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 <linux/kthread.h>
41 #include <lustre/lustre_idl.h>
42 #include <lustre_disk.h>
43 #include <dt_object.h>
44 #include <linux/xattr.h>
46 #include "osd_internal.h"
48 #include "osd_scrub.h"
50 #define HALF_SEC msecs_to_jiffies(MSEC_PER_SEC >> 1)
52 #define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL
54 #define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */
55 #define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */
56 #define SCRUB_NEXT_EXIT 3 /* exit all the loops */
57 #define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */
58 #define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */
59 #define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */
60 #define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */
61 #define SCRUB_NEXT_NOLMA 8 /* the inode has no FID-in-LMA */
62 #define SCRUB_NEXT_OSTOBJ 9 /* for OST-object */
63 #define SCRUB_NEXT_OSTOBJ_OLD 10 /* old OST-object, no LMA or no FID-on-OST
68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
70 return container_of0(scrub, struct osd_device, od_scrub);
73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
75 return osd_sb(osd_scrub2dev(scrub));
78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
79 struct osd_otable_cache *ooc)
81 return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
86 return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
90 * update/insert/delete the specified OI mapping (@fid @id) according to the ops
92 * \retval 1, changed nothing
93 * \retval 0, changed successfully
94 * \retval -ve, on error
96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
97 struct osd_device *dev,
98 const struct lu_fid *fid,
99 const struct osd_inode_id *id,
101 enum oi_check_flags flags)
107 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
110 /* DTO_INDEX_INSERT is enough for other two ops:
111 * delete/update, but save stack. */
112 th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
113 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
116 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
117 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
118 PFID(fid), id->oii_ino, id->oii_gen, rc);
123 case DTO_INDEX_UPDATE:
124 rc = osd_oi_update(info, dev, fid, id, th, flags);
125 if (unlikely(rc == -ENOENT)) {
126 /* Some unlink thread may removed the OI mapping. */
130 case DTO_INDEX_INSERT:
131 rc = osd_oi_insert(info, dev, fid, id, th, flags);
132 if (unlikely(rc == -EEXIST)) {
134 /* XXX: There are trouble things when adding OI
135 * mapping for IGIF object, which may cause
136 * multiple objects to be mapped to the same
137 * IGIF formatted FID. Consider the following
140 * 1) The MDT is upgrading from 1.8 device.
141 * The OI scrub generates IGIF FID1 for the
142 * OBJ1 and adds the OI mapping.
144 * 2) For some reason, the OI scrub does not
145 * process all the IGIF objects completely.
147 * 3) The MDT is backuped and restored against
150 * 4) When the MDT mounts up, the OI scrub will
151 * try to rebuild the OI files. For some IGIF
152 * object, OBJ2, which was not processed by the
153 * OI scrub before the backup/restore, and the
154 * new generated IGIF formatted FID may be just
155 * the FID1, the same as OBJ1.
157 * Under such case, the OI scrub cannot know how
158 * to generate new FID for the OBJ2.
160 * Currently, we do nothing for that. One possible
161 * solution is to generate new normal FID for the
164 * Anyway, it is rare, only exists in theory. */
167 case DTO_INDEX_DELETE:
168 rc = osd_oi_delete(info, dev, fid, th, flags);
170 /* It is normal that the unlink thread has removed the
171 * OI mapping already. */
176 LASSERTF(0, "Unexpected ops %d\n", ops);
180 ldiskfs_journal_stop(th);
182 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
183 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
184 PFID(fid), id->oii_ino, id->oii_gen, rc);
189 /* OI_scrub file ops */
191 static void osd_scrub_file_to_cpu(struct scrub_file *des,
192 struct scrub_file *src)
194 memcpy(des->sf_uuid, src->sf_uuid, 16);
195 des->sf_flags = le64_to_cpu(src->sf_flags);
196 des->sf_magic = le32_to_cpu(src->sf_magic);
197 des->sf_status = le16_to_cpu(src->sf_status);
198 des->sf_param = le16_to_cpu(src->sf_param);
199 des->sf_time_last_complete =
200 le64_to_cpu(src->sf_time_last_complete);
201 des->sf_time_latest_start =
202 le64_to_cpu(src->sf_time_latest_start);
203 des->sf_time_last_checkpoint =
204 le64_to_cpu(src->sf_time_last_checkpoint);
205 des->sf_pos_latest_start =
206 le64_to_cpu(src->sf_pos_latest_start);
207 des->sf_pos_last_checkpoint =
208 le64_to_cpu(src->sf_pos_last_checkpoint);
209 des->sf_pos_first_inconsistent =
210 le64_to_cpu(src->sf_pos_first_inconsistent);
211 des->sf_items_checked =
212 le64_to_cpu(src->sf_items_checked);
213 des->sf_items_updated =
214 le64_to_cpu(src->sf_items_updated);
215 des->sf_items_failed =
216 le64_to_cpu(src->sf_items_failed);
217 des->sf_items_updated_prior =
218 le64_to_cpu(src->sf_items_updated_prior);
219 des->sf_run_time = le32_to_cpu(src->sf_run_time);
220 des->sf_success_count = le32_to_cpu(src->sf_success_count);
221 des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
222 des->sf_internal_flags = le16_to_cpu(src->sf_internal_flags);
223 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
226 static void osd_scrub_file_to_le(struct scrub_file *des,
227 struct scrub_file *src)
229 memcpy(des->sf_uuid, src->sf_uuid, 16);
230 des->sf_flags = cpu_to_le64(src->sf_flags);
231 des->sf_magic = cpu_to_le32(src->sf_magic);
232 des->sf_status = cpu_to_le16(src->sf_status);
233 des->sf_param = cpu_to_le16(src->sf_param);
234 des->sf_time_last_complete =
235 cpu_to_le64(src->sf_time_last_complete);
236 des->sf_time_latest_start =
237 cpu_to_le64(src->sf_time_latest_start);
238 des->sf_time_last_checkpoint =
239 cpu_to_le64(src->sf_time_last_checkpoint);
240 des->sf_pos_latest_start =
241 cpu_to_le64(src->sf_pos_latest_start);
242 des->sf_pos_last_checkpoint =
243 cpu_to_le64(src->sf_pos_last_checkpoint);
244 des->sf_pos_first_inconsistent =
245 cpu_to_le64(src->sf_pos_first_inconsistent);
246 des->sf_items_checked =
247 cpu_to_le64(src->sf_items_checked);
248 des->sf_items_updated =
249 cpu_to_le64(src->sf_items_updated);
250 des->sf_items_failed =
251 cpu_to_le64(src->sf_items_failed);
252 des->sf_items_updated_prior =
253 cpu_to_le64(src->sf_items_updated_prior);
254 des->sf_run_time = cpu_to_le32(src->sf_run_time);
255 des->sf_success_count = cpu_to_le32(src->sf_success_count);
256 des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
257 des->sf_internal_flags = cpu_to_le16(src->sf_internal_flags);
258 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
261 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
263 struct scrub_file *sf = &scrub->os_file;
265 memset(sf, 0, sizeof(*sf));
266 memcpy(sf->sf_uuid, uuid, 16);
267 sf->sf_magic = SCRUB_MAGIC_V1;
268 sf->sf_status = SS_INIT;
271 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
273 struct scrub_file *sf = &scrub->os_file;
275 CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
276 LPX64", add flags = "LPX64"\n",
277 osd_scrub2name(scrub), sf->sf_flags, flags);
279 memcpy(sf->sf_uuid, uuid, 16);
280 sf->sf_status = SS_INIT;
281 sf->sf_flags |= flags;
282 sf->sf_flags &= ~SF_AUTO;
284 sf->sf_time_latest_start = 0;
285 sf->sf_time_last_checkpoint = 0;
286 sf->sf_pos_latest_start = 0;
287 sf->sf_pos_last_checkpoint = 0;
288 sf->sf_pos_first_inconsistent = 0;
289 sf->sf_items_checked = 0;
290 sf->sf_items_updated = 0;
291 sf->sf_items_failed = 0;
292 if (!scrub->os_in_join)
293 sf->sf_items_updated_prior = 0;
295 sf->sf_items_noscrub = 0;
296 sf->sf_items_igif = 0;
299 static int osd_scrub_file_load(struct osd_scrub *scrub)
302 int len = sizeof(scrub->os_file_disk);
305 rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
307 struct scrub_file *sf = &scrub->os_file;
309 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
310 if (sf->sf_magic != SCRUB_MAGIC_V1) {
311 CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
312 "0x%x != 0x%x\n", osd_scrub2name(scrub),
313 sf->sf_magic, SCRUB_MAGIC_V1);
314 /* Process it as new scrub file. */
319 } else if (rc != 0) {
320 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
321 "expected = %d: rc = %d\n",
322 osd_scrub2name(scrub), len, rc);
326 /* return -ENOENT for empty scrub file case. */
333 int osd_scrub_file_store(struct osd_scrub *scrub)
335 struct osd_device *dev;
338 int len = sizeof(scrub->os_file_disk);
342 dev = container_of0(scrub, struct osd_device, od_scrub);
343 credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
344 osd_dto_credits_noquota[DTO_WRITE_BLOCK];
345 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
348 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
349 "rc = %d\n", osd_scrub2name(scrub), rc);
353 osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
354 rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
356 ldiskfs_journal_stop(jh);
358 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
359 "expected = %d: rc = %d\n",
360 osd_scrub2name(scrub), len, rc);
362 scrub->os_time_last_checkpoint = cfs_time_current();
363 scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
364 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
369 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
370 struct inode *inode, const struct lu_fid *fid)
372 struct filter_fid_old *ff = &info->oti_ff;
373 struct dentry *dentry = &info->oti_obj_dentry;
374 struct lu_fid *tfid = &info->oti_fid;
378 bool removed = false;
382 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
385 if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
386 struct ost_id *oi = &info->oti_ostid;
388 fid_to_ostid(fid, oi);
389 ostid_to_fid(tfid, oi, 0);
394 /* We want the LMA to fit into the 256-byte OST inode, so operate
396 * 1) read old XATTR_NAME_FID and save the parent FID;
397 * 2) delete the old XATTR_NAME_FID;
398 * 3) make new LMA and add it;
399 * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
401 * Making the LMA to fit into the 256-byte OST inode can save time for
402 * normal osd_check_lma() and for other OI scrub scanning in future.
403 * So it is worth to make some slow conversion here. */
404 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
405 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
408 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
409 DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
413 /* 1) read old XATTR_NAME_FID and save the parent FID */
414 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
415 if (rc == sizeof(*ff)) {
416 /* 2) delete the old XATTR_NAME_FID */
417 ll_vfs_dq_init(inode);
418 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
423 } else if (unlikely(rc == -ENODATA)) {
425 } else if (rc != sizeof(struct filter_fid)) {
426 GOTO(stop, rc = -EINVAL);
429 /* 3) make new LMA and add it */
430 rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
431 if (rc == 0 && reset)
432 size = sizeof(struct filter_fid);
433 else if (rc != 0 && removed)
434 /* If failed, we should try to add the old back. */
435 size = sizeof(struct filter_fid_old);
437 /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
441 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
443 if (rc1 != 0 && rc == 0)
450 ldiskfs_journal_stop(jh);
452 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
453 osd_name(dev), PFID(tfid), rc);
458 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
459 struct osd_idmap_cache *oic, int val)
461 struct osd_scrub *scrub = &dev->od_scrub;
462 struct scrub_file *sf = &scrub->os_file;
463 struct lu_fid *fid = &oic->oic_fid;
464 struct osd_inode_id *lid = &oic->oic_lid;
465 struct osd_inode_id *lid2 = &info->oti_id;
466 struct osd_inconsistent_item *oii = NULL;
467 struct inode *inode = NULL;
468 int ops = DTO_INDEX_UPDATE;
471 bool converted = false;
474 down_write(&scrub->os_rwsem);
475 scrub->os_new_checked++;
479 if (scrub->os_in_prior)
480 oii = list_entry(oic, struct osd_inconsistent_item,
483 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
486 if (fid_is_igif(fid))
489 if (val == SCRUB_NEXT_OSTOBJ_OLD) {
490 inode = osd_iget(info, dev, lid);
493 /* Someone removed the inode. */
494 if (rc == -ENOENT || rc == -ESTALE)
499 sf->sf_flags |= SF_UPGRADE;
500 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
501 dev->od_check_ff = 1;
502 rc = osd_scrub_convert_ff(info, dev, inode, fid);
509 if ((val == SCRUB_NEXT_NOLMA) &&
510 (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
513 if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
514 ops = DTO_INDEX_INSERT;
519 rc = osd_oi_lookup(info, dev, fid, lid2,
520 (val == SCRUB_NEXT_OSTOBJ ||
521 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
524 ops = DTO_INDEX_INSERT;
525 else if (rc != -ESTALE)
530 inode = osd_iget(info, dev, lid);
533 /* Someone removed the inode. */
534 if (rc == -ENOENT || rc == -ESTALE)
540 if (!scrub->os_partial_scan)
541 scrub->os_full_speed = 1;
543 idx = osd_oi_fid2idx(dev, fid);
545 case SCRUB_NEXT_NOLMA:
546 sf->sf_flags |= SF_UPGRADE;
547 if (!(sf->sf_param & SP_DRYRUN)) {
548 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
553 if (!(sf->sf_flags & SF_INCONSISTENT))
554 dev->od_igif_inoi = 0;
556 case SCRUB_NEXT_OSTOBJ:
557 sf->sf_flags |= SF_INCONSISTENT;
558 case SCRUB_NEXT_OSTOBJ_OLD:
561 sf->sf_flags |= SF_RECREATED;
562 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
563 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
566 } else if (osd_id_eq(lid, lid2)) {
568 sf->sf_items_updated++;
572 if (!scrub->os_partial_scan)
573 scrub->os_full_speed = 1;
575 sf->sf_flags |= SF_INCONSISTENT;
577 /* XXX: If the device is restored from file-level backup, then
578 * some IGIFs may have been already in OI files, and some
579 * may be not yet. Means upgrading from 1.8 may be partly
580 * processed, but some clients may hold some immobilized
581 * IGIFs, and use them to access related objects. Under
582 * such case, OSD does not know whether an given IGIF has
583 * been processed or to be processed, and it also cannot
584 * generate local ino#/gen# directly from the immobilized
585 * IGIF because of the backup/restore. Then force OSD to
586 * lookup the given IGIF in OI files, and if no entry,
587 * then ask the client to retry after upgrading completed.
588 * No better choice. */
589 dev->od_igif_inoi = 1;
592 rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
593 (val == SCRUB_NEXT_OSTOBJ ||
594 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
596 if (scrub->os_in_prior)
597 sf->sf_items_updated_prior++;
599 sf->sf_items_updated++;
606 sf->sf_items_failed++;
607 if (sf->sf_pos_first_inconsistent == 0 ||
608 sf->sf_pos_first_inconsistent > lid->oii_ino)
609 sf->sf_pos_first_inconsistent = lid->oii_ino;
614 /* There may be conflict unlink during the OI scrub,
615 * if happend, then remove the new added OI mapping. */
616 if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
617 unlikely(inode->i_nlink == 0))
618 osd_scrub_refresh_mapping(info, dev, fid, lid,
619 DTO_INDEX_DELETE, false,
620 (val == SCRUB_NEXT_OSTOBJ ||
621 val == SCRUB_NEXT_OSTOBJ_OLD) ?
622 OI_KNOWN_ON_OST : 0);
623 up_write(&scrub->os_rwsem);
625 if (inode != NULL && !IS_ERR(inode))
629 LASSERT(!list_empty(&oii->oii_list));
631 spin_lock(&scrub->os_lock);
632 list_del_init(&oii->oii_list);
633 spin_unlock(&scrub->os_lock);
636 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
641 static int osd_scrub_prep(struct osd_device *dev)
643 struct osd_scrub *scrub = &dev->od_scrub;
644 struct ptlrpc_thread *thread = &scrub->os_thread;
645 struct scrub_file *sf = &scrub->os_file;
646 __u32 flags = scrub->os_start_flags;
648 bool drop_dryrun = false;
651 CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
652 osd_scrub2name(scrub), flags);
654 down_write(&scrub->os_rwsem);
655 if (flags & SS_SET_FAILOUT)
656 sf->sf_param |= SP_FAILOUT;
657 else if (flags & SS_CLEAR_FAILOUT)
658 sf->sf_param &= ~SP_FAILOUT;
660 if (flags & SS_SET_DRYRUN) {
661 sf->sf_param |= SP_DRYRUN;
662 } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
663 sf->sf_param &= ~SP_DRYRUN;
667 if (flags & SS_RESET)
668 osd_scrub_file_reset(scrub,
669 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
671 if (flags & SS_AUTO_FULL) {
672 scrub->os_full_speed = 1;
673 scrub->os_partial_scan = 0;
674 sf->sf_flags |= SF_AUTO;
675 } else if (flags & SS_AUTO_PARTIAL) {
676 scrub->os_full_speed = 0;
677 scrub->os_partial_scan = 1;
678 sf->sf_flags |= SF_AUTO;
679 } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
681 scrub->os_full_speed = 1;
682 scrub->os_partial_scan = 0;
684 scrub->os_full_speed = 0;
685 scrub->os_partial_scan = 0;
688 spin_lock(&scrub->os_lock);
689 scrub->os_in_prior = 0;
690 scrub->os_waiting = 0;
691 scrub->os_paused = 0;
692 scrub->os_in_join = 0;
693 scrub->os_full_scrub = 0;
694 spin_unlock(&scrub->os_lock);
695 scrub->os_new_checked = 0;
696 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
697 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
698 else if (sf->sf_pos_last_checkpoint != 0)
699 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
701 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
703 scrub->os_pos_current = sf->sf_pos_latest_start;
704 sf->sf_status = SS_SCANNING;
705 sf->sf_time_latest_start = cfs_time_current_sec();
706 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
707 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
708 rc = osd_scrub_file_store(scrub);
710 spin_lock(&scrub->os_lock);
711 thread_set_flags(thread, SVC_RUNNING);
712 spin_unlock(&scrub->os_lock);
713 wake_up_all(&thread->t_ctl_waitq);
715 up_write(&scrub->os_rwsem);
720 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
722 struct scrub_file *sf = &scrub->os_file;
725 if (likely(cfs_time_before(cfs_time_current(),
726 scrub->os_time_next_checkpoint) ||
727 scrub->os_new_checked == 0))
730 down_write(&scrub->os_rwsem);
731 sf->sf_items_checked += scrub->os_new_checked;
732 scrub->os_new_checked = 0;
733 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
734 sf->sf_time_last_checkpoint = cfs_time_current_sec();
735 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
736 scrub->os_time_last_checkpoint);
737 rc = osd_scrub_file_store(scrub);
738 up_write(&scrub->os_rwsem);
743 static void osd_scrub_post(struct osd_scrub *scrub, int result)
745 struct scrub_file *sf = &scrub->os_file;
748 CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
749 osd_scrub2name(scrub), result);
751 down_write(&scrub->os_rwsem);
752 spin_lock(&scrub->os_lock);
753 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
754 spin_unlock(&scrub->os_lock);
755 if (scrub->os_new_checked > 0) {
756 sf->sf_items_checked += scrub->os_new_checked;
757 scrub->os_new_checked = 0;
758 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
760 sf->sf_time_last_checkpoint = cfs_time_current_sec();
762 struct osd_device *dev =
763 container_of0(scrub, struct osd_device, od_scrub);
765 dev->od_igif_inoi = 1;
766 dev->od_check_ff = 0;
767 sf->sf_status = SS_COMPLETED;
768 if (!(sf->sf_param & SP_DRYRUN)) {
769 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
770 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
771 SF_UPGRADE | SF_AUTO);
773 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
774 sf->sf_success_count++;
775 } else if (result == 0) {
776 if (scrub->os_paused)
777 sf->sf_status = SS_PAUSED;
779 sf->sf_status = SS_STOPPED;
781 sf->sf_status = SS_FAILED;
783 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
784 scrub->os_time_last_checkpoint);
785 result = osd_scrub_file_store(scrub);
786 up_write(&scrub->os_rwsem);
791 /* iteration engine */
793 struct osd_iit_param {
794 struct super_block *sb;
795 struct buffer_head *bitmap;
801 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
802 struct osd_device *dev,
803 struct osd_iit_param *param,
804 struct osd_idmap_cache **oic,
807 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
808 struct osd_device *dev,
809 struct osd_iit_param *param,
810 struct osd_idmap_cache *oic,
811 bool *noslot, int rc);
813 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
815 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
816 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
817 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
818 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
819 return SCRUB_NEXT_BREAK;
821 *pos = param->gbase + param->offset;
827 * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
828 * \retval 0: FID-on-MDT
830 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
831 struct osd_device *dev,
834 /* XXX: The initial OI scrub will scan the top level /O to generate
835 * a small local FLDB according to the <seq>. If the given FID
836 * is in the local FLDB, then it is FID-on-OST; otherwise it's
837 * quite possible for FID-on-MDT. */
839 return SCRUB_NEXT_OSTOBJ_OLD;
844 static int osd_scrub_get_fid(struct osd_thread_info *info,
845 struct osd_device *dev, struct inode *inode,
846 struct lu_fid *fid, bool scrub)
848 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
850 bool has_lma = false;
852 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
855 if (lma->lma_compat & LMAC_NOT_IN_OI ||
856 lma->lma_incompat & LMAI_AGENT)
857 return SCRUB_NEXT_CONTINUE;
859 *fid = lma->lma_self_fid;
863 if (lma->lma_compat & LMAC_FID_ON_OST)
864 return SCRUB_NEXT_OSTOBJ;
866 if (fid_is_idif(fid))
867 return SCRUB_NEXT_OSTOBJ_OLD;
869 /* For local object. */
870 if (fid_is_internal(fid))
873 /* For external visible MDT-object with non-normal FID. */
874 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
877 /* For the object with normal FID, it may be MDT-object,
878 * or may be 2.4 OST-object, need further distinguish.
879 * Fall through to next section. */
882 if (rc == -ENODATA || rc == 0) {
883 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
886 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
887 rc = SCRUB_NEXT_OSTOBJ_OLD;
893 /* It is FID-on-OST, but we do not know how
894 * to generate its FID, ignore it directly. */
895 rc = SCRUB_NEXT_CONTINUE;
897 /* It is 2.4 OST-object. */
898 rc = SCRUB_NEXT_OSTOBJ_OLD;
906 if (dev->od_scrub.os_convert_igif) {
907 lu_igif_build(fid, inode->i_ino,
908 inode->i_generation);
910 rc = SCRUB_NEXT_NOLMA;
914 /* It may be FID-on-OST, or may be FID for
915 * non-MDT0, anyway, we do not know how to
916 * generate its FID, ignore it directly. */
917 rc = SCRUB_NEXT_CONTINUE;
922 /* For OI scrub case only: the object has LMA but has no ff
923 * (or ff crashed). It may be MDT-object, may be OST-object
924 * with crashed ff. The last check is local FLDB. */
925 rc = osd_scrub_check_local_fldb(info, dev, fid);
931 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
932 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
933 struct super_block *sb, bool scrub)
939 /* Not handle the backend root object and agent parent object.
940 * They are neither visible to namespace nor have OI mappings. */
941 if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
942 pos == osd_remote_parent_ino(dev)))
943 RETURN(SCRUB_NEXT_CONTINUE);
945 osd_id_gen(lid, pos, OSD_OII_NOGEN);
946 inode = osd_iget(info, dev, lid);
949 /* The inode may be removed after bitmap searching, or the
950 * file is new created without inode initialized yet. */
951 if (rc == -ENOENT || rc == -ESTALE)
952 RETURN(SCRUB_NEXT_CONTINUE);
954 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
955 "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
961 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
962 /* Only skip it for the first OI scrub accessing. */
963 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
964 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
967 rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
976 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
977 struct osd_iit_param *param,
978 struct osd_idmap_cache **oic, const bool noslot)
980 struct osd_scrub *scrub = &dev->od_scrub;
981 struct ptlrpc_thread *thread = &scrub->os_thread;
983 struct osd_inode_id *lid;
986 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
987 struct l_wait_info lwi;
989 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
990 if (likely(lwi.lwi_timeout > 0))
991 l_wait_event(thread->t_ctl_waitq,
992 !list_empty(&scrub->os_inconsistent_items) ||
993 !thread_is_running(thread),
997 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
998 spin_lock(&scrub->os_lock);
999 thread_set_flags(thread, SVC_STOPPING);
1000 spin_unlock(&scrub->os_lock);
1001 return SCRUB_NEXT_CRASH;
1004 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1005 return SCRUB_NEXT_FATAL;
1007 if (unlikely(!thread_is_running(thread)))
1008 return SCRUB_NEXT_EXIT;
1010 if (!list_empty(&scrub->os_inconsistent_items)) {
1011 struct osd_inconsistent_item *oii;
1013 oii = list_entry(scrub->os_inconsistent_items.next,
1014 struct osd_inconsistent_item, oii_list);
1015 *oic = &oii->oii_cache;
1016 scrub->os_in_prior = 1;
1021 return SCRUB_NEXT_WAIT;
1023 rc = osd_iit_next(param, &scrub->os_pos_current);
1027 *oic = &scrub->os_oic;
1028 fid = &(*oic)->oic_fid;
1029 lid = &(*oic)->oic_lid;
1030 rc = osd_iit_iget(info, dev, fid, lid,
1031 scrub->os_pos_current, param->sb, true);
1035 static int osd_preload_next(struct osd_thread_info *info,
1036 struct osd_device *dev, struct osd_iit_param *param,
1037 struct osd_idmap_cache **oic, const bool noslot)
1039 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1040 struct osd_scrub *scrub;
1041 struct ptlrpc_thread *thread;
1044 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1048 scrub = &dev->od_scrub;
1049 thread = &scrub->os_thread;
1050 if (thread_is_running(thread) &&
1051 ooc->ooc_pos_preload >= scrub->os_pos_current)
1052 return SCRUB_NEXT_EXIT;
1054 rc = osd_iit_iget(info, dev,
1055 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1056 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1057 ooc->ooc_pos_preload, param->sb, false);
1058 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1059 * ignore the failure, so it still need to skip the inode next time. */
1060 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1065 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1067 spin_lock(&scrub->os_lock);
1068 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1069 !list_empty(&scrub->os_inconsistent_items) ||
1070 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1071 scrub->os_waiting = 0;
1073 scrub->os_waiting = 1;
1074 spin_unlock(&scrub->os_lock);
1076 return !scrub->os_waiting;
1079 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1080 struct osd_iit_param *param,
1081 struct osd_idmap_cache *oic, bool *noslot, int rc)
1083 struct l_wait_info lwi = { 0 };
1084 struct osd_scrub *scrub = &dev->od_scrub;
1085 struct scrub_file *sf = &scrub->os_file;
1086 struct ptlrpc_thread *thread = &scrub->os_thread;
1087 struct osd_otable_it *it = dev->od_otable_it;
1088 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
1091 case SCRUB_NEXT_CONTINUE:
1093 case SCRUB_NEXT_WAIT:
1095 case SCRUB_NEXT_NOSCRUB:
1096 down_write(&scrub->os_rwsem);
1097 scrub->os_new_checked++;
1098 sf->sf_items_noscrub++;
1099 up_write(&scrub->os_rwsem);
1103 rc = osd_scrub_check_update(info, dev, oic, rc);
1107 rc = osd_scrub_checkpoint(scrub);
1109 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1110 "rc = %d\n", osd_scrub2name(scrub),
1111 scrub->os_pos_current, rc);
1112 /* Continue, as long as the scrub itself can go ahead. */
1115 if (scrub->os_in_prior) {
1116 scrub->os_in_prior = 0;
1121 scrub->os_pos_current = param->gbase + ++(param->offset);
1124 if (it != NULL && it->ooi_waiting && ooc != NULL &&
1125 ooc->ooc_pos_preload < scrub->os_pos_current) {
1126 spin_lock(&scrub->os_lock);
1127 it->ooi_waiting = 0;
1128 wake_up_all(&thread->t_ctl_waitq);
1129 spin_unlock(&scrub->os_lock);
1132 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1135 if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1141 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1144 if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1151 static int osd_preload_exec(struct osd_thread_info *info,
1152 struct osd_device *dev, struct osd_iit_param *param,
1153 struct osd_idmap_cache *oic, bool *noslot, int rc)
1155 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1158 ooc->ooc_cached_items++;
1159 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1160 ~OSD_OTABLE_IT_CACHE_MASK;
1162 return rc > 0 ? 0 : rc;
1165 #define SCRUB_IT_ALL 1
1166 #define SCRUB_IT_CRASH 2
1168 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1171 struct osd_scrub *scrub = &dev->od_scrub;
1172 struct ptlrpc_thread *thread = &scrub->os_thread;
1173 struct scrub_file *sf = &scrub->os_file;
1177 LASSERT(!(flags & SS_AUTO_PARTIAL));
1179 down_write(&scrub->os_rwsem);
1180 scrub->os_in_join = 1;
1181 if (flags & SS_SET_FAILOUT)
1182 sf->sf_param |= SP_FAILOUT;
1183 else if (flags & SS_CLEAR_FAILOUT)
1184 sf->sf_param &= ~SP_FAILOUT;
1186 if (flags & SS_SET_DRYRUN)
1187 sf->sf_param |= SP_DRYRUN;
1188 else if (flags & SS_CLEAR_DRYRUN)
1189 sf->sf_param &= ~SP_DRYRUN;
1191 if (flags & SS_RESET) {
1192 osd_scrub_file_reset(scrub,
1193 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1194 inconsistent ? SF_INCONSISTENT : 0);
1195 sf->sf_status = SS_SCANNING;
1198 if (flags & SS_AUTO_FULL) {
1199 sf->sf_flags |= SF_AUTO;
1200 scrub->os_full_speed = 1;
1203 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1204 scrub->os_full_speed = 1;
1206 scrub->os_full_speed = 0;
1208 scrub->os_new_checked = 0;
1209 if (sf->sf_pos_last_checkpoint != 0)
1210 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1212 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1214 scrub->os_pos_current = sf->sf_pos_latest_start;
1215 sf->sf_time_latest_start = cfs_time_current_sec();
1216 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1217 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1218 rc = osd_scrub_file_store(scrub);
1220 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1221 "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1223 spin_lock(&scrub->os_lock);
1224 scrub->os_waiting = 0;
1225 scrub->os_paused = 0;
1226 scrub->os_partial_scan = 0;
1227 scrub->os_in_join = 0;
1228 scrub->os_full_scrub = 0;
1229 spin_unlock(&scrub->os_lock);
1230 wake_up_all(&thread->t_ctl_waitq);
1231 up_write(&scrub->os_rwsem);
1236 static int osd_inode_iteration(struct osd_thread_info *info,
1237 struct osd_device *dev, __u32 max, bool preload)
1239 struct osd_scrub *scrub = &dev->od_scrub;
1240 struct ptlrpc_thread *thread = &scrub->os_thread;
1241 struct scrub_file *sf = &scrub->os_file;
1242 osd_iit_next_policy next;
1243 osd_iit_exec_policy exec;
1246 struct osd_iit_param param = { NULL };
1247 struct l_wait_info lwi = { 0 };
1253 param.sb = osd_sb(dev);
1257 while (scrub->os_partial_scan && !scrub->os_in_join) {
1258 struct osd_idmap_cache *oic = NULL;
1260 rc = osd_scrub_next(info, dev, ¶m, &oic, noslot);
1262 case SCRUB_NEXT_EXIT:
1264 case SCRUB_NEXT_CRASH:
1265 RETURN(SCRUB_IT_CRASH);
1266 case SCRUB_NEXT_FATAL:
1268 case SCRUB_NEXT_WAIT: {
1269 struct kstatfs *ksfs = &info->oti_ksfs;
1272 if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1273 unlikely(sf->sf_items_updated_prior == 0))
1276 if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1277 scrub->os_full_scrub) {
1278 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1283 rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1285 __u64 used = ksfs->f_files - ksfs->f_ffree;
1287 do_div(used, sf->sf_items_updated_prior);
1288 /* If we hit too much inconsistent OI
1289 * mappings during the partial scan,
1290 * then scan the device completely. */
1291 if (used < dev->od_full_scrub_ratio) {
1293 SS_AUTO_FULL | SS_RESET, true);
1299 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1303 saved_flags = sf->sf_flags;
1304 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1305 SF_UPGRADE | SF_AUTO);
1306 sf->sf_status = SS_COMPLETED;
1307 l_wait_event(thread->t_ctl_waitq,
1308 !thread_is_running(thread) ||
1309 !scrub->os_partial_scan ||
1310 scrub->os_in_join ||
1311 !list_empty(&scrub->os_inconsistent_items),
1313 sf->sf_flags = saved_flags;
1314 sf->sf_status = SS_SCANNING;
1316 if (unlikely(!thread_is_running(thread)))
1319 if (!scrub->os_partial_scan || scrub->os_in_join)
1325 LASSERTF(rc == 0, "rc = %d\n", rc);
1327 osd_scrub_exec(info, dev, ¶m, oic, &noslot, rc);
1334 l_wait_event(thread->t_ctl_waitq,
1335 !thread_is_running(thread) || !scrub->os_in_join,
1338 if (unlikely(!thread_is_running(thread)))
1344 next = osd_scrub_next;
1345 exec = osd_scrub_exec;
1346 pos = &scrub->os_pos_current;
1347 count = &scrub->os_new_checked;
1349 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1351 next = osd_preload_next;
1352 exec = osd_preload_exec;
1353 pos = &ooc->ooc_pos_preload;
1354 count = &ooc->ooc_cached_items;
1356 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1358 while (*pos <= limit && *count < max) {
1359 struct osd_idmap_cache *oic = NULL;
1360 struct ldiskfs_group_desc *desc;
1362 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1363 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1367 ldiskfs_lock_group(param.sb, param.bg);
1368 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1369 ldiskfs_unlock_group(param.sb, param.bg);
1370 *pos = 1 + (param.bg + 1) *
1371 LDISKFS_INODES_PER_GROUP(param.sb);
1374 ldiskfs_unlock_group(param.sb, param.bg);
1376 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1377 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1378 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1379 if (param.bitmap == NULL) {
1380 CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1381 "scrub will stop, urgent mode\n",
1382 osd_scrub2name(scrub), (__u32)param.bg);
1386 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1389 ldiskfs_itable_unused_count(param.sb, desc) >
1390 LDISKFS_INODES_PER_GROUP(param.sb))
1393 rc = next(info, dev, ¶m, &oic, noslot);
1395 case SCRUB_NEXT_BREAK:
1397 case SCRUB_NEXT_EXIT:
1398 brelse(param.bitmap);
1400 case SCRUB_NEXT_CRASH:
1401 brelse(param.bitmap);
1402 RETURN(SCRUB_IT_CRASH);
1403 case SCRUB_NEXT_FATAL:
1404 brelse(param.bitmap);
1408 rc = exec(info, dev, ¶m, oic, &noslot, rc);
1410 brelse(param.bitmap);
1416 brelse(param.bitmap);
1420 RETURN(SCRUB_IT_ALL);
1424 static int osd_otable_it_preload(const struct lu_env *env,
1425 struct osd_otable_it *it)
1427 struct osd_device *dev = it->ooi_dev;
1428 struct osd_scrub *scrub = &dev->od_scrub;
1429 struct osd_otable_cache *ooc = &it->ooi_cache;
1433 rc = osd_inode_iteration(osd_oti_get(env), dev,
1434 OSD_OTABLE_IT_CACHE_SIZE, true);
1435 if (rc == SCRUB_IT_ALL)
1436 it->ooi_all_cached = 1;
1438 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1439 spin_lock(&scrub->os_lock);
1440 scrub->os_waiting = 0;
1441 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1442 spin_unlock(&scrub->os_lock);
1445 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1448 static int osd_scrub_main(void *args)
1451 struct osd_device *dev = (struct osd_device *)args;
1452 struct osd_scrub *scrub = &dev->od_scrub;
1453 struct ptlrpc_thread *thread = &scrub->os_thread;
1457 rc = lu_env_init(&env, LCT_LOCAL);
1459 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1460 osd_scrub2name(scrub), rc);
1464 rc = osd_scrub_prep(dev);
1466 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1467 osd_scrub2name(scrub), rc);
1471 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1472 struct l_wait_info lwi = { 0 };
1473 struct osd_otable_it *it = dev->od_otable_it;
1474 struct osd_otable_cache *ooc = &it->ooi_cache;
1476 l_wait_event(thread->t_ctl_waitq,
1477 it->ooi_user_ready || !thread_is_running(thread),
1479 if (unlikely(!thread_is_running(thread)))
1482 scrub->os_pos_current = ooc->ooc_pos_preload;
1485 CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1486 osd_scrub2name(scrub), scrub->os_start_flags,
1487 scrub->os_pos_current);
1489 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1490 if (unlikely(rc == SCRUB_IT_CRASH))
1491 GOTO(out, rc = -EINVAL);
1495 osd_scrub_post(scrub, rc);
1496 CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1497 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1500 while (!list_empty(&scrub->os_inconsistent_items)) {
1501 struct osd_inconsistent_item *oii;
1503 oii = list_entry(scrub->os_inconsistent_items.next,
1504 struct osd_inconsistent_item, oii_list);
1505 list_del_init(&oii->oii_list);
1511 spin_lock(&scrub->os_lock);
1512 thread_set_flags(thread, SVC_STOPPED);
1513 wake_up_all(&thread->t_ctl_waitq);
1514 spin_unlock(&scrub->os_lock);
1518 /* initial OI scrub */
1520 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1521 struct dentry *, filldir_t filldir);
1523 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1524 loff_t offset, __u64 ino, unsigned d_type);
1525 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1526 loff_t offset, __u64 ino, unsigned d_type);
1527 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1528 loff_t offset, __u64 ino, unsigned d_type);
1529 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1530 loff_t offset, __u64 ino, unsigned d_type);
1533 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1534 struct dentry *dentry, filldir_t filldir);
1536 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1537 struct dentry *dentry, filldir_t filldir);
1540 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1541 struct dentry *dentry, filldir_t filldir);
1544 OLF_SCAN_SUBITEMS = 0x0001,
1545 OLF_HIDE_FID = 0x0002,
1546 OLF_SHOW_NAME = 0x0004,
1548 OLF_IDX_IN_FID = 0x0010,
1553 struct lu_fid olm_fid;
1556 scandir_t olm_scandir;
1557 filldir_t olm_filldir;
1560 /* Add the new introduced local files in the list in the future. */
1561 static const struct osd_lf_map osd_lf_maps[] = {
1563 { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1564 sizeof(CATLIST) - 1, NULL, NULL },
1567 { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1568 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1569 osd_ios_general_scan, osd_ios_varfid_fill },
1571 /* NIDTBL_VERSIONS */
1572 { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1573 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1574 osd_ios_varfid_fill },
1577 { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1580 { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1581 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1582 osd_ios_ROOT_scan, NULL },
1584 /* changelog_catalog */
1585 { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1588 /* changelog_users */
1589 { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1593 { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1594 sizeof("fld") - 1, NULL, NULL },
1597 { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1598 sizeof(LAST_RCVD) - 1, NULL, NULL },
1601 { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1602 sizeof(REPLY_DATA) - 1, NULL, NULL },
1605 { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1606 sizeof(LOV_OBJID) - 1, NULL, NULL },
1609 { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1610 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1613 { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1614 osd_ios_general_scan, osd_ios_varfid_fill },
1617 { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1618 osd_ios_general_scan, osd_ios_varfid_fill },
1621 { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1622 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1625 { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1626 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1629 { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1630 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1633 { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1634 osd_ios_general_scan, osd_ios_varfid_fill },
1636 /* lfsck_bookmark */
1637 { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1641 { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1644 /* lfsck_namespace */
1645 { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1648 /* OBJECTS, upgrade from old device */
1649 { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1650 osd_ios_OBJECTS_scan, NULL },
1652 /* lquota_v2.user, upgrade from old device */
1653 { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1656 /* lquota_v2.group, upgrade from old device */
1657 { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1660 /* LAST_GROUP, upgrade from old device */
1661 { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1662 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1664 /* committed batchid for cross-MDT operation */
1665 { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1666 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1668 /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1669 * and f_oid = index for their log files. See lu_update_log{_dir}_fid()
1670 * for more details. */
1673 { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1674 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1677 /* update_log_dir */
1678 { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1679 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1680 sizeof("update_log_dir") - 1,
1681 osd_ios_general_scan, osd_ios_uld_fill },
1684 { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1685 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1686 osd_ios_general_scan, osd_ios_lf_fill },
1688 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1691 /* Add the new introduced files under .lustre/ in the list in the future. */
1692 static const struct osd_lf_map osd_dl_maps[] = {
1694 { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1695 sizeof("fid") - 1, NULL, NULL },
1697 /* .lustre/lost+found */
1698 { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1699 sizeof("lost+found") - 1, NULL, NULL },
1701 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1704 struct osd_ios_item {
1705 struct list_head oii_list;
1706 struct dentry *oii_dentry;
1707 scandir_t oii_scandir;
1708 filldir_t oii_filldir;
1711 struct osd_ios_filldir_buf {
1712 #ifdef HAVE_DIR_CONTEXT
1713 /* please keep it as first member */
1714 struct dir_context ctx;
1716 struct osd_thread_info *oifb_info;
1717 struct osd_device *oifb_dev;
1718 struct dentry *oifb_dentry;
1721 static inline struct dentry *
1722 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1724 struct dentry *dentry;
1726 dentry = ll_lookup_one_len(name, parent, namelen);
1727 if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1729 return ERR_PTR(-ENOENT);
1736 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1737 scandir_t scandir, filldir_t filldir)
1739 struct osd_ios_item *item;
1742 OBD_ALLOC_PTR(item);
1746 INIT_LIST_HEAD(&item->oii_list);
1747 item->oii_dentry = dget(dentry);
1748 item->oii_scandir = scandir;
1749 item->oii_filldir = filldir;
1750 list_add_tail(&item->oii_list, &dev->od_ios_list);
1756 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1758 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1759 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1760 * reference the inode, or fixed if it is missing or references another inode.
1763 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1764 struct inode *inode, const struct lu_fid *fid, int flags)
1766 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1767 struct osd_inode_id *id = &info->oti_id;
1768 struct osd_inode_id *id2 = &info->oti_id2;
1769 struct osd_scrub *scrub = &dev->od_scrub;
1770 struct scrub_file *sf = &scrub->os_file;
1775 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1776 if (rc != 0 && rc != -ENODATA) {
1777 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1778 "rc = %d\n", osd_name(dev), rc);
1783 osd_id_gen(id, inode->i_ino, inode->i_generation);
1784 if (rc == -ENODATA) {
1785 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1786 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1789 if (flags & OLF_IDX_IN_FID) {
1790 LASSERT(dev->od_index >= 0);
1792 tfid.f_oid = dev->od_index;
1795 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1797 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1798 "scrub: rc = %d\n", osd_name(dev), rc);
1803 if (lma->lma_compat & LMAC_NOT_IN_OI)
1806 tfid = lma->lma_self_fid;
1809 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1814 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1815 DTO_INDEX_INSERT, true, 0);
1822 if (osd_id_eq_strict(id, id2))
1825 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1826 osd_scrub_file_reset(scrub,
1827 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1829 rc = osd_scrub_file_store(scrub);
1834 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1835 DTO_INDEX_UPDATE, true, 0);
1843 * It scans the /lost+found, and for the OST-object (with filter_fid
1844 * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1846 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1847 loff_t offset, __u64 ino, unsigned d_type)
1849 struct osd_ios_filldir_buf *fill_buf = buf;
1850 struct osd_thread_info *info = fill_buf->oifb_info;
1851 struct osd_device *dev = fill_buf->oifb_dev;
1852 struct lu_fid *fid = &info->oti_fid;
1853 struct osd_scrub *scrub = &dev->od_scrub;
1854 struct dentry *parent = fill_buf->oifb_dentry;
1855 struct dentry *child;
1856 struct inode *dir = parent->d_inode;
1857 struct inode *inode;
1861 /* skip any '.' started names */
1865 scrub->os_lf_scanned++;
1866 child = osd_ios_lookup_one_len(name, parent, namelen);
1867 if (IS_ERR(child)) {
1868 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1869 osd_name(dev), namelen, name, (int)PTR_ERR(child));
1873 inode = child->d_inode;
1874 if (S_ISDIR(inode->i_mode)) {
1875 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1878 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1879 "rc = %d\n", osd_name(dev), namelen, name, rc);
1883 if (!S_ISREG(inode->i_mode))
1886 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1887 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1888 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1890 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1891 "/lost+found.\n", namelen, name, PFID(fid));
1892 scrub->os_lf_repaired++;
1894 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1896 osd_name(dev), namelen, name, PFID(fid), rc);
1900 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1901 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1902 * can process them in furtuer. */
1908 scrub->os_lf_failed++;
1910 /* skip the failure to make the scanning to continue. */
1914 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1915 loff_t offset, __u64 ino, unsigned d_type)
1917 struct osd_ios_filldir_buf *fill_buf = buf;
1918 struct osd_device *dev = fill_buf->oifb_dev;
1919 struct dentry *child;
1923 /* skip any '.' started names */
1927 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1929 RETURN(PTR_ERR(child));
1931 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1933 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1934 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1935 osd_ios_varfid_fill);
1941 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1942 loff_t offset, __u64 ino, unsigned d_type)
1944 struct osd_ios_filldir_buf *fill_buf = buf;
1945 struct osd_device *dev = fill_buf->oifb_dev;
1946 const struct osd_lf_map *map;
1947 struct dentry *child;
1951 /* skip any '.' started names */
1955 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1956 if (map->olm_namelen != namelen)
1959 if (strncmp(map->olm_name, name, namelen) == 0)
1963 if (map->olm_name == NULL)
1966 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1968 RETURN(PTR_ERR(child));
1970 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1971 &map->olm_fid, map->olm_flags);
1977 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1978 loff_t offset, __u64 ino, unsigned d_type)
1980 struct osd_ios_filldir_buf *fill_buf = buf;
1981 struct dentry *child;
1986 /* skip any non-DFID format name */
1990 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1992 RETURN(PTR_ERR(child));
1994 /* skip the start '[' */
1995 sscanf(&name[1], SFID, RFID(&tfid));
1996 if (fid_is_sane(&tfid))
1997 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
1998 child->d_inode, &tfid, 0);
2006 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2007 loff_t offset, __u64 ino, unsigned d_type)
2009 struct osd_ios_filldir_buf *fill_buf = buf;
2010 struct osd_device *dev = fill_buf->oifb_dev;
2011 const struct osd_lf_map *map;
2012 struct dentry *child;
2016 /* skip any '.' started names */
2020 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2021 if (map->olm_namelen != namelen)
2024 if (strncmp(map->olm_name, name, namelen) == 0)
2028 if (map->olm_name == NULL)
2031 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2033 RETURN(PTR_ERR(child));
2035 if (!(map->olm_flags & OLF_NO_OI))
2036 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2037 &map->olm_fid, map->olm_flags);
2038 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2039 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2047 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2048 struct dentry *dentry, filldir_t filldir)
2050 struct osd_ios_filldir_buf buf = {
2051 #ifdef HAVE_DIR_CONTEXT
2052 .ctx.actor = filldir,
2056 .oifb_dentry = dentry };
2057 struct file *filp = &info->oti_file;
2058 struct inode *inode = dentry->d_inode;
2059 const struct file_operations *fops = inode->i_fop;
2063 LASSERT(filldir != NULL);
2066 filp->f_path.dentry = dentry;
2067 filp->f_mode = FMODE_64BITHASH;
2068 filp->f_mapping = inode->i_mapping;
2070 filp->private_data = NULL;
2071 set_file_inode(filp, inode);
2073 #ifdef HAVE_DIR_CONTEXT
2074 buf.ctx.pos = filp->f_pos;
2075 rc = fops->iterate(filp, &buf.ctx);
2076 filp->f_pos = buf.ctx.pos;
2078 rc = fops->readdir(filp, &buf, filldir);
2080 fops->release(inode, filp);
2086 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2087 struct dentry *dentry, filldir_t filldir)
2089 struct osd_scrub *scrub = &dev->od_scrub;
2090 struct scrub_file *sf = &scrub->os_file;
2091 struct dentry *child;
2095 /* It is existing MDT0 device. We only allow the case of object without
2096 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2097 * can generate IGIF mode FID for the object and related OI mapping. If
2098 * it is on other MDTs, then becuase file-level backup/restore, related
2099 * OI mapping may be invalid already, we do not know which is the right
2100 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2102 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2103 * "fid" xattr, then something crashed. We cannot re-generate the
2104 * FID directly, instead, the OI scrub will scan the OI structure
2105 * and try to re-generate the LMA from the OI mapping. But if the
2106 * OI mapping crashed or lost also, then we have to give up under
2107 * double failure cases. */
2108 scrub->os_convert_igif = 1;
2109 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2110 strlen(dot_lustre_name));
2111 if (IS_ERR(child)) {
2112 rc = PTR_ERR(child);
2113 if (rc == -ENOENT) {
2114 /* It is 1.8 MDT device. */
2115 if (!(sf->sf_flags & SF_UPGRADE)) {
2116 osd_scrub_file_reset(scrub,
2117 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2119 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2120 rc = osd_scrub_file_store(scrub);
2126 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2127 * so the client will get IGIF for the ".lustre" object when
2130 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2131 * it does not know whether there are some old clients cached
2132 * the ".lustre" IGIF during the upgrading. Two choices:
2134 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2135 * It will allow the old connected clients to access the
2136 * ".lustre" with cached IGIF. But it will cause others
2137 * on the MDT failed to check "fid_is_dot_lustre()".
2139 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2140 * for ".lustre" in spite of whether there are some clients
2141 * cached the ".lustre" IGIF or not. It enables the check
2142 * "fid_is_dot_lustre()" on the MDT, although it will cause
2143 * that the old connected clients cannot access the ".lustre"
2144 * with the cached IGIF.
2146 * Usually, it is rare case for the old connected clients
2147 * to access the ".lustre" with cached IGIF. So we prefer
2148 * to the solution 2). */
2149 rc = osd_ios_scan_one(info, dev, child->d_inode,
2150 &LU_DOT_LUSTRE_FID, 0);
2152 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2161 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2162 struct dentry *dentry, filldir_t filldir)
2164 struct osd_scrub *scrub = &dev->od_scrub;
2165 struct scrub_file *sf = &scrub->os_file;
2166 struct dentry *child;
2170 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2171 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2172 rc = osd_scrub_file_store(scrub);
2177 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2178 if (!IS_ERR(child)) {
2179 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2182 rc = PTR_ERR(child);
2185 if (rc != 0 && rc != -ENOENT)
2188 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2189 if (!IS_ERR(child)) {
2190 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2193 rc = PTR_ERR(child);
2202 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2203 struct osd_device *dev)
2205 struct osd_ios_item *item = NULL;
2206 scandir_t scandir = osd_ios_general_scan;
2207 filldir_t filldir = osd_ios_root_fill;
2208 struct dentry *dentry = osd_sb(dev)->s_root;
2209 const struct osd_lf_map *map = osd_lf_maps;
2213 /* Lookup IGIF in OI by force for initial OI scrub. */
2214 dev->od_igif_inoi = 1;
2217 rc = scandir(info, dev, dentry, filldir);
2219 dput(item->oii_dentry);
2226 if (list_empty(&dev->od_ios_list))
2229 item = list_entry(dev->od_ios_list.next,
2230 struct osd_ios_item, oii_list);
2231 list_del_init(&item->oii_list);
2233 LASSERT(item->oii_scandir != NULL);
2234 scandir = item->oii_scandir;
2235 filldir = item->oii_filldir;
2236 dentry = item->oii_dentry;
2239 while (!list_empty(&dev->od_ios_list)) {
2240 item = list_entry(dev->od_ios_list.next,
2241 struct osd_ios_item, oii_list);
2242 list_del_init(&item->oii_list);
2243 dput(item->oii_dentry);
2250 /* There maybe the case that the object has been removed, but its OI
2251 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2252 * file-level backup/restore. So here cleanup the stale OI mappings. */
2253 while (map->olm_name != NULL) {
2254 struct dentry *child;
2256 if (fid_is_zero(&map->olm_fid)) {
2261 child = osd_ios_lookup_one_len(map->olm_name,
2262 osd_sb(dev)->s_root,
2266 else if (PTR_ERR(child) == -ENOENT)
2267 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2268 NULL, DTO_INDEX_DELETE,
2276 char *osd_lf_fid2name(const struct lu_fid *fid)
2278 const struct osd_lf_map *map = osd_lf_maps;
2280 while (map->olm_name != NULL) {
2281 if (!lu_fid_eq(fid, &map->olm_fid)) {
2286 if (map->olm_flags & OLF_SHOW_NAME)
2287 return map->olm_name;
2295 /* OI scrub start/stop */
2297 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2299 struct osd_scrub *scrub = &dev->od_scrub;
2300 struct ptlrpc_thread *thread = &scrub->os_thread;
2301 struct l_wait_info lwi = { 0 };
2302 struct task_struct *task;
2306 /* os_lock: sync status between stop and scrub thread */
2307 spin_lock(&scrub->os_lock);
2310 if (thread_is_running(thread)) {
2311 spin_unlock(&scrub->os_lock);
2312 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2313 (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2316 osd_scrub_join(dev, flags, false);
2317 spin_lock(&scrub->os_lock);
2318 if (!thread_is_running(thread))
2321 spin_unlock(&scrub->os_lock);
2325 if (unlikely(thread_is_stopping(thread))) {
2326 spin_unlock(&scrub->os_lock);
2327 l_wait_event(thread->t_ctl_waitq,
2328 thread_is_stopped(thread),
2330 spin_lock(&scrub->os_lock);
2333 spin_unlock(&scrub->os_lock);
2335 if (scrub->os_file.sf_status == SS_COMPLETED) {
2336 if (!(flags & SS_SET_FAILOUT))
2337 flags |= SS_CLEAR_FAILOUT;
2339 if (!(flags & SS_SET_DRYRUN))
2340 flags |= SS_CLEAR_DRYRUN;
2345 scrub->os_start_flags = flags;
2346 thread_set_flags(thread, 0);
2347 task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2350 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2351 osd_scrub2name(scrub), rc);
2355 l_wait_event(thread->t_ctl_waitq,
2356 thread_is_running(thread) || thread_is_stopped(thread),
2362 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2367 /* od_otable_mutex: prevent curcurrent start/stop */
2368 mutex_lock(&dev->od_otable_mutex);
2369 rc = do_osd_scrub_start(dev, flags);
2370 mutex_unlock(&dev->od_otable_mutex);
2372 RETURN(rc == -EALREADY ? 0 : rc);
2375 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2377 struct ptlrpc_thread *thread = &scrub->os_thread;
2378 struct l_wait_info lwi = { 0 };
2380 /* os_lock: sync status between stop and scrub thread */
2381 spin_lock(&scrub->os_lock);
2382 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2383 thread_set_flags(thread, SVC_STOPPING);
2384 spin_unlock(&scrub->os_lock);
2385 wake_up_all(&thread->t_ctl_waitq);
2386 l_wait_event(thread->t_ctl_waitq,
2387 thread_is_stopped(thread),
2389 /* Do not skip the last lock/unlock, which can guarantee that
2390 * the caller cannot return until the OI scrub thread exit. */
2391 spin_lock(&scrub->os_lock);
2393 spin_unlock(&scrub->os_lock);
2396 static void osd_scrub_stop(struct osd_device *dev)
2398 /* od_otable_mutex: prevent curcurrent start/stop */
2399 mutex_lock(&dev->od_otable_mutex);
2400 dev->od_scrub.os_paused = 1;
2401 do_osd_scrub_stop(&dev->od_scrub);
2402 mutex_unlock(&dev->od_otable_mutex);
2405 /* OI scrub setup/cleanup */
2407 static const char osd_scrub_name[] = "OI_scrub";
2409 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2411 struct osd_thread_info *info = osd_oti_get(env);
2412 struct osd_scrub *scrub = &dev->od_scrub;
2413 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
2414 struct scrub_file *sf = &scrub->os_file;
2415 struct super_block *sb = osd_sb(dev);
2416 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2417 struct lvfs_run_ctxt saved;
2419 struct inode *inode;
2420 struct lu_fid *fid = &info->oti_fid;
2425 memset(scrub, 0, sizeof(*scrub));
2426 OBD_SET_CTXT_MAGIC(ctxt);
2427 ctxt->pwdmnt = dev->od_mnt;
2428 ctxt->pwd = dev->od_mnt->mnt_root;
2429 ctxt->fs = get_ds();
2431 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2432 init_rwsem(&scrub->os_rwsem);
2433 spin_lock_init(&scrub->os_lock);
2434 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2436 push_ctxt(&saved, ctxt);
2437 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2439 pop_ctxt(&saved, ctxt);
2440 RETURN(PTR_ERR(filp));
2443 inode = filp->f_path.dentry->d_inode;
2444 /* 'What the @fid is' is not imporatant, because the object
2445 * has no OI mapping, and only is visible inside the OSD.*/
2446 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2447 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2449 filp_close(filp, NULL);
2450 pop_ctxt(&saved, ctxt);
2454 scrub->os_inode = igrab(inode);
2455 filp_close(filp, NULL);
2456 pop_ctxt(&saved, ctxt);
2458 rc = osd_scrub_file_load(scrub);
2459 if (rc == -ENOENT) {
2460 osd_scrub_file_init(scrub, es->s_uuid);
2461 /* If the "/O" dir does not exist when mount (indicated by
2462 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2463 * then it is quite probably that the device is a new one,
2464 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2466 * For the rare case that "/O" and "OI_scrub" both lost on
2467 * an old device, it can be found and cleared later.
2469 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2470 * need to check "filter_fid_old" and to convert it to
2471 * "filter_fid" for each object, and all the IGIF should
2472 * have their FID mapping in OI files already. */
2473 if (dev->od_maybe_new)
2474 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2476 } else if (rc != 0) {
2477 GOTO(cleanup_inode, rc);
2479 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2480 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2482 } else if (sf->sf_status == SS_SCANNING) {
2483 sf->sf_status = SS_CRASHED;
2488 if (sf->sf_pos_last_checkpoint != 0)
2489 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2491 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2494 rc = osd_scrub_file_store(scrub);
2496 GOTO(cleanup_inode, rc);
2499 /* Initialize OI files. */
2500 rc = osd_oi_init(info, dev);
2502 GOTO(cleanup_inode, rc);
2504 rc = osd_initial_OI_scrub(info, dev);
2506 GOTO(cleanup_oi, rc);
2508 if (sf->sf_flags & SF_UPGRADE ||
2509 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2510 sf->sf_success_count > 0)) {
2511 dev->od_igif_inoi = 0;
2512 dev->od_check_ff = dev->od_is_ost;
2514 dev->od_igif_inoi = 1;
2515 dev->od_check_ff = 0;
2518 if (sf->sf_flags & SF_INCONSISTENT)
2519 /* The 'od_igif_inoi' will be set under the
2521 * 1) new created system, or
2522 * 2) restored from file-level backup, or
2523 * 3) the upgrading completed.
2525 * The 'od_igif_inoi' may be cleared by OI scrub
2526 * later if found that the system is upgrading. */
2527 dev->od_igif_inoi = 1;
2529 if (!dev->od_noscrub &&
2530 ((sf->sf_status == SS_PAUSED) ||
2531 (sf->sf_status == SS_CRASHED &&
2532 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2533 SF_UPGRADE | SF_AUTO)) ||
2534 (sf->sf_status == SS_INIT &&
2535 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2537 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2540 GOTO(cleanup_oi, rc);
2542 /* it is possible that dcache entries may keep objects after they are
2543 * deleted by OSD. While it looks safe this can cause object data to
2544 * stay until umount causing failures in tests calculating free space,
2545 * e.g. replay-ost-single. Since those dcache entries are not used
2546 * anymore let's just free them after use here */
2547 shrink_dcache_sb(sb);
2551 osd_oi_fini(info, dev);
2553 iput(scrub->os_inode);
2554 scrub->os_inode = NULL;
2559 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2561 struct osd_scrub *scrub = &dev->od_scrub;
2563 LASSERT(dev->od_otable_it == NULL);
2565 if (scrub->os_inode != NULL) {
2566 osd_scrub_stop(dev);
2567 iput(scrub->os_inode);
2568 scrub->os_inode = NULL;
2570 if (dev->od_oi_table != NULL)
2571 osd_oi_fini(osd_oti_get(env), dev);
2574 /* object table based iteration APIs */
2576 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2577 struct dt_object *dt, __u32 attr)
2579 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2580 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2581 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2582 struct osd_scrub *scrub = &dev->od_scrub;
2583 struct osd_otable_it *it;
2588 /* od_otable_mutex: prevent curcurrent init/fini */
2589 mutex_lock(&dev->od_otable_mutex);
2590 if (dev->od_otable_it != NULL)
2591 GOTO(out, it = ERR_PTR(-EALREADY));
2595 GOTO(out, it = ERR_PTR(-ENOMEM));
2597 dev->od_otable_it = it;
2599 it->ooi_cache.ooc_consumer_idx = -1;
2600 if (flags & DOIF_OUTUSED)
2601 it->ooi_used_outside = 1;
2603 if (flags & DOIF_RESET)
2606 if (valid & DOIV_ERROR_HANDLE) {
2607 if (flags & DOIF_FAILOUT)
2608 start |= SS_SET_FAILOUT;
2610 start |= SS_CLEAR_FAILOUT;
2613 if (valid & DOIV_DRYRUN) {
2614 if (flags & DOIF_DRYRUN)
2615 start |= SS_SET_DRYRUN;
2617 start |= SS_CLEAR_DRYRUN;
2620 rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2621 if (rc < 0 && rc != -EALREADY) {
2622 dev->od_otable_it = NULL;
2624 GOTO(out, it = ERR_PTR(rc));
2627 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2632 mutex_unlock(&dev->od_otable_mutex);
2633 return (struct dt_it *)it;
2636 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2638 struct osd_otable_it *it = (struct osd_otable_it *)di;
2639 struct osd_device *dev = it->ooi_dev;
2641 /* od_otable_mutex: prevent curcurrent init/fini */
2642 mutex_lock(&dev->od_otable_mutex);
2643 do_osd_scrub_stop(&dev->od_scrub);
2644 LASSERT(dev->od_otable_it == it);
2646 dev->od_otable_it = NULL;
2647 mutex_unlock(&dev->od_otable_mutex);
2651 static int osd_otable_it_get(const struct lu_env *env,
2652 struct dt_it *di, const struct dt_key *key)
2657 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2662 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2664 spin_lock(&scrub->os_lock);
2665 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2666 scrub->os_waiting ||
2667 !thread_is_running(&scrub->os_thread))
2668 it->ooi_waiting = 0;
2670 it->ooi_waiting = 1;
2671 spin_unlock(&scrub->os_lock);
2673 return !it->ooi_waiting;
2676 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2678 struct osd_otable_it *it = (struct osd_otable_it *)di;
2679 struct osd_device *dev = it->ooi_dev;
2680 struct osd_scrub *scrub = &dev->od_scrub;
2681 struct osd_otable_cache *ooc = &it->ooi_cache;
2682 struct ptlrpc_thread *thread = &scrub->os_thread;
2683 struct l_wait_info lwi = { 0 };
2687 LASSERT(it->ooi_user_ready);
2690 if (!thread_is_running(thread) && !it->ooi_used_outside)
2693 if (ooc->ooc_cached_items > 0) {
2694 ooc->ooc_cached_items--;
2695 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2696 ~OSD_OTABLE_IT_CACHE_MASK;
2700 if (it->ooi_all_cached) {
2701 l_wait_event(thread->t_ctl_waitq,
2702 !thread_is_running(thread),
2707 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2708 spin_lock(&scrub->os_lock);
2709 scrub->os_waiting = 0;
2710 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2711 spin_unlock(&scrub->os_lock);
2714 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2715 l_wait_event(thread->t_ctl_waitq,
2716 osd_otable_it_wakeup(scrub, it),
2719 if (!thread_is_running(thread) && !it->ooi_used_outside)
2722 rc = osd_otable_it_preload(env, it);
2729 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2730 const struct dt_it *di)
2735 static int osd_otable_it_key_size(const struct lu_env *env,
2736 const struct dt_it *di)
2738 return sizeof(__u64);
2741 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2742 struct dt_rec *rec, __u32 attr)
2744 struct osd_otable_it *it = (struct osd_otable_it *)di;
2745 struct osd_otable_cache *ooc = &it->ooi_cache;
2747 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2749 /* Filter out Invald FID already. */
2750 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2751 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2752 PFID((struct lu_fid *)rec),
2753 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2758 static __u64 osd_otable_it_store(const struct lu_env *env,
2759 const struct dt_it *di)
2761 struct osd_otable_it *it = (struct osd_otable_it *)di;
2762 struct osd_otable_cache *ooc = &it->ooi_cache;
2765 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2766 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2768 hash = ooc->ooc_pos_preload;
2773 * Set the OSD layer iteration start position as the specified hash.
2775 static int osd_otable_it_load(const struct lu_env *env,
2776 const struct dt_it *di, __u64 hash)
2778 struct osd_otable_it *it = (struct osd_otable_it *)di;
2779 struct osd_device *dev = it->ooi_dev;
2780 struct osd_otable_cache *ooc = &it->ooi_cache;
2781 struct osd_scrub *scrub = &dev->od_scrub;
2785 /* Forbid to set iteration position after iteration started. */
2786 if (it->ooi_user_ready)
2789 if (hash > OSD_OTABLE_MAX_HASH)
2790 hash = OSD_OTABLE_MAX_HASH;
2792 ooc->ooc_pos_preload = hash;
2793 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2794 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2796 it->ooi_user_ready = 1;
2797 if (!scrub->os_full_speed)
2798 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2800 /* Unplug OSD layer iteration by the first next() call. */
2801 rc = osd_otable_it_next(env, (struct dt_it *)it);
2806 static int osd_otable_it_key_rec(const struct lu_env *env,
2807 const struct dt_it *di, void *key_rec)
2812 const struct dt_index_operations osd_otable_ops = {
2814 .init = osd_otable_it_init,
2815 .fini = osd_otable_it_fini,
2816 .get = osd_otable_it_get,
2817 .put = osd_otable_it_put,
2818 .next = osd_otable_it_next,
2819 .key = osd_otable_it_key,
2820 .key_size = osd_otable_it_key_size,
2821 .rec = osd_otable_it_rec,
2822 .store = osd_otable_it_store,
2823 .load = osd_otable_it_load,
2824 .key_rec = osd_otable_it_key_rec,
2828 /* high priority inconsistent items list APIs */
2830 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
2832 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2835 struct osd_inconsistent_item *oii;
2836 struct osd_scrub *scrub = &dev->od_scrub;
2837 struct ptlrpc_thread *thread = &scrub->os_thread;
2842 if (unlikely(oii == NULL))
2845 INIT_LIST_HEAD(&oii->oii_list);
2846 oii->oii_cache = *oic;
2847 oii->oii_insert = insert;
2849 if (scrub->os_partial_scan) {
2850 __u64 now = cfs_time_current_sec();
2852 /* If there haven't been errors in a long time,
2853 * decay old count until either the errors are
2854 * gone or we reach the current interval. */
2855 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2856 scrub->os_bad_oimap_time +
2857 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2858 scrub->os_bad_oimap_count >>= 1;
2859 scrub->os_bad_oimap_time +=
2860 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2863 scrub->os_bad_oimap_time = now;
2864 if (++scrub->os_bad_oimap_count >
2865 dev->od_full_scrub_threshold_rate)
2866 scrub->os_full_scrub = 1;
2869 spin_lock(&scrub->os_lock);
2870 if (unlikely(!thread_is_running(thread))) {
2871 spin_unlock(&scrub->os_lock);
2876 if (list_empty(&scrub->os_inconsistent_items))
2878 list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2879 spin_unlock(&scrub->os_lock);
2882 wake_up_all(&thread->t_ctl_waitq);
2887 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2888 struct osd_inode_id *id)
2890 struct osd_scrub *scrub = &dev->od_scrub;
2891 struct osd_inconsistent_item *oii;
2894 spin_lock(&scrub->os_lock);
2895 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2896 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2897 *id = oii->oii_cache.oic_lid;
2898 spin_unlock(&scrub->os_lock);
2902 spin_unlock(&scrub->os_lock);
2909 static const char *scrub_status_names[] = {
2920 static const char *scrub_flags_names[] = {
2928 static const char *scrub_param_names[] = {
2934 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2941 rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2945 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2948 rc = seq_printf(m, "%s%c", names[i],
2949 bits != 0 ? ',' : '\n');
2957 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2962 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2963 cfs_time_current_sec() - time);
2965 rc = seq_printf(m, "%s: N/A\n", prefix);
2969 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2974 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2976 rc = seq_printf(m, "%s: N/A\n", prefix);
2980 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2982 struct osd_scrub *scrub = &dev->od_scrub;
2983 struct scrub_file *sf = &scrub->os_file;
2988 down_read(&scrub->os_rwsem);
2989 rc = seq_printf(m, "name: OI_scrub\n"
2993 sf->sf_magic, (int)sf->sf_oi_count,
2994 scrub_status_names[sf->sf_status]);
2998 rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
3003 rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
3008 rc = scrub_time_dump(m, sf->sf_time_last_complete,
3009 "time_since_last_completed");
3013 rc = scrub_time_dump(m, sf->sf_time_latest_start,
3014 "time_since_latest_start");
3018 rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
3019 "time_since_last_checkpoint");
3023 rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
3024 "latest_start_position");
3028 rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3029 "last_checkpoint_position");
3033 rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3034 "first_failure_position");
3038 checked = sf->sf_items_checked + scrub->os_new_checked;
3039 rc = seq_printf(m, "checked: "LPU64"\n"
3040 "updated: "LPU64"\n"
3042 "prior_updated: "LPU64"\n"
3043 "noscrub: "LPU64"\n"
3045 "success_count: %u\n",
3046 checked, sf->sf_items_updated, sf->sf_items_failed,
3047 sf->sf_items_updated_prior, sf->sf_items_noscrub,
3048 sf->sf_items_igif, sf->sf_success_count);
3053 if (thread_is_running(&scrub->os_thread)) {
3054 cfs_duration_t duration = cfs_time_current() -
3055 scrub->os_time_last_checkpoint;
3056 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3058 __u32 rtime = sf->sf_run_time +
3059 cfs_duration_sec(duration + HALF_SEC);
3062 do_div(new_checked, duration);
3064 do_div(speed, rtime);
3065 rc = seq_printf(m, "run_time: %u seconds\n"
3066 "average_speed: "LPU64" objects/sec\n"
3067 "real-time_speed: "LPU64" objects/sec\n"
3068 "current_position: %u\n"
3069 "lf_scanned: "LPU64"\n"
3070 "lf_reparied: "LPU64"\n"
3071 "lf_failed: "LPU64"\n",
3072 rtime, speed, new_checked, scrub->os_pos_current,
3073 scrub->os_lf_scanned, scrub->os_lf_repaired,
3074 scrub->os_lf_failed);
3076 if (sf->sf_run_time != 0)
3077 do_div(speed, sf->sf_run_time);
3078 rc = seq_printf(m, "run_time: %u seconds\n"
3079 "average_speed: "LPU64" objects/sec\n"
3080 "real-time_speed: N/A\n"
3081 "current_position: N/A\n"
3082 "lf_scanned: "LPU64"\n"
3083 "lf_reparied: "LPU64"\n"
3084 "lf_failed: "LPU64"\n",
3085 sf->sf_run_time, speed, scrub->os_lf_scanned,
3086 scrub->os_lf_repaired, scrub->os_lf_failed);
3090 up_read(&scrub->os_rwsem);
3091 return (rc < 0 ? -ENOSPC : 0);