != NULL)
2488 OBD_FREE_PTR(old_uuid);
2489 if (new_uuid != NULL)
2490 OBD_FREE_PTR(new_uuid);
2491 } else if (sf->sf_status == SS_SCANNING) {
2492 sf->sf_status = SS_CRASHED;
2496 if ((sf->sf_oi_count & (sf->sf_oi_count - 1)) != 0) {
2497 LCONSOLE_WARN("%s: invalid oi count %d, set it to %d\n",
2498 osd_dev2name(dev), sf->sf_oi_count,
2500 sf->sf_oi_count = osd_oi_count;
2505 if (sf->sf_pos_last_checkpoint != 0)
2506 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2508 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2511 rc = scrub_file_store(env, scrub);
2513 GOTO(cleanup_obj, rc);
2516 /* Initialize OI files. */
2517 rc = osd_oi_init(info, dev, restored);
2519 GOTO(cleanup_obj, rc);
2521 if (!dev->od_dt_dev.dd_rdonly) {
2522 rc = osd_initial_OI_scrub(info, dev);
2524 GOTO(cleanup_oi, rc);
2527 if (sf->sf_flags & SF_UPGRADE ||
2528 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2529 sf->sf_success_count > 0)) {
2530 dev->od_igif_inoi = 0;
2531 dev->od_check_ff = dev->od_is_ost;
2533 dev->od_igif_inoi = 1;
2534 dev->od_check_ff = 0;
2537 if (sf->sf_flags & SF_INCONSISTENT)
2538 /* The 'od_igif_inoi' will be set under the
2540 * 1) new created system, or
2541 * 2) restored from file-level backup, or
2542 * 3) the upgrading completed.
2544 * The 'od_igif_inoi' may be cleared by OI scrub
2545 * later if found that the system is upgrading. */
2546 dev->od_igif_inoi = 1;
2548 if (!dev->od_dt_dev.dd_rdonly &&
2549 dev->od_auto_scrub_interval != AS_NEVER &&
2550 ((sf->sf_status == SS_PAUSED) ||
2551 (sf->sf_status == SS_CRASHED &&
2552 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2553 SF_UPGRADE | SF_AUTO)) ||
2554 (sf->sf_status == SS_INIT &&
2555 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2557 rc = osd_scrub_start(env, dev, SS_AUTO_FULL);
2560 GOTO(cleanup_oi, rc);
2562 /* it is possible that dcache entries may keep objects after they are
2563 * deleted by OSD. While it looks safe this can cause object data to
2564 * stay until umount causing failures in tests calculating free space,
2565 * e.g. replay-ost-single. Since those dcache entries are not used
2566 * anymore let's just free them after use here */
2567 shrink_dcache_sb(sb);
2571 osd_oi_fini(info, dev);
2573 dt_object_put_nocache(env, scrub->os_obj);
2574 scrub->os_obj = NULL;
2579 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2581 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2583 LASSERT(dev->od_otable_it == NULL);
2585 if (scrub->os_obj != NULL) {
2586 osd_scrub_stop(dev);
2587 dt_object_put_nocache(env, scrub->os_obj);
2588 scrub->os_obj = NULL;
2590 if (dev->od_oi_table != NULL)
2591 osd_oi_fini(osd_oti_get(env), dev);
2594 /* object table based iteration APIs */
2596 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2597 struct dt_object *dt, __u32 attr)
2599 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2600 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2601 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2602 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2603 struct osd_otable_it *it;
2608 /* od_otable_mutex: prevent curcurrent init/fini */
2609 mutex_lock(&dev->od_otable_mutex);
2610 if (dev->od_otable_it != NULL)
2611 GOTO(out, it = ERR_PTR(-EALREADY));
2615 GOTO(out, it = ERR_PTR(-ENOMEM));
2617 dev->od_otable_it = it;
2619 it->ooi_cache.ooc_consumer_idx = -1;
2620 if (flags & DOIF_OUTUSED)
2621 it->ooi_used_outside = 1;
2623 if (flags & DOIF_RESET)
2626 if (valid & DOIV_ERROR_HANDLE) {
2627 if (flags & DOIF_FAILOUT)
2628 start |= SS_SET_FAILOUT;
2630 start |= SS_CLEAR_FAILOUT;
2633 if (valid & DOIV_DRYRUN) {
2634 if (flags & DOIF_DRYRUN)
2635 start |= SS_SET_DRYRUN;
2637 start |= SS_CLEAR_DRYRUN;
2640 rc = scrub_start(osd_scrub_main, scrub, dev, start & ~SS_AUTO_PARTIAL);
2641 if (rc == -EALREADY) {
2642 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2643 } else if (rc < 0) {
2644 dev->od_otable_it = NULL;
2648 /* We have to start from the begining. */
2649 it->ooi_cache.ooc_pos_preload =
2650 LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2656 mutex_unlock(&dev->od_otable_mutex);
2657 return (struct dt_it *)it;
2660 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2662 struct osd_otable_it *it = (struct osd_otable_it *)di;
2663 struct osd_device *dev = it->ooi_dev;
2665 /* od_otable_mutex: prevent curcurrent init/fini */
2666 mutex_lock(&dev->od_otable_mutex);
2667 scrub_stop(&dev->od_scrub.os_scrub);
2668 LASSERT(dev->od_otable_it == it);
2670 dev->od_otable_it = NULL;
2671 mutex_unlock(&dev->od_otable_mutex);
2675 static int osd_otable_it_get(const struct lu_env *env,
2676 struct dt_it *di, const struct dt_key *key)
2681 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2686 osd_otable_it_wakeup(struct lustre_scrub *scrub, struct osd_otable_it *it)
2688 spin_lock(&scrub->os_lock);
2689 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2690 scrub->os_waiting ||
2691 !thread_is_running(&scrub->os_thread))
2692 it->ooi_waiting = 0;
2694 it->ooi_waiting = 1;
2695 spin_unlock(&scrub->os_lock);
2697 return !it->ooi_waiting;
2700 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2702 struct osd_otable_it *it = (struct osd_otable_it *)di;
2703 struct osd_device *dev = it->ooi_dev;
2704 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2705 struct osd_otable_cache *ooc = &it->ooi_cache;
2706 struct ptlrpc_thread *thread = &scrub->os_thread;
2707 struct l_wait_info lwi = { 0 };
2711 LASSERT(it->ooi_user_ready);
2714 if (!thread_is_running(thread) && !it->ooi_used_outside)
2717 if (ooc->ooc_cached_items > 0) {
2718 ooc->ooc_cached_items--;
2719 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2720 ~OSD_OTABLE_IT_CACHE_MASK;
2724 if (it->ooi_all_cached) {
2725 l_wait_event(thread->t_ctl_waitq,
2726 !thread_is_running(thread),
2731 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2732 spin_lock(&scrub->os_lock);
2733 scrub->os_waiting = 0;
2734 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2735 spin_unlock(&scrub->os_lock);
2738 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2739 l_wait_event(thread->t_ctl_waitq,
2740 osd_otable_it_wakeup(scrub, it),
2743 if (!thread_is_running(thread) && !it->ooi_used_outside)
2746 rc = osd_otable_it_preload(env, it);
2753 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2754 const struct dt_it *di)
2759 static int osd_otable_it_key_size(const struct lu_env *env,
2760 const struct dt_it *di)
2762 return sizeof(__u64);
2765 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2766 struct dt_rec *rec, __u32 attr)
2768 struct osd_otable_it *it = (struct osd_otable_it *)di;
2769 struct osd_otable_cache *ooc = &it->ooi_cache;
2771 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2773 /* Filter out Invald FID already. */
2774 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2775 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2776 PFID((struct lu_fid *)rec),
2777 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2782 static __u64 osd_otable_it_store(const struct lu_env *env,
2783 const struct dt_it *di)
2785 struct osd_otable_it *it = (struct osd_otable_it *)di;
2786 struct osd_otable_cache *ooc = &it->ooi_cache;
2789 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2790 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2792 hash = ooc->ooc_pos_preload;
2797 * Set the OSD layer iteration start position as the specified hash.
2799 static int osd_otable_it_load(const struct lu_env *env,
2800 const struct dt_it *di, __u64 hash)
2802 struct osd_otable_it *it = (struct osd_otable_it *)di;
2803 struct osd_device *dev = it->ooi_dev;
2804 struct osd_otable_cache *ooc = &it->ooi_cache;
2805 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2806 struct osd_iit_param *param = &it->ooi_iit_param;
2810 /* Forbid to set iteration position after iteration started. */
2811 if (it->ooi_user_ready)
2814 LASSERT(!scrub->os_partial_scan);
2816 if (hash > OSD_OTABLE_MAX_HASH)
2817 hash = OSD_OTABLE_MAX_HASH;
2819 /* The hash is the last checkpoint position,
2820 * we will start from the next one. */
2821 ooc->ooc_pos_preload = hash + 1;
2822 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2823 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2825 it->ooi_user_ready = 1;
2826 if (!scrub->os_full_speed)
2827 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2829 memset(param, 0, sizeof(*param));
2830 param->sb = osd_sb(dev);
2831 param->start = ooc->ooc_pos_preload;
2832 param->bg = (ooc->ooc_pos_preload - 1) /
2833 LDISKFS_INODES_PER_GROUP(param->sb);
2834 param->offset = (ooc->ooc_pos_preload - 1) %
2835 LDISKFS_INODES_PER_GROUP(param->sb);
2836 param->gbase = 1 + param->bg * LDISKFS_INODES_PER_GROUP(param->sb);
2838 /* Unplug OSD layer iteration by the first next() call. */
2839 rc = osd_otable_it_next(env, (struct dt_it *)it);
2844 static int osd_otable_it_key_rec(const struct lu_env *env,
2845 const struct dt_it *di, void *key_rec)
2850 const struct dt_index_operations osd_otable_ops = {
2852 .init = osd_otable_it_init,
2853 .fini = osd_otable_it_fini,
2854 .get = osd_otable_it_get,
2855 .put = osd_otable_it_put,
2856 .next = osd_otable_it_next,
2857 .key = osd_otable_it_key,
2858 .key_size = osd_otable_it_key_size,
2859 .rec = osd_otable_it_rec,
2860 .store = osd_otable_it_store,
2861 .load = osd_otable_it_load,
2862 .key_rec = osd_otable_it_key_rec,
2866 /* high priority inconsistent items list APIs */
2868 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
2870 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2873 struct osd_inconsistent_item *oii;
2874 struct osd_scrub *oscrub = &dev->od_scrub;
2875 struct lustre_scrub *lscrub = &oscrub->os_scrub;
2876 struct ptlrpc_thread *thread = &lscrub->os_thread;
2881 if (unlikely(oii == NULL))
2884 INIT_LIST_HEAD(&oii->oii_list);
2885 oii->oii_cache = *oic;
2886 oii->oii_insert = insert;
2888 if (lscrub->os_partial_scan) {
2889 __u64 now = ktime_get_real_seconds();
2891 /* If there haven't been errors in a long time,
2892 * decay old count until either the errors are
2893 * gone or we reach the current interval. */
2894 while (unlikely(oscrub->os_bad_oimap_count > 0 &&
2895 oscrub->os_bad_oimap_time +
2896 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2897 oscrub->os_bad_oimap_count >>= 1;
2898 oscrub->os_bad_oimap_time +=
2899 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2902 oscrub->os_bad_oimap_time = now;
2903 if (++oscrub->os_bad_oimap_count >
2904 dev->od_full_scrub_threshold_rate)
2905 lscrub->os_full_scrub = 1;
2908 spin_lock(&lscrub->os_lock);
2909 if (unlikely(!thread_is_running(thread))) {
2910 spin_unlock(&lscrub->os_lock);
2915 if (list_empty(&lscrub->os_inconsistent_items))
2917 list_add_tail(&oii->oii_list, &lscrub->os_inconsistent_items);
2918 spin_unlock(&lscrub->os_lock);
2921 wake_up_all(&thread->t_ctl_waitq);
2926 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2927 struct osd_inode_id *id)
2929 struct lustre_scrub *scrub = &dev->od_scrub.os_scrub;
2930 struct osd_inconsistent_item *oii;
2933 spin_lock(&scrub->os_lock);
2934 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2935 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2936 *id = oii->oii_cache.oic_lid;
2937 spin_unlock(&scrub->os_lock);
2941 spin_unlock(&scrub->os_lock);
2946 void osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2948 struct osd_scrub *scrub = &dev->od_scrub;
2950 scrub_dump(m, &scrub->os_scrub);
2951 seq_printf(m, "lf_scanned: %llu\n"
2953 "lf_failed: %llu\n",
2954 scrub->os_lf_scanned,
2955 scrub->os_scrub.os_file.sf_param & SP_DRYRUN ?
2956 "inconsistent" : "repaired",
2957 scrub->os_lf_repaired,
2958 scrub->os_lf_failed);