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, 2015, 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, bool *exist)
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 ? id->oii_ino : -1, id ? id->oii_gen : -1,
124 case DTO_INDEX_UPDATE:
125 rc = osd_oi_update(info, dev, fid, id, th, flags);
126 if (unlikely(rc == -ENOENT)) {
127 /* Some unlink thread may removed the OI mapping. */
131 case DTO_INDEX_INSERT:
132 rc = osd_oi_insert(info, dev, fid, id, th, flags, exist);
133 if (unlikely(rc == -EEXIST)) {
135 /* XXX: There are trouble things when adding OI
136 * mapping for IGIF object, which may cause
137 * multiple objects to be mapped to the same
138 * IGIF formatted FID. Consider the following
141 * 1) The MDT is upgrading from 1.8 device.
142 * The OI scrub generates IGIF FID1 for the
143 * OBJ1 and adds the OI mapping.
145 * 2) For some reason, the OI scrub does not
146 * process all the IGIF objects completely.
148 * 3) The MDT is backuped and restored against
151 * 4) When the MDT mounts up, the OI scrub will
152 * try to rebuild the OI files. For some IGIF
153 * object, OBJ2, which was not processed by the
154 * OI scrub before the backup/restore, and the
155 * new generated IGIF formatted FID may be just
156 * the FID1, the same as OBJ1.
158 * Under such case, the OI scrub cannot know how
159 * to generate new FID for the OBJ2.
161 * Currently, we do nothing for that. One possible
162 * solution is to generate new normal FID for the
165 * Anyway, it is rare, only exists in theory. */
168 case DTO_INDEX_DELETE:
169 rc = osd_oi_delete(info, dev, fid, th, flags);
171 /* It is normal that the unlink thread has removed the
172 * OI mapping already. */
177 LASSERTF(0, "Unexpected ops %d\n", ops);
181 ldiskfs_journal_stop(th);
183 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
184 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
185 PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
191 /* OI_scrub file ops */
193 static void osd_scrub_file_to_cpu(struct scrub_file *des,
194 struct scrub_file *src)
196 memcpy(des->sf_uuid, src->sf_uuid, 16);
197 des->sf_flags = le64_to_cpu(src->sf_flags);
198 des->sf_magic = le32_to_cpu(src->sf_magic);
199 des->sf_status = le16_to_cpu(src->sf_status);
200 des->sf_param = le16_to_cpu(src->sf_param);
201 des->sf_time_last_complete =
202 le64_to_cpu(src->sf_time_last_complete);
203 des->sf_time_latest_start =
204 le64_to_cpu(src->sf_time_latest_start);
205 des->sf_time_last_checkpoint =
206 le64_to_cpu(src->sf_time_last_checkpoint);
207 des->sf_pos_latest_start =
208 le64_to_cpu(src->sf_pos_latest_start);
209 des->sf_pos_last_checkpoint =
210 le64_to_cpu(src->sf_pos_last_checkpoint);
211 des->sf_pos_first_inconsistent =
212 le64_to_cpu(src->sf_pos_first_inconsistent);
213 des->sf_items_checked =
214 le64_to_cpu(src->sf_items_checked);
215 des->sf_items_updated =
216 le64_to_cpu(src->sf_items_updated);
217 des->sf_items_failed =
218 le64_to_cpu(src->sf_items_failed);
219 des->sf_items_updated_prior =
220 le64_to_cpu(src->sf_items_updated_prior);
221 des->sf_run_time = le32_to_cpu(src->sf_run_time);
222 des->sf_success_count = le32_to_cpu(src->sf_success_count);
223 des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
224 des->sf_internal_flags = le16_to_cpu(src->sf_internal_flags);
225 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
228 static void osd_scrub_file_to_le(struct scrub_file *des,
229 struct scrub_file *src)
231 memcpy(des->sf_uuid, src->sf_uuid, 16);
232 des->sf_flags = cpu_to_le64(src->sf_flags);
233 des->sf_magic = cpu_to_le32(src->sf_magic);
234 des->sf_status = cpu_to_le16(src->sf_status);
235 des->sf_param = cpu_to_le16(src->sf_param);
236 des->sf_time_last_complete =
237 cpu_to_le64(src->sf_time_last_complete);
238 des->sf_time_latest_start =
239 cpu_to_le64(src->sf_time_latest_start);
240 des->sf_time_last_checkpoint =
241 cpu_to_le64(src->sf_time_last_checkpoint);
242 des->sf_pos_latest_start =
243 cpu_to_le64(src->sf_pos_latest_start);
244 des->sf_pos_last_checkpoint =
245 cpu_to_le64(src->sf_pos_last_checkpoint);
246 des->sf_pos_first_inconsistent =
247 cpu_to_le64(src->sf_pos_first_inconsistent);
248 des->sf_items_checked =
249 cpu_to_le64(src->sf_items_checked);
250 des->sf_items_updated =
251 cpu_to_le64(src->sf_items_updated);
252 des->sf_items_failed =
253 cpu_to_le64(src->sf_items_failed);
254 des->sf_items_updated_prior =
255 cpu_to_le64(src->sf_items_updated_prior);
256 des->sf_run_time = cpu_to_le32(src->sf_run_time);
257 des->sf_success_count = cpu_to_le32(src->sf_success_count);
258 des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
259 des->sf_internal_flags = cpu_to_le16(src->sf_internal_flags);
260 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
263 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
265 struct scrub_file *sf = &scrub->os_file;
267 memset(sf, 0, sizeof(*sf));
268 memcpy(sf->sf_uuid, uuid, 16);
269 sf->sf_magic = SCRUB_MAGIC_V1;
270 sf->sf_status = SS_INIT;
273 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
275 struct scrub_file *sf = &scrub->os_file;
277 CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
278 "%#llx, add flags = %#llx\n",
279 osd_scrub2name(scrub), sf->sf_flags, flags);
281 memcpy(sf->sf_uuid, uuid, 16);
282 sf->sf_status = SS_INIT;
283 sf->sf_flags |= flags;
284 sf->sf_flags &= ~SF_AUTO;
286 sf->sf_time_latest_start = 0;
287 sf->sf_time_last_checkpoint = 0;
288 sf->sf_pos_latest_start = 0;
289 sf->sf_pos_last_checkpoint = 0;
290 sf->sf_pos_first_inconsistent = 0;
291 sf->sf_items_checked = 0;
292 sf->sf_items_updated = 0;
293 sf->sf_items_failed = 0;
294 if (!scrub->os_in_join)
295 sf->sf_items_updated_prior = 0;
297 sf->sf_items_noscrub = 0;
298 sf->sf_items_igif = 0;
301 static int osd_scrub_file_load(struct osd_scrub *scrub)
304 int len = sizeof(scrub->os_file_disk);
307 rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
309 struct scrub_file *sf = &scrub->os_file;
311 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
312 if (sf->sf_magic != SCRUB_MAGIC_V1) {
313 CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
314 "0x%x != 0x%x\n", osd_scrub2name(scrub),
315 sf->sf_magic, SCRUB_MAGIC_V1);
316 /* Process it as new scrub file. */
321 } else if (rc != 0) {
322 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
323 "expected = %d: rc = %d\n",
324 osd_scrub2name(scrub), len, rc);
328 /* return -ENOENT for empty scrub file case. */
335 int osd_scrub_file_store(struct osd_scrub *scrub)
337 struct osd_device *dev;
340 int len = sizeof(scrub->os_file_disk);
344 dev = container_of0(scrub, struct osd_device, od_scrub);
345 credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
346 osd_dto_credits_noquota[DTO_WRITE_BLOCK];
347 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
350 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
351 "rc = %d\n", osd_scrub2name(scrub), rc);
355 osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
356 rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
358 ldiskfs_journal_stop(jh);
360 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
361 "expected = %d: rc = %d\n",
362 osd_scrub2name(scrub), len, rc);
364 scrub->os_time_last_checkpoint = cfs_time_current();
365 scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
366 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
371 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
372 struct inode *inode, const struct lu_fid *fid)
374 struct filter_fid_old *ff = &info->oti_ff;
375 struct dentry *dentry = &info->oti_obj_dentry;
376 struct lu_fid *tfid = &info->oti_fid;
380 bool removed = false;
384 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
387 if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
388 struct ost_id *oi = &info->oti_ostid;
390 fid_to_ostid(fid, oi);
391 ostid_to_fid(tfid, oi, 0);
396 /* We want the LMA to fit into the 256-byte OST inode, so operate
398 * 1) read old XATTR_NAME_FID and save the parent FID;
399 * 2) delete the old XATTR_NAME_FID;
400 * 3) make new LMA and add it;
401 * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
403 * Making the LMA to fit into the 256-byte OST inode can save time for
404 * normal osd_check_lma() and for other OI scrub scanning in future.
405 * So it is worth to make some slow conversion here. */
406 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
407 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
410 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
411 DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
415 /* 1) read old XATTR_NAME_FID and save the parent FID */
416 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
417 if (rc == sizeof(*ff)) {
418 /* 2) delete the old XATTR_NAME_FID */
419 ll_vfs_dq_init(inode);
420 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
425 } else if (unlikely(rc == -ENODATA)) {
427 } else if (rc != sizeof(struct filter_fid)) {
428 GOTO(stop, rc = -EINVAL);
431 /* 3) make new LMA and add it */
432 rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
433 if (rc == 0 && reset)
434 size = sizeof(struct filter_fid);
435 else if (rc != 0 && removed)
436 /* If failed, we should try to add the old back. */
437 size = sizeof(struct filter_fid_old);
439 /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
443 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
445 if (rc1 != 0 && rc == 0)
452 ldiskfs_journal_stop(jh);
454 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
455 osd_name(dev), PFID(tfid), rc);
460 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
461 struct osd_idmap_cache *oic, int val)
463 struct osd_scrub *scrub = &dev->od_scrub;
464 struct scrub_file *sf = &scrub->os_file;
465 struct lu_fid *fid = &oic->oic_fid;
466 struct osd_inode_id *lid = &oic->oic_lid;
467 struct osd_inode_id *lid2 = &info->oti_id;
468 struct osd_inconsistent_item *oii = NULL;
469 struct inode *inode = NULL;
470 int ops = DTO_INDEX_UPDATE;
472 bool converted = false;
476 down_write(&scrub->os_rwsem);
477 scrub->os_new_checked++;
481 if (scrub->os_in_prior)
482 oii = list_entry(oic, struct osd_inconsistent_item,
485 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
488 if (fid_is_igif(fid))
491 if (val == SCRUB_NEXT_OSTOBJ_OLD) {
492 inode = osd_iget(info, dev, lid);
495 /* Someone removed the inode. */
496 if (rc == -ENOENT || rc == -ESTALE)
501 /* The inode has been reused as EA inode, ignore it. */
502 if (unlikely(osd_is_ea_inode(inode)))
505 sf->sf_flags |= SF_UPGRADE;
506 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
507 dev->od_check_ff = 1;
508 rc = osd_scrub_convert_ff(info, dev, inode, fid);
515 if ((val == SCRUB_NEXT_NOLMA) &&
516 (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
519 if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
520 ops = DTO_INDEX_INSERT;
525 rc = osd_oi_lookup(info, dev, fid, lid2,
526 (val == SCRUB_NEXT_OSTOBJ ||
527 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
530 ops = DTO_INDEX_INSERT;
531 else if (rc != -ESTALE)
536 inode = osd_iget(info, dev, lid);
539 /* Someone removed the inode. */
540 if (rc == -ENOENT || rc == -ESTALE)
545 /* The inode has been reused as EA inode, ignore it. */
546 if (unlikely(osd_is_ea_inode(inode)))
550 if (!scrub->os_partial_scan)
551 scrub->os_full_speed = 1;
554 case SCRUB_NEXT_NOLMA:
555 sf->sf_flags |= SF_UPGRADE;
556 if (!(sf->sf_param & SP_DRYRUN)) {
557 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
562 if (!(sf->sf_flags & SF_INCONSISTENT))
563 dev->od_igif_inoi = 0;
565 case SCRUB_NEXT_OSTOBJ:
566 sf->sf_flags |= SF_INCONSISTENT;
567 case SCRUB_NEXT_OSTOBJ_OLD:
572 } else if (osd_id_eq(lid, lid2)) {
574 sf->sf_items_updated++;
578 if (!scrub->os_partial_scan)
579 scrub->os_full_speed = 1;
581 sf->sf_flags |= SF_INCONSISTENT;
583 /* XXX: If the device is restored from file-level backup, then
584 * some IGIFs may have been already in OI files, and some
585 * may be not yet. Means upgrading from 1.8 may be partly
586 * processed, but some clients may hold some immobilized
587 * IGIFs, and use them to access related objects. Under
588 * such case, OSD does not know whether an given IGIF has
589 * been processed or to be processed, and it also cannot
590 * generate local ino#/gen# directly from the immobilized
591 * IGIF because of the backup/restore. Then force OSD to
592 * lookup the given IGIF in OI files, and if no entry,
593 * then ask the client to retry after upgrading completed.
594 * No better choice. */
595 dev->od_igif_inoi = 1;
598 rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
599 (val == SCRUB_NEXT_OSTOBJ ||
600 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0,
603 if (scrub->os_in_prior)
604 sf->sf_items_updated_prior++;
606 sf->sf_items_updated++;
608 if (ops == DTO_INDEX_INSERT && val == 0 && !exist) {
609 int idx = osd_oi_fid2idx(dev, fid);
611 sf->sf_flags |= SF_RECREATED;
612 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
613 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
621 sf->sf_items_failed++;
622 if (sf->sf_pos_first_inconsistent == 0 ||
623 sf->sf_pos_first_inconsistent > lid->oii_ino)
624 sf->sf_pos_first_inconsistent = lid->oii_ino;
629 /* There may be conflict unlink during the OI scrub,
630 * if happend, then remove the new added OI mapping. */
631 if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
632 unlikely(ldiskfs_test_inode_state(inode,
633 LDISKFS_STATE_LUSTRE_DESTROY)))
634 osd_scrub_refresh_mapping(info, dev, fid, lid,
635 DTO_INDEX_DELETE, false,
636 (val == SCRUB_NEXT_OSTOBJ ||
637 val == SCRUB_NEXT_OSTOBJ_OLD) ?
638 OI_KNOWN_ON_OST : 0, NULL);
639 up_write(&scrub->os_rwsem);
641 if (inode != NULL && !IS_ERR(inode))
645 spin_lock(&scrub->os_lock);
646 if (likely(!list_empty(&oii->oii_list)))
647 list_del(&oii->oii_list);
648 spin_unlock(&scrub->os_lock);
653 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
658 static int osd_scrub_prep(struct osd_device *dev)
660 struct osd_scrub *scrub = &dev->od_scrub;
661 struct ptlrpc_thread *thread = &scrub->os_thread;
662 struct scrub_file *sf = &scrub->os_file;
663 __u32 flags = scrub->os_start_flags;
665 bool drop_dryrun = false;
668 CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
669 osd_scrub2name(scrub), flags);
671 down_write(&scrub->os_rwsem);
672 if (flags & SS_SET_FAILOUT)
673 sf->sf_param |= SP_FAILOUT;
674 else if (flags & SS_CLEAR_FAILOUT)
675 sf->sf_param &= ~SP_FAILOUT;
677 if (flags & SS_SET_DRYRUN) {
678 sf->sf_param |= SP_DRYRUN;
679 } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
680 sf->sf_param &= ~SP_DRYRUN;
684 if (flags & SS_RESET)
685 osd_scrub_file_reset(scrub,
686 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
688 if (flags & SS_AUTO_FULL) {
689 scrub->os_full_speed = 1;
690 scrub->os_partial_scan = 0;
691 sf->sf_flags |= SF_AUTO;
692 } else if (flags & SS_AUTO_PARTIAL) {
693 scrub->os_full_speed = 0;
694 scrub->os_partial_scan = 1;
695 sf->sf_flags |= SF_AUTO;
696 } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
698 scrub->os_full_speed = 1;
699 scrub->os_partial_scan = 0;
701 scrub->os_full_speed = 0;
702 scrub->os_partial_scan = 0;
705 spin_lock(&scrub->os_lock);
706 scrub->os_in_prior = 0;
707 scrub->os_waiting = 0;
708 scrub->os_paused = 0;
709 scrub->os_in_join = 0;
710 scrub->os_full_scrub = 0;
711 spin_unlock(&scrub->os_lock);
712 scrub->os_new_checked = 0;
713 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
714 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
715 else if (sf->sf_pos_last_checkpoint != 0)
716 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
718 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
720 scrub->os_pos_current = sf->sf_pos_latest_start;
721 sf->sf_status = SS_SCANNING;
722 sf->sf_time_latest_start = cfs_time_current_sec();
723 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
724 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
725 rc = osd_scrub_file_store(scrub);
727 spin_lock(&scrub->os_lock);
728 thread_set_flags(thread, SVC_RUNNING);
729 spin_unlock(&scrub->os_lock);
730 wake_up_all(&thread->t_ctl_waitq);
732 up_write(&scrub->os_rwsem);
737 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
739 struct scrub_file *sf = &scrub->os_file;
742 if (likely(cfs_time_before(cfs_time_current(),
743 scrub->os_time_next_checkpoint) ||
744 scrub->os_new_checked == 0))
747 down_write(&scrub->os_rwsem);
748 sf->sf_items_checked += scrub->os_new_checked;
749 scrub->os_new_checked = 0;
750 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
751 sf->sf_time_last_checkpoint = cfs_time_current_sec();
752 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
753 scrub->os_time_last_checkpoint);
754 rc = osd_scrub_file_store(scrub);
755 up_write(&scrub->os_rwsem);
760 static int osd_scrub_post(struct osd_scrub *scrub, int result)
762 struct scrub_file *sf = &scrub->os_file;
766 CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
767 osd_scrub2name(scrub), result);
769 down_write(&scrub->os_rwsem);
770 spin_lock(&scrub->os_lock);
771 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
772 spin_unlock(&scrub->os_lock);
773 if (scrub->os_new_checked > 0) {
774 sf->sf_items_checked += scrub->os_new_checked;
775 scrub->os_new_checked = 0;
776 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
778 sf->sf_time_last_checkpoint = cfs_time_current_sec();
780 struct osd_device *dev =
781 container_of0(scrub, struct osd_device, od_scrub);
783 dev->od_igif_inoi = 1;
784 dev->od_check_ff = 0;
785 sf->sf_status = SS_COMPLETED;
786 if (!(sf->sf_param & SP_DRYRUN)) {
787 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
788 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
789 SF_UPGRADE | SF_AUTO);
791 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
792 sf->sf_success_count++;
793 } else if (result == 0) {
794 if (scrub->os_paused)
795 sf->sf_status = SS_PAUSED;
797 sf->sf_status = SS_STOPPED;
799 sf->sf_status = SS_FAILED;
801 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
802 scrub->os_time_last_checkpoint);
803 rc = osd_scrub_file_store(scrub);
804 up_write(&scrub->os_rwsem);
806 RETURN(rc < 0 ? rc : result);
809 /* iteration engine */
811 struct osd_iit_param {
812 struct super_block *sb;
813 struct buffer_head *bitmap;
819 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
820 struct osd_device *dev,
821 struct osd_iit_param *param,
822 struct osd_idmap_cache **oic,
825 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
826 struct osd_device *dev,
827 struct osd_iit_param *param,
828 struct osd_idmap_cache *oic,
829 bool *noslot, int rc);
831 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
833 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
834 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
835 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
836 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
837 return SCRUB_NEXT_BREAK;
839 *pos = param->gbase + param->offset;
845 * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
846 * \retval 0: FID-on-MDT
848 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
849 struct osd_device *dev,
852 /* XXX: The initial OI scrub will scan the top level /O to generate
853 * a small local FLDB according to the <seq>. If the given FID
854 * is in the local FLDB, then it is FID-on-OST; otherwise it's
855 * quite possible for FID-on-MDT. */
857 return SCRUB_NEXT_OSTOBJ_OLD;
862 static int osd_scrub_get_fid(struct osd_thread_info *info,
863 struct osd_device *dev, struct inode *inode,
864 struct lu_fid *fid, bool scrub)
866 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
868 bool has_lma = false;
870 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
873 if (lma->lma_compat & LMAC_NOT_IN_OI ||
874 lma->lma_incompat & LMAI_AGENT)
875 return SCRUB_NEXT_CONTINUE;
877 *fid = lma->lma_self_fid;
881 if (lma->lma_compat & LMAC_FID_ON_OST)
882 return SCRUB_NEXT_OSTOBJ;
884 if (fid_is_idif(fid))
885 return SCRUB_NEXT_OSTOBJ_OLD;
887 /* For local object. */
888 if (fid_is_internal(fid))
891 /* For external visible MDT-object with non-normal FID. */
892 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
895 /* For the object with normal FID, it may be MDT-object,
896 * or may be 2.4 OST-object, need further distinguish.
897 * Fall through to next section. */
900 if (rc == -ENODATA || rc == 0) {
901 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
904 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
905 rc = SCRUB_NEXT_OSTOBJ_OLD;
911 /* It is FID-on-OST, but we do not know how
912 * to generate its FID, ignore it directly. */
913 rc = SCRUB_NEXT_CONTINUE;
915 /* It is 2.4 OST-object. */
916 rc = SCRUB_NEXT_OSTOBJ_OLD;
924 if (dev->od_scrub.os_convert_igif) {
925 lu_igif_build(fid, inode->i_ino,
926 inode->i_generation);
928 rc = SCRUB_NEXT_NOLMA;
932 /* It may be FID-on-OST, or may be FID for
933 * non-MDT0, anyway, we do not know how to
934 * generate its FID, ignore it directly. */
935 rc = SCRUB_NEXT_CONTINUE;
940 /* For OI scrub case only: the object has LMA but has no ff
941 * (or ff crashed). It may be MDT-object, may be OST-object
942 * with crashed ff. The last check is local FLDB. */
943 rc = osd_scrub_check_local_fldb(info, dev, fid);
949 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
950 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
951 struct super_block *sb, bool scrub)
957 /* Not handle the backend root object and agent parent object.
958 * They are neither visible to namespace nor have OI mappings. */
959 if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
960 pos == osd_remote_parent_ino(dev)))
961 RETURN(SCRUB_NEXT_CONTINUE);
963 osd_id_gen(lid, pos, OSD_OII_NOGEN);
964 inode = osd_iget(info, dev, lid);
967 /* The inode may be removed after bitmap searching, or the
968 * file is new created without inode initialized yet. */
969 if (rc == -ENOENT || rc == -ESTALE)
970 RETURN(SCRUB_NEXT_CONTINUE);
972 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
973 "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
978 /* It is an EA inode, no OI mapping for it, skip it. */
979 if (osd_is_ea_inode(inode))
980 GOTO(put, rc = SCRUB_NEXT_CONTINUE);
983 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
984 /* Only skip it for the first OI scrub accessing. */
985 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
986 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
989 rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
998 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
999 struct osd_iit_param *param,
1000 struct osd_idmap_cache **oic, const bool noslot)
1002 struct osd_scrub *scrub = &dev->od_scrub;
1003 struct ptlrpc_thread *thread = &scrub->os_thread;
1005 struct osd_inode_id *lid;
1008 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
1009 struct l_wait_info lwi;
1011 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
1012 if (likely(lwi.lwi_timeout > 0))
1013 l_wait_event(thread->t_ctl_waitq,
1014 !list_empty(&scrub->os_inconsistent_items) ||
1015 !thread_is_running(thread),
1019 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
1020 spin_lock(&scrub->os_lock);
1021 thread_set_flags(thread, SVC_STOPPING);
1022 spin_unlock(&scrub->os_lock);
1023 return SCRUB_NEXT_CRASH;
1026 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1027 return SCRUB_NEXT_FATAL;
1029 if (unlikely(!thread_is_running(thread)))
1030 return SCRUB_NEXT_EXIT;
1032 if (!list_empty(&scrub->os_inconsistent_items)) {
1033 spin_lock(&scrub->os_lock);
1034 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1035 struct osd_inconsistent_item *oii;
1037 oii = list_entry(scrub->os_inconsistent_items.next,
1038 struct osd_inconsistent_item, oii_list);
1039 spin_unlock(&scrub->os_lock);
1041 *oic = &oii->oii_cache;
1042 scrub->os_in_prior = 1;
1046 spin_unlock(&scrub->os_lock);
1050 return SCRUB_NEXT_WAIT;
1052 rc = osd_iit_next(param, &scrub->os_pos_current);
1056 *oic = &scrub->os_oic;
1057 fid = &(*oic)->oic_fid;
1058 lid = &(*oic)->oic_lid;
1059 rc = osd_iit_iget(info, dev, fid, lid,
1060 scrub->os_pos_current, param->sb, true);
1064 static int osd_preload_next(struct osd_thread_info *info,
1065 struct osd_device *dev, struct osd_iit_param *param,
1066 struct osd_idmap_cache **oic, const bool noslot)
1068 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1069 struct osd_scrub *scrub;
1070 struct ptlrpc_thread *thread;
1073 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1077 scrub = &dev->od_scrub;
1078 thread = &scrub->os_thread;
1079 if (thread_is_running(thread) &&
1080 ooc->ooc_pos_preload >= scrub->os_pos_current)
1081 return SCRUB_NEXT_EXIT;
1083 rc = osd_iit_iget(info, dev,
1084 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1085 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1086 ooc->ooc_pos_preload, param->sb, false);
1087 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1088 * ignore the failure, so it still need to skip the inode next time. */
1089 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1094 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1096 spin_lock(&scrub->os_lock);
1097 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1098 !list_empty(&scrub->os_inconsistent_items) ||
1099 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1100 scrub->os_waiting = 0;
1102 scrub->os_waiting = 1;
1103 spin_unlock(&scrub->os_lock);
1105 return !scrub->os_waiting;
1108 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1109 struct osd_iit_param *param,
1110 struct osd_idmap_cache *oic, bool *noslot, int rc)
1112 struct l_wait_info lwi = { 0 };
1113 struct osd_scrub *scrub = &dev->od_scrub;
1114 struct scrub_file *sf = &scrub->os_file;
1115 struct ptlrpc_thread *thread = &scrub->os_thread;
1116 struct osd_otable_it *it = dev->od_otable_it;
1117 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
1120 case SCRUB_NEXT_CONTINUE:
1122 case SCRUB_NEXT_WAIT:
1124 case SCRUB_NEXT_NOSCRUB:
1125 down_write(&scrub->os_rwsem);
1126 scrub->os_new_checked++;
1127 sf->sf_items_noscrub++;
1128 up_write(&scrub->os_rwsem);
1132 rc = osd_scrub_check_update(info, dev, oic, rc);
1134 scrub->os_in_prior = 0;
1138 rc = osd_scrub_checkpoint(scrub);
1140 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1141 "rc = %d\n", osd_scrub2name(scrub),
1142 scrub->os_pos_current, rc);
1143 /* Continue, as long as the scrub itself can go ahead. */
1146 if (scrub->os_in_prior) {
1147 scrub->os_in_prior = 0;
1152 scrub->os_pos_current = param->gbase + ++(param->offset);
1155 if (it != NULL && it->ooi_waiting && ooc != NULL &&
1156 ooc->ooc_pos_preload < scrub->os_pos_current) {
1157 spin_lock(&scrub->os_lock);
1158 it->ooi_waiting = 0;
1159 wake_up_all(&thread->t_ctl_waitq);
1160 spin_unlock(&scrub->os_lock);
1163 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1166 if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1172 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1175 if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1182 static int osd_preload_exec(struct osd_thread_info *info,
1183 struct osd_device *dev, struct osd_iit_param *param,
1184 struct osd_idmap_cache *oic, bool *noslot, int rc)
1186 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1189 ooc->ooc_cached_items++;
1190 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1191 ~OSD_OTABLE_IT_CACHE_MASK;
1193 return rc > 0 ? 0 : rc;
1196 #define SCRUB_IT_ALL 1
1197 #define SCRUB_IT_CRASH 2
1199 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1202 struct osd_scrub *scrub = &dev->od_scrub;
1203 struct ptlrpc_thread *thread = &scrub->os_thread;
1204 struct scrub_file *sf = &scrub->os_file;
1208 LASSERT(!(flags & SS_AUTO_PARTIAL));
1210 down_write(&scrub->os_rwsem);
1211 scrub->os_in_join = 1;
1212 if (flags & SS_SET_FAILOUT)
1213 sf->sf_param |= SP_FAILOUT;
1214 else if (flags & SS_CLEAR_FAILOUT)
1215 sf->sf_param &= ~SP_FAILOUT;
1217 if (flags & SS_SET_DRYRUN)
1218 sf->sf_param |= SP_DRYRUN;
1219 else if (flags & SS_CLEAR_DRYRUN)
1220 sf->sf_param &= ~SP_DRYRUN;
1222 if (flags & SS_RESET) {
1223 osd_scrub_file_reset(scrub,
1224 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1225 inconsistent ? SF_INCONSISTENT : 0);
1226 sf->sf_status = SS_SCANNING;
1229 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1230 scrub->os_full_speed = 1;
1232 scrub->os_full_speed = 0;
1234 if (flags & SS_AUTO_FULL) {
1235 sf->sf_flags |= SF_AUTO;
1236 scrub->os_full_speed = 1;
1239 scrub->os_new_checked = 0;
1240 if (sf->sf_pos_last_checkpoint != 0)
1241 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1243 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1245 scrub->os_pos_current = sf->sf_pos_latest_start;
1246 sf->sf_time_latest_start = cfs_time_current_sec();
1247 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1248 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1249 rc = osd_scrub_file_store(scrub);
1251 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1252 "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1254 spin_lock(&scrub->os_lock);
1255 scrub->os_waiting = 0;
1256 scrub->os_paused = 0;
1257 scrub->os_partial_scan = 0;
1258 scrub->os_in_join = 0;
1259 scrub->os_full_scrub = 0;
1260 spin_unlock(&scrub->os_lock);
1261 wake_up_all(&thread->t_ctl_waitq);
1262 up_write(&scrub->os_rwsem);
1267 static int osd_inode_iteration(struct osd_thread_info *info,
1268 struct osd_device *dev, __u32 max, bool preload)
1270 struct osd_scrub *scrub = &dev->od_scrub;
1271 struct ptlrpc_thread *thread = &scrub->os_thread;
1272 struct scrub_file *sf = &scrub->os_file;
1273 osd_iit_next_policy next;
1274 osd_iit_exec_policy exec;
1277 struct osd_iit_param param = { NULL };
1278 struct l_wait_info lwi = { 0 };
1284 param.sb = osd_sb(dev);
1288 while (scrub->os_partial_scan && !scrub->os_in_join) {
1289 struct osd_idmap_cache *oic = NULL;
1291 rc = osd_scrub_next(info, dev, ¶m, &oic, noslot);
1293 case SCRUB_NEXT_EXIT:
1295 case SCRUB_NEXT_CRASH:
1296 RETURN(SCRUB_IT_CRASH);
1297 case SCRUB_NEXT_FATAL:
1299 case SCRUB_NEXT_WAIT: {
1300 struct kstatfs *ksfs = &info->oti_ksfs;
1303 if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1304 unlikely(sf->sf_items_updated_prior == 0))
1307 if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1308 scrub->os_full_scrub) {
1309 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1314 rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1316 __u64 used = ksfs->f_files - ksfs->f_ffree;
1318 do_div(used, sf->sf_items_updated_prior);
1319 /* If we hit too much inconsistent OI
1320 * mappings during the partial scan,
1321 * then scan the device completely. */
1322 if (used < dev->od_full_scrub_ratio) {
1324 SS_AUTO_FULL | SS_RESET, true);
1330 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1334 saved_flags = sf->sf_flags;
1335 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1336 SF_UPGRADE | SF_AUTO);
1337 sf->sf_status = SS_COMPLETED;
1338 l_wait_event(thread->t_ctl_waitq,
1339 !thread_is_running(thread) ||
1340 !scrub->os_partial_scan ||
1341 scrub->os_in_join ||
1342 !list_empty(&scrub->os_inconsistent_items),
1344 sf->sf_flags = saved_flags;
1345 sf->sf_status = SS_SCANNING;
1347 if (unlikely(!thread_is_running(thread)))
1350 if (!scrub->os_partial_scan || scrub->os_in_join)
1356 LASSERTF(rc == 0, "rc = %d\n", rc);
1358 osd_scrub_exec(info, dev, ¶m, oic, &noslot, rc);
1365 l_wait_event(thread->t_ctl_waitq,
1366 !thread_is_running(thread) || !scrub->os_in_join,
1369 if (unlikely(!thread_is_running(thread)))
1375 next = osd_scrub_next;
1376 exec = osd_scrub_exec;
1377 pos = &scrub->os_pos_current;
1378 count = &scrub->os_new_checked;
1380 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1382 next = osd_preload_next;
1383 exec = osd_preload_exec;
1384 pos = &ooc->ooc_pos_preload;
1385 count = &ooc->ooc_cached_items;
1387 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1389 while (*pos <= limit && *count < max) {
1390 struct osd_idmap_cache *oic = NULL;
1391 struct ldiskfs_group_desc *desc;
1393 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1394 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1398 ldiskfs_lock_group(param.sb, param.bg);
1399 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1400 ldiskfs_unlock_group(param.sb, param.bg);
1401 *pos = 1 + (param.bg + 1) *
1402 LDISKFS_INODES_PER_GROUP(param.sb);
1405 ldiskfs_unlock_group(param.sb, param.bg);
1407 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1408 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1409 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1410 if (param.bitmap == NULL) {
1411 CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1412 "scrub will stop, urgent mode\n",
1413 osd_scrub2name(scrub), (__u32)param.bg);
1417 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1420 ldiskfs_itable_unused_count(param.sb, desc) >
1421 LDISKFS_INODES_PER_GROUP(param.sb)) {
1422 *pos = 1 + (param.bg + 1) *
1423 LDISKFS_INODES_PER_GROUP(param.sb);
1427 rc = next(info, dev, ¶m, &oic, noslot);
1429 case SCRUB_NEXT_BREAK:
1431 case SCRUB_NEXT_EXIT:
1432 brelse(param.bitmap);
1434 case SCRUB_NEXT_CRASH:
1435 brelse(param.bitmap);
1436 RETURN(SCRUB_IT_CRASH);
1437 case SCRUB_NEXT_FATAL:
1438 brelse(param.bitmap);
1442 rc = exec(info, dev, ¶m, oic, &noslot, rc);
1444 brelse(param.bitmap);
1450 brelse(param.bitmap);
1454 RETURN(SCRUB_IT_ALL);
1458 static int osd_otable_it_preload(const struct lu_env *env,
1459 struct osd_otable_it *it)
1461 struct osd_device *dev = it->ooi_dev;
1462 struct osd_scrub *scrub = &dev->od_scrub;
1463 struct osd_otable_cache *ooc = &it->ooi_cache;
1467 rc = osd_inode_iteration(osd_oti_get(env), dev,
1468 OSD_OTABLE_IT_CACHE_SIZE, true);
1469 if (rc == SCRUB_IT_ALL)
1470 it->ooi_all_cached = 1;
1472 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1473 spin_lock(&scrub->os_lock);
1474 scrub->os_waiting = 0;
1475 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1476 spin_unlock(&scrub->os_lock);
1479 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1482 static int osd_scrub_main(void *args)
1485 struct osd_device *dev = (struct osd_device *)args;
1486 struct osd_scrub *scrub = &dev->od_scrub;
1487 struct ptlrpc_thread *thread = &scrub->os_thread;
1491 rc = lu_env_init(&env, LCT_LOCAL);
1493 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1494 osd_scrub2name(scrub), rc);
1498 rc = osd_scrub_prep(dev);
1500 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1501 osd_scrub2name(scrub), rc);
1505 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1506 struct l_wait_info lwi = { 0 };
1507 struct osd_otable_it *it = dev->od_otable_it;
1508 struct osd_otable_cache *ooc = &it->ooi_cache;
1510 l_wait_event(thread->t_ctl_waitq,
1511 it->ooi_user_ready || !thread_is_running(thread),
1513 if (unlikely(!thread_is_running(thread)))
1516 scrub->os_pos_current = ooc->ooc_pos_preload;
1519 CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1520 osd_scrub2name(scrub), scrub->os_start_flags,
1521 scrub->os_pos_current);
1523 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1524 if (unlikely(rc == SCRUB_IT_CRASH))
1525 GOTO(out, rc = -EINVAL);
1529 rc = osd_scrub_post(scrub, rc);
1530 CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1531 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1534 while (!list_empty(&scrub->os_inconsistent_items)) {
1535 struct osd_inconsistent_item *oii;
1537 oii = list_entry(scrub->os_inconsistent_items.next,
1538 struct osd_inconsistent_item, oii_list);
1539 list_del_init(&oii->oii_list);
1545 spin_lock(&scrub->os_lock);
1546 thread_set_flags(thread, SVC_STOPPED);
1547 wake_up_all(&thread->t_ctl_waitq);
1548 spin_unlock(&scrub->os_lock);
1552 /* initial OI scrub */
1554 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1555 struct dentry *, filldir_t filldir);
1557 #ifdef HAVE_FILLDIR_USE_CTX
1558 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1559 int namelen, loff_t offset, __u64 ino,
1561 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1562 int namelen, loff_t offset, __u64 ino,
1564 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1565 int namelen, loff_t offset, __u64 ino,
1567 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1568 int namelen, loff_t offset, __u64 ino,
1571 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1572 loff_t offset, __u64 ino, unsigned d_type);
1573 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1574 loff_t offset, __u64 ino, unsigned d_type);
1575 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1576 loff_t offset, __u64 ino, unsigned d_type);
1577 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1578 loff_t offset, __u64 ino, unsigned d_type);
1582 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1583 struct dentry *dentry, filldir_t filldir);
1585 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1586 struct dentry *dentry, filldir_t filldir);
1589 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1590 struct dentry *dentry, filldir_t filldir);
1593 OLF_SCAN_SUBITEMS = 0x0001,
1594 OLF_HIDE_FID = 0x0002,
1595 OLF_SHOW_NAME = 0x0004,
1597 OLF_IDX_IN_FID = 0x0010,
1602 struct lu_fid olm_fid;
1605 scandir_t olm_scandir;
1606 filldir_t olm_filldir;
1609 /* Add the new introduced local files in the list in the future. */
1610 static const struct osd_lf_map osd_lf_maps[] = {
1612 { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1613 sizeof(CATLIST) - 1, NULL, NULL },
1616 { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1617 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1618 osd_ios_general_scan, osd_ios_varfid_fill },
1620 /* NIDTBL_VERSIONS */
1621 { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1622 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1623 osd_ios_varfid_fill },
1626 { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1629 { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1630 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1631 osd_ios_ROOT_scan, NULL },
1633 /* changelog_catalog */
1634 { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1637 /* changelog_users */
1638 { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1642 { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1643 sizeof("fld") - 1, NULL, NULL },
1646 { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1647 sizeof(LAST_RCVD) - 1, NULL, NULL },
1650 { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1651 sizeof(REPLY_DATA) - 1, NULL, NULL },
1654 { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1655 sizeof(LOV_OBJID) - 1, NULL, NULL },
1658 { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1659 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1662 { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1663 osd_ios_general_scan, osd_ios_varfid_fill },
1666 { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1667 osd_ios_general_scan, osd_ios_varfid_fill },
1670 { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1671 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1674 { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1675 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1678 { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1679 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1682 { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1683 osd_ios_general_scan, osd_ios_varfid_fill },
1685 /* lfsck_bookmark */
1686 { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1690 { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1693 /* lfsck_namespace */
1694 { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1697 /* OBJECTS, upgrade from old device */
1698 { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1699 osd_ios_OBJECTS_scan, NULL },
1701 /* lquota_v2.user, upgrade from old device */
1702 { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1705 /* lquota_v2.group, upgrade from old device */
1706 { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1709 /* LAST_GROUP, upgrade from old device */
1710 { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1711 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1713 /* committed batchid for cross-MDT operation */
1714 { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1715 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1717 /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1718 * and f_oid = index for their log files. See lu_update_log{_dir}_fid()
1719 * for more details. */
1722 { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1723 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1726 /* update_log_dir */
1727 { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1728 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1729 sizeof("update_log_dir") - 1,
1730 osd_ios_general_scan, osd_ios_uld_fill },
1733 { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1734 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1735 osd_ios_general_scan, osd_ios_lf_fill },
1737 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1740 /* Add the new introduced files under .lustre/ in the list in the future. */
1741 static const struct osd_lf_map osd_dl_maps[] = {
1743 { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1744 sizeof("fid") - 1, NULL, NULL },
1746 /* .lustre/lost+found */
1747 { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1748 sizeof("lost+found") - 1, NULL, NULL },
1750 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1753 struct osd_ios_item {
1754 struct list_head oii_list;
1755 struct dentry *oii_dentry;
1756 scandir_t oii_scandir;
1757 filldir_t oii_filldir;
1760 struct osd_ios_filldir_buf {
1761 #ifdef HAVE_DIR_CONTEXT
1762 /* please keep it as first member */
1763 struct dir_context ctx;
1765 struct osd_thread_info *oifb_info;
1766 struct osd_device *oifb_dev;
1767 struct dentry *oifb_dentry;
1770 static inline struct dentry *
1771 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1773 struct dentry *dentry;
1775 dentry = ll_lookup_one_len(name, parent, namelen);
1776 if (IS_ERR(dentry)) {
1777 int rc = PTR_ERR(dentry);
1780 CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1781 namelen, name, parent->d_name.len,
1782 parent->d_name.name, parent->d_inode->i_ino,
1783 parent->d_inode->i_generation, rc);
1788 if (dentry->d_inode == NULL) {
1790 return ERR_PTR(-ENOENT);
1797 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1798 scandir_t scandir, filldir_t filldir)
1800 struct osd_ios_item *item;
1803 OBD_ALLOC_PTR(item);
1807 INIT_LIST_HEAD(&item->oii_list);
1808 item->oii_dentry = dget(dentry);
1809 item->oii_scandir = scandir;
1810 item->oii_filldir = filldir;
1811 list_add_tail(&item->oii_list, &dev->od_ios_list);
1817 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1819 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1820 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1821 * reference the inode, or fixed if it is missing or references another inode.
1824 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1825 struct inode *inode, const struct lu_fid *fid, int flags)
1827 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1828 struct osd_inode_id *id = &info->oti_id;
1829 struct osd_inode_id *id2 = &info->oti_id2;
1830 struct osd_scrub *scrub = &dev->od_scrub;
1831 struct scrub_file *sf = &scrub->os_file;
1836 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1837 if (rc != 0 && rc != -ENODATA) {
1838 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1839 "rc = %d\n", osd_name(dev), rc);
1844 osd_id_gen(id, inode->i_ino, inode->i_generation);
1845 if (rc == -ENODATA) {
1846 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1847 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1850 if (flags & OLF_IDX_IN_FID) {
1851 LASSERT(dev->od_index >= 0);
1853 tfid.f_oid = dev->od_index;
1856 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1858 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1859 "scrub: rc = %d\n", osd_name(dev), rc);
1864 if (lma->lma_compat & LMAC_NOT_IN_OI)
1867 tfid = lma->lma_self_fid;
1870 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1875 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1876 DTO_INDEX_INSERT, true, 0, NULL);
1883 if (osd_id_eq_strict(id, id2))
1886 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1887 osd_scrub_file_reset(scrub,
1888 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1890 rc = osd_scrub_file_store(scrub);
1895 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1896 DTO_INDEX_UPDATE, true, 0, NULL);
1904 * It scans the /lost+found, and for the OST-object (with filter_fid
1905 * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1907 #ifdef HAVE_FILLDIR_USE_CTX
1908 static int osd_ios_lf_fill(struct dir_context *buf,
1910 static int osd_ios_lf_fill(void *buf,
1912 const char *name, int namelen,
1913 loff_t offset, __u64 ino, unsigned d_type)
1915 struct osd_ios_filldir_buf *fill_buf =
1916 (struct osd_ios_filldir_buf *)buf;
1917 struct osd_thread_info *info = fill_buf->oifb_info;
1918 struct osd_device *dev = fill_buf->oifb_dev;
1919 struct lu_fid *fid = &info->oti_fid;
1920 struct osd_scrub *scrub = &dev->od_scrub;
1921 struct dentry *parent = fill_buf->oifb_dentry;
1922 struct dentry *child;
1923 struct inode *dir = parent->d_inode;
1924 struct inode *inode;
1928 /* skip any '.' started names */
1932 scrub->os_lf_scanned++;
1933 child = osd_ios_lookup_one_len(name, parent, namelen);
1934 if (IS_ERR(child)) {
1935 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1936 osd_name(dev), namelen, name, (int)PTR_ERR(child));
1940 inode = child->d_inode;
1941 if (S_ISDIR(inode->i_mode)) {
1942 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1945 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1946 "rc = %d\n", osd_name(dev), namelen, name, rc);
1950 if (!S_ISREG(inode->i_mode))
1953 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1954 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1955 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1957 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1958 "/lost+found.\n", namelen, name, PFID(fid));
1959 scrub->os_lf_repaired++;
1961 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1963 osd_name(dev), namelen, name, PFID(fid), rc);
1967 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1968 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1969 * can process them in furtuer. */
1975 scrub->os_lf_failed++;
1977 /* skip the failure to make the scanning to continue. */
1981 #ifdef HAVE_FILLDIR_USE_CTX
1982 static int osd_ios_varfid_fill(struct dir_context *buf,
1984 static int osd_ios_varfid_fill(void *buf,
1986 const char *name, int namelen,
1987 loff_t offset, __u64 ino, unsigned d_type)
1989 struct osd_ios_filldir_buf *fill_buf =
1990 (struct osd_ios_filldir_buf *)buf;
1991 struct osd_device *dev = fill_buf->oifb_dev;
1992 struct dentry *child;
1996 /* skip any '.' started names */
2000 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2002 RETURN(PTR_ERR(child));
2004 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2006 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2007 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2008 osd_ios_varfid_fill);
2014 #ifdef HAVE_FILLDIR_USE_CTX
2015 static int osd_ios_dl_fill(struct dir_context *buf,
2017 static int osd_ios_dl_fill(void *buf,
2019 const char *name, int namelen,
2020 loff_t offset, __u64 ino, unsigned d_type)
2022 struct osd_ios_filldir_buf *fill_buf =
2023 (struct osd_ios_filldir_buf *)buf;
2024 struct osd_device *dev = fill_buf->oifb_dev;
2025 const struct osd_lf_map *map;
2026 struct dentry *child;
2030 /* skip any '.' started names */
2034 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2035 if (map->olm_namelen != namelen)
2038 if (strncmp(map->olm_name, name, namelen) == 0)
2042 if (map->olm_name == NULL)
2045 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2047 RETURN(PTR_ERR(child));
2049 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2050 &map->olm_fid, map->olm_flags);
2056 #ifdef HAVE_FILLDIR_USE_CTX
2057 static int osd_ios_uld_fill(struct dir_context *buf,
2059 static int osd_ios_uld_fill(void *buf,
2061 const char *name, int namelen,
2062 loff_t offset, __u64 ino, unsigned d_type)
2064 struct osd_ios_filldir_buf *fill_buf =
2065 (struct osd_ios_filldir_buf *)buf;
2066 struct dentry *child;
2071 /* skip any non-DFID format name */
2075 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2077 RETURN(PTR_ERR(child));
2079 /* skip the start '[' */
2080 sscanf(&name[1], SFID, RFID(&tfid));
2081 if (fid_is_sane(&tfid))
2082 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2083 child->d_inode, &tfid, 0);
2091 #ifdef HAVE_FILLDIR_USE_CTX
2092 static int osd_ios_root_fill(struct dir_context *buf,
2094 static int osd_ios_root_fill(void *buf,
2096 const char *name, int namelen,
2097 loff_t offset, __u64 ino, unsigned d_type)
2099 struct osd_ios_filldir_buf *fill_buf =
2100 (struct osd_ios_filldir_buf *)buf;
2101 struct osd_device *dev = fill_buf->oifb_dev;
2102 const struct osd_lf_map *map;
2103 struct dentry *child;
2107 /* skip any '.' started names */
2111 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2112 if (map->olm_namelen != namelen)
2115 if (strncmp(map->olm_name, name, namelen) == 0)
2119 if (map->olm_name == NULL)
2122 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2124 RETURN(PTR_ERR(child));
2126 if (!(map->olm_flags & OLF_NO_OI))
2127 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2128 &map->olm_fid, map->olm_flags);
2129 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2130 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2138 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2139 struct dentry *dentry, filldir_t filldir)
2141 struct osd_ios_filldir_buf buf = {
2142 #ifdef HAVE_DIR_CONTEXT
2143 .ctx.actor = filldir,
2147 .oifb_dentry = dentry };
2148 struct file *filp = &info->oti_file;
2149 struct inode *inode = dentry->d_inode;
2150 const struct file_operations *fops = inode->i_fop;
2154 LASSERT(filldir != NULL);
2157 filp->f_path.dentry = dentry;
2158 filp->f_mode = FMODE_64BITHASH;
2159 filp->f_mapping = inode->i_mapping;
2161 filp->private_data = NULL;
2162 set_file_inode(filp, inode);
2164 #ifdef HAVE_DIR_CONTEXT
2165 buf.ctx.pos = filp->f_pos;
2166 rc = fops->iterate(filp, &buf.ctx);
2167 filp->f_pos = buf.ctx.pos;
2169 rc = fops->readdir(filp, &buf, filldir);
2171 fops->release(inode, filp);
2177 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2178 struct dentry *dentry, filldir_t filldir)
2180 struct osd_scrub *scrub = &dev->od_scrub;
2181 struct scrub_file *sf = &scrub->os_file;
2182 struct dentry *child;
2186 /* It is existing MDT0 device. We only allow the case of object without
2187 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2188 * can generate IGIF mode FID for the object and related OI mapping. If
2189 * it is on other MDTs, then becuase file-level backup/restore, related
2190 * OI mapping may be invalid already, we do not know which is the right
2191 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2193 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2194 * "fid" xattr, then something crashed. We cannot re-generate the
2195 * FID directly, instead, the OI scrub will scan the OI structure
2196 * and try to re-generate the LMA from the OI mapping. But if the
2197 * OI mapping crashed or lost also, then we have to give up under
2198 * double failure cases. */
2199 scrub->os_convert_igif = 1;
2200 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2201 strlen(dot_lustre_name));
2202 if (IS_ERR(child)) {
2203 rc = PTR_ERR(child);
2204 if (rc == -ENOENT) {
2205 /* It is 1.8 MDT device. */
2206 if (!(sf->sf_flags & SF_UPGRADE)) {
2207 osd_scrub_file_reset(scrub,
2208 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2210 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2211 rc = osd_scrub_file_store(scrub);
2217 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2218 * so the client will get IGIF for the ".lustre" object when
2221 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2222 * it does not know whether there are some old clients cached
2223 * the ".lustre" IGIF during the upgrading. Two choices:
2225 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2226 * It will allow the old connected clients to access the
2227 * ".lustre" with cached IGIF. But it will cause others
2228 * on the MDT failed to check "fid_is_dot_lustre()".
2230 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2231 * for ".lustre" in spite of whether there are some clients
2232 * cached the ".lustre" IGIF or not. It enables the check
2233 * "fid_is_dot_lustre()" on the MDT, although it will cause
2234 * that the old connected clients cannot access the ".lustre"
2235 * with the cached IGIF.
2237 * Usually, it is rare case for the old connected clients
2238 * to access the ".lustre" with cached IGIF. So we prefer
2239 * to the solution 2). */
2240 rc = osd_ios_scan_one(info, dev, child->d_inode,
2241 &LU_DOT_LUSTRE_FID, 0);
2243 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2252 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2253 struct dentry *dentry, filldir_t filldir)
2255 struct osd_scrub *scrub = &dev->od_scrub;
2256 struct scrub_file *sf = &scrub->os_file;
2257 struct dentry *child;
2261 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2262 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2263 rc = osd_scrub_file_store(scrub);
2268 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2269 if (!IS_ERR(child)) {
2270 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2273 rc = PTR_ERR(child);
2276 if (rc != 0 && rc != -ENOENT)
2279 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2280 if (!IS_ERR(child)) {
2281 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2284 rc = PTR_ERR(child);
2293 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2294 struct osd_device *dev)
2296 struct osd_ios_item *item = NULL;
2297 scandir_t scandir = osd_ios_general_scan;
2298 filldir_t filldir = osd_ios_root_fill;
2299 struct dentry *dentry = osd_sb(dev)->s_root;
2300 const struct osd_lf_map *map = osd_lf_maps;
2304 /* Lookup IGIF in OI by force for initial OI scrub. */
2305 dev->od_igif_inoi = 1;
2308 rc = scandir(info, dev, dentry, filldir);
2310 dput(item->oii_dentry);
2317 if (list_empty(&dev->od_ios_list))
2320 item = list_entry(dev->od_ios_list.next,
2321 struct osd_ios_item, oii_list);
2322 list_del_init(&item->oii_list);
2324 LASSERT(item->oii_scandir != NULL);
2325 scandir = item->oii_scandir;
2326 filldir = item->oii_filldir;
2327 dentry = item->oii_dentry;
2330 while (!list_empty(&dev->od_ios_list)) {
2331 item = list_entry(dev->od_ios_list.next,
2332 struct osd_ios_item, oii_list);
2333 list_del_init(&item->oii_list);
2334 dput(item->oii_dentry);
2341 /* There maybe the case that the object has been removed, but its OI
2342 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2343 * file-level backup/restore. So here cleanup the stale OI mappings. */
2344 while (map->olm_name != NULL) {
2345 struct dentry *child;
2347 if (fid_is_zero(&map->olm_fid)) {
2352 child = osd_ios_lookup_one_len(map->olm_name,
2353 osd_sb(dev)->s_root,
2357 else if (PTR_ERR(child) == -ENOENT)
2358 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2359 NULL, DTO_INDEX_DELETE,
2367 char *osd_lf_fid2name(const struct lu_fid *fid)
2369 const struct osd_lf_map *map = osd_lf_maps;
2371 while (map->olm_name != NULL) {
2372 if (!lu_fid_eq(fid, &map->olm_fid)) {
2377 if (map->olm_flags & OLF_SHOW_NAME)
2378 return map->olm_name;
2386 /* OI scrub start/stop */
2388 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2390 struct osd_scrub *scrub = &dev->od_scrub;
2391 struct ptlrpc_thread *thread = &scrub->os_thread;
2392 struct l_wait_info lwi = { 0 };
2393 struct task_struct *task;
2397 /* os_lock: sync status between stop and scrub thread */
2398 spin_lock(&scrub->os_lock);
2401 if (thread_is_running(thread)) {
2402 spin_unlock(&scrub->os_lock);
2403 if (!(scrub->os_file.sf_flags & SF_AUTO ||
2404 scrub->os_partial_scan) ||
2405 (flags & SS_AUTO_PARTIAL))
2408 osd_scrub_join(dev, flags, false);
2409 spin_lock(&scrub->os_lock);
2410 if (!thread_is_running(thread))
2413 spin_unlock(&scrub->os_lock);
2417 if (unlikely(thread_is_stopping(thread))) {
2418 spin_unlock(&scrub->os_lock);
2419 l_wait_event(thread->t_ctl_waitq,
2420 thread_is_stopped(thread),
2422 spin_lock(&scrub->os_lock);
2425 spin_unlock(&scrub->os_lock);
2427 if (scrub->os_file.sf_status == SS_COMPLETED) {
2428 if (!(flags & SS_SET_FAILOUT))
2429 flags |= SS_CLEAR_FAILOUT;
2431 if (!(flags & SS_SET_DRYRUN))
2432 flags |= SS_CLEAR_DRYRUN;
2437 scrub->os_start_flags = flags;
2438 thread_set_flags(thread, 0);
2439 task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2442 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2443 osd_scrub2name(scrub), rc);
2447 l_wait_event(thread->t_ctl_waitq,
2448 thread_is_running(thread) || thread_is_stopped(thread),
2454 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2459 /* od_otable_mutex: prevent curcurrent start/stop */
2460 mutex_lock(&dev->od_otable_mutex);
2461 rc = do_osd_scrub_start(dev, flags);
2462 mutex_unlock(&dev->od_otable_mutex);
2464 RETURN(rc == -EALREADY ? 0 : rc);
2467 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2469 struct ptlrpc_thread *thread = &scrub->os_thread;
2470 struct l_wait_info lwi = { 0 };
2472 /* os_lock: sync status between stop and scrub thread */
2473 spin_lock(&scrub->os_lock);
2474 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2475 thread_set_flags(thread, SVC_STOPPING);
2476 spin_unlock(&scrub->os_lock);
2477 wake_up_all(&thread->t_ctl_waitq);
2478 l_wait_event(thread->t_ctl_waitq,
2479 thread_is_stopped(thread),
2481 /* Do not skip the last lock/unlock, which can guarantee that
2482 * the caller cannot return until the OI scrub thread exit. */
2483 spin_lock(&scrub->os_lock);
2485 spin_unlock(&scrub->os_lock);
2488 static void osd_scrub_stop(struct osd_device *dev)
2490 /* od_otable_mutex: prevent curcurrent start/stop */
2491 mutex_lock(&dev->od_otable_mutex);
2492 dev->od_scrub.os_paused = 1;
2493 do_osd_scrub_stop(&dev->od_scrub);
2494 mutex_unlock(&dev->od_otable_mutex);
2497 /* OI scrub setup/cleanup */
2499 static const char osd_scrub_name[] = "OI_scrub";
2501 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2503 struct osd_thread_info *info = osd_oti_get(env);
2504 struct osd_scrub *scrub = &dev->od_scrub;
2505 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
2506 struct scrub_file *sf = &scrub->os_file;
2507 struct super_block *sb = osd_sb(dev);
2508 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2509 struct lvfs_run_ctxt saved;
2511 struct inode *inode;
2512 struct lu_fid *fid = &info->oti_fid;
2514 bool restored = false;
2518 memset(scrub, 0, sizeof(*scrub));
2519 OBD_SET_CTXT_MAGIC(ctxt);
2520 ctxt->pwdmnt = dev->od_mnt;
2521 ctxt->pwd = dev->od_mnt->mnt_root;
2522 ctxt->fs = get_ds();
2524 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2525 init_rwsem(&scrub->os_rwsem);
2526 spin_lock_init(&scrub->os_lock);
2527 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2529 push_ctxt(&saved, ctxt);
2530 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2532 pop_ctxt(&saved, ctxt);
2533 RETURN(PTR_ERR(filp));
2536 inode = file_inode(filp);
2537 /* 'What the @fid is' is not imporatant, because the object
2538 * has no OI mapping, and only is visible inside the OSD.*/
2539 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2540 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2542 filp_close(filp, NULL);
2543 pop_ctxt(&saved, ctxt);
2547 scrub->os_inode = igrab(inode);
2548 filp_close(filp, NULL);
2549 pop_ctxt(&saved, ctxt);
2551 rc = osd_scrub_file_load(scrub);
2552 if (rc == -ENOENT) {
2553 osd_scrub_file_init(scrub, es->s_uuid);
2554 /* If the "/O" dir does not exist when mount (indicated by
2555 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2556 * then it is quite probably that the device is a new one,
2557 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2559 * For the rare case that "/O" and "OI_scrub" both lost on
2560 * an old device, it can be found and cleared later.
2562 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2563 * need to check "filter_fid_old" and to convert it to
2564 * "filter_fid" for each object, and all the IGIF should
2565 * have their FID mapping in OI files already. */
2566 if (dev->od_maybe_new)
2567 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2569 } else if (rc != 0) {
2570 GOTO(cleanup_inode, rc);
2572 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2573 struct obd_uuid *old_uuid;
2574 struct obd_uuid *new_uuid;
2576 OBD_ALLOC_PTR(old_uuid);
2577 OBD_ALLOC_PTR(new_uuid);
2578 if (old_uuid == NULL || new_uuid == NULL) {
2579 CERROR("%.16s: UUID has been changed, but"
2580 "failed to allocate RAM for report\n",
2581 LDISKFS_SB(sb)->s_es->s_volume_name);
2583 class_uuid_unparse(sf->sf_uuid, old_uuid);
2584 class_uuid_unparse(es->s_uuid, new_uuid);
2585 CERROR("%.16s: UUID has been changed from "
2587 LDISKFS_SB(sb)->s_es->s_volume_name,
2588 old_uuid->uuid, new_uuid->uuid);
2590 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2593 if (old_uuid != NULL)
2594 OBD_FREE_PTR(old_uuid);
2595 if (new_uuid != NULL)
2596 OBD_FREE_PTR(new_uuid);
2597 } else if (sf->sf_status == SS_SCANNING) {
2598 sf->sf_status = SS_CRASHED;
2603 if (sf->sf_pos_last_checkpoint != 0)
2604 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2606 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2609 rc = osd_scrub_file_store(scrub);
2611 GOTO(cleanup_inode, rc);
2614 /* Initialize OI files. */
2615 rc = osd_oi_init(info, dev, restored);
2617 GOTO(cleanup_inode, rc);
2619 rc = osd_initial_OI_scrub(info, dev);
2621 GOTO(cleanup_oi, rc);
2623 if (sf->sf_flags & SF_UPGRADE ||
2624 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2625 sf->sf_success_count > 0)) {
2626 dev->od_igif_inoi = 0;
2627 dev->od_check_ff = dev->od_is_ost;
2629 dev->od_igif_inoi = 1;
2630 dev->od_check_ff = 0;
2633 if (sf->sf_flags & SF_INCONSISTENT)
2634 /* The 'od_igif_inoi' will be set under the
2636 * 1) new created system, or
2637 * 2) restored from file-level backup, or
2638 * 3) the upgrading completed.
2640 * The 'od_igif_inoi' may be cleared by OI scrub
2641 * later if found that the system is upgrading. */
2642 dev->od_igif_inoi = 1;
2644 if (!dev->od_noscrub &&
2645 ((sf->sf_status == SS_PAUSED) ||
2646 (sf->sf_status == SS_CRASHED &&
2647 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2648 SF_UPGRADE | SF_AUTO)) ||
2649 (sf->sf_status == SS_INIT &&
2650 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2652 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2655 GOTO(cleanup_oi, rc);
2657 /* it is possible that dcache entries may keep objects after they are
2658 * deleted by OSD. While it looks safe this can cause object data to
2659 * stay until umount causing failures in tests calculating free space,
2660 * e.g. replay-ost-single. Since those dcache entries are not used
2661 * anymore let's just free them after use here */
2662 shrink_dcache_sb(sb);
2666 osd_oi_fini(info, dev);
2668 iput(scrub->os_inode);
2669 scrub->os_inode = NULL;
2674 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2676 struct osd_scrub *scrub = &dev->od_scrub;
2678 LASSERT(dev->od_otable_it == NULL);
2680 if (scrub->os_inode != NULL) {
2681 osd_scrub_stop(dev);
2682 iput(scrub->os_inode);
2683 scrub->os_inode = NULL;
2685 if (dev->od_oi_table != NULL)
2686 osd_oi_fini(osd_oti_get(env), dev);
2689 /* object table based iteration APIs */
2691 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2692 struct dt_object *dt, __u32 attr)
2694 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2695 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2696 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2697 struct osd_scrub *scrub = &dev->od_scrub;
2698 struct osd_otable_it *it;
2703 /* od_otable_mutex: prevent curcurrent init/fini */
2704 mutex_lock(&dev->od_otable_mutex);
2705 if (dev->od_otable_it != NULL)
2706 GOTO(out, it = ERR_PTR(-EALREADY));
2710 GOTO(out, it = ERR_PTR(-ENOMEM));
2712 dev->od_otable_it = it;
2714 it->ooi_cache.ooc_consumer_idx = -1;
2715 if (flags & DOIF_OUTUSED)
2716 it->ooi_used_outside = 1;
2718 if (flags & DOIF_RESET)
2721 if (valid & DOIV_ERROR_HANDLE) {
2722 if (flags & DOIF_FAILOUT)
2723 start |= SS_SET_FAILOUT;
2725 start |= SS_CLEAR_FAILOUT;
2728 if (valid & DOIV_DRYRUN) {
2729 if (flags & DOIF_DRYRUN)
2730 start |= SS_SET_DRYRUN;
2732 start |= SS_CLEAR_DRYRUN;
2735 rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2736 if (rc < 0 && rc != -EALREADY) {
2737 dev->od_otable_it = NULL;
2739 GOTO(out, it = ERR_PTR(rc));
2742 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2747 mutex_unlock(&dev->od_otable_mutex);
2748 return (struct dt_it *)it;
2751 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2753 struct osd_otable_it *it = (struct osd_otable_it *)di;
2754 struct osd_device *dev = it->ooi_dev;
2756 /* od_otable_mutex: prevent curcurrent init/fini */
2757 mutex_lock(&dev->od_otable_mutex);
2758 do_osd_scrub_stop(&dev->od_scrub);
2759 LASSERT(dev->od_otable_it == it);
2761 dev->od_otable_it = NULL;
2762 mutex_unlock(&dev->od_otable_mutex);
2766 static int osd_otable_it_get(const struct lu_env *env,
2767 struct dt_it *di, const struct dt_key *key)
2772 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2777 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2779 spin_lock(&scrub->os_lock);
2780 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2781 scrub->os_waiting ||
2782 !thread_is_running(&scrub->os_thread))
2783 it->ooi_waiting = 0;
2785 it->ooi_waiting = 1;
2786 spin_unlock(&scrub->os_lock);
2788 return !it->ooi_waiting;
2791 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2793 struct osd_otable_it *it = (struct osd_otable_it *)di;
2794 struct osd_device *dev = it->ooi_dev;
2795 struct osd_scrub *scrub = &dev->od_scrub;
2796 struct osd_otable_cache *ooc = &it->ooi_cache;
2797 struct ptlrpc_thread *thread = &scrub->os_thread;
2798 struct l_wait_info lwi = { 0 };
2802 LASSERT(it->ooi_user_ready);
2805 if (!thread_is_running(thread) && !it->ooi_used_outside)
2808 if (ooc->ooc_cached_items > 0) {
2809 ooc->ooc_cached_items--;
2810 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2811 ~OSD_OTABLE_IT_CACHE_MASK;
2815 if (it->ooi_all_cached) {
2816 l_wait_event(thread->t_ctl_waitq,
2817 !thread_is_running(thread),
2822 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2823 spin_lock(&scrub->os_lock);
2824 scrub->os_waiting = 0;
2825 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2826 spin_unlock(&scrub->os_lock);
2829 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2830 l_wait_event(thread->t_ctl_waitq,
2831 osd_otable_it_wakeup(scrub, it),
2834 if (!thread_is_running(thread) && !it->ooi_used_outside)
2837 rc = osd_otable_it_preload(env, it);
2844 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2845 const struct dt_it *di)
2850 static int osd_otable_it_key_size(const struct lu_env *env,
2851 const struct dt_it *di)
2853 return sizeof(__u64);
2856 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2857 struct dt_rec *rec, __u32 attr)
2859 struct osd_otable_it *it = (struct osd_otable_it *)di;
2860 struct osd_otable_cache *ooc = &it->ooi_cache;
2862 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2864 /* Filter out Invald FID already. */
2865 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2866 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2867 PFID((struct lu_fid *)rec),
2868 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2873 static __u64 osd_otable_it_store(const struct lu_env *env,
2874 const struct dt_it *di)
2876 struct osd_otable_it *it = (struct osd_otable_it *)di;
2877 struct osd_otable_cache *ooc = &it->ooi_cache;
2880 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2881 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2883 hash = ooc->ooc_pos_preload;
2888 * Set the OSD layer iteration start position as the specified hash.
2890 static int osd_otable_it_load(const struct lu_env *env,
2891 const struct dt_it *di, __u64 hash)
2893 struct osd_otable_it *it = (struct osd_otable_it *)di;
2894 struct osd_device *dev = it->ooi_dev;
2895 struct osd_otable_cache *ooc = &it->ooi_cache;
2896 struct osd_scrub *scrub = &dev->od_scrub;
2900 /* Forbid to set iteration position after iteration started. */
2901 if (it->ooi_user_ready)
2904 LASSERT(!scrub->os_partial_scan);
2906 if (hash > OSD_OTABLE_MAX_HASH)
2907 hash = OSD_OTABLE_MAX_HASH;
2909 ooc->ooc_pos_preload = hash;
2910 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2911 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2913 it->ooi_user_ready = 1;
2914 if (!scrub->os_full_speed)
2915 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2917 /* Unplug OSD layer iteration by the first next() call. */
2918 rc = osd_otable_it_next(env, (struct dt_it *)it);
2923 static int osd_otable_it_key_rec(const struct lu_env *env,
2924 const struct dt_it *di, void *key_rec)
2929 const struct dt_index_operations osd_otable_ops = {
2931 .init = osd_otable_it_init,
2932 .fini = osd_otable_it_fini,
2933 .get = osd_otable_it_get,
2934 .put = osd_otable_it_put,
2935 .next = osd_otable_it_next,
2936 .key = osd_otable_it_key,
2937 .key_size = osd_otable_it_key_size,
2938 .rec = osd_otable_it_rec,
2939 .store = osd_otable_it_store,
2940 .load = osd_otable_it_load,
2941 .key_rec = osd_otable_it_key_rec,
2945 /* high priority inconsistent items list APIs */
2947 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
2949 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2952 struct osd_inconsistent_item *oii;
2953 struct osd_scrub *scrub = &dev->od_scrub;
2954 struct ptlrpc_thread *thread = &scrub->os_thread;
2959 if (unlikely(oii == NULL))
2962 INIT_LIST_HEAD(&oii->oii_list);
2963 oii->oii_cache = *oic;
2964 oii->oii_insert = insert;
2966 if (scrub->os_partial_scan) {
2967 __u64 now = cfs_time_current_sec();
2969 /* If there haven't been errors in a long time,
2970 * decay old count until either the errors are
2971 * gone or we reach the current interval. */
2972 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2973 scrub->os_bad_oimap_time +
2974 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2975 scrub->os_bad_oimap_count >>= 1;
2976 scrub->os_bad_oimap_time +=
2977 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2980 scrub->os_bad_oimap_time = now;
2981 if (++scrub->os_bad_oimap_count >
2982 dev->od_full_scrub_threshold_rate)
2983 scrub->os_full_scrub = 1;
2986 spin_lock(&scrub->os_lock);
2987 if (unlikely(!thread_is_running(thread))) {
2988 spin_unlock(&scrub->os_lock);
2993 if (list_empty(&scrub->os_inconsistent_items))
2995 list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2996 spin_unlock(&scrub->os_lock);
2999 wake_up_all(&thread->t_ctl_waitq);
3004 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3005 struct osd_inode_id *id)
3007 struct osd_scrub *scrub = &dev->od_scrub;
3008 struct osd_inconsistent_item *oii;
3011 spin_lock(&scrub->os_lock);
3012 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3013 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3014 *id = oii->oii_cache.oic_lid;
3015 spin_unlock(&scrub->os_lock);
3019 spin_unlock(&scrub->os_lock);
3026 static const char *scrub_status_names[] = {
3037 static const char *scrub_flags_names[] = {
3045 static const char *scrub_param_names[] = {
3051 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3057 seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3059 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3062 seq_printf(m, "%s%c", names[i],
3063 bits != 0 ? ',' : '\n');
3068 static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3071 seq_printf(m, "%s: %llu seconds\n", prefix,
3072 cfs_time_current_sec() - time);
3074 seq_printf(m, "%s: N/A\n", prefix);
3077 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3080 seq_printf(m, "%s: %llu\n", prefix, pos);
3082 seq_printf(m, "%s: N/A\n", prefix);
3085 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3087 struct osd_scrub *scrub = &dev->od_scrub;
3088 struct scrub_file *sf = &scrub->os_file;
3092 down_read(&scrub->os_rwsem);
3093 seq_printf(m, "name: OI_scrub\n"
3097 sf->sf_magic, (int)sf->sf_oi_count,
3098 scrub_status_names[sf->sf_status]);
3100 scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
3102 scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
3104 scrub_time_dump(m, sf->sf_time_last_complete,
3105 "time_since_last_completed");
3107 scrub_time_dump(m, sf->sf_time_latest_start,
3108 "time_since_latest_start");
3110 scrub_time_dump(m, sf->sf_time_last_checkpoint,
3111 "time_since_last_checkpoint");
3113 scrub_pos_dump(m, sf->sf_pos_latest_start,
3114 "latest_start_position");
3116 scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3117 "last_checkpoint_position");
3119 scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3120 "first_failure_position");
3122 checked = sf->sf_items_checked + scrub->os_new_checked;
3123 seq_printf(m, "checked: %llu\n"
3126 "prior_updated: %llu\n"
3129 "success_count: %u\n",
3130 checked, sf->sf_items_updated, sf->sf_items_failed,
3131 sf->sf_items_updated_prior, sf->sf_items_noscrub,
3132 sf->sf_items_igif, sf->sf_success_count);
3135 if (thread_is_running(&scrub->os_thread)) {
3136 cfs_duration_t duration = cfs_time_current() -
3137 scrub->os_time_last_checkpoint;
3138 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3140 __u32 rtime = sf->sf_run_time +
3141 cfs_duration_sec(duration + HALF_SEC);
3144 do_div(new_checked, duration);
3146 do_div(speed, rtime);
3147 seq_printf(m, "run_time: %u seconds\n"
3148 "average_speed: %llu objects/sec\n"
3149 "real-time_speed: %llu objects/sec\n"
3150 "current_position: %u\n"
3151 "lf_scanned: %llu\n"
3152 "lf_repaired: %llu\n"
3153 "lf_failed: %llu\n",
3154 rtime, speed, new_checked, scrub->os_pos_current,
3155 scrub->os_lf_scanned, scrub->os_lf_repaired,
3156 scrub->os_lf_failed);
3158 if (sf->sf_run_time != 0)
3159 do_div(speed, sf->sf_run_time);
3160 seq_printf(m, "run_time: %u seconds\n"
3161 "average_speed: %llu objects/sec\n"
3162 "real-time_speed: N/A\n"
3163 "current_position: N/A\n"
3164 "lf_scanned: %llu\n"
3165 "lf_repaired: %llu\n"
3166 "lf_failed: %llu\n",
3167 sf->sf_run_time, speed, scrub->os_lf_scanned,
3168 scrub->os_lf_repaired, scrub->os_lf_failed);
3171 up_read(&scrub->os_rwsem);