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;
373 struct lu_fid *tfid = &info->oti_fid;
377 bool removed = false;
381 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
384 if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
385 struct ost_id *oi = &info->oti_ostid;
387 fid_to_ostid(fid, oi);
388 ostid_to_fid(tfid, oi, 0);
393 /* We want the LMA to fit into the 256-byte OST inode, so operate
395 * 1) read old XATTR_NAME_FID and save the parent FID;
396 * 2) delete the old XATTR_NAME_FID;
397 * 3) make new LMA and add it;
398 * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
400 * Making the LMA to fit into the 256-byte OST inode can save time for
401 * normal osd_check_lma() and for other OI scrub scanning in future.
402 * So it is worth to make some slow conversion here. */
403 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
404 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
407 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
408 DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
412 /* 1) read old XATTR_NAME_FID and save the parent FID */
413 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
414 if (rc == sizeof(*ff)) {
415 /* 2) delete the old XATTR_NAME_FID */
416 ll_vfs_dq_init(inode);
417 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
422 } else if (unlikely(rc == -ENODATA)) {
424 } else if (rc != sizeof(struct filter_fid)) {
425 GOTO(stop, rc = -EINVAL);
428 /* 3) make new LMA and add it */
429 rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
430 if (rc == 0 && reset)
431 size = sizeof(struct filter_fid);
432 else if (rc != 0 && removed)
433 /* If failed, we should try to add the old back. */
434 size = sizeof(struct filter_fid_old);
436 /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
440 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
442 if (rc1 != 0 && rc == 0)
449 ldiskfs_journal_stop(jh);
451 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
452 osd_name(dev), PFID(tfid), rc);
457 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
458 struct osd_idmap_cache *oic, int val)
460 struct osd_scrub *scrub = &dev->od_scrub;
461 struct scrub_file *sf = &scrub->os_file;
462 struct lu_fid *fid = &oic->oic_fid;
463 struct osd_inode_id *lid = &oic->oic_lid;
464 struct osd_inode_id *lid2 = &info->oti_id;
465 struct osd_inconsistent_item *oii = NULL;
466 struct inode *inode = NULL;
467 int ops = DTO_INDEX_UPDATE;
470 bool converted = false;
473 down_write(&scrub->os_rwsem);
474 scrub->os_new_checked++;
478 if (scrub->os_in_prior)
479 oii = list_entry(oic, struct osd_inconsistent_item,
482 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
485 if (fid_is_igif(fid))
488 if (val == SCRUB_NEXT_OSTOBJ_OLD) {
489 inode = osd_iget(info, dev, lid);
492 /* Someone removed the inode. */
493 if (rc == -ENOENT || rc == -ESTALE)
498 sf->sf_flags |= SF_UPGRADE;
499 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
500 dev->od_check_ff = 1;
501 rc = osd_scrub_convert_ff(info, dev, inode, fid);
508 if ((val == SCRUB_NEXT_NOLMA) &&
509 (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
512 if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
515 rc = osd_oi_lookup(info, dev, fid, lid2,
516 (val == SCRUB_NEXT_OSTOBJ ||
517 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
519 if (rc != -ENOENT && rc != -ESTALE)
524 inode = osd_iget(info, dev, lid);
527 /* Someone removed the inode. */
528 if (rc == -ENOENT || rc == -ESTALE)
534 if (!scrub->os_partial_scan)
535 scrub->os_full_speed = 1;
537 ops = DTO_INDEX_INSERT;
538 idx = osd_oi_fid2idx(dev, fid);
540 case SCRUB_NEXT_NOLMA:
541 sf->sf_flags |= SF_UPGRADE;
542 if (!(sf->sf_param & SP_DRYRUN)) {
543 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
548 if (!(sf->sf_flags & SF_INCONSISTENT))
549 dev->od_igif_inoi = 0;
551 case SCRUB_NEXT_OSTOBJ:
552 sf->sf_flags |= SF_INCONSISTENT;
553 case SCRUB_NEXT_OSTOBJ_OLD:
556 sf->sf_flags |= SF_RECREATED;
557 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
558 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
561 } else if (osd_id_eq(lid, lid2)) {
563 sf->sf_items_updated++;
567 if (!scrub->os_partial_scan)
568 scrub->os_full_speed = 1;
570 sf->sf_flags |= SF_INCONSISTENT;
572 /* XXX: If the device is restored from file-level backup, then
573 * some IGIFs may have been already in OI files, and some
574 * may be not yet. Means upgrading from 1.8 may be partly
575 * processed, but some clients may hold some immobilized
576 * IGIFs, and use them to access related objects. Under
577 * such case, OSD does not know whether an given IGIF has
578 * been processed or to be processed, and it also cannot
579 * generate local ino#/gen# directly from the immobilized
580 * IGIF because of the backup/restore. Then force OSD to
581 * lookup the given IGIF in OI files, and if no entry,
582 * then ask the client to retry after upgrading completed.
583 * No better choice. */
584 dev->od_igif_inoi = 1;
587 rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
588 (val == SCRUB_NEXT_OSTOBJ ||
589 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
591 if (scrub->os_in_prior)
592 sf->sf_items_updated_prior++;
594 sf->sf_items_updated++;
601 sf->sf_items_failed++;
602 if (sf->sf_pos_first_inconsistent == 0 ||
603 sf->sf_pos_first_inconsistent > lid->oii_ino)
604 sf->sf_pos_first_inconsistent = lid->oii_ino;
609 /* There may be conflict unlink during the OI scrub,
610 * if happend, then remove the new added OI mapping. */
611 if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
612 unlikely(inode->i_nlink == 0))
613 osd_scrub_refresh_mapping(info, dev, fid, lid,
614 DTO_INDEX_DELETE, false,
615 (val == SCRUB_NEXT_OSTOBJ ||
616 val == SCRUB_NEXT_OSTOBJ_OLD) ?
617 OI_KNOWN_ON_OST : 0);
618 up_write(&scrub->os_rwsem);
620 if (inode != NULL && !IS_ERR(inode))
624 LASSERT(!list_empty(&oii->oii_list));
626 spin_lock(&scrub->os_lock);
627 list_del_init(&oii->oii_list);
628 spin_unlock(&scrub->os_lock);
631 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
636 static int osd_scrub_prep(struct osd_device *dev)
638 struct osd_scrub *scrub = &dev->od_scrub;
639 struct ptlrpc_thread *thread = &scrub->os_thread;
640 struct scrub_file *sf = &scrub->os_file;
641 __u32 flags = scrub->os_start_flags;
643 bool drop_dryrun = false;
646 CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
647 osd_scrub2name(scrub), flags);
649 down_write(&scrub->os_rwsem);
650 if (flags & SS_SET_FAILOUT)
651 sf->sf_param |= SP_FAILOUT;
652 else if (flags & SS_CLEAR_FAILOUT)
653 sf->sf_param &= ~SP_FAILOUT;
655 if (flags & SS_SET_DRYRUN) {
656 sf->sf_param |= SP_DRYRUN;
657 } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
658 sf->sf_param &= ~SP_DRYRUN;
662 if (flags & SS_RESET)
663 osd_scrub_file_reset(scrub,
664 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
666 if (flags & SS_AUTO_FULL) {
667 scrub->os_full_speed = 1;
668 scrub->os_partial_scan = 0;
669 sf->sf_flags |= SF_AUTO;
670 } else if (flags & SS_AUTO_PARTIAL) {
671 scrub->os_full_speed = 0;
672 scrub->os_partial_scan = 1;
673 sf->sf_flags |= SF_AUTO;
674 } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
676 scrub->os_full_speed = 1;
677 scrub->os_partial_scan = 0;
679 scrub->os_full_speed = 0;
680 scrub->os_partial_scan = 0;
683 spin_lock(&scrub->os_lock);
684 scrub->os_in_prior = 0;
685 scrub->os_waiting = 0;
686 scrub->os_paused = 0;
687 scrub->os_in_join = 0;
688 scrub->os_full_scrub = 0;
689 spin_unlock(&scrub->os_lock);
690 scrub->os_new_checked = 0;
691 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
692 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
693 else if (sf->sf_pos_last_checkpoint != 0)
694 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
696 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
698 scrub->os_pos_current = sf->sf_pos_latest_start;
699 sf->sf_status = SS_SCANNING;
700 sf->sf_time_latest_start = cfs_time_current_sec();
701 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
702 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
703 rc = osd_scrub_file_store(scrub);
705 spin_lock(&scrub->os_lock);
706 thread_set_flags(thread, SVC_RUNNING);
707 spin_unlock(&scrub->os_lock);
708 wake_up_all(&thread->t_ctl_waitq);
710 up_write(&scrub->os_rwsem);
715 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
717 struct scrub_file *sf = &scrub->os_file;
720 if (likely(cfs_time_before(cfs_time_current(),
721 scrub->os_time_next_checkpoint) ||
722 scrub->os_new_checked == 0))
725 down_write(&scrub->os_rwsem);
726 sf->sf_items_checked += scrub->os_new_checked;
727 scrub->os_new_checked = 0;
728 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
729 sf->sf_time_last_checkpoint = cfs_time_current_sec();
730 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
731 scrub->os_time_last_checkpoint);
732 rc = osd_scrub_file_store(scrub);
733 up_write(&scrub->os_rwsem);
738 static void osd_scrub_post(struct osd_scrub *scrub, int result)
740 struct scrub_file *sf = &scrub->os_file;
743 CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
744 osd_scrub2name(scrub), result);
746 down_write(&scrub->os_rwsem);
747 spin_lock(&scrub->os_lock);
748 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
749 spin_unlock(&scrub->os_lock);
750 if (scrub->os_new_checked > 0) {
751 sf->sf_items_checked += scrub->os_new_checked;
752 scrub->os_new_checked = 0;
753 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
755 sf->sf_time_last_checkpoint = cfs_time_current_sec();
757 struct osd_device *dev =
758 container_of0(scrub, struct osd_device, od_scrub);
760 dev->od_igif_inoi = 1;
761 dev->od_check_ff = 0;
762 sf->sf_status = SS_COMPLETED;
763 if (!(sf->sf_param & SP_DRYRUN)) {
764 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
765 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
766 SF_UPGRADE | SF_AUTO);
768 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
769 sf->sf_success_count++;
770 } else if (result == 0) {
771 if (scrub->os_paused)
772 sf->sf_status = SS_PAUSED;
774 sf->sf_status = SS_STOPPED;
776 sf->sf_status = SS_FAILED;
778 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
779 scrub->os_time_last_checkpoint);
780 result = osd_scrub_file_store(scrub);
781 up_write(&scrub->os_rwsem);
786 /* iteration engine */
788 struct osd_iit_param {
789 struct super_block *sb;
790 struct buffer_head *bitmap;
796 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
797 struct osd_device *dev,
798 struct osd_iit_param *param,
799 struct osd_idmap_cache **oic,
802 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
803 struct osd_device *dev,
804 struct osd_iit_param *param,
805 struct osd_idmap_cache *oic,
806 bool *noslot, int rc);
808 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
810 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
811 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
812 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
813 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
814 return SCRUB_NEXT_BREAK;
816 *pos = param->gbase + param->offset;
822 * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
823 * \retval 0: FID-on-MDT
825 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
826 struct osd_device *dev,
829 /* XXX: The initial OI scrub will scan the top level /O to generate
830 * a small local FLDB according to the <seq>. If the given FID
831 * is in the local FLDB, then it is FID-on-OST; otherwise it's
832 * quite possible for FID-on-MDT. */
834 return SCRUB_NEXT_OSTOBJ_OLD;
839 static int osd_scrub_get_fid(struct osd_thread_info *info,
840 struct osd_device *dev, struct inode *inode,
841 struct lu_fid *fid, bool scrub)
843 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
845 bool has_lma = false;
847 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
850 if (lma->lma_compat & LMAC_NOT_IN_OI ||
851 lma->lma_incompat & LMAI_AGENT)
852 return SCRUB_NEXT_CONTINUE;
854 *fid = lma->lma_self_fid;
858 if (lma->lma_compat & LMAC_FID_ON_OST)
859 return SCRUB_NEXT_OSTOBJ;
861 if (fid_is_idif(fid))
862 return SCRUB_NEXT_OSTOBJ_OLD;
864 /* For local object. */
865 if (fid_is_internal(fid))
868 /* For external visible MDT-object with non-normal FID. */
869 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
872 /* For the object with normal FID, it may be MDT-object,
873 * or may be 2.4 OST-object, need further distinguish.
874 * Fall through to next section. */
877 if (rc == -ENODATA || rc == 0) {
878 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
881 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
882 rc = SCRUB_NEXT_OSTOBJ_OLD;
888 /* It is FID-on-OST, but we do not know how
889 * to generate its FID, ignore it directly. */
890 rc = SCRUB_NEXT_CONTINUE;
892 /* It is 2.4 OST-object. */
893 rc = SCRUB_NEXT_OSTOBJ_OLD;
901 if (dev->od_scrub.os_convert_igif) {
902 lu_igif_build(fid, inode->i_ino,
903 inode->i_generation);
905 rc = SCRUB_NEXT_NOLMA;
909 /* It may be FID-on-OST, or may be FID for
910 * non-MDT0, anyway, we do not know how to
911 * generate its FID, ignore it directly. */
912 rc = SCRUB_NEXT_CONTINUE;
917 /* For OI scrub case only: the object has LMA but has no ff
918 * (or ff crashed). It may be MDT-object, may be OST-object
919 * with crashed ff. The last check is local FLDB. */
920 rc = osd_scrub_check_local_fldb(info, dev, fid);
926 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
927 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
928 struct super_block *sb, bool scrub)
934 /* Not handle the backend root object and agent parent object.
935 * They are neither visible to namespace nor have OI mappings. */
936 if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
937 pos == osd_remote_parent_ino(dev)))
938 RETURN(SCRUB_NEXT_CONTINUE);
940 osd_id_gen(lid, pos, OSD_OII_NOGEN);
941 inode = osd_iget(info, dev, lid);
944 /* The inode may be removed after bitmap searching, or the
945 * file is new created without inode initialized yet. */
946 if (rc == -ENOENT || rc == -ESTALE)
947 RETURN(SCRUB_NEXT_CONTINUE);
949 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
950 "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
956 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
957 /* Only skip it for the first OI scrub accessing. */
958 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
959 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
962 rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
971 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
972 struct osd_iit_param *param,
973 struct osd_idmap_cache **oic, const bool noslot)
975 struct osd_scrub *scrub = &dev->od_scrub;
976 struct ptlrpc_thread *thread = &scrub->os_thread;
978 struct osd_inode_id *lid;
981 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
982 struct l_wait_info lwi;
984 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
985 if (likely(lwi.lwi_timeout > 0))
986 l_wait_event(thread->t_ctl_waitq,
987 !list_empty(&scrub->os_inconsistent_items) ||
988 !thread_is_running(thread),
992 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
993 spin_lock(&scrub->os_lock);
994 thread_set_flags(thread, SVC_STOPPING);
995 spin_unlock(&scrub->os_lock);
996 return SCRUB_NEXT_CRASH;
999 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1000 return SCRUB_NEXT_FATAL;
1002 if (unlikely(!thread_is_running(thread)))
1003 return SCRUB_NEXT_EXIT;
1005 if (!list_empty(&scrub->os_inconsistent_items)) {
1006 struct osd_inconsistent_item *oii;
1008 oii = list_entry(scrub->os_inconsistent_items.next,
1009 struct osd_inconsistent_item, oii_list);
1010 *oic = &oii->oii_cache;
1011 scrub->os_in_prior = 1;
1016 return SCRUB_NEXT_WAIT;
1018 rc = osd_iit_next(param, &scrub->os_pos_current);
1022 *oic = &scrub->os_oic;
1023 fid = &(*oic)->oic_fid;
1024 lid = &(*oic)->oic_lid;
1025 rc = osd_iit_iget(info, dev, fid, lid,
1026 scrub->os_pos_current, param->sb, true);
1030 static int osd_preload_next(struct osd_thread_info *info,
1031 struct osd_device *dev, struct osd_iit_param *param,
1032 struct osd_idmap_cache **oic, const bool noslot)
1034 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1035 struct osd_scrub *scrub;
1036 struct ptlrpc_thread *thread;
1039 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1043 scrub = &dev->od_scrub;
1044 thread = &scrub->os_thread;
1045 if (thread_is_running(thread) &&
1046 ooc->ooc_pos_preload >= scrub->os_pos_current)
1047 return SCRUB_NEXT_EXIT;
1049 rc = osd_iit_iget(info, dev,
1050 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1051 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1052 ooc->ooc_pos_preload, param->sb, false);
1053 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1054 * ignore the failure, so it still need to skip the inode next time. */
1055 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1060 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1062 spin_lock(&scrub->os_lock);
1063 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1064 !list_empty(&scrub->os_inconsistent_items) ||
1065 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1066 scrub->os_waiting = 0;
1068 scrub->os_waiting = 1;
1069 spin_unlock(&scrub->os_lock);
1071 return !scrub->os_waiting;
1074 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1075 struct osd_iit_param *param,
1076 struct osd_idmap_cache *oic, bool *noslot, int rc)
1078 struct l_wait_info lwi = { 0 };
1079 struct osd_scrub *scrub = &dev->od_scrub;
1080 struct scrub_file *sf = &scrub->os_file;
1081 struct ptlrpc_thread *thread = &scrub->os_thread;
1082 struct osd_otable_it *it = dev->od_otable_it;
1083 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
1086 case SCRUB_NEXT_CONTINUE:
1088 case SCRUB_NEXT_WAIT:
1090 case SCRUB_NEXT_NOSCRUB:
1091 down_write(&scrub->os_rwsem);
1092 scrub->os_new_checked++;
1093 sf->sf_items_noscrub++;
1094 up_write(&scrub->os_rwsem);
1098 rc = osd_scrub_check_update(info, dev, oic, rc);
1102 rc = osd_scrub_checkpoint(scrub);
1104 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1105 "rc = %d\n", osd_scrub2name(scrub),
1106 scrub->os_pos_current, rc);
1107 /* Continue, as long as the scrub itself can go ahead. */
1110 if (scrub->os_in_prior) {
1111 scrub->os_in_prior = 0;
1116 scrub->os_pos_current = param->gbase + ++(param->offset);
1119 if (it != NULL && it->ooi_waiting && ooc != NULL &&
1120 ooc->ooc_pos_preload < scrub->os_pos_current) {
1121 spin_lock(&scrub->os_lock);
1122 it->ooi_waiting = 0;
1123 wake_up_all(&thread->t_ctl_waitq);
1124 spin_unlock(&scrub->os_lock);
1127 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1130 if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1136 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1139 if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1146 static int osd_preload_exec(struct osd_thread_info *info,
1147 struct osd_device *dev, struct osd_iit_param *param,
1148 struct osd_idmap_cache *oic, bool *noslot, int rc)
1150 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1153 ooc->ooc_cached_items++;
1154 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1155 ~OSD_OTABLE_IT_CACHE_MASK;
1157 return rc > 0 ? 0 : rc;
1160 #define SCRUB_IT_ALL 1
1161 #define SCRUB_IT_CRASH 2
1163 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1166 struct osd_scrub *scrub = &dev->od_scrub;
1167 struct ptlrpc_thread *thread = &scrub->os_thread;
1168 struct scrub_file *sf = &scrub->os_file;
1172 LASSERT(!(flags & SS_AUTO_PARTIAL));
1174 down_write(&scrub->os_rwsem);
1175 scrub->os_in_join = 1;
1176 if (flags & SS_SET_FAILOUT)
1177 sf->sf_param |= SP_FAILOUT;
1178 else if (flags & SS_CLEAR_FAILOUT)
1179 sf->sf_param &= ~SP_FAILOUT;
1181 if (flags & SS_SET_DRYRUN)
1182 sf->sf_param |= SP_DRYRUN;
1183 else if (flags & SS_CLEAR_DRYRUN)
1184 sf->sf_param &= ~SP_DRYRUN;
1186 if (flags & SS_RESET) {
1187 osd_scrub_file_reset(scrub,
1188 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1189 inconsistent ? SF_INCONSISTENT : 0);
1190 sf->sf_status = SS_SCANNING;
1193 if (flags & SS_AUTO_FULL) {
1194 sf->sf_flags |= SF_AUTO;
1195 scrub->os_full_speed = 1;
1198 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1199 scrub->os_full_speed = 1;
1201 scrub->os_full_speed = 0;
1203 scrub->os_new_checked = 0;
1204 if (sf->sf_pos_last_checkpoint != 0)
1205 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1207 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1209 scrub->os_pos_current = sf->sf_pos_latest_start;
1210 sf->sf_time_latest_start = cfs_time_current_sec();
1211 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1212 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1213 rc = osd_scrub_file_store(scrub);
1215 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1216 "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1218 spin_lock(&scrub->os_lock);
1219 scrub->os_waiting = 0;
1220 scrub->os_paused = 0;
1221 scrub->os_partial_scan = 0;
1222 scrub->os_in_join = 0;
1223 scrub->os_full_scrub = 0;
1224 spin_unlock(&scrub->os_lock);
1225 wake_up_all(&thread->t_ctl_waitq);
1226 up_write(&scrub->os_rwsem);
1231 static int osd_inode_iteration(struct osd_thread_info *info,
1232 struct osd_device *dev, __u32 max, bool preload)
1234 struct osd_scrub *scrub = &dev->od_scrub;
1235 struct ptlrpc_thread *thread = &scrub->os_thread;
1236 struct scrub_file *sf = &scrub->os_file;
1237 osd_iit_next_policy next;
1238 osd_iit_exec_policy exec;
1241 struct osd_iit_param param = { NULL };
1242 struct l_wait_info lwi = { 0 };
1248 param.sb = osd_sb(dev);
1252 while (scrub->os_partial_scan && !scrub->os_in_join) {
1253 struct osd_idmap_cache *oic = NULL;
1255 rc = osd_scrub_next(info, dev, ¶m, &oic, noslot);
1257 case SCRUB_NEXT_EXIT:
1259 case SCRUB_NEXT_CRASH:
1260 RETURN(SCRUB_IT_CRASH);
1261 case SCRUB_NEXT_FATAL:
1263 case SCRUB_NEXT_WAIT: {
1264 struct kstatfs *ksfs = &info->oti_ksfs;
1267 if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1268 unlikely(sf->sf_items_updated_prior == 0))
1271 if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1272 scrub->os_full_scrub) {
1273 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1278 rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1280 __u64 used = ksfs->f_files - ksfs->f_ffree;
1282 do_div(used, sf->sf_items_updated_prior);
1283 /* If we hit too much inconsistent OI
1284 * mappings during the partial scan,
1285 * then scan the device completely. */
1286 if (used < dev->od_full_scrub_ratio) {
1288 SS_AUTO_FULL | SS_RESET, true);
1294 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1298 saved_flags = sf->sf_flags;
1299 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1300 SF_UPGRADE | SF_AUTO);
1301 sf->sf_status = SS_COMPLETED;
1302 l_wait_event(thread->t_ctl_waitq,
1303 !thread_is_running(thread) ||
1304 !scrub->os_partial_scan ||
1305 scrub->os_in_join ||
1306 !list_empty(&scrub->os_inconsistent_items),
1308 sf->sf_flags = saved_flags;
1309 sf->sf_status = SS_SCANNING;
1311 if (unlikely(!thread_is_running(thread)))
1314 if (!scrub->os_partial_scan || scrub->os_in_join)
1320 LASSERTF(rc == 0, "rc = %d\n", rc);
1322 osd_scrub_exec(info, dev, ¶m, oic, &noslot, rc);
1329 l_wait_event(thread->t_ctl_waitq,
1330 !thread_is_running(thread) || !scrub->os_in_join,
1333 if (unlikely(!thread_is_running(thread)))
1339 next = osd_scrub_next;
1340 exec = osd_scrub_exec;
1341 pos = &scrub->os_pos_current;
1342 count = &scrub->os_new_checked;
1344 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1346 next = osd_preload_next;
1347 exec = osd_preload_exec;
1348 pos = &ooc->ooc_pos_preload;
1349 count = &ooc->ooc_cached_items;
1351 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1353 while (*pos <= limit && *count < max) {
1354 struct osd_idmap_cache *oic = NULL;
1355 struct ldiskfs_group_desc *desc;
1357 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1358 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1362 ldiskfs_lock_group(param.sb, param.bg);
1363 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1364 ldiskfs_unlock_group(param.sb, param.bg);
1365 *pos = 1 + (param.bg + 1) *
1366 LDISKFS_INODES_PER_GROUP(param.sb);
1369 ldiskfs_unlock_group(param.sb, param.bg);
1371 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1372 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1373 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1374 if (param.bitmap == NULL) {
1375 CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1376 "scrub will stop, urgent mode\n",
1377 osd_scrub2name(scrub), (__u32)param.bg);
1381 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1384 ldiskfs_itable_unused_count(param.sb, desc) >
1385 LDISKFS_INODES_PER_GROUP(param.sb))
1388 rc = next(info, dev, ¶m, &oic, noslot);
1390 case SCRUB_NEXT_BREAK:
1392 case SCRUB_NEXT_EXIT:
1393 brelse(param.bitmap);
1395 case SCRUB_NEXT_CRASH:
1396 brelse(param.bitmap);
1397 RETURN(SCRUB_IT_CRASH);
1398 case SCRUB_NEXT_FATAL:
1399 brelse(param.bitmap);
1403 rc = exec(info, dev, ¶m, oic, &noslot, rc);
1405 brelse(param.bitmap);
1411 brelse(param.bitmap);
1415 RETURN(SCRUB_IT_ALL);
1419 static int osd_otable_it_preload(const struct lu_env *env,
1420 struct osd_otable_it *it)
1422 struct osd_device *dev = it->ooi_dev;
1423 struct osd_scrub *scrub = &dev->od_scrub;
1424 struct osd_otable_cache *ooc = &it->ooi_cache;
1428 rc = osd_inode_iteration(osd_oti_get(env), dev,
1429 OSD_OTABLE_IT_CACHE_SIZE, true);
1430 if (rc == SCRUB_IT_ALL)
1431 it->ooi_all_cached = 1;
1433 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1434 spin_lock(&scrub->os_lock);
1435 scrub->os_waiting = 0;
1436 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1437 spin_unlock(&scrub->os_lock);
1440 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1443 static int osd_scrub_main(void *args)
1446 struct osd_device *dev = (struct osd_device *)args;
1447 struct osd_scrub *scrub = &dev->od_scrub;
1448 struct ptlrpc_thread *thread = &scrub->os_thread;
1452 rc = lu_env_init(&env, LCT_LOCAL);
1454 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1455 osd_scrub2name(scrub), rc);
1459 rc = osd_scrub_prep(dev);
1461 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1462 osd_scrub2name(scrub), rc);
1466 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1467 struct l_wait_info lwi = { 0 };
1468 struct osd_otable_it *it = dev->od_otable_it;
1469 struct osd_otable_cache *ooc = &it->ooi_cache;
1471 l_wait_event(thread->t_ctl_waitq,
1472 it->ooi_user_ready || !thread_is_running(thread),
1474 if (unlikely(!thread_is_running(thread)))
1477 scrub->os_pos_current = ooc->ooc_pos_preload;
1480 CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1481 osd_scrub2name(scrub), scrub->os_start_flags,
1482 scrub->os_pos_current);
1484 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1485 if (unlikely(rc == SCRUB_IT_CRASH))
1486 GOTO(out, rc = -EINVAL);
1490 osd_scrub_post(scrub, rc);
1491 CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1492 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1495 while (!list_empty(&scrub->os_inconsistent_items)) {
1496 struct osd_inconsistent_item *oii;
1498 oii = list_entry(scrub->os_inconsistent_items.next,
1499 struct osd_inconsistent_item, oii_list);
1500 list_del_init(&oii->oii_list);
1506 spin_lock(&scrub->os_lock);
1507 thread_set_flags(thread, SVC_STOPPED);
1508 wake_up_all(&thread->t_ctl_waitq);
1509 spin_unlock(&scrub->os_lock);
1513 /* initial OI scrub */
1515 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1516 struct dentry *, filldir_t filldir);
1518 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1519 loff_t offset, __u64 ino, unsigned d_type);
1520 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1521 loff_t offset, __u64 ino, unsigned d_type);
1522 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1523 loff_t offset, __u64 ino, unsigned d_type);
1526 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1527 struct dentry *dentry, filldir_t filldir);
1529 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1530 struct dentry *dentry, filldir_t filldir);
1533 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1534 struct dentry *dentry, filldir_t filldir);
1537 OLF_SCAN_SUBITEMS = 0x0001,
1538 OLF_HIDE_FID = 0x0002,
1539 OLF_SHOW_NAME = 0x0004,
1545 struct lu_fid olm_fid;
1548 scandir_t olm_scandir;
1549 filldir_t olm_filldir;
1552 /* Add the new introduced local files in the list in the future. */
1553 static const struct osd_lf_map osd_lf_maps[] = {
1555 { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1556 sizeof(CATLIST) - 1, NULL, NULL },
1559 { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1560 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1561 osd_ios_general_scan, osd_ios_varfid_fill },
1563 /* NIDTBL_VERSIONS */
1564 { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1565 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1566 osd_ios_varfid_fill },
1569 { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1572 { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1573 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1574 osd_ios_ROOT_scan, NULL },
1576 /* changelog_catalog */
1577 { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1580 /* changelog_users */
1581 { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1585 { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1586 sizeof("fld") - 1, NULL, NULL },
1589 { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1590 sizeof(LAST_RCVD) - 1, NULL, NULL },
1593 { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1594 sizeof(LOV_OBJID) - 1, NULL, NULL },
1597 { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1598 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1601 { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1602 osd_ios_general_scan, osd_ios_varfid_fill },
1605 { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1606 osd_ios_general_scan, osd_ios_varfid_fill },
1609 { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1610 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1613 { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1614 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1617 { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1618 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1621 { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1622 osd_ios_general_scan, osd_ios_varfid_fill },
1624 /* lfsck_bookmark */
1625 { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1629 { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1632 /* lfsck_namespace */
1633 { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1636 /* OBJECTS, upgrade from old device */
1637 { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1638 osd_ios_OBJECTS_scan, NULL },
1640 /* lquota_v2.user, upgrade from old device */
1641 { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1644 /* lquota_v2.group, upgrade from old device */
1645 { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1648 /* LAST_GROUP, upgrade from old device */
1649 { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1650 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1652 /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1653 { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1654 OLF_SHOW_NAME, sizeof("SLAVE_LOG") - 1, NULL, NULL },
1657 { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1658 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1659 osd_ios_general_scan, osd_ios_lf_fill },
1661 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1664 /* Add the new introduced files under .lustre/ in the list in the future. */
1665 static const struct osd_lf_map osd_dl_maps[] = {
1667 { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1668 sizeof("fid") - 1, NULL, NULL },
1670 /* .lustre/lost+found */
1671 { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1672 sizeof("lost+found") - 1, NULL, NULL },
1674 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1677 struct osd_ios_item {
1678 struct list_head oii_list;
1679 struct dentry *oii_dentry;
1680 scandir_t oii_scandir;
1681 filldir_t oii_filldir;
1684 struct osd_ios_filldir_buf {
1685 #ifdef HAVE_DIR_CONTEXT
1686 /* please keep it as first member */
1687 struct dir_context ctx;
1689 struct osd_thread_info *oifb_info;
1690 struct osd_device *oifb_dev;
1691 struct dentry *oifb_dentry;
1694 static inline struct dentry *
1695 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1697 struct dentry *dentry;
1699 dentry = ll_lookup_one_len(name, parent, namelen);
1700 if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1702 return ERR_PTR(-ENOENT);
1709 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1710 scandir_t scandir, filldir_t filldir)
1712 struct osd_ios_item *item;
1715 OBD_ALLOC_PTR(item);
1719 INIT_LIST_HEAD(&item->oii_list);
1720 item->oii_dentry = dget(dentry);
1721 item->oii_scandir = scandir;
1722 item->oii_filldir = filldir;
1723 list_add_tail(&item->oii_list, &dev->od_ios_list);
1729 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1731 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1732 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1733 * reference the inode, or fixed if it is missing or references another inode.
1736 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1737 struct inode *inode, const struct lu_fid *fid, int flags)
1739 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1740 struct osd_inode_id *id = &info->oti_id;
1741 struct osd_inode_id *id2 = &info->oti_id2;
1742 struct osd_scrub *scrub = &dev->od_scrub;
1743 struct scrub_file *sf = &scrub->os_file;
1748 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1749 if (rc != 0 && rc != -ENODATA) {
1750 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1751 "rc = %d\n", osd_name(dev), rc);
1756 osd_id_gen(id, inode->i_ino, inode->i_generation);
1757 if (rc == -ENODATA) {
1758 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1759 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1762 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1764 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1765 "scrub: rc = %d\n", osd_name(dev), rc);
1770 if (lma->lma_compat & LMAC_NOT_IN_OI)
1773 tfid = lma->lma_self_fid;
1776 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1781 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1782 DTO_INDEX_INSERT, true, 0);
1789 if (osd_id_eq_strict(id, id2))
1792 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1793 osd_scrub_file_reset(scrub,
1794 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1796 rc = osd_scrub_file_store(scrub);
1801 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1802 DTO_INDEX_UPDATE, true, 0);
1810 * It scans the /lost+found, and for the OST-object (with filter_fid
1811 * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1813 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1814 loff_t offset, __u64 ino, unsigned d_type)
1816 struct osd_ios_filldir_buf *fill_buf = buf;
1817 struct osd_thread_info *info = fill_buf->oifb_info;
1818 struct osd_device *dev = fill_buf->oifb_dev;
1819 struct lu_fid *fid = &info->oti_fid;
1820 struct osd_scrub *scrub = &dev->od_scrub;
1821 struct dentry *parent = fill_buf->oifb_dentry;
1822 struct dentry *child;
1823 struct inode *dir = parent->d_inode;
1824 struct inode *inode;
1828 /* skip any '.' started names */
1832 scrub->os_lf_scanned++;
1833 child = osd_ios_lookup_one_len(name, parent, namelen);
1834 if (IS_ERR(child)) {
1835 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1836 osd_name(dev), namelen, name, (int)PTR_ERR(child));
1840 inode = child->d_inode;
1841 if (S_ISDIR(inode->i_mode)) {
1842 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1845 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1846 "rc = %d\n", osd_name(dev), namelen, name, rc);
1850 if (!S_ISREG(inode->i_mode))
1853 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1854 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1855 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1857 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1858 "/lost+found.\n", namelen, name, PFID(fid));
1859 scrub->os_lf_repaired++;
1861 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1863 osd_name(dev), namelen, name, PFID(fid), rc);
1867 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1868 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1869 * can process them in furtuer. */
1875 scrub->os_lf_failed++;
1877 /* skip the failure to make the scanning to continue. */
1881 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1882 loff_t offset, __u64 ino, unsigned d_type)
1884 struct osd_ios_filldir_buf *fill_buf = buf;
1885 struct osd_device *dev = fill_buf->oifb_dev;
1886 struct dentry *child;
1890 /* skip any '.' started names */
1894 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1896 RETURN(PTR_ERR(child));
1898 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1900 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1901 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1902 osd_ios_varfid_fill);
1908 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1909 loff_t offset, __u64 ino, unsigned d_type)
1911 struct osd_ios_filldir_buf *fill_buf = buf;
1912 struct osd_device *dev = fill_buf->oifb_dev;
1913 const struct osd_lf_map *map;
1914 struct dentry *child;
1918 /* skip any '.' started names */
1922 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1923 if (map->olm_namelen != namelen)
1926 if (strncmp(map->olm_name, name, namelen) == 0)
1930 if (map->olm_name == NULL)
1933 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1935 RETURN(PTR_ERR(child));
1937 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1938 &map->olm_fid, map->olm_flags);
1944 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1945 loff_t offset, __u64 ino, unsigned d_type)
1947 struct osd_ios_filldir_buf *fill_buf = buf;
1948 struct osd_device *dev = fill_buf->oifb_dev;
1949 const struct osd_lf_map *map;
1950 struct dentry *child;
1954 /* skip any '.' started names */
1958 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1959 if (map->olm_namelen != namelen)
1962 if (strncmp(map->olm_name, name, namelen) == 0)
1966 if (map->olm_name == NULL)
1969 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1971 RETURN(PTR_ERR(child));
1973 if (!(map->olm_flags & OLF_NO_OI))
1974 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1975 &map->olm_fid, map->olm_flags);
1976 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1977 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1985 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1986 struct dentry *dentry, filldir_t filldir)
1988 struct osd_ios_filldir_buf buf = {
1989 #ifdef HAVE_DIR_CONTEXT
1990 .ctx.actor = filldir,
1994 .oifb_dentry = dentry };
1995 struct file *filp = &info->oti_file;
1996 struct inode *inode = dentry->d_inode;
1997 const struct file_operations *fops = inode->i_fop;
2001 LASSERT(filldir != NULL);
2004 filp->f_dentry = dentry;
2005 filp->f_mode = FMODE_64BITHASH;
2006 filp->f_mapping = inode->i_mapping;
2008 filp->private_data = NULL;
2009 set_file_inode(filp, inode);
2011 #ifdef HAVE_DIR_CONTEXT
2012 buf.ctx.pos = filp->f_pos;
2013 rc = fops->iterate(filp, &buf.ctx);
2014 filp->f_pos = buf.ctx.pos;
2016 rc = fops->readdir(filp, &buf, filldir);
2018 fops->release(inode, filp);
2024 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2025 struct dentry *dentry, filldir_t filldir)
2027 struct osd_scrub *scrub = &dev->od_scrub;
2028 struct scrub_file *sf = &scrub->os_file;
2029 struct dentry *child;
2033 /* It is existing MDT0 device. We only allow the case of object without
2034 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2035 * can generate IGIF mode FID for the object and related OI mapping. If
2036 * it is on other MDTs, then becuase file-level backup/restore, related
2037 * OI mapping may be invalid already, we do not know which is the right
2038 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2040 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2041 * "fid" xattr, then something crashed. We cannot re-generate the
2042 * FID directly, instead, the OI scrub will scan the OI structure
2043 * and try to re-generate the LMA from the OI mapping. But if the
2044 * OI mapping crashed or lost also, then we have to give up under
2045 * double failure cases. */
2046 scrub->os_convert_igif = 1;
2047 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2048 strlen(dot_lustre_name));
2049 if (IS_ERR(child)) {
2050 rc = PTR_ERR(child);
2051 if (rc == -ENOENT) {
2052 /* It is 1.8 MDT device. */
2053 if (!(sf->sf_flags & SF_UPGRADE)) {
2054 osd_scrub_file_reset(scrub,
2055 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2057 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2058 rc = osd_scrub_file_store(scrub);
2064 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2065 * so the client will get IGIF for the ".lustre" object when
2068 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2069 * it does not know whether there are some old clients cached
2070 * the ".lustre" IGIF during the upgrading. Two choices:
2072 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2073 * It will allow the old connected clients to access the
2074 * ".lustre" with cached IGIF. But it will cause others
2075 * on the MDT failed to check "fid_is_dot_lustre()".
2077 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2078 * for ".lustre" in spite of whether there are some clients
2079 * cached the ".lustre" IGIF or not. It enables the check
2080 * "fid_is_dot_lustre()" on the MDT, although it will cause
2081 * that the old connected clients cannot access the ".lustre"
2082 * with the cached IGIF.
2084 * Usually, it is rare case for the old connected clients
2085 * to access the ".lustre" with cached IGIF. So we prefer
2086 * to the solution 2). */
2087 rc = osd_ios_scan_one(info, dev, child->d_inode,
2088 &LU_DOT_LUSTRE_FID, 0);
2090 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2099 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2100 struct dentry *dentry, filldir_t filldir)
2102 struct osd_scrub *scrub = &dev->od_scrub;
2103 struct scrub_file *sf = &scrub->os_file;
2104 struct dentry *child;
2108 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2109 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2110 rc = osd_scrub_file_store(scrub);
2115 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2116 if (!IS_ERR(child)) {
2117 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2120 rc = PTR_ERR(child);
2123 if (rc != 0 && rc != -ENOENT)
2126 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2127 if (!IS_ERR(child)) {
2128 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2131 rc = PTR_ERR(child);
2140 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2141 struct osd_device *dev)
2143 struct osd_ios_item *item = NULL;
2144 scandir_t scandir = osd_ios_general_scan;
2145 filldir_t filldir = osd_ios_root_fill;
2146 struct dentry *dentry = osd_sb(dev)->s_root;
2147 const struct osd_lf_map *map = osd_lf_maps;
2151 /* Lookup IGIF in OI by force for initial OI scrub. */
2152 dev->od_igif_inoi = 1;
2155 rc = scandir(info, dev, dentry, filldir);
2157 dput(item->oii_dentry);
2164 if (list_empty(&dev->od_ios_list))
2167 item = list_entry(dev->od_ios_list.next,
2168 struct osd_ios_item, oii_list);
2169 list_del_init(&item->oii_list);
2171 LASSERT(item->oii_scandir != NULL);
2172 scandir = item->oii_scandir;
2173 filldir = item->oii_filldir;
2174 dentry = item->oii_dentry;
2177 while (!list_empty(&dev->od_ios_list)) {
2178 item = list_entry(dev->od_ios_list.next,
2179 struct osd_ios_item, oii_list);
2180 list_del_init(&item->oii_list);
2181 dput(item->oii_dentry);
2188 /* There maybe the case that the object has been removed, but its OI
2189 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2190 * file-level backup/restore. So here cleanup the stale OI mappings. */
2191 while (map->olm_name != NULL) {
2192 struct dentry *child;
2194 if (fid_is_zero(&map->olm_fid)) {
2199 child = osd_ios_lookup_one_len(map->olm_name,
2200 osd_sb(dev)->s_root,
2204 else if (PTR_ERR(child) == -ENOENT)
2205 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2206 NULL, DTO_INDEX_DELETE,
2214 char *osd_lf_fid2name(const struct lu_fid *fid)
2216 const struct osd_lf_map *map = osd_lf_maps;
2218 while (map->olm_name != NULL) {
2219 if (!lu_fid_eq(fid, &map->olm_fid)) {
2224 if (map->olm_flags & OLF_SHOW_NAME)
2225 return map->olm_name;
2233 /* OI scrub start/stop */
2235 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2237 struct osd_scrub *scrub = &dev->od_scrub;
2238 struct ptlrpc_thread *thread = &scrub->os_thread;
2239 struct l_wait_info lwi = { 0 };
2240 struct task_struct *task;
2244 /* os_lock: sync status between stop and scrub thread */
2245 spin_lock(&scrub->os_lock);
2248 if (thread_is_running(thread)) {
2249 spin_unlock(&scrub->os_lock);
2250 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2251 (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2254 osd_scrub_join(dev, flags, false);
2255 spin_lock(&scrub->os_lock);
2256 if (!thread_is_running(thread))
2259 spin_unlock(&scrub->os_lock);
2263 if (unlikely(thread_is_stopping(thread))) {
2264 spin_unlock(&scrub->os_lock);
2265 l_wait_event(thread->t_ctl_waitq,
2266 thread_is_stopped(thread),
2268 spin_lock(&scrub->os_lock);
2271 spin_unlock(&scrub->os_lock);
2273 if (scrub->os_file.sf_status == SS_COMPLETED) {
2274 if (!(flags & SS_SET_FAILOUT))
2275 flags |= SS_CLEAR_FAILOUT;
2277 if (!(flags & SS_SET_DRYRUN))
2278 flags |= SS_CLEAR_DRYRUN;
2283 scrub->os_start_flags = flags;
2284 thread_set_flags(thread, 0);
2285 task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2288 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2289 osd_scrub2name(scrub), rc);
2293 l_wait_event(thread->t_ctl_waitq,
2294 thread_is_running(thread) || thread_is_stopped(thread),
2300 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2305 /* od_otable_mutex: prevent curcurrent start/stop */
2306 mutex_lock(&dev->od_otable_mutex);
2307 rc = do_osd_scrub_start(dev, flags);
2308 mutex_unlock(&dev->od_otable_mutex);
2310 RETURN(rc == -EALREADY ? 0 : rc);
2313 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2315 struct ptlrpc_thread *thread = &scrub->os_thread;
2316 struct l_wait_info lwi = { 0 };
2318 /* os_lock: sync status between stop and scrub thread */
2319 spin_lock(&scrub->os_lock);
2320 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2321 thread_set_flags(thread, SVC_STOPPING);
2322 spin_unlock(&scrub->os_lock);
2323 wake_up_all(&thread->t_ctl_waitq);
2324 l_wait_event(thread->t_ctl_waitq,
2325 thread_is_stopped(thread),
2327 /* Do not skip the last lock/unlock, which can guarantee that
2328 * the caller cannot return until the OI scrub thread exit. */
2329 spin_lock(&scrub->os_lock);
2331 spin_unlock(&scrub->os_lock);
2334 static void osd_scrub_stop(struct osd_device *dev)
2336 /* od_otable_mutex: prevent curcurrent start/stop */
2337 mutex_lock(&dev->od_otable_mutex);
2338 dev->od_scrub.os_paused = 1;
2339 do_osd_scrub_stop(&dev->od_scrub);
2340 mutex_unlock(&dev->od_otable_mutex);
2343 /* OI scrub setup/cleanup */
2345 static const char osd_scrub_name[] = "OI_scrub";
2347 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2349 struct osd_thread_info *info = osd_oti_get(env);
2350 struct osd_scrub *scrub = &dev->od_scrub;
2351 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
2352 struct scrub_file *sf = &scrub->os_file;
2353 struct super_block *sb = osd_sb(dev);
2354 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2355 struct lvfs_run_ctxt saved;
2357 struct inode *inode;
2358 struct lu_fid *fid = &info->oti_fid;
2363 memset(scrub, 0, sizeof(*scrub));
2364 OBD_SET_CTXT_MAGIC(ctxt);
2365 ctxt->pwdmnt = dev->od_mnt;
2366 ctxt->pwd = dev->od_mnt->mnt_root;
2367 ctxt->fs = get_ds();
2369 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2370 init_rwsem(&scrub->os_rwsem);
2371 spin_lock_init(&scrub->os_lock);
2372 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2374 push_ctxt(&saved, ctxt);
2375 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2377 pop_ctxt(&saved, ctxt);
2378 RETURN(PTR_ERR(filp));
2381 inode = filp->f_dentry->d_inode;
2382 /* 'What the @fid is' is not imporatant, because the object
2383 * has no OI mapping, and only is visible inside the OSD.*/
2384 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2385 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2387 filp_close(filp, NULL);
2388 pop_ctxt(&saved, ctxt);
2392 scrub->os_inode = igrab(inode);
2393 filp_close(filp, NULL);
2394 pop_ctxt(&saved, ctxt);
2396 rc = osd_scrub_file_load(scrub);
2397 if (rc == -ENOENT) {
2398 osd_scrub_file_init(scrub, es->s_uuid);
2399 /* If the "/O" dir does not exist when mount (indicated by
2400 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2401 * then it is quite probably that the device is a new one,
2402 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2404 * For the rare case that "/O" and "OI_scrub" both lost on
2405 * an old device, it can be found and cleared later.
2407 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2408 * need to check "filter_fid_old" and to convert it to
2409 * "filter_fid" for each object, and all the IGIF should
2410 * have their FID mapping in OI files already. */
2411 if (dev->od_maybe_new)
2412 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2414 } else if (rc != 0) {
2415 GOTO(cleanup_inode, rc);
2417 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2418 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2420 } else if (sf->sf_status == SS_SCANNING) {
2421 sf->sf_status = SS_CRASHED;
2426 if (sf->sf_pos_last_checkpoint != 0)
2427 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2429 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2432 rc = osd_scrub_file_store(scrub);
2434 GOTO(cleanup_inode, rc);
2437 /* Initialize OI files. */
2438 rc = osd_oi_init(info, dev);
2440 GOTO(cleanup_inode, rc);
2442 rc = osd_initial_OI_scrub(info, dev);
2444 GOTO(cleanup_oi, rc);
2446 if (sf->sf_flags & SF_UPGRADE ||
2447 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2448 sf->sf_success_count > 0)) {
2449 dev->od_igif_inoi = 0;
2450 dev->od_check_ff = dev->od_is_ost;
2452 dev->od_igif_inoi = 1;
2453 dev->od_check_ff = 0;
2456 if (sf->sf_flags & SF_INCONSISTENT)
2457 /* The 'od_igif_inoi' will be set under the
2459 * 1) new created system, or
2460 * 2) restored from file-level backup, or
2461 * 3) the upgrading completed.
2463 * The 'od_igif_inoi' may be cleared by OI scrub
2464 * later if found that the system is upgrading. */
2465 dev->od_igif_inoi = 1;
2467 if (!dev->od_noscrub &&
2468 ((sf->sf_status == SS_PAUSED) ||
2469 (sf->sf_status == SS_CRASHED &&
2470 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2471 SF_UPGRADE | SF_AUTO)) ||
2472 (sf->sf_status == SS_INIT &&
2473 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2475 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2478 GOTO(cleanup_oi, rc);
2480 /* it is possible that dcache entries may keep objects after they are
2481 * deleted by OSD. While it looks safe this can cause object data to
2482 * stay until umount causing failures in tests calculating free space,
2483 * e.g. replay-ost-single. Since those dcache entries are not used
2484 * anymore let's just free them after use here */
2485 shrink_dcache_sb(sb);
2489 osd_oi_fini(info, dev);
2491 iput(scrub->os_inode);
2492 scrub->os_inode = NULL;
2497 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2499 struct osd_scrub *scrub = &dev->od_scrub;
2501 LASSERT(dev->od_otable_it == NULL);
2503 if (scrub->os_inode != NULL) {
2504 osd_scrub_stop(dev);
2505 iput(scrub->os_inode);
2506 scrub->os_inode = NULL;
2508 if (dev->od_oi_table != NULL)
2509 osd_oi_fini(osd_oti_get(env), dev);
2512 /* object table based iteration APIs */
2514 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2515 struct dt_object *dt, __u32 attr,
2516 struct lustre_capa *capa)
2518 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2519 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2520 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2521 struct osd_scrub *scrub = &dev->od_scrub;
2522 struct osd_otable_it *it;
2527 /* od_otable_mutex: prevent curcurrent init/fini */
2528 mutex_lock(&dev->od_otable_mutex);
2529 if (dev->od_otable_it != NULL)
2530 GOTO(out, it = ERR_PTR(-EALREADY));
2534 GOTO(out, it = ERR_PTR(-ENOMEM));
2536 dev->od_otable_it = it;
2538 it->ooi_cache.ooc_consumer_idx = -1;
2539 if (flags & DOIF_OUTUSED)
2540 it->ooi_used_outside = 1;
2542 if (flags & DOIF_RESET)
2545 if (valid & DOIV_ERROR_HANDLE) {
2546 if (flags & DOIF_FAILOUT)
2547 start |= SS_SET_FAILOUT;
2549 start |= SS_CLEAR_FAILOUT;
2552 if (valid & DOIV_DRYRUN) {
2553 if (flags & DOIF_DRYRUN)
2554 start |= SS_SET_DRYRUN;
2556 start |= SS_CLEAR_DRYRUN;
2559 rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2560 if (rc < 0 && rc != -EALREADY) {
2561 dev->od_otable_it = NULL;
2563 GOTO(out, it = ERR_PTR(rc));
2566 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2571 mutex_unlock(&dev->od_otable_mutex);
2572 return (struct dt_it *)it;
2575 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2577 struct osd_otable_it *it = (struct osd_otable_it *)di;
2578 struct osd_device *dev = it->ooi_dev;
2580 /* od_otable_mutex: prevent curcurrent init/fini */
2581 mutex_lock(&dev->od_otable_mutex);
2582 do_osd_scrub_stop(&dev->od_scrub);
2583 LASSERT(dev->od_otable_it == it);
2585 dev->od_otable_it = NULL;
2586 mutex_unlock(&dev->od_otable_mutex);
2590 static int osd_otable_it_get(const struct lu_env *env,
2591 struct dt_it *di, const struct dt_key *key)
2596 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2601 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2603 spin_lock(&scrub->os_lock);
2604 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2605 scrub->os_waiting ||
2606 !thread_is_running(&scrub->os_thread))
2607 it->ooi_waiting = 0;
2609 it->ooi_waiting = 1;
2610 spin_unlock(&scrub->os_lock);
2612 return !it->ooi_waiting;
2615 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2617 struct osd_otable_it *it = (struct osd_otable_it *)di;
2618 struct osd_device *dev = it->ooi_dev;
2619 struct osd_scrub *scrub = &dev->od_scrub;
2620 struct osd_otable_cache *ooc = &it->ooi_cache;
2621 struct ptlrpc_thread *thread = &scrub->os_thread;
2622 struct l_wait_info lwi = { 0 };
2626 LASSERT(it->ooi_user_ready);
2629 if (!thread_is_running(thread) && !it->ooi_used_outside)
2632 if (ooc->ooc_cached_items > 0) {
2633 ooc->ooc_cached_items--;
2634 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2635 ~OSD_OTABLE_IT_CACHE_MASK;
2639 if (it->ooi_all_cached) {
2640 l_wait_event(thread->t_ctl_waitq,
2641 !thread_is_running(thread),
2646 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2647 spin_lock(&scrub->os_lock);
2648 scrub->os_waiting = 0;
2649 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2650 spin_unlock(&scrub->os_lock);
2653 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2654 l_wait_event(thread->t_ctl_waitq,
2655 osd_otable_it_wakeup(scrub, it),
2658 if (!thread_is_running(thread) && !it->ooi_used_outside)
2661 rc = osd_otable_it_preload(env, it);
2668 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2669 const struct dt_it *di)
2674 static int osd_otable_it_key_size(const struct lu_env *env,
2675 const struct dt_it *di)
2677 return sizeof(__u64);
2680 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2681 struct dt_rec *rec, __u32 attr)
2683 struct osd_otable_it *it = (struct osd_otable_it *)di;
2684 struct osd_otable_cache *ooc = &it->ooi_cache;
2686 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2688 /* Filter out Invald FID already. */
2689 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2690 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2691 PFID((struct lu_fid *)rec),
2692 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2697 static __u64 osd_otable_it_store(const struct lu_env *env,
2698 const struct dt_it *di)
2700 struct osd_otable_it *it = (struct osd_otable_it *)di;
2701 struct osd_otable_cache *ooc = &it->ooi_cache;
2704 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2705 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2707 hash = ooc->ooc_pos_preload;
2712 * Set the OSD layer iteration start position as the specified hash.
2714 static int osd_otable_it_load(const struct lu_env *env,
2715 const struct dt_it *di, __u64 hash)
2717 struct osd_otable_it *it = (struct osd_otable_it *)di;
2718 struct osd_device *dev = it->ooi_dev;
2719 struct osd_otable_cache *ooc = &it->ooi_cache;
2720 struct osd_scrub *scrub = &dev->od_scrub;
2724 /* Forbid to set iteration position after iteration started. */
2725 if (it->ooi_user_ready)
2728 if (hash > OSD_OTABLE_MAX_HASH)
2729 hash = OSD_OTABLE_MAX_HASH;
2731 ooc->ooc_pos_preload = hash;
2732 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2733 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2735 it->ooi_user_ready = 1;
2736 if (!scrub->os_full_speed)
2737 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2739 /* Unplug OSD layer iteration by the first next() call. */
2740 rc = osd_otable_it_next(env, (struct dt_it *)it);
2745 static int osd_otable_it_key_rec(const struct lu_env *env,
2746 const struct dt_it *di, void *key_rec)
2751 const struct dt_index_operations osd_otable_ops = {
2753 .init = osd_otable_it_init,
2754 .fini = osd_otable_it_fini,
2755 .get = osd_otable_it_get,
2756 .put = osd_otable_it_put,
2757 .next = osd_otable_it_next,
2758 .key = osd_otable_it_key,
2759 .key_size = osd_otable_it_key_size,
2760 .rec = osd_otable_it_rec,
2761 .store = osd_otable_it_store,
2762 .load = osd_otable_it_load,
2763 .key_rec = osd_otable_it_key_rec,
2767 /* high priority inconsistent items list APIs */
2769 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
2771 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2774 struct osd_inconsistent_item *oii;
2775 struct osd_scrub *scrub = &dev->od_scrub;
2776 struct ptlrpc_thread *thread = &scrub->os_thread;
2781 if (unlikely(oii == NULL))
2784 INIT_LIST_HEAD(&oii->oii_list);
2785 oii->oii_cache = *oic;
2786 oii->oii_insert = insert;
2788 if (scrub->os_partial_scan) {
2789 __u64 now = cfs_time_current_sec();
2791 /* If there haven't been errors in a long time,
2792 * decay old count until either the errors are
2793 * gone or we reach the current interval. */
2794 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2795 scrub->os_bad_oimap_time +
2796 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2797 scrub->os_bad_oimap_count >>= 1;
2798 scrub->os_bad_oimap_time +=
2799 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2802 scrub->os_bad_oimap_time = now;
2803 if (++scrub->os_bad_oimap_count >
2804 dev->od_full_scrub_threshold_rate)
2805 scrub->os_full_scrub = 1;
2808 spin_lock(&scrub->os_lock);
2809 if (unlikely(!thread_is_running(thread))) {
2810 spin_unlock(&scrub->os_lock);
2815 if (list_empty(&scrub->os_inconsistent_items))
2817 list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2818 spin_unlock(&scrub->os_lock);
2821 wake_up_all(&thread->t_ctl_waitq);
2826 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2827 struct osd_inode_id *id)
2829 struct osd_scrub *scrub = &dev->od_scrub;
2830 struct osd_inconsistent_item *oii;
2833 spin_lock(&scrub->os_lock);
2834 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2835 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2836 *id = oii->oii_cache.oic_lid;
2837 spin_unlock(&scrub->os_lock);
2841 spin_unlock(&scrub->os_lock);
2848 static const char *scrub_status_names[] = {
2859 static const char *scrub_flags_names[] = {
2867 static const char *scrub_param_names[] = {
2873 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2880 rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2884 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2887 rc = seq_printf(m, "%s%c", names[i],
2888 bits != 0 ? ',' : '\n');
2896 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2901 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2902 cfs_time_current_sec() - time);
2904 rc = seq_printf(m, "%s: N/A\n", prefix);
2908 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2913 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2915 rc = seq_printf(m, "%s: N/A\n", prefix);
2919 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2921 struct osd_scrub *scrub = &dev->od_scrub;
2922 struct scrub_file *sf = &scrub->os_file;
2927 down_read(&scrub->os_rwsem);
2928 rc = seq_printf(m, "name: OI_scrub\n"
2932 sf->sf_magic, (int)sf->sf_oi_count,
2933 scrub_status_names[sf->sf_status]);
2937 rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2942 rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2947 rc = scrub_time_dump(m, sf->sf_time_last_complete,
2948 "time_since_last_completed");
2952 rc = scrub_time_dump(m, sf->sf_time_latest_start,
2953 "time_since_latest_start");
2957 rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2958 "time_since_last_checkpoint");
2962 rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2963 "latest_start_position");
2967 rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2968 "last_checkpoint_position");
2972 rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2973 "first_failure_position");
2977 checked = sf->sf_items_checked + scrub->os_new_checked;
2978 rc = seq_printf(m, "checked: "LPU64"\n"
2979 "updated: "LPU64"\n"
2981 "prior_updated: "LPU64"\n"
2982 "noscrub: "LPU64"\n"
2984 "success_count: %u\n",
2985 checked, sf->sf_items_updated, sf->sf_items_failed,
2986 sf->sf_items_updated_prior, sf->sf_items_noscrub,
2987 sf->sf_items_igif, sf->sf_success_count);
2992 if (thread_is_running(&scrub->os_thread)) {
2993 cfs_duration_t duration = cfs_time_current() -
2994 scrub->os_time_last_checkpoint;
2995 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
2997 __u32 rtime = sf->sf_run_time +
2998 cfs_duration_sec(duration + HALF_SEC);
3001 do_div(new_checked, duration);
3003 do_div(speed, rtime);
3004 rc = seq_printf(m, "run_time: %u seconds\n"
3005 "average_speed: "LPU64" objects/sec\n"
3006 "real-time_speed: "LPU64" objects/sec\n"
3007 "current_position: %u\n"
3008 "lf_scanned: "LPU64"\n"
3009 "lf_reparied: "LPU64"\n"
3010 "lf_failed: "LPU64"\n",
3011 rtime, speed, new_checked, scrub->os_pos_current,
3012 scrub->os_lf_scanned, scrub->os_lf_repaired,
3013 scrub->os_lf_failed);
3015 if (sf->sf_run_time != 0)
3016 do_div(speed, sf->sf_run_time);
3017 rc = seq_printf(m, "run_time: %u seconds\n"
3018 "average_speed: "LPU64" objects/sec\n"
3019 "real-time_speed: N/A\n"
3020 "current_position: N/A\n"
3021 "lf_scanned: "LPU64"\n"
3022 "lf_reparied: "LPU64"\n"
3023 "lf_failed: "LPU64"\n",
3024 sf->sf_run_time, speed, scrub->os_lf_scanned,
3025 scrub->os_lf_repaired, scrub->os_lf_failed);
3029 up_read(&scrub->os_rwsem);
3030 return (rc < 0 ? -ENOSPC : 0);