4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2012, 2014, 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 <lustre/lustre_idl.h>
41 #include <lustre_disk.h>
42 #include <dt_object.h>
43 #include <linux/xattr.h>
45 #include "osd_internal.h"
47 #include "osd_scrub.h"
49 #define HALF_SEC msecs_to_jiffies(MSEC_PER_SEC >> 1)
51 #define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL
53 #define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */
54 #define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */
55 #define SCRUB_NEXT_EXIT 3 /* exit all the loops */
56 #define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */
57 #define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */
58 #define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */
59 #define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */
60 #define SCRUB_NEXT_NOLMA 8 /* the inode has no FID-in-LMA */
61 #define SCRUB_NEXT_OSTOBJ 9 /* for OST-object */
62 #define SCRUB_NEXT_OSTOBJ_OLD 10 /* old OST-object, no LMA or no FID-on-OST
67 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
69 return container_of0(scrub, struct osd_device, od_scrub);
72 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
74 return osd_sb(osd_scrub2dev(scrub));
77 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
78 struct osd_otable_cache *ooc)
80 return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
83 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
85 return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
89 * update/insert/delete the specified OI mapping (@fid @id) according to the ops
91 * \retval 1, changed nothing
92 * \retval 0, changed successfully
93 * \retval -ve, on error
95 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
96 struct osd_device *dev,
97 const struct lu_fid *fid,
98 const struct osd_inode_id *id,
100 enum oi_check_flags flags)
106 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
109 /* DTO_INDEX_INSERT is enough for other two ops:
110 * delete/update, but save stack. */
111 th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
112 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
115 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
116 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
117 PFID(fid), id->oii_ino, id->oii_gen, rc);
122 case DTO_INDEX_UPDATE:
123 rc = osd_oi_update(info, dev, fid, id, th, flags);
124 if (unlikely(rc == -ENOENT)) {
125 /* Some unlink thread may removed the OI mapping. */
129 case DTO_INDEX_INSERT:
130 rc = osd_oi_insert(info, dev, fid, id, th, flags);
131 if (unlikely(rc == -EEXIST)) {
133 /* XXX: There are trouble things when adding OI
134 * mapping for IGIF object, which may cause
135 * multiple objects to be mapped to the same
136 * IGIF formatted FID. Consider the following
139 * 1) The MDT is upgrading from 1.8 device.
140 * The OI scrub generates IGIF FID1 for the
141 * OBJ1 and adds the OI mapping.
143 * 2) For some reason, the OI scrub does not
144 * process all the IGIF objects completely.
146 * 3) The MDT is backuped and restored against
149 * 4) When the MDT mounts up, the OI scrub will
150 * try to rebuild the OI files. For some IGIF
151 * object, OBJ2, which was not processed by the
152 * OI scrub before the backup/restore, and the
153 * new generated IGIF formatted FID may be just
154 * the FID1, the same as OBJ1.
156 * Under such case, the OI scrub cannot know how
157 * to generate new FID for the OBJ2.
159 * Currently, we do nothing for that. One possible
160 * solution is to generate new normal FID for the
163 * Anyway, it is rare, only exists in theory. */
166 case DTO_INDEX_DELETE:
167 rc = osd_oi_delete(info, dev, fid, th, flags);
169 /* It is normal that the unlink thread has removed the
170 * OI mapping already. */
175 LASSERTF(0, "Unexpected ops %d\n", ops);
179 ldiskfs_journal_stop(th);
181 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
182 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
183 PFID(fid), id->oii_ino, id->oii_gen, rc);
188 /* OI_scrub file ops */
190 static void osd_scrub_file_to_cpu(struct scrub_file *des,
191 struct scrub_file *src)
193 memcpy(des->sf_uuid, src->sf_uuid, 16);
194 des->sf_flags = le64_to_cpu(src->sf_flags);
195 des->sf_magic = le32_to_cpu(src->sf_magic);
196 des->sf_status = le16_to_cpu(src->sf_status);
197 des->sf_param = le16_to_cpu(src->sf_param);
198 des->sf_time_last_complete =
199 le64_to_cpu(src->sf_time_last_complete);
200 des->sf_time_latest_start =
201 le64_to_cpu(src->sf_time_latest_start);
202 des->sf_time_last_checkpoint =
203 le64_to_cpu(src->sf_time_last_checkpoint);
204 des->sf_pos_latest_start =
205 le64_to_cpu(src->sf_pos_latest_start);
206 des->sf_pos_last_checkpoint =
207 le64_to_cpu(src->sf_pos_last_checkpoint);
208 des->sf_pos_first_inconsistent =
209 le64_to_cpu(src->sf_pos_first_inconsistent);
210 des->sf_items_checked =
211 le64_to_cpu(src->sf_items_checked);
212 des->sf_items_updated =
213 le64_to_cpu(src->sf_items_updated);
214 des->sf_items_failed =
215 le64_to_cpu(src->sf_items_failed);
216 des->sf_items_updated_prior =
217 le64_to_cpu(src->sf_items_updated_prior);
218 des->sf_run_time = le32_to_cpu(src->sf_run_time);
219 des->sf_success_count = le32_to_cpu(src->sf_success_count);
220 des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
221 des->sf_internal_flags = le16_to_cpu(src->sf_internal_flags);
222 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
225 static void osd_scrub_file_to_le(struct scrub_file *des,
226 struct scrub_file *src)
228 memcpy(des->sf_uuid, src->sf_uuid, 16);
229 des->sf_flags = cpu_to_le64(src->sf_flags);
230 des->sf_magic = cpu_to_le32(src->sf_magic);
231 des->sf_status = cpu_to_le16(src->sf_status);
232 des->sf_param = cpu_to_le16(src->sf_param);
233 des->sf_time_last_complete =
234 cpu_to_le64(src->sf_time_last_complete);
235 des->sf_time_latest_start =
236 cpu_to_le64(src->sf_time_latest_start);
237 des->sf_time_last_checkpoint =
238 cpu_to_le64(src->sf_time_last_checkpoint);
239 des->sf_pos_latest_start =
240 cpu_to_le64(src->sf_pos_latest_start);
241 des->sf_pos_last_checkpoint =
242 cpu_to_le64(src->sf_pos_last_checkpoint);
243 des->sf_pos_first_inconsistent =
244 cpu_to_le64(src->sf_pos_first_inconsistent);
245 des->sf_items_checked =
246 cpu_to_le64(src->sf_items_checked);
247 des->sf_items_updated =
248 cpu_to_le64(src->sf_items_updated);
249 des->sf_items_failed =
250 cpu_to_le64(src->sf_items_failed);
251 des->sf_items_updated_prior =
252 cpu_to_le64(src->sf_items_updated_prior);
253 des->sf_run_time = cpu_to_le32(src->sf_run_time);
254 des->sf_success_count = cpu_to_le32(src->sf_success_count);
255 des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
256 des->sf_internal_flags = cpu_to_le16(src->sf_internal_flags);
257 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
260 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
262 struct scrub_file *sf = &scrub->os_file;
264 memset(sf, 0, sizeof(*sf));
265 memcpy(sf->sf_uuid, uuid, 16);
266 sf->sf_magic = SCRUB_MAGIC_V1;
267 sf->sf_status = SS_INIT;
270 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
272 struct scrub_file *sf = &scrub->os_file;
274 CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
275 LPX64", add flags = "LPX64"\n",
276 osd_scrub2name(scrub), sf->sf_flags, flags);
278 memcpy(sf->sf_uuid, uuid, 16);
279 sf->sf_status = SS_INIT;
280 sf->sf_flags |= flags;
281 sf->sf_flags &= ~SF_AUTO;
283 sf->sf_time_latest_start = 0;
284 sf->sf_time_last_checkpoint = 0;
285 sf->sf_pos_latest_start = 0;
286 sf->sf_pos_last_checkpoint = 0;
287 sf->sf_pos_first_inconsistent = 0;
288 sf->sf_items_checked = 0;
289 sf->sf_items_updated = 0;
290 sf->sf_items_failed = 0;
291 if (!scrub->os_in_join)
292 sf->sf_items_updated_prior = 0;
294 sf->sf_items_noscrub = 0;
295 sf->sf_items_igif = 0;
298 static int osd_scrub_file_load(struct osd_scrub *scrub)
301 int len = sizeof(scrub->os_file_disk);
304 rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
306 struct scrub_file *sf = &scrub->os_file;
308 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
309 if (sf->sf_magic != SCRUB_MAGIC_V1) {
310 CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
311 "0x%x != 0x%x\n", osd_scrub2name(scrub),
312 sf->sf_magic, SCRUB_MAGIC_V1);
313 /* Process it as new scrub file. */
318 } else if (rc != 0) {
319 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
320 "expected = %d: rc = %d\n",
321 osd_scrub2name(scrub), len, rc);
325 /* return -ENOENT for empty scrub file case. */
332 int osd_scrub_file_store(struct osd_scrub *scrub)
334 struct osd_device *dev;
337 int len = sizeof(scrub->os_file_disk);
341 dev = container_of0(scrub, struct osd_device, od_scrub);
342 credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
343 osd_dto_credits_noquota[DTO_WRITE_BLOCK];
344 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
347 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
348 "rc = %d\n", osd_scrub2name(scrub), rc);
352 osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
353 rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
355 ldiskfs_journal_stop(jh);
357 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
358 "expected = %d: rc = %d\n",
359 osd_scrub2name(scrub), len, rc);
361 scrub->os_time_last_checkpoint = cfs_time_current();
362 scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
363 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
368 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
369 struct inode *inode, const struct lu_fid *fid)
371 struct filter_fid_old *ff = &info->oti_ff;
372 struct dentry *dentry = &info->oti_obj_dentry;
376 bool removed = false;
380 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
383 /* We want the LMA to fit into the 256-byte OST inode, so operate
385 * 1) read old XATTR_NAME_FID and save the parent FID;
386 * 2) delete the old XATTR_NAME_FID;
387 * 3) make new LMA and add it;
388 * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
390 * Making the LMA to fit into the 256-byte OST inode can save time for
391 * normal osd_check_lma() and for other OI scrub scanning in future.
392 * So it is worth to make some slow conversion here. */
393 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
394 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
397 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
398 DFID": rc = %d\n", osd_name(dev), PFID(fid), rc);
402 /* 1) read old XATTR_NAME_FID and save the parent FID */
403 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
404 if (rc == sizeof(*ff)) {
405 /* 2) delete the old XATTR_NAME_FID */
406 ll_vfs_dq_init(inode);
407 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
412 } else if (unlikely(rc == -ENODATA)) {
414 } else if (rc != sizeof(struct filter_fid)) {
415 GOTO(stop, rc = -EINVAL);
418 /* 3) make new LMA and add it */
419 rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0);
420 if (rc == 0 && reset)
421 size = sizeof(struct filter_fid);
422 else if (rc != 0 && removed)
423 /* If failed, we should try to add the old back. */
424 size = sizeof(struct filter_fid_old);
426 /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
430 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
432 if (rc1 != 0 && rc == 0)
439 ldiskfs_journal_stop(jh);
441 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
442 osd_name(dev), PFID(fid), rc);
447 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
448 struct osd_idmap_cache *oic, int val)
450 struct osd_scrub *scrub = &dev->od_scrub;
451 struct scrub_file *sf = &scrub->os_file;
452 struct lu_fid *fid = &oic->oic_fid;
453 struct osd_inode_id *lid = &oic->oic_lid;
454 struct osd_inode_id *lid2 = &info->oti_id;
455 struct osd_inconsistent_item *oii = NULL;
456 struct inode *inode = NULL;
457 int ops = DTO_INDEX_UPDATE;
460 bool converted = false;
463 down_write(&scrub->os_rwsem);
464 scrub->os_new_checked++;
468 if (scrub->os_in_prior)
469 oii = list_entry(oic, struct osd_inconsistent_item,
472 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
475 if (fid_is_igif(fid))
478 if (val == SCRUB_NEXT_OSTOBJ_OLD) {
479 inode = osd_iget(info, dev, lid);
482 /* Someone removed the inode. */
483 if (rc == -ENOENT || rc == -ESTALE)
488 sf->sf_flags |= SF_UPGRADE;
489 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
490 dev->od_check_ff = 1;
491 rc = osd_scrub_convert_ff(info, dev, inode, fid);
498 if ((val == SCRUB_NEXT_NOLMA) &&
499 (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
502 if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
505 rc = osd_oi_lookup(info, dev, fid, lid2,
506 (val == SCRUB_NEXT_OSTOBJ ||
507 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
509 if (rc != -ENOENT && rc != -ESTALE)
514 inode = osd_iget(info, dev, lid);
517 /* Someone removed the inode. */
518 if (rc == -ENOENT || rc == -ESTALE)
524 if (!scrub->os_partial_scan)
525 scrub->os_full_speed = 1;
527 ops = DTO_INDEX_INSERT;
528 idx = osd_oi_fid2idx(dev, fid);
530 case SCRUB_NEXT_NOLMA:
531 sf->sf_flags |= SF_UPGRADE;
532 if (!(sf->sf_param & SP_DRYRUN)) {
533 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
538 if (!(sf->sf_flags & SF_INCONSISTENT))
539 dev->od_igif_inoi = 0;
541 case SCRUB_NEXT_OSTOBJ:
542 sf->sf_flags |= SF_INCONSISTENT;
543 case SCRUB_NEXT_OSTOBJ_OLD:
546 sf->sf_flags |= SF_RECREATED;
547 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
548 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
551 } else if (osd_id_eq(lid, lid2)) {
553 sf->sf_items_updated++;
557 if (!scrub->os_partial_scan)
558 scrub->os_full_speed = 1;
560 sf->sf_flags |= SF_INCONSISTENT;
562 /* XXX: If the device is restored from file-level backup, then
563 * some IGIFs may have been already in OI files, and some
564 * may be not yet. Means upgrading from 1.8 may be partly
565 * processed, but some clients may hold some immobilized
566 * IGIFs, and use them to access related objects. Under
567 * such case, OSD does not know whether an given IGIF has
568 * been processed or to be processed, and it also cannot
569 * generate local ino#/gen# directly from the immobilized
570 * IGIF because of the backup/restore. Then force OSD to
571 * lookup the given IGIF in OI files, and if no entry,
572 * then ask the client to retry after upgrading completed.
573 * No better choice. */
574 dev->od_igif_inoi = 1;
577 rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
578 (val == SCRUB_NEXT_OSTOBJ ||
579 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
581 if (scrub->os_in_prior)
582 sf->sf_items_updated_prior++;
584 sf->sf_items_updated++;
591 sf->sf_items_failed++;
592 if (sf->sf_pos_first_inconsistent == 0 ||
593 sf->sf_pos_first_inconsistent > lid->oii_ino)
594 sf->sf_pos_first_inconsistent = lid->oii_ino;
599 /* There may be conflict unlink during the OI scrub,
600 * if happend, then remove the new added OI mapping. */
601 if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
602 unlikely(inode->i_nlink == 0))
603 osd_scrub_refresh_mapping(info, dev, fid, lid,
604 DTO_INDEX_DELETE, false,
605 (val == SCRUB_NEXT_OSTOBJ ||
606 val == SCRUB_NEXT_OSTOBJ_OLD) ?
607 OI_KNOWN_ON_OST : 0);
608 up_write(&scrub->os_rwsem);
610 if (inode != NULL && !IS_ERR(inode))
614 LASSERT(!list_empty(&oii->oii_list));
616 spin_lock(&scrub->os_lock);
617 list_del_init(&oii->oii_list);
618 spin_unlock(&scrub->os_lock);
621 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
626 static int osd_scrub_prep(struct osd_device *dev)
628 struct osd_scrub *scrub = &dev->od_scrub;
629 struct ptlrpc_thread *thread = &scrub->os_thread;
630 struct scrub_file *sf = &scrub->os_file;
631 __u32 flags = scrub->os_start_flags;
633 bool drop_dryrun = false;
636 CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
637 osd_scrub2name(scrub), flags);
639 down_write(&scrub->os_rwsem);
640 if (flags & SS_SET_FAILOUT)
641 sf->sf_param |= SP_FAILOUT;
642 else if (flags & SS_CLEAR_FAILOUT)
643 sf->sf_param &= ~SP_FAILOUT;
645 if (flags & SS_SET_DRYRUN) {
646 sf->sf_param |= SP_DRYRUN;
647 } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
648 sf->sf_param &= ~SP_DRYRUN;
652 if (flags & SS_RESET)
653 osd_scrub_file_reset(scrub,
654 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
656 if (flags & SS_AUTO_FULL) {
657 scrub->os_full_speed = 1;
658 scrub->os_partial_scan = 0;
659 sf->sf_flags |= SF_AUTO;
660 } else if (flags & SS_AUTO_PARTIAL) {
661 scrub->os_full_speed = 0;
662 scrub->os_partial_scan = 1;
663 sf->sf_flags |= SF_AUTO;
664 } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
666 scrub->os_full_speed = 1;
667 scrub->os_partial_scan = 0;
669 scrub->os_full_speed = 0;
670 scrub->os_partial_scan = 0;
673 spin_lock(&scrub->os_lock);
674 scrub->os_in_prior = 0;
675 scrub->os_waiting = 0;
676 scrub->os_paused = 0;
677 scrub->os_in_join = 0;
678 scrub->os_full_scrub = 0;
679 spin_unlock(&scrub->os_lock);
680 scrub->os_new_checked = 0;
681 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
682 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
683 else if (sf->sf_pos_last_checkpoint != 0)
684 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
686 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
688 scrub->os_pos_current = sf->sf_pos_latest_start;
689 sf->sf_status = SS_SCANNING;
690 sf->sf_time_latest_start = cfs_time_current_sec();
691 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
692 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
693 rc = osd_scrub_file_store(scrub);
695 spin_lock(&scrub->os_lock);
696 thread_set_flags(thread, SVC_RUNNING);
697 spin_unlock(&scrub->os_lock);
698 wake_up_all(&thread->t_ctl_waitq);
700 up_write(&scrub->os_rwsem);
705 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
707 struct scrub_file *sf = &scrub->os_file;
710 if (likely(cfs_time_before(cfs_time_current(),
711 scrub->os_time_next_checkpoint) ||
712 scrub->os_new_checked == 0))
715 down_write(&scrub->os_rwsem);
716 sf->sf_items_checked += scrub->os_new_checked;
717 scrub->os_new_checked = 0;
718 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
719 sf->sf_time_last_checkpoint = cfs_time_current_sec();
720 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
721 scrub->os_time_last_checkpoint);
722 rc = osd_scrub_file_store(scrub);
723 up_write(&scrub->os_rwsem);
728 static void osd_scrub_post(struct osd_scrub *scrub, int result)
730 struct scrub_file *sf = &scrub->os_file;
733 CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
734 osd_scrub2name(scrub), result);
736 down_write(&scrub->os_rwsem);
737 spin_lock(&scrub->os_lock);
738 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
739 spin_unlock(&scrub->os_lock);
740 if (scrub->os_new_checked > 0) {
741 sf->sf_items_checked += scrub->os_new_checked;
742 scrub->os_new_checked = 0;
743 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
745 sf->sf_time_last_checkpoint = cfs_time_current_sec();
747 struct osd_device *dev =
748 container_of0(scrub, struct osd_device, od_scrub);
750 dev->od_igif_inoi = 1;
751 dev->od_check_ff = 0;
752 sf->sf_status = SS_COMPLETED;
753 if (!(sf->sf_param & SP_DRYRUN)) {
754 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
755 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
756 SF_UPGRADE | SF_AUTO);
758 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
759 sf->sf_success_count++;
760 } else if (result == 0) {
761 if (scrub->os_paused)
762 sf->sf_status = SS_PAUSED;
764 sf->sf_status = SS_STOPPED;
766 sf->sf_status = SS_FAILED;
768 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
769 scrub->os_time_last_checkpoint);
770 result = osd_scrub_file_store(scrub);
771 up_write(&scrub->os_rwsem);
776 /* iteration engine */
778 struct osd_iit_param {
779 struct super_block *sb;
780 struct buffer_head *bitmap;
786 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
787 struct osd_device *dev,
788 struct osd_iit_param *param,
789 struct osd_idmap_cache **oic,
792 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
793 struct osd_device *dev,
794 struct osd_iit_param *param,
795 struct osd_idmap_cache *oic,
796 bool *noslot, int rc);
798 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
800 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
801 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
802 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
803 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
804 return SCRUB_NEXT_BREAK;
806 *pos = param->gbase + param->offset;
812 * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
813 * \retval 0: FID-on-MDT
815 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
816 struct osd_device *dev,
819 /* XXX: The initial OI scrub will scan the top level /O to generate
820 * a small local FLDB according to the <seq>. If the given FID
821 * is in the local FLDB, then it is FID-on-OST; otherwise it's
822 * quite possible for FID-on-MDT. */
824 return SCRUB_NEXT_OSTOBJ_OLD;
829 static int osd_scrub_get_fid(struct osd_thread_info *info,
830 struct osd_device *dev, struct inode *inode,
831 struct lu_fid *fid, bool scrub)
833 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
835 bool has_lma = false;
837 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
840 if (lma->lma_compat & LMAC_NOT_IN_OI ||
841 lma->lma_incompat & LMAI_AGENT)
842 return SCRUB_NEXT_CONTINUE;
844 *fid = lma->lma_self_fid;
848 if (lma->lma_compat & LMAC_FID_ON_OST)
849 return SCRUB_NEXT_OSTOBJ;
851 if (fid_is_idif(fid))
852 return SCRUB_NEXT_OSTOBJ_OLD;
854 /* For local object. */
855 if (fid_is_internal(fid))
858 /* For external visible MDT-object with non-normal FID. */
859 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
862 /* For the object with normal FID, it may be MDT-object,
863 * or may be 2.4 OST-object, need further distinguish.
864 * Fall through to next section. */
867 if (rc == -ENODATA || rc == 0) {
868 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
871 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
872 rc = SCRUB_NEXT_OSTOBJ_OLD;
878 /* It is FID-on-OST, but we do not know how
879 * to generate its FID, ignore it directly. */
880 rc = SCRUB_NEXT_CONTINUE;
882 /* It is 2.4 OST-object. */
883 rc = SCRUB_NEXT_OSTOBJ_OLD;
891 if (dev->od_scrub.os_convert_igif) {
892 lu_igif_build(fid, inode->i_ino,
893 inode->i_generation);
895 rc = SCRUB_NEXT_NOLMA;
899 /* It may be FID-on-OST, or may be FID for
900 * non-MDT0, anyway, we do not know how to
901 * generate its FID, ignore it directly. */
902 rc = SCRUB_NEXT_CONTINUE;
907 /* For OI scrub case only: the object has LMA but has no ff
908 * (or ff crashed). It may be MDT-object, may be OST-object
909 * with crashed ff. The last check is local FLDB. */
910 rc = osd_scrub_check_local_fldb(info, dev, fid);
916 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
917 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
918 struct super_block *sb, bool scrub)
924 /* Not handle the backend root object and agent parent object.
925 * They are neither visible to namespace nor have OI mappings. */
926 if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
927 pos == osd_remote_parent_ino(dev)))
928 RETURN(SCRUB_NEXT_CONTINUE);
930 osd_id_gen(lid, pos, OSD_OII_NOGEN);
931 inode = osd_iget(info, dev, lid);
934 /* The inode may be removed after bitmap searching, or the
935 * file is new created without inode initialized yet. */
936 if (rc == -ENOENT || rc == -ESTALE)
937 RETURN(SCRUB_NEXT_CONTINUE);
939 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
940 "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
946 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
947 /* Only skip it for the first OI scrub accessing. */
948 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
949 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
952 rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
961 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
962 struct osd_iit_param *param,
963 struct osd_idmap_cache **oic, const bool noslot)
965 struct osd_scrub *scrub = &dev->od_scrub;
966 struct ptlrpc_thread *thread = &scrub->os_thread;
968 struct osd_inode_id *lid;
971 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
972 struct l_wait_info lwi;
974 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
975 if (likely(lwi.lwi_timeout > 0))
976 l_wait_event(thread->t_ctl_waitq,
977 !list_empty(&scrub->os_inconsistent_items) ||
978 !thread_is_running(thread),
982 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
983 spin_lock(&scrub->os_lock);
984 thread_set_flags(thread, SVC_STOPPING);
985 spin_unlock(&scrub->os_lock);
986 return SCRUB_NEXT_CRASH;
989 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
990 return SCRUB_NEXT_FATAL;
992 if (unlikely(!thread_is_running(thread)))
993 return SCRUB_NEXT_EXIT;
995 if (!list_empty(&scrub->os_inconsistent_items)) {
996 struct osd_inconsistent_item *oii;
998 oii = list_entry(scrub->os_inconsistent_items.next,
999 struct osd_inconsistent_item, oii_list);
1000 *oic = &oii->oii_cache;
1001 scrub->os_in_prior = 1;
1006 return SCRUB_NEXT_WAIT;
1008 rc = osd_iit_next(param, &scrub->os_pos_current);
1012 *oic = &scrub->os_oic;
1013 fid = &(*oic)->oic_fid;
1014 lid = &(*oic)->oic_lid;
1015 rc = osd_iit_iget(info, dev, fid, lid,
1016 scrub->os_pos_current, param->sb, true);
1020 static int osd_preload_next(struct osd_thread_info *info,
1021 struct osd_device *dev, struct osd_iit_param *param,
1022 struct osd_idmap_cache **oic, const bool noslot)
1024 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1025 struct osd_scrub *scrub;
1026 struct ptlrpc_thread *thread;
1029 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1033 scrub = &dev->od_scrub;
1034 thread = &scrub->os_thread;
1035 if (thread_is_running(thread) &&
1036 ooc->ooc_pos_preload >= scrub->os_pos_current)
1037 return SCRUB_NEXT_EXIT;
1039 rc = osd_iit_iget(info, dev,
1040 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1041 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1042 ooc->ooc_pos_preload, param->sb, false);
1043 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1044 * ignore the failure, so it still need to skip the inode next time. */
1045 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1050 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1052 spin_lock(&scrub->os_lock);
1053 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1054 !list_empty(&scrub->os_inconsistent_items) ||
1055 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1056 scrub->os_waiting = 0;
1058 scrub->os_waiting = 1;
1059 spin_unlock(&scrub->os_lock);
1061 return !scrub->os_waiting;
1064 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1065 struct osd_iit_param *param,
1066 struct osd_idmap_cache *oic, bool *noslot, int rc)
1068 struct l_wait_info lwi = { 0 };
1069 struct osd_scrub *scrub = &dev->od_scrub;
1070 struct scrub_file *sf = &scrub->os_file;
1071 struct ptlrpc_thread *thread = &scrub->os_thread;
1072 struct osd_otable_it *it = dev->od_otable_it;
1073 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
1076 case SCRUB_NEXT_CONTINUE:
1078 case SCRUB_NEXT_WAIT:
1080 case SCRUB_NEXT_NOSCRUB:
1081 down_write(&scrub->os_rwsem);
1082 scrub->os_new_checked++;
1083 sf->sf_items_noscrub++;
1084 up_write(&scrub->os_rwsem);
1088 rc = osd_scrub_check_update(info, dev, oic, rc);
1092 rc = osd_scrub_checkpoint(scrub);
1094 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1095 "rc = %d\n", osd_scrub2name(scrub),
1096 scrub->os_pos_current, rc);
1097 /* Continue, as long as the scrub itself can go ahead. */
1100 if (scrub->os_in_prior) {
1101 scrub->os_in_prior = 0;
1106 scrub->os_pos_current = param->gbase + ++(param->offset);
1109 if (it != NULL && it->ooi_waiting && ooc != NULL &&
1110 ooc->ooc_pos_preload < scrub->os_pos_current) {
1111 spin_lock(&scrub->os_lock);
1112 it->ooi_waiting = 0;
1113 wake_up_all(&thread->t_ctl_waitq);
1114 spin_unlock(&scrub->os_lock);
1117 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1120 if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1126 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1129 if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1136 static int osd_preload_exec(struct osd_thread_info *info,
1137 struct osd_device *dev, struct osd_iit_param *param,
1138 struct osd_idmap_cache *oic, bool *noslot, int rc)
1140 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1143 ooc->ooc_cached_items++;
1144 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1145 ~OSD_OTABLE_IT_CACHE_MASK;
1147 return rc > 0 ? 0 : rc;
1150 #define SCRUB_IT_ALL 1
1151 #define SCRUB_IT_CRASH 2
1153 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1156 struct osd_scrub *scrub = &dev->od_scrub;
1157 struct ptlrpc_thread *thread = &scrub->os_thread;
1158 struct scrub_file *sf = &scrub->os_file;
1162 LASSERT(!(flags & SS_AUTO_PARTIAL));
1164 down_write(&scrub->os_rwsem);
1165 scrub->os_in_join = 1;
1166 if (flags & SS_SET_FAILOUT)
1167 sf->sf_param |= SP_FAILOUT;
1168 else if (flags & SS_CLEAR_FAILOUT)
1169 sf->sf_param &= ~SP_FAILOUT;
1171 if (flags & SS_SET_DRYRUN)
1172 sf->sf_param |= SP_DRYRUN;
1173 else if (flags & SS_CLEAR_DRYRUN)
1174 sf->sf_param &= ~SP_DRYRUN;
1176 if (flags & SS_RESET) {
1177 osd_scrub_file_reset(scrub,
1178 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1179 inconsistent ? SF_INCONSISTENT : 0);
1180 sf->sf_status = SS_SCANNING;
1183 if (flags & SS_AUTO_FULL) {
1184 sf->sf_flags |= SF_AUTO;
1185 scrub->os_full_speed = 1;
1188 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1189 scrub->os_full_speed = 1;
1191 scrub->os_full_speed = 0;
1193 scrub->os_new_checked = 0;
1194 if (sf->sf_pos_last_checkpoint != 0)
1195 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1197 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1199 scrub->os_pos_current = sf->sf_pos_latest_start;
1200 sf->sf_time_latest_start = cfs_time_current_sec();
1201 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1202 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1203 rc = osd_scrub_file_store(scrub);
1205 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1206 "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1208 spin_lock(&scrub->os_lock);
1209 scrub->os_waiting = 0;
1210 scrub->os_paused = 0;
1211 scrub->os_partial_scan = 0;
1212 scrub->os_in_join = 0;
1213 scrub->os_full_scrub = 0;
1214 spin_unlock(&scrub->os_lock);
1215 wake_up_all(&thread->t_ctl_waitq);
1216 up_write(&scrub->os_rwsem);
1221 static int osd_inode_iteration(struct osd_thread_info *info,
1222 struct osd_device *dev, __u32 max, bool preload)
1224 struct osd_scrub *scrub = &dev->od_scrub;
1225 struct ptlrpc_thread *thread = &scrub->os_thread;
1226 struct scrub_file *sf = &scrub->os_file;
1227 osd_iit_next_policy next;
1228 osd_iit_exec_policy exec;
1231 struct osd_iit_param param = { NULL };
1232 struct l_wait_info lwi = { 0 };
1238 param.sb = osd_sb(dev);
1242 while (scrub->os_partial_scan && !scrub->os_in_join) {
1243 struct osd_idmap_cache *oic = NULL;
1245 rc = osd_scrub_next(info, dev, ¶m, &oic, noslot);
1247 case SCRUB_NEXT_EXIT:
1249 case SCRUB_NEXT_CRASH:
1250 RETURN(SCRUB_IT_CRASH);
1251 case SCRUB_NEXT_FATAL:
1253 case SCRUB_NEXT_WAIT: {
1254 struct kstatfs *ksfs = &info->oti_ksfs;
1257 if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1258 unlikely(sf->sf_items_updated_prior == 0))
1261 if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1262 scrub->os_full_scrub) {
1263 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1268 rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1270 __u64 used = ksfs->f_files - ksfs->f_ffree;
1272 do_div(used, sf->sf_items_updated_prior);
1273 /* If we hit too much inconsistent OI
1274 * mappings during the partial scan,
1275 * then scan the device completely. */
1276 if (used < dev->od_full_scrub_ratio) {
1278 SS_AUTO_FULL | SS_RESET, true);
1284 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1288 saved_flags = sf->sf_flags;
1289 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1290 SF_UPGRADE | SF_AUTO);
1291 sf->sf_status = SS_COMPLETED;
1292 l_wait_event(thread->t_ctl_waitq,
1293 !thread_is_running(thread) ||
1294 !scrub->os_partial_scan ||
1295 scrub->os_in_join ||
1296 !list_empty(&scrub->os_inconsistent_items),
1298 sf->sf_flags = saved_flags;
1299 sf->sf_status = SS_SCANNING;
1301 if (unlikely(!thread_is_running(thread)))
1304 if (!scrub->os_partial_scan || scrub->os_in_join)
1310 LASSERTF(rc == 0, "rc = %d\n", rc);
1312 osd_scrub_exec(info, dev, ¶m, oic, &noslot, rc);
1319 l_wait_event(thread->t_ctl_waitq,
1320 !thread_is_running(thread) || !scrub->os_in_join,
1323 if (unlikely(!thread_is_running(thread)))
1329 next = osd_scrub_next;
1330 exec = osd_scrub_exec;
1331 pos = &scrub->os_pos_current;
1332 count = &scrub->os_new_checked;
1334 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1336 next = osd_preload_next;
1337 exec = osd_preload_exec;
1338 pos = &ooc->ooc_pos_preload;
1339 count = &ooc->ooc_cached_items;
1341 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1343 while (*pos <= limit && *count < max) {
1344 struct osd_idmap_cache *oic = NULL;
1345 struct ldiskfs_group_desc *desc;
1347 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1348 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1352 ldiskfs_lock_group(param.sb, param.bg);
1353 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1354 ldiskfs_unlock_group(param.sb, param.bg);
1355 *pos = 1 + (param.bg + 1) *
1356 LDISKFS_INODES_PER_GROUP(param.sb);
1359 ldiskfs_unlock_group(param.sb, param.bg);
1361 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1362 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1363 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1364 if (param.bitmap == NULL) {
1365 CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1366 "scrub will stop, urgent mode\n",
1367 osd_scrub2name(scrub), (__u32)param.bg);
1371 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1374 ldiskfs_itable_unused_count(param.sb, desc) >
1375 LDISKFS_INODES_PER_GROUP(param.sb))
1378 rc = next(info, dev, ¶m, &oic, noslot);
1380 case SCRUB_NEXT_BREAK:
1382 case SCRUB_NEXT_EXIT:
1383 brelse(param.bitmap);
1385 case SCRUB_NEXT_CRASH:
1386 brelse(param.bitmap);
1387 RETURN(SCRUB_IT_CRASH);
1388 case SCRUB_NEXT_FATAL:
1389 brelse(param.bitmap);
1393 rc = exec(info, dev, ¶m, oic, &noslot, rc);
1395 brelse(param.bitmap);
1401 brelse(param.bitmap);
1405 RETURN(SCRUB_IT_ALL);
1409 static int osd_otable_it_preload(const struct lu_env *env,
1410 struct osd_otable_it *it)
1412 struct osd_device *dev = it->ooi_dev;
1413 struct osd_scrub *scrub = &dev->od_scrub;
1414 struct osd_otable_cache *ooc = &it->ooi_cache;
1418 rc = osd_inode_iteration(osd_oti_get(env), dev,
1419 OSD_OTABLE_IT_CACHE_SIZE, true);
1420 if (rc == SCRUB_IT_ALL)
1421 it->ooi_all_cached = 1;
1423 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1424 spin_lock(&scrub->os_lock);
1425 scrub->os_waiting = 0;
1426 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1427 spin_unlock(&scrub->os_lock);
1430 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1433 static int osd_scrub_main(void *args)
1436 struct osd_device *dev = (struct osd_device *)args;
1437 struct osd_scrub *scrub = &dev->od_scrub;
1438 struct ptlrpc_thread *thread = &scrub->os_thread;
1442 rc = lu_env_init(&env, LCT_LOCAL);
1444 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1445 osd_scrub2name(scrub), rc);
1449 rc = osd_scrub_prep(dev);
1451 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1452 osd_scrub2name(scrub), rc);
1456 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1457 struct l_wait_info lwi = { 0 };
1458 struct osd_otable_it *it = dev->od_otable_it;
1459 struct osd_otable_cache *ooc = &it->ooi_cache;
1461 l_wait_event(thread->t_ctl_waitq,
1462 it->ooi_user_ready || !thread_is_running(thread),
1464 if (unlikely(!thread_is_running(thread)))
1467 scrub->os_pos_current = ooc->ooc_pos_preload;
1470 CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1471 osd_scrub2name(scrub), scrub->os_start_flags,
1472 scrub->os_pos_current);
1474 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1475 if (unlikely(rc == SCRUB_IT_CRASH))
1476 GOTO(out, rc = -EINVAL);
1480 osd_scrub_post(scrub, rc);
1481 CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1482 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1485 while (!list_empty(&scrub->os_inconsistent_items)) {
1486 struct osd_inconsistent_item *oii;
1488 oii = list_entry(scrub->os_inconsistent_items.next,
1489 struct osd_inconsistent_item, oii_list);
1490 list_del_init(&oii->oii_list);
1496 spin_lock(&scrub->os_lock);
1497 thread_set_flags(thread, SVC_STOPPED);
1498 wake_up_all(&thread->t_ctl_waitq);
1499 spin_unlock(&scrub->os_lock);
1503 /* initial OI scrub */
1505 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1506 struct dentry *, filldir_t filldir);
1508 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1509 loff_t offset, __u64 ino, unsigned d_type);
1510 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1511 loff_t offset, __u64 ino, unsigned d_type);
1512 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1513 loff_t offset, __u64 ino, unsigned d_type);
1516 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1517 struct dentry *dentry, filldir_t filldir);
1519 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1520 struct dentry *dentry, filldir_t filldir);
1523 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1524 struct dentry *dentry, filldir_t filldir);
1527 OLF_SCAN_SUBITEMS = 0x0001,
1528 OLF_HIDE_FID = 0x0002,
1529 OLF_SHOW_NAME = 0x0004,
1535 struct lu_fid olm_fid;
1538 scandir_t olm_scandir;
1539 filldir_t olm_filldir;
1542 /* Add the new introduced local files in the list in the future. */
1543 static const struct osd_lf_map osd_lf_maps[] = {
1545 { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1546 sizeof(CATLIST) - 1, NULL, NULL },
1549 { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1550 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1551 osd_ios_general_scan, osd_ios_varfid_fill },
1553 /* NIDTBL_VERSIONS */
1554 { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1555 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1556 osd_ios_varfid_fill },
1559 { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1562 { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1563 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1564 osd_ios_ROOT_scan, NULL },
1566 /* changelog_catalog */
1567 { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1570 /* changelog_users */
1571 { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1575 { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1576 sizeof("fld") - 1, NULL, NULL },
1579 { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1580 sizeof(LAST_RCVD) - 1, NULL, NULL },
1583 { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1584 sizeof(LOV_OBJID) - 1, NULL, NULL },
1587 { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1588 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1591 { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1592 osd_ios_general_scan, osd_ios_varfid_fill },
1595 { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1596 osd_ios_general_scan, osd_ios_varfid_fill },
1599 { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1600 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1603 { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1604 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1607 { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1608 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1611 { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1612 osd_ios_general_scan, osd_ios_varfid_fill },
1614 /* lfsck_bookmark */
1615 { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1619 { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1622 /* lfsck_namespace */
1623 { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1626 /* OBJECTS, upgrade from old device */
1627 { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1628 osd_ios_OBJECTS_scan, NULL },
1630 /* lquota_v2.user, upgrade from old device */
1631 { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1634 /* lquota_v2.group, upgrade from old device */
1635 { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1638 /* LAST_GROUP, upgrade from old device */
1639 { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1640 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1642 /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1643 { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1644 OLF_SHOW_NAME, sizeof("SLAVE_LOG") - 1, NULL, NULL },
1647 { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1648 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1649 osd_ios_general_scan, osd_ios_lf_fill },
1651 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1654 /* Add the new introduced files under .lustre/ in the list in the future. */
1655 static const struct osd_lf_map osd_dl_maps[] = {
1657 { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1658 sizeof("fid") - 1, NULL, NULL },
1660 /* .lustre/lost+found */
1661 { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1662 sizeof("lost+found") - 1, NULL, NULL },
1664 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1667 struct osd_ios_item {
1668 struct list_head oii_list;
1669 struct dentry *oii_dentry;
1670 scandir_t oii_scandir;
1671 filldir_t oii_filldir;
1674 struct osd_ios_filldir_buf {
1675 #ifdef HAVE_DIR_CONTEXT
1676 /* please keep it as first member */
1677 struct dir_context ctx;
1679 struct osd_thread_info *oifb_info;
1680 struct osd_device *oifb_dev;
1681 struct dentry *oifb_dentry;
1684 static inline struct dentry *
1685 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1687 struct dentry *dentry;
1689 dentry = ll_lookup_one_len(name, parent, namelen);
1690 if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1692 return ERR_PTR(-ENOENT);
1699 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1700 scandir_t scandir, filldir_t filldir)
1702 struct osd_ios_item *item;
1705 OBD_ALLOC_PTR(item);
1709 INIT_LIST_HEAD(&item->oii_list);
1710 item->oii_dentry = dget(dentry);
1711 item->oii_scandir = scandir;
1712 item->oii_filldir = filldir;
1713 list_add_tail(&item->oii_list, &dev->od_ios_list);
1719 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1721 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1722 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1723 * reference the inode, or fixed if it is missing or references another inode.
1726 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1727 struct inode *inode, const struct lu_fid *fid, int flags)
1729 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1730 struct osd_inode_id *id = &info->oti_id;
1731 struct osd_inode_id *id2 = &info->oti_id2;
1732 struct osd_scrub *scrub = &dev->od_scrub;
1733 struct scrub_file *sf = &scrub->os_file;
1738 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1739 if (rc != 0 && rc != -ENODATA) {
1740 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1741 "rc = %d\n", osd_name(dev), rc);
1746 osd_id_gen(id, inode->i_ino, inode->i_generation);
1747 if (rc == -ENODATA) {
1748 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1749 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1752 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1754 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1755 "scrub: rc = %d\n", osd_name(dev), rc);
1760 if (lma->lma_compat & LMAC_NOT_IN_OI)
1763 tfid = lma->lma_self_fid;
1766 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1771 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1772 DTO_INDEX_INSERT, true, 0);
1779 if (osd_id_eq_strict(id, id2))
1782 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1783 osd_scrub_file_reset(scrub,
1784 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1786 rc = osd_scrub_file_store(scrub);
1791 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1792 DTO_INDEX_UPDATE, true, 0);
1800 * It scans the /lost+found, and for the OST-object (with filter_fid
1801 * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1803 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1804 loff_t offset, __u64 ino, unsigned d_type)
1806 struct osd_ios_filldir_buf *fill_buf = buf;
1807 struct osd_thread_info *info = fill_buf->oifb_info;
1808 struct osd_device *dev = fill_buf->oifb_dev;
1809 struct lu_fid *fid = &info->oti_fid;
1810 struct osd_scrub *scrub = &dev->od_scrub;
1811 struct dentry *parent = fill_buf->oifb_dentry;
1812 struct dentry *child;
1813 struct inode *dir = parent->d_inode;
1814 struct inode *inode;
1818 /* skip any '.' started names */
1822 scrub->os_lf_scanned++;
1823 child = osd_ios_lookup_one_len(name, parent, namelen);
1824 if (IS_ERR(child)) {
1825 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1826 osd_name(dev), namelen, name, (int)PTR_ERR(child));
1830 inode = child->d_inode;
1831 if (S_ISDIR(inode->i_mode)) {
1832 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1835 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1836 "rc = %d\n", osd_name(dev), namelen, name, rc);
1840 if (!S_ISREG(inode->i_mode))
1843 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1844 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1845 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1847 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1848 "/lost+found.\n", namelen, name, PFID(fid));
1849 scrub->os_lf_repaired++;
1851 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1853 osd_name(dev), namelen, name, PFID(fid), rc);
1857 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1858 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1859 * can process them in furtuer. */
1865 scrub->os_lf_failed++;
1867 /* skip the failure to make the scanning to continue. */
1871 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1872 loff_t offset, __u64 ino, unsigned d_type)
1874 struct osd_ios_filldir_buf *fill_buf = buf;
1875 struct osd_device *dev = fill_buf->oifb_dev;
1876 struct dentry *child;
1880 /* skip any '.' started names */
1884 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1886 RETURN(PTR_ERR(child));
1888 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1890 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1891 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1892 osd_ios_varfid_fill);
1898 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1899 loff_t offset, __u64 ino, unsigned d_type)
1901 struct osd_ios_filldir_buf *fill_buf = buf;
1902 struct osd_device *dev = fill_buf->oifb_dev;
1903 const struct osd_lf_map *map;
1904 struct dentry *child;
1908 /* skip any '.' started names */
1912 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1913 if (map->olm_namelen != namelen)
1916 if (strncmp(map->olm_name, name, namelen) == 0)
1920 if (map->olm_name == NULL)
1923 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1925 RETURN(PTR_ERR(child));
1927 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1928 &map->olm_fid, map->olm_flags);
1934 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1935 loff_t offset, __u64 ino, unsigned d_type)
1937 struct osd_ios_filldir_buf *fill_buf = buf;
1938 struct osd_device *dev = fill_buf->oifb_dev;
1939 const struct osd_lf_map *map;
1940 struct dentry *child;
1944 /* skip any '.' started names */
1948 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1949 if (map->olm_namelen != namelen)
1952 if (strncmp(map->olm_name, name, namelen) == 0)
1956 if (map->olm_name == NULL)
1959 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1961 RETURN(PTR_ERR(child));
1963 if (!(map->olm_flags & OLF_NO_OI))
1964 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1965 &map->olm_fid, map->olm_flags);
1966 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1967 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1975 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1976 struct dentry *dentry, filldir_t filldir)
1978 struct osd_ios_filldir_buf buf = {
1979 #ifdef HAVE_DIR_CONTEXT
1980 .ctx.actor = filldir,
1984 .oifb_dentry = dentry };
1985 struct file *filp = &info->oti_file;
1986 struct inode *inode = dentry->d_inode;
1987 const struct file_operations *fops = inode->i_fop;
1991 LASSERT(filldir != NULL);
1994 filp->f_dentry = dentry;
1995 filp->f_mode = FMODE_64BITHASH;
1996 filp->f_mapping = inode->i_mapping;
1998 filp->private_data = NULL;
1999 set_file_inode(filp, inode);
2001 #ifdef HAVE_DIR_CONTEXT
2002 buf.ctx.pos = filp->f_pos;
2003 rc = fops->iterate(filp, &buf.ctx);
2004 filp->f_pos = buf.ctx.pos;
2006 rc = fops->readdir(filp, &buf, filldir);
2008 fops->release(inode, filp);
2014 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2015 struct dentry *dentry, filldir_t filldir)
2017 struct osd_scrub *scrub = &dev->od_scrub;
2018 struct scrub_file *sf = &scrub->os_file;
2019 struct dentry *child;
2023 /* It is existing MDT0 device. We only allow the case of object without
2024 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2025 * can generate IGIF mode FID for the object and related OI mapping. If
2026 * it is on other MDTs, then becuase file-level backup/restore, related
2027 * OI mapping may be invalid already, we do not know which is the right
2028 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2030 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2031 * "fid" xattr, then something crashed. We cannot re-generate the
2032 * FID directly, instead, the OI scrub will scan the OI structure
2033 * and try to re-generate the LMA from the OI mapping. But if the
2034 * OI mapping crashed or lost also, then we have to give up under
2035 * double failure cases. */
2036 scrub->os_convert_igif = 1;
2037 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2038 strlen(dot_lustre_name));
2039 if (IS_ERR(child)) {
2040 rc = PTR_ERR(child);
2041 if (rc == -ENOENT) {
2042 /* It is 1.8 MDT device. */
2043 if (!(sf->sf_flags & SF_UPGRADE)) {
2044 osd_scrub_file_reset(scrub,
2045 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2047 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2048 rc = osd_scrub_file_store(scrub);
2054 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2055 * so the client will get IGIF for the ".lustre" object when
2058 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2059 * it does not know whether there are some old clients cached
2060 * the ".lustre" IGIF during the upgrading. Two choices:
2062 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2063 * It will allow the old connected clients to access the
2064 * ".lustre" with cached IGIF. But it will cause others
2065 * on the MDT failed to check "fid_is_dot_lustre()".
2067 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2068 * for ".lustre" in spite of whether there are some clients
2069 * cached the ".lustre" IGIF or not. It enables the check
2070 * "fid_is_dot_lustre()" on the MDT, although it will cause
2071 * that the old connected clients cannot access the ".lustre"
2072 * with the cached IGIF.
2074 * Usually, it is rare case for the old connected clients
2075 * to access the ".lustre" with cached IGIF. So we prefer
2076 * to the solution 2). */
2077 rc = osd_ios_scan_one(info, dev, child->d_inode,
2078 &LU_DOT_LUSTRE_FID, 0);
2080 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2089 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2090 struct dentry *dentry, filldir_t filldir)
2092 struct osd_scrub *scrub = &dev->od_scrub;
2093 struct scrub_file *sf = &scrub->os_file;
2094 struct dentry *child;
2098 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2099 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2100 rc = osd_scrub_file_store(scrub);
2105 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2106 if (!IS_ERR(child)) {
2107 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2110 rc = PTR_ERR(child);
2113 if (rc != 0 && rc != -ENOENT)
2116 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2117 if (!IS_ERR(child)) {
2118 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2121 rc = PTR_ERR(child);
2130 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2131 struct osd_device *dev)
2133 struct osd_ios_item *item = NULL;
2134 scandir_t scandir = osd_ios_general_scan;
2135 filldir_t filldir = osd_ios_root_fill;
2136 struct dentry *dentry = osd_sb(dev)->s_root;
2137 const struct osd_lf_map *map = osd_lf_maps;
2141 /* Lookup IGIF in OI by force for initial OI scrub. */
2142 dev->od_igif_inoi = 1;
2145 rc = scandir(info, dev, dentry, filldir);
2147 dput(item->oii_dentry);
2154 if (list_empty(&dev->od_ios_list))
2157 item = list_entry(dev->od_ios_list.next,
2158 struct osd_ios_item, oii_list);
2159 list_del_init(&item->oii_list);
2161 LASSERT(item->oii_scandir != NULL);
2162 scandir = item->oii_scandir;
2163 filldir = item->oii_filldir;
2164 dentry = item->oii_dentry;
2167 while (!list_empty(&dev->od_ios_list)) {
2168 item = list_entry(dev->od_ios_list.next,
2169 struct osd_ios_item, oii_list);
2170 list_del_init(&item->oii_list);
2171 dput(item->oii_dentry);
2178 /* There maybe the case that the object has been removed, but its OI
2179 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2180 * file-level backup/restore. So here cleanup the stale OI mappings. */
2181 while (map->olm_name != NULL) {
2182 struct dentry *child;
2184 if (fid_is_zero(&map->olm_fid)) {
2189 child = osd_ios_lookup_one_len(map->olm_name,
2190 osd_sb(dev)->s_root,
2194 else if (PTR_ERR(child) == -ENOENT)
2195 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2196 NULL, DTO_INDEX_DELETE,
2204 char *osd_lf_fid2name(const struct lu_fid *fid)
2206 const struct osd_lf_map *map = osd_lf_maps;
2208 while (map->olm_name != NULL) {
2209 if (!lu_fid_eq(fid, &map->olm_fid)) {
2214 if (map->olm_flags & OLF_SHOW_NAME)
2215 return map->olm_name;
2223 /* OI scrub start/stop */
2225 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2227 struct osd_scrub *scrub = &dev->od_scrub;
2228 struct ptlrpc_thread *thread = &scrub->os_thread;
2229 struct l_wait_info lwi = { 0 };
2230 struct task_struct *task;
2234 /* os_lock: sync status between stop and scrub thread */
2235 spin_lock(&scrub->os_lock);
2238 if (thread_is_running(thread)) {
2239 spin_unlock(&scrub->os_lock);
2240 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2241 (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2244 osd_scrub_join(dev, flags, false);
2245 spin_lock(&scrub->os_lock);
2246 if (!thread_is_running(thread))
2249 spin_unlock(&scrub->os_lock);
2253 if (unlikely(thread_is_stopping(thread))) {
2254 spin_unlock(&scrub->os_lock);
2255 l_wait_event(thread->t_ctl_waitq,
2256 thread_is_stopped(thread),
2258 spin_lock(&scrub->os_lock);
2261 spin_unlock(&scrub->os_lock);
2263 if (scrub->os_file.sf_status == SS_COMPLETED) {
2264 if (!(flags & SS_SET_FAILOUT))
2265 flags |= SS_CLEAR_FAILOUT;
2267 if (!(flags & SS_SET_DRYRUN))
2268 flags |= SS_CLEAR_DRYRUN;
2273 scrub->os_start_flags = flags;
2274 thread_set_flags(thread, 0);
2275 task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2278 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2279 osd_scrub2name(scrub), rc);
2283 l_wait_event(thread->t_ctl_waitq,
2284 thread_is_running(thread) || thread_is_stopped(thread),
2290 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2295 /* od_otable_mutex: prevent curcurrent start/stop */
2296 mutex_lock(&dev->od_otable_mutex);
2297 rc = do_osd_scrub_start(dev, flags);
2298 mutex_unlock(&dev->od_otable_mutex);
2300 RETURN(rc == -EALREADY ? 0 : rc);
2303 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2305 struct ptlrpc_thread *thread = &scrub->os_thread;
2306 struct l_wait_info lwi = { 0 };
2308 /* os_lock: sync status between stop and scrub thread */
2309 spin_lock(&scrub->os_lock);
2310 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2311 thread_set_flags(thread, SVC_STOPPING);
2312 spin_unlock(&scrub->os_lock);
2313 wake_up_all(&thread->t_ctl_waitq);
2314 l_wait_event(thread->t_ctl_waitq,
2315 thread_is_stopped(thread),
2317 /* Do not skip the last lock/unlock, which can guarantee that
2318 * the caller cannot return until the OI scrub thread exit. */
2319 spin_lock(&scrub->os_lock);
2321 spin_unlock(&scrub->os_lock);
2324 static void osd_scrub_stop(struct osd_device *dev)
2326 /* od_otable_mutex: prevent curcurrent start/stop */
2327 mutex_lock(&dev->od_otable_mutex);
2328 dev->od_scrub.os_paused = 1;
2329 do_osd_scrub_stop(&dev->od_scrub);
2330 mutex_unlock(&dev->od_otable_mutex);
2333 /* OI scrub setup/cleanup */
2335 static const char osd_scrub_name[] = "OI_scrub";
2337 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2339 struct osd_thread_info *info = osd_oti_get(env);
2340 struct osd_scrub *scrub = &dev->od_scrub;
2341 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
2342 struct scrub_file *sf = &scrub->os_file;
2343 struct super_block *sb = osd_sb(dev);
2344 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2345 struct lvfs_run_ctxt saved;
2347 struct inode *inode;
2348 struct lu_fid *fid = &info->oti_fid;
2353 memset(scrub, 0, sizeof(*scrub));
2354 OBD_SET_CTXT_MAGIC(ctxt);
2355 ctxt->pwdmnt = dev->od_mnt;
2356 ctxt->pwd = dev->od_mnt->mnt_root;
2357 ctxt->fs = get_ds();
2359 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2360 init_rwsem(&scrub->os_rwsem);
2361 spin_lock_init(&scrub->os_lock);
2362 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2364 push_ctxt(&saved, ctxt);
2365 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2367 pop_ctxt(&saved, ctxt);
2368 RETURN(PTR_ERR(filp));
2371 inode = filp->f_dentry->d_inode;
2372 /* 'What the @fid is' is not imporatant, because the object
2373 * has no OI mapping, and only is visible inside the OSD.*/
2374 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2375 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2377 filp_close(filp, NULL);
2378 pop_ctxt(&saved, ctxt);
2382 scrub->os_inode = igrab(inode);
2383 filp_close(filp, NULL);
2384 pop_ctxt(&saved, ctxt);
2386 rc = osd_scrub_file_load(scrub);
2387 if (rc == -ENOENT) {
2388 osd_scrub_file_init(scrub, es->s_uuid);
2389 /* If the "/O" dir does not exist when mount (indicated by
2390 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2391 * then it is quite probably that the device is a new one,
2392 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2394 * For the rare case that "/O" and "OI_scrub" both lost on
2395 * an old device, it can be found and cleared later.
2397 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2398 * need to check "filter_fid_old" and to convert it to
2399 * "filter_fid" for each object, and all the IGIF should
2400 * have their FID mapping in OI files already. */
2401 if (dev->od_maybe_new)
2402 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2404 } else if (rc != 0) {
2405 GOTO(cleanup_inode, rc);
2407 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2408 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2410 } else if (sf->sf_status == SS_SCANNING) {
2411 sf->sf_status = SS_CRASHED;
2416 if (sf->sf_pos_last_checkpoint != 0)
2417 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2419 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2422 rc = osd_scrub_file_store(scrub);
2424 GOTO(cleanup_inode, rc);
2427 /* Initialize OI files. */
2428 rc = osd_oi_init(info, dev);
2430 GOTO(cleanup_inode, rc);
2432 rc = osd_initial_OI_scrub(info, dev);
2434 GOTO(cleanup_oi, rc);
2436 if (sf->sf_flags & SF_UPGRADE ||
2437 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2438 sf->sf_success_count > 0)) {
2439 dev->od_igif_inoi = 0;
2440 dev->od_check_ff = dev->od_is_ost;
2442 dev->od_igif_inoi = 1;
2443 dev->od_check_ff = 0;
2446 if (sf->sf_flags & SF_INCONSISTENT)
2447 /* The 'od_igif_inoi' will be set under the
2449 * 1) new created system, or
2450 * 2) restored from file-level backup, or
2451 * 3) the upgrading completed.
2453 * The 'od_igif_inoi' may be cleared by OI scrub
2454 * later if found that the system is upgrading. */
2455 dev->od_igif_inoi = 1;
2457 if (!dev->od_noscrub &&
2458 ((sf->sf_status == SS_PAUSED) ||
2459 (sf->sf_status == SS_CRASHED &&
2460 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2461 SF_UPGRADE | SF_AUTO)) ||
2462 (sf->sf_status == SS_INIT &&
2463 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2465 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2468 GOTO(cleanup_oi, rc);
2470 /* it is possible that dcache entries may keep objects after they are
2471 * deleted by OSD. While it looks safe this can cause object data to
2472 * stay until umount causing failures in tests calculating free space,
2473 * e.g. replay-ost-single. Since those dcache entries are not used
2474 * anymore let's just free them after use here */
2475 shrink_dcache_sb(sb);
2479 osd_oi_fini(info, dev);
2481 iput(scrub->os_inode);
2482 scrub->os_inode = NULL;
2487 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2489 struct osd_scrub *scrub = &dev->od_scrub;
2491 LASSERT(dev->od_otable_it == NULL);
2493 if (scrub->os_inode != NULL) {
2494 osd_scrub_stop(dev);
2495 iput(scrub->os_inode);
2496 scrub->os_inode = NULL;
2498 if (dev->od_oi_table != NULL)
2499 osd_oi_fini(osd_oti_get(env), dev);
2502 /* object table based iteration APIs */
2504 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2505 struct dt_object *dt, __u32 attr,
2506 struct lustre_capa *capa)
2508 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2509 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2510 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2511 struct osd_scrub *scrub = &dev->od_scrub;
2512 struct osd_otable_it *it;
2517 /* od_otable_mutex: prevent curcurrent init/fini */
2518 mutex_lock(&dev->od_otable_mutex);
2519 if (dev->od_otable_it != NULL)
2520 GOTO(out, it = ERR_PTR(-EALREADY));
2524 GOTO(out, it = ERR_PTR(-ENOMEM));
2526 dev->od_otable_it = it;
2528 it->ooi_cache.ooc_consumer_idx = -1;
2529 if (flags & DOIF_OUTUSED)
2530 it->ooi_used_outside = 1;
2532 if (flags & DOIF_RESET)
2535 if (valid & DOIV_ERROR_HANDLE) {
2536 if (flags & DOIF_FAILOUT)
2537 start |= SS_SET_FAILOUT;
2539 start |= SS_CLEAR_FAILOUT;
2542 if (valid & DOIV_DRYRUN) {
2543 if (flags & DOIF_DRYRUN)
2544 start |= SS_SET_DRYRUN;
2546 start |= SS_CLEAR_DRYRUN;
2549 rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2550 if (rc < 0 && rc != -EALREADY) {
2551 dev->od_otable_it = NULL;
2553 GOTO(out, it = ERR_PTR(rc));
2556 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2561 mutex_unlock(&dev->od_otable_mutex);
2562 return (struct dt_it *)it;
2565 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2567 struct osd_otable_it *it = (struct osd_otable_it *)di;
2568 struct osd_device *dev = it->ooi_dev;
2570 /* od_otable_mutex: prevent curcurrent init/fini */
2571 mutex_lock(&dev->od_otable_mutex);
2572 do_osd_scrub_stop(&dev->od_scrub);
2573 LASSERT(dev->od_otable_it == it);
2575 dev->od_otable_it = NULL;
2576 mutex_unlock(&dev->od_otable_mutex);
2580 static int osd_otable_it_get(const struct lu_env *env,
2581 struct dt_it *di, const struct dt_key *key)
2586 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2591 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2593 spin_lock(&scrub->os_lock);
2594 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2595 scrub->os_waiting ||
2596 !thread_is_running(&scrub->os_thread))
2597 it->ooi_waiting = 0;
2599 it->ooi_waiting = 1;
2600 spin_unlock(&scrub->os_lock);
2602 return !it->ooi_waiting;
2605 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2607 struct osd_otable_it *it = (struct osd_otable_it *)di;
2608 struct osd_device *dev = it->ooi_dev;
2609 struct osd_scrub *scrub = &dev->od_scrub;
2610 struct osd_otable_cache *ooc = &it->ooi_cache;
2611 struct ptlrpc_thread *thread = &scrub->os_thread;
2612 struct l_wait_info lwi = { 0 };
2616 LASSERT(it->ooi_user_ready);
2619 if (!thread_is_running(thread) && !it->ooi_used_outside)
2622 if (ooc->ooc_cached_items > 0) {
2623 ooc->ooc_cached_items--;
2624 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2625 ~OSD_OTABLE_IT_CACHE_MASK;
2629 if (it->ooi_all_cached) {
2630 l_wait_event(thread->t_ctl_waitq,
2631 !thread_is_running(thread),
2636 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2637 spin_lock(&scrub->os_lock);
2638 scrub->os_waiting = 0;
2639 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2640 spin_unlock(&scrub->os_lock);
2643 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2644 l_wait_event(thread->t_ctl_waitq,
2645 osd_otable_it_wakeup(scrub, it),
2648 if (!thread_is_running(thread) && !it->ooi_used_outside)
2651 rc = osd_otable_it_preload(env, it);
2658 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2659 const struct dt_it *di)
2664 static int osd_otable_it_key_size(const struct lu_env *env,
2665 const struct dt_it *di)
2667 return sizeof(__u64);
2670 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2671 struct dt_rec *rec, __u32 attr)
2673 struct osd_otable_it *it = (struct osd_otable_it *)di;
2674 struct osd_otable_cache *ooc = &it->ooi_cache;
2676 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2678 /* Filter out Invald FID already. */
2679 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2680 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2681 PFID((struct lu_fid *)rec),
2682 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2687 static __u64 osd_otable_it_store(const struct lu_env *env,
2688 const struct dt_it *di)
2690 struct osd_otable_it *it = (struct osd_otable_it *)di;
2691 struct osd_otable_cache *ooc = &it->ooi_cache;
2694 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2695 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2697 hash = ooc->ooc_pos_preload;
2702 * Set the OSD layer iteration start position as the specified hash.
2704 static int osd_otable_it_load(const struct lu_env *env,
2705 const struct dt_it *di, __u64 hash)
2707 struct osd_otable_it *it = (struct osd_otable_it *)di;
2708 struct osd_device *dev = it->ooi_dev;
2709 struct osd_otable_cache *ooc = &it->ooi_cache;
2710 struct osd_scrub *scrub = &dev->od_scrub;
2714 /* Forbid to set iteration position after iteration started. */
2715 if (it->ooi_user_ready)
2718 if (hash > OSD_OTABLE_MAX_HASH)
2719 hash = OSD_OTABLE_MAX_HASH;
2721 ooc->ooc_pos_preload = hash;
2722 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2723 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2725 it->ooi_user_ready = 1;
2726 if (!scrub->os_full_speed)
2727 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2729 /* Unplug OSD layer iteration by the first next() call. */
2730 rc = osd_otable_it_next(env, (struct dt_it *)it);
2735 static int osd_otable_it_key_rec(const struct lu_env *env,
2736 const struct dt_it *di, void *key_rec)
2741 const struct dt_index_operations osd_otable_ops = {
2743 .init = osd_otable_it_init,
2744 .fini = osd_otable_it_fini,
2745 .get = osd_otable_it_get,
2746 .put = osd_otable_it_put,
2747 .next = osd_otable_it_next,
2748 .key = osd_otable_it_key,
2749 .key_size = osd_otable_it_key_size,
2750 .rec = osd_otable_it_rec,
2751 .store = osd_otable_it_store,
2752 .load = osd_otable_it_load,
2753 .key_rec = osd_otable_it_key_rec,
2757 /* high priority inconsistent items list APIs */
2759 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
2761 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2764 struct osd_inconsistent_item *oii;
2765 struct osd_scrub *scrub = &dev->od_scrub;
2766 struct ptlrpc_thread *thread = &scrub->os_thread;
2771 if (unlikely(oii == NULL))
2774 INIT_LIST_HEAD(&oii->oii_list);
2775 oii->oii_cache = *oic;
2776 oii->oii_insert = insert;
2778 if (scrub->os_partial_scan) {
2779 __u64 now = cfs_time_current_sec();
2781 /* If there haven't been errors in a long time,
2782 * decay old count until either the errors are
2783 * gone or we reach the current interval. */
2784 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2785 scrub->os_bad_oimap_time +
2786 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2787 scrub->os_bad_oimap_count >>= 1;
2788 scrub->os_bad_oimap_time +=
2789 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2792 scrub->os_bad_oimap_time = now;
2793 if (++scrub->os_bad_oimap_count >
2794 dev->od_full_scrub_threshold_rate)
2795 scrub->os_full_scrub = 1;
2798 spin_lock(&scrub->os_lock);
2799 if (unlikely(!thread_is_running(thread))) {
2800 spin_unlock(&scrub->os_lock);
2805 if (list_empty(&scrub->os_inconsistent_items))
2807 list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2808 spin_unlock(&scrub->os_lock);
2811 wake_up_all(&thread->t_ctl_waitq);
2816 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2817 struct osd_inode_id *id)
2819 struct osd_scrub *scrub = &dev->od_scrub;
2820 struct osd_inconsistent_item *oii;
2823 spin_lock(&scrub->os_lock);
2824 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2825 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2826 *id = oii->oii_cache.oic_lid;
2827 spin_unlock(&scrub->os_lock);
2831 spin_unlock(&scrub->os_lock);
2838 static const char *scrub_status_names[] = {
2849 static const char *scrub_flags_names[] = {
2857 static const char *scrub_param_names[] = {
2863 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2870 rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2874 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2877 rc = seq_printf(m, "%s%c", names[i],
2878 bits != 0 ? ',' : '\n');
2886 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2891 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2892 cfs_time_current_sec() - time);
2894 rc = seq_printf(m, "%s: N/A\n", prefix);
2898 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2903 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2905 rc = seq_printf(m, "%s: N/A\n", prefix);
2909 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2911 struct osd_scrub *scrub = &dev->od_scrub;
2912 struct scrub_file *sf = &scrub->os_file;
2917 down_read(&scrub->os_rwsem);
2918 rc = seq_printf(m, "name: OI_scrub\n"
2922 sf->sf_magic, (int)sf->sf_oi_count,
2923 scrub_status_names[sf->sf_status]);
2927 rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2932 rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2937 rc = scrub_time_dump(m, sf->sf_time_last_complete,
2938 "time_since_last_completed");
2942 rc = scrub_time_dump(m, sf->sf_time_latest_start,
2943 "time_since_latest_start");
2947 rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2948 "time_since_last_checkpoint");
2952 rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2953 "latest_start_position");
2957 rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2958 "last_checkpoint_position");
2962 rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2963 "first_failure_position");
2967 checked = sf->sf_items_checked + scrub->os_new_checked;
2968 rc = seq_printf(m, "checked: "LPU64"\n"
2969 "updated: "LPU64"\n"
2971 "prior_updated: "LPU64"\n"
2972 "noscrub: "LPU64"\n"
2974 "success_count: %u\n",
2975 checked, sf->sf_items_updated, sf->sf_items_failed,
2976 sf->sf_items_updated_prior, sf->sf_items_noscrub,
2977 sf->sf_items_igif, sf->sf_success_count);
2982 if (thread_is_running(&scrub->os_thread)) {
2983 cfs_duration_t duration = cfs_time_current() -
2984 scrub->os_time_last_checkpoint;
2985 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
2987 __u32 rtime = sf->sf_run_time +
2988 cfs_duration_sec(duration + HALF_SEC);
2991 do_div(new_checked, duration);
2993 do_div(speed, rtime);
2994 rc = seq_printf(m, "run_time: %u seconds\n"
2995 "average_speed: "LPU64" objects/sec\n"
2996 "real-time_speed: "LPU64" objects/sec\n"
2997 "current_position: %u\n"
2998 "lf_scanned: "LPU64"\n"
2999 "lf_reparied: "LPU64"\n"
3000 "lf_failed: "LPU64"\n",
3001 rtime, speed, new_checked, scrub->os_pos_current,
3002 scrub->os_lf_scanned, scrub->os_lf_repaired,
3003 scrub->os_lf_failed);
3005 if (sf->sf_run_time != 0)
3006 do_div(speed, sf->sf_run_time);
3007 rc = seq_printf(m, "run_time: %u seconds\n"
3008 "average_speed: "LPU64" objects/sec\n"
3009 "real-time_speed: N/A\n"
3010 "current_position: N/A\n"
3011 "lf_scanned: "LPU64"\n"
3012 "lf_reparied: "LPU64"\n"
3013 "lf_failed: "LPU64"\n",
3014 sf->sf_run_time, speed, scrub->os_lf_scanned,
3015 scrub->os_lf_repaired, scrub->os_lf_failed);
3019 up_read(&scrub->os_rwsem);
3020 return (rc < 0 ? -ENOSPC : 0);