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, 2016, 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;
1388 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1389 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1390 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1391 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1393 while (*pos <= limit && *count < max) {
1394 struct ldiskfs_group_desc *desc;
1395 bool next_group = false;
1397 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1401 ldiskfs_lock_group(param.sb, param.bg);
1402 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1403 ldiskfs_unlock_group(param.sb, param.bg);
1407 ldiskfs_unlock_group(param.sb, param.bg);
1409 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1410 if (!param.bitmap) {
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) &&
1419 struct osd_idmap_cache *oic = NULL;
1422 ldiskfs_itable_unused_count(param.sb, desc) >
1423 LDISKFS_INODES_PER_GROUP(param.sb)) {
1428 rc = next(info, dev, ¶m, &oic, noslot);
1430 case SCRUB_NEXT_BREAK:
1433 case SCRUB_NEXT_EXIT:
1434 brelse(param.bitmap);
1436 case SCRUB_NEXT_CRASH:
1437 brelse(param.bitmap);
1438 RETURN(SCRUB_IT_CRASH);
1439 case SCRUB_NEXT_FATAL:
1440 brelse(param.bitmap);
1444 rc = exec(info, dev, ¶m, oic, &noslot, rc);
1446 brelse(param.bitmap);
1453 brelse(param.bitmap);
1454 param.bitmap = NULL;
1461 param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1467 RETURN(SCRUB_IT_ALL);
1471 static int osd_otable_it_preload(const struct lu_env *env,
1472 struct osd_otable_it *it)
1474 struct osd_device *dev = it->ooi_dev;
1475 struct osd_scrub *scrub = &dev->od_scrub;
1476 struct osd_otable_cache *ooc = &it->ooi_cache;
1480 rc = osd_inode_iteration(osd_oti_get(env), dev,
1481 OSD_OTABLE_IT_CACHE_SIZE, true);
1482 if (rc == SCRUB_IT_ALL)
1483 it->ooi_all_cached = 1;
1485 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1486 spin_lock(&scrub->os_lock);
1487 scrub->os_waiting = 0;
1488 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1489 spin_unlock(&scrub->os_lock);
1492 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1495 static int osd_scrub_main(void *args)
1498 struct osd_device *dev = (struct osd_device *)args;
1499 struct osd_scrub *scrub = &dev->od_scrub;
1500 struct ptlrpc_thread *thread = &scrub->os_thread;
1504 rc = lu_env_init(&env, LCT_LOCAL);
1506 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1507 osd_scrub2name(scrub), rc);
1511 rc = osd_scrub_prep(dev);
1513 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1514 osd_scrub2name(scrub), rc);
1518 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1519 struct l_wait_info lwi = { 0 };
1520 struct osd_otable_it *it = dev->od_otable_it;
1521 struct osd_otable_cache *ooc = &it->ooi_cache;
1523 l_wait_event(thread->t_ctl_waitq,
1524 it->ooi_user_ready || !thread_is_running(thread),
1526 if (unlikely(!thread_is_running(thread)))
1529 scrub->os_pos_current = ooc->ooc_pos_preload;
1532 CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1533 osd_scrub2name(scrub), scrub->os_start_flags,
1534 scrub->os_pos_current);
1536 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1537 if (unlikely(rc == SCRUB_IT_CRASH))
1538 GOTO(out, rc = -EINVAL);
1542 rc = osd_scrub_post(scrub, rc);
1543 CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1544 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1547 while (!list_empty(&scrub->os_inconsistent_items)) {
1548 struct osd_inconsistent_item *oii;
1550 oii = list_entry(scrub->os_inconsistent_items.next,
1551 struct osd_inconsistent_item, oii_list);
1552 list_del_init(&oii->oii_list);
1558 spin_lock(&scrub->os_lock);
1559 thread_set_flags(thread, SVC_STOPPED);
1560 wake_up_all(&thread->t_ctl_waitq);
1561 spin_unlock(&scrub->os_lock);
1565 /* initial OI scrub */
1567 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1568 struct dentry *, filldir_t filldir);
1570 #ifdef HAVE_FILLDIR_USE_CTX
1571 static int osd_ios_varfid_fill(struct dir_context *buf, const char *name,
1572 int namelen, loff_t offset, __u64 ino,
1574 static int osd_ios_lf_fill(struct dir_context *buf, const char *name,
1575 int namelen, loff_t offset, __u64 ino,
1577 static int osd_ios_dl_fill(struct dir_context *buf, const char *name,
1578 int namelen, loff_t offset, __u64 ino,
1580 static int osd_ios_uld_fill(struct dir_context *buf, const char *name,
1581 int namelen, loff_t offset, __u64 ino,
1584 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1585 loff_t offset, __u64 ino, unsigned d_type);
1586 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1587 loff_t offset, __u64 ino, unsigned d_type);
1588 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1589 loff_t offset, __u64 ino, unsigned d_type);
1590 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1591 loff_t offset, __u64 ino, unsigned d_type);
1595 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1596 struct dentry *dentry, filldir_t filldir);
1598 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1599 struct dentry *dentry, filldir_t filldir);
1602 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1603 struct dentry *dentry, filldir_t filldir);
1606 OLF_SCAN_SUBITEMS = 0x0001,
1607 OLF_HIDE_FID = 0x0002,
1608 OLF_SHOW_NAME = 0x0004,
1610 OLF_IDX_IN_FID = 0x0010,
1615 struct lu_fid olm_fid;
1618 scandir_t olm_scandir;
1619 filldir_t olm_filldir;
1622 /* Add the new introduced local files in the list in the future. */
1623 static const struct osd_lf_map osd_lf_maps[] = {
1626 .olm_name = CATLIST,
1628 .f_seq = FID_SEQ_LOCAL_FILE,
1629 .f_oid = LLOG_CATALOGS_OID,
1631 .olm_flags = OLF_SHOW_NAME,
1632 .olm_namelen = sizeof(CATLIST) - 1,
1637 .olm_name = MOUNT_CONFIGS_DIR,
1639 .f_seq = FID_SEQ_LOCAL_FILE,
1640 .f_oid = MGS_CONFIGS_OID,
1642 .olm_flags = OLF_SCAN_SUBITEMS,
1643 .olm_namelen = sizeof(MOUNT_CONFIGS_DIR) - 1,
1644 .olm_scandir = osd_ios_general_scan,
1645 .olm_filldir = osd_ios_varfid_fill,
1648 /* NIDTBL_VERSIONS */
1650 .olm_name = MGS_NIDTBL_DIR,
1651 .olm_flags = OLF_SCAN_SUBITEMS,
1652 .olm_namelen = sizeof(MGS_NIDTBL_DIR) - 1,
1653 .olm_scandir = osd_ios_general_scan,
1654 .olm_filldir = osd_ios_varfid_fill,
1659 .olm_name = "PENDING",
1660 .olm_namelen = sizeof("PENDING") - 1,
1667 .f_seq = FID_SEQ_ROOT,
1668 .f_oid = FID_OID_ROOT,
1670 .olm_flags = OLF_SCAN_SUBITEMS | OLF_HIDE_FID,
1671 .olm_namelen = sizeof("ROOT") - 1,
1672 .olm_scandir = osd_ios_ROOT_scan,
1675 /* changelog_catalog */
1677 .olm_name = CHANGELOG_CATALOG,
1678 .olm_namelen = sizeof(CHANGELOG_CATALOG) - 1,
1681 /* changelog_users */
1683 .olm_name = CHANGELOG_USERS,
1684 .olm_namelen = sizeof(CHANGELOG_USERS) - 1,
1691 .f_seq = FID_SEQ_LOCAL_FILE,
1692 .f_oid = FLD_INDEX_OID,
1694 .olm_flags = OLF_SHOW_NAME,
1695 .olm_namelen = sizeof("fld") - 1,
1700 .olm_name = LAST_RCVD,
1702 .f_seq = FID_SEQ_LOCAL_FILE,
1703 .f_oid = LAST_RECV_OID,
1705 .olm_flags = OLF_SHOW_NAME,
1706 .olm_namelen = sizeof(LAST_RCVD) - 1,
1711 .olm_name = REPLY_DATA,
1713 .f_seq = FID_SEQ_LOCAL_FILE,
1714 .f_oid = REPLY_DATA_OID,
1716 .olm_flags = OLF_SHOW_NAME,
1717 .olm_namelen = sizeof(REPLY_DATA) - 1,
1722 .olm_name = LOV_OBJID,
1724 .f_seq = FID_SEQ_LOCAL_FILE,
1725 .f_oid = MDD_LOV_OBJ_OID,
1727 .olm_flags = OLF_SHOW_NAME,
1728 .olm_namelen = sizeof(LOV_OBJID) - 1,
1733 .olm_name = LOV_OBJSEQ,
1735 .f_seq = FID_SEQ_LOCAL_FILE,
1736 .f_oid = MDD_LOV_OBJ_OSEQ,
1738 .olm_flags = OLF_SHOW_NAME,
1739 .olm_namelen = sizeof(LOV_OBJSEQ) - 1,
1744 .olm_name = QMT_DIR,
1745 .olm_flags = OLF_SCAN_SUBITEMS,
1746 .olm_namelen = sizeof(QMT_DIR) - 1,
1747 .olm_scandir = osd_ios_general_scan,
1748 .olm_filldir = osd_ios_varfid_fill,
1753 .olm_name = QSD_DIR,
1754 .olm_flags = OLF_SCAN_SUBITEMS,
1755 .olm_namelen = sizeof(QSD_DIR) - 1,
1756 .olm_scandir = osd_ios_general_scan,
1757 .olm_filldir = osd_ios_varfid_fill,
1762 .olm_name = "seq_ctl",
1764 .f_seq = FID_SEQ_LOCAL_FILE,
1765 .f_oid = FID_SEQ_CTL_OID,
1767 .olm_flags = OLF_SHOW_NAME,
1768 .olm_namelen = sizeof("seq_ctl") - 1,
1773 .olm_name = "seq_srv",
1775 .f_seq = FID_SEQ_LOCAL_FILE,
1776 .f_oid = FID_SEQ_SRV_OID,
1778 .olm_flags = OLF_SHOW_NAME,
1779 .olm_namelen = sizeof("seq_srv") - 1,
1784 .olm_name = HEALTH_CHECK,
1786 .f_seq = FID_SEQ_LOCAL_FILE,
1787 .f_oid = OFD_HEALTH_CHECK_OID,
1789 .olm_flags = OLF_SHOW_NAME,
1790 .olm_namelen = sizeof(HEALTH_CHECK) - 1,
1795 .olm_name = LFSCK_DIR,
1796 .olm_namelen = sizeof(LFSCK_DIR) - 1,
1797 .olm_scandir = osd_ios_general_scan,
1798 .olm_filldir = osd_ios_varfid_fill,
1801 /* lfsck_bookmark */
1803 .olm_name = LFSCK_BOOKMARK,
1804 .olm_namelen = sizeof(LFSCK_BOOKMARK) - 1,
1809 .olm_name = LFSCK_LAYOUT,
1810 .olm_namelen = sizeof(LFSCK_LAYOUT) - 1,
1813 /* lfsck_namespace */
1815 .olm_name = LFSCK_NAMESPACE,
1816 .olm_namelen = sizeof(LFSCK_NAMESPACE) - 1,
1819 /* OBJECTS, upgrade from old device */
1821 .olm_name = OBJECTS,
1822 .olm_flags = OLF_SCAN_SUBITEMS,
1823 .olm_namelen = sizeof(OBJECTS) - 1,
1824 .olm_scandir = osd_ios_OBJECTS_scan,
1827 /* lquota_v2.user, upgrade from old device */
1829 .olm_name = "lquota_v2.user",
1830 .olm_namelen = sizeof("lquota_v2.user") - 1,
1833 /* lquota_v2.group, upgrade from old device */
1835 .olm_name = "lquota_v2.group",
1836 .olm_namelen = sizeof("lquota_v2.group") - 1,
1839 /* LAST_GROUP, upgrade from old device */
1841 .olm_name = "LAST_GROUP",
1843 .f_seq = FID_SEQ_LOCAL_FILE,
1844 .f_oid = OFD_LAST_GROUP_OID,
1846 .olm_flags = OLF_SHOW_NAME,
1847 .olm_namelen = sizeof("LAST_GROUP") - 1,
1850 /* committed batchid for cross-MDT operation */
1852 .olm_name = "BATCHID",
1854 .f_seq = FID_SEQ_LOCAL_FILE,
1855 .f_oid = BATCHID_COMMITTED_OID,
1857 .olm_flags = OLF_SHOW_NAME,
1858 .olm_namelen = sizeof("BATCHID") - 1,
1861 /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1862 * and f_oid = index for their log files. See lu_update_log{_dir}_fid()
1863 * for more details. */
1867 .olm_name = "update_log",
1869 .f_seq = FID_SEQ_UPDATE_LOG,
1871 .olm_flags = OLF_SHOW_NAME | OLF_IDX_IN_FID,
1872 .olm_namelen = sizeof("update_log") - 1,
1875 /* update_log_dir */
1877 .olm_name = "update_log_dir",
1879 .f_seq = FID_SEQ_UPDATE_LOG_DIR,
1881 .olm_flags = OLF_SHOW_NAME | OLF_SCAN_SUBITEMS |
1883 .olm_namelen = sizeof("update_log_dir") - 1,
1884 .olm_scandir = osd_ios_general_scan,
1885 .olm_filldir = osd_ios_uld_fill,
1890 .olm_name = "lost+found",
1892 .f_seq = FID_SEQ_LOCAL_FILE,
1893 .f_oid = OSD_LPF_OID,
1895 .olm_flags = OLF_SCAN_SUBITEMS,
1896 .olm_namelen = sizeof("lost+found") - 1,
1897 .olm_scandir = osd_ios_general_scan,
1898 .olm_filldir = osd_ios_lf_fill,
1906 /* Add the new introduced files under .lustre/ in the list in the future. */
1907 static const struct osd_lf_map osd_dl_maps[] = {
1912 .f_seq = FID_SEQ_DOT_LUSTRE,
1913 .f_oid = FID_OID_DOT_LUSTRE_OBF,
1915 .olm_namelen = sizeof("fid") - 1,
1917 /* .lustre/lost+found */
1919 .olm_name = "lost+found",
1921 .f_seq = FID_SEQ_DOT_LUSTRE,
1922 .f_oid = FID_OID_DOT_LUSTRE_LPF,
1924 .olm_namelen = sizeof("lost+found") - 1,
1931 struct osd_ios_item {
1932 struct list_head oii_list;
1933 struct dentry *oii_dentry;
1934 scandir_t oii_scandir;
1935 filldir_t oii_filldir;
1938 struct osd_ios_filldir_buf {
1939 #ifdef HAVE_DIR_CONTEXT
1940 /* please keep it as first member */
1941 struct dir_context ctx;
1943 struct osd_thread_info *oifb_info;
1944 struct osd_device *oifb_dev;
1945 struct dentry *oifb_dentry;
1948 static inline struct dentry *
1949 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1951 struct dentry *dentry;
1953 dentry = ll_lookup_one_len(name, parent, namelen);
1954 if (IS_ERR(dentry)) {
1955 int rc = PTR_ERR(dentry);
1958 CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1959 namelen, name, parent->d_name.len,
1960 parent->d_name.name, parent->d_inode->i_ino,
1961 parent->d_inode->i_generation, rc);
1966 if (dentry->d_inode == NULL) {
1968 return ERR_PTR(-ENOENT);
1975 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1976 scandir_t scandir, filldir_t filldir)
1978 struct osd_ios_item *item;
1981 OBD_ALLOC_PTR(item);
1985 INIT_LIST_HEAD(&item->oii_list);
1986 item->oii_dentry = dget(dentry);
1987 item->oii_scandir = scandir;
1988 item->oii_filldir = filldir;
1989 list_add_tail(&item->oii_list, &dev->od_ios_list);
1995 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1997 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1998 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1999 * reference the inode, or fixed if it is missing or references another inode.
2002 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
2003 struct inode *inode, const struct lu_fid *fid, int flags)
2005 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
2006 struct osd_inode_id *id = &info->oti_id;
2007 struct osd_inode_id *id2 = &info->oti_id2;
2008 struct osd_scrub *scrub = &dev->od_scrub;
2009 struct scrub_file *sf = &scrub->os_file;
2014 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
2015 if (rc != 0 && rc != -ENODATA) {
2016 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
2017 "rc = %d\n", osd_name(dev), rc);
2022 osd_id_gen(id, inode->i_ino, inode->i_generation);
2023 if (rc == -ENODATA) {
2024 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
2025 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
2028 if (flags & OLF_IDX_IN_FID) {
2029 LASSERT(dev->od_index >= 0);
2031 tfid.f_oid = dev->od_index;
2034 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
2036 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
2037 "scrub: rc = %d\n", osd_name(dev), rc);
2042 if (lma->lma_compat & LMAC_NOT_IN_OI)
2045 tfid = lma->lma_self_fid;
2048 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
2053 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2054 DTO_INDEX_INSERT, true, 0, NULL);
2061 if (osd_id_eq_strict(id, id2))
2064 if (!(sf->sf_flags & SF_INCONSISTENT)) {
2065 osd_scrub_file_reset(scrub,
2066 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2068 rc = osd_scrub_file_store(scrub);
2073 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
2074 DTO_INDEX_UPDATE, true, 0, NULL);
2082 * It scans the /lost+found, and for the OST-object (with filter_fid
2083 * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
2085 #ifdef HAVE_FILLDIR_USE_CTX
2086 static int osd_ios_lf_fill(struct dir_context *buf,
2088 static int osd_ios_lf_fill(void *buf,
2090 const char *name, int namelen,
2091 loff_t offset, __u64 ino, unsigned d_type)
2093 struct osd_ios_filldir_buf *fill_buf =
2094 (struct osd_ios_filldir_buf *)buf;
2095 struct osd_thread_info *info = fill_buf->oifb_info;
2096 struct osd_device *dev = fill_buf->oifb_dev;
2097 struct lu_fid *fid = &info->oti_fid;
2098 struct osd_scrub *scrub = &dev->od_scrub;
2099 struct dentry *parent = fill_buf->oifb_dentry;
2100 struct dentry *child;
2101 struct inode *dir = parent->d_inode;
2102 struct inode *inode;
2106 /* skip any '.' started names */
2110 scrub->os_lf_scanned++;
2111 child = osd_ios_lookup_one_len(name, parent, namelen);
2112 if (IS_ERR(child)) {
2113 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
2114 osd_name(dev), namelen, name, (int)PTR_ERR(child));
2118 inode = child->d_inode;
2119 if (S_ISDIR(inode->i_mode)) {
2120 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2123 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
2124 "rc = %d\n", osd_name(dev), namelen, name, rc);
2128 if (!S_ISREG(inode->i_mode))
2131 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
2132 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
2133 rc = osd_obj_map_recover(info, dev, dir, child, fid);
2135 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
2136 "/lost+found.\n", namelen, name, PFID(fid));
2137 scrub->os_lf_repaired++;
2139 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
2141 osd_name(dev), namelen, name, PFID(fid), rc);
2145 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
2146 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
2147 * can process them in furtuer. */
2153 scrub->os_lf_failed++;
2155 /* skip the failure to make the scanning to continue. */
2159 #ifdef HAVE_FILLDIR_USE_CTX
2160 static int osd_ios_varfid_fill(struct dir_context *buf,
2162 static int osd_ios_varfid_fill(void *buf,
2164 const char *name, int namelen,
2165 loff_t offset, __u64 ino, unsigned d_type)
2167 struct osd_ios_filldir_buf *fill_buf =
2168 (struct osd_ios_filldir_buf *)buf;
2169 struct osd_device *dev = fill_buf->oifb_dev;
2170 struct dentry *child;
2174 /* skip any '.' started names */
2178 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2180 RETURN(PTR_ERR(child));
2182 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2184 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
2185 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2186 osd_ios_varfid_fill);
2192 #ifdef HAVE_FILLDIR_USE_CTX
2193 static int osd_ios_dl_fill(struct dir_context *buf,
2195 static int osd_ios_dl_fill(void *buf,
2197 const char *name, int namelen,
2198 loff_t offset, __u64 ino, unsigned d_type)
2200 struct osd_ios_filldir_buf *fill_buf =
2201 (struct osd_ios_filldir_buf *)buf;
2202 struct osd_device *dev = fill_buf->oifb_dev;
2203 const struct osd_lf_map *map;
2204 struct dentry *child;
2208 /* skip any '.' started names */
2212 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
2213 if (map->olm_namelen != namelen)
2216 if (strncmp(map->olm_name, name, namelen) == 0)
2220 if (map->olm_name == NULL)
2223 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2225 RETURN(PTR_ERR(child));
2227 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2228 &map->olm_fid, map->olm_flags);
2234 #ifdef HAVE_FILLDIR_USE_CTX
2235 static int osd_ios_uld_fill(struct dir_context *buf,
2237 static int osd_ios_uld_fill(void *buf,
2239 const char *name, int namelen,
2240 loff_t offset, __u64 ino, unsigned d_type)
2242 struct osd_ios_filldir_buf *fill_buf =
2243 (struct osd_ios_filldir_buf *)buf;
2244 struct dentry *child;
2249 /* skip any non-DFID format name */
2253 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2255 RETURN(PTR_ERR(child));
2257 /* skip the start '[' */
2258 sscanf(&name[1], SFID, RFID(&tfid));
2259 if (fid_is_sane(&tfid))
2260 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2261 child->d_inode, &tfid, 0);
2269 #ifdef HAVE_FILLDIR_USE_CTX
2270 static int osd_ios_root_fill(struct dir_context *buf,
2272 static int osd_ios_root_fill(void *buf,
2274 const char *name, int namelen,
2275 loff_t offset, __u64 ino, unsigned d_type)
2277 struct osd_ios_filldir_buf *fill_buf =
2278 (struct osd_ios_filldir_buf *)buf;
2279 struct osd_device *dev = fill_buf->oifb_dev;
2280 const struct osd_lf_map *map;
2281 struct dentry *child;
2285 /* skip any '.' started names */
2289 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2290 if (map->olm_namelen != namelen)
2293 if (strncmp(map->olm_name, name, namelen) == 0)
2297 if (map->olm_name == NULL)
2300 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2302 RETURN(PTR_ERR(child));
2304 if (!(map->olm_flags & OLF_NO_OI))
2305 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2306 &map->olm_fid, map->olm_flags);
2307 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2308 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2316 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2317 struct dentry *dentry, filldir_t filldir)
2319 struct osd_ios_filldir_buf buf = {
2320 #ifdef HAVE_DIR_CONTEXT
2321 .ctx.actor = filldir,
2325 .oifb_dentry = dentry };
2326 struct file *filp = &info->oti_file;
2327 struct inode *inode = dentry->d_inode;
2328 const struct file_operations *fops = inode->i_fop;
2332 LASSERT(filldir != NULL);
2335 filp->f_path.dentry = dentry;
2336 filp->f_mode = FMODE_64BITHASH;
2337 filp->f_mapping = inode->i_mapping;
2339 filp->private_data = NULL;
2340 set_file_inode(filp, inode);
2342 #ifdef HAVE_DIR_CONTEXT
2343 buf.ctx.pos = filp->f_pos;
2344 rc = fops->iterate(filp, &buf.ctx);
2345 filp->f_pos = buf.ctx.pos;
2347 rc = fops->readdir(filp, &buf, filldir);
2349 fops->release(inode, filp);
2355 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2356 struct dentry *dentry, filldir_t filldir)
2358 struct osd_scrub *scrub = &dev->od_scrub;
2359 struct scrub_file *sf = &scrub->os_file;
2360 struct dentry *child;
2364 /* It is existing MDT0 device. We only allow the case of object without
2365 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2366 * can generate IGIF mode FID for the object and related OI mapping. If
2367 * it is on other MDTs, then becuase file-level backup/restore, related
2368 * OI mapping may be invalid already, we do not know which is the right
2369 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2371 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2372 * "fid" xattr, then something crashed. We cannot re-generate the
2373 * FID directly, instead, the OI scrub will scan the OI structure
2374 * and try to re-generate the LMA from the OI mapping. But if the
2375 * OI mapping crashed or lost also, then we have to give up under
2376 * double failure cases. */
2377 scrub->os_convert_igif = 1;
2378 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2379 strlen(dot_lustre_name));
2380 if (IS_ERR(child)) {
2381 rc = PTR_ERR(child);
2382 if (rc == -ENOENT) {
2383 /* It is 1.8 MDT device. */
2384 if (!(sf->sf_flags & SF_UPGRADE)) {
2385 osd_scrub_file_reset(scrub,
2386 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2388 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2389 rc = osd_scrub_file_store(scrub);
2395 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2396 * so the client will get IGIF for the ".lustre" object when
2399 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2400 * it does not know whether there are some old clients cached
2401 * the ".lustre" IGIF during the upgrading. Two choices:
2403 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2404 * It will allow the old connected clients to access the
2405 * ".lustre" with cached IGIF. But it will cause others
2406 * on the MDT failed to check "fid_is_dot_lustre()".
2408 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2409 * for ".lustre" in spite of whether there are some clients
2410 * cached the ".lustre" IGIF or not. It enables the check
2411 * "fid_is_dot_lustre()" on the MDT, although it will cause
2412 * that the old connected clients cannot access the ".lustre"
2413 * with the cached IGIF.
2415 * Usually, it is rare case for the old connected clients
2416 * to access the ".lustre" with cached IGIF. So we prefer
2417 * to the solution 2). */
2418 rc = osd_ios_scan_one(info, dev, child->d_inode,
2419 &LU_DOT_LUSTRE_FID, 0);
2421 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2430 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2431 struct dentry *dentry, filldir_t filldir)
2433 struct osd_scrub *scrub = &dev->od_scrub;
2434 struct scrub_file *sf = &scrub->os_file;
2435 struct dentry *child;
2439 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2440 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2441 rc = osd_scrub_file_store(scrub);
2446 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2447 if (!IS_ERR(child)) {
2448 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2451 rc = PTR_ERR(child);
2454 if (rc != 0 && rc != -ENOENT)
2457 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2458 if (!IS_ERR(child)) {
2459 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2462 rc = PTR_ERR(child);
2471 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2472 struct osd_device *dev)
2474 struct osd_ios_item *item = NULL;
2475 scandir_t scandir = osd_ios_general_scan;
2476 filldir_t filldir = osd_ios_root_fill;
2477 struct dentry *dentry = osd_sb(dev)->s_root;
2478 const struct osd_lf_map *map = osd_lf_maps;
2482 /* Lookup IGIF in OI by force for initial OI scrub. */
2483 dev->od_igif_inoi = 1;
2486 rc = scandir(info, dev, dentry, filldir);
2488 dput(item->oii_dentry);
2495 if (list_empty(&dev->od_ios_list))
2498 item = list_entry(dev->od_ios_list.next,
2499 struct osd_ios_item, oii_list);
2500 list_del_init(&item->oii_list);
2502 LASSERT(item->oii_scandir != NULL);
2503 scandir = item->oii_scandir;
2504 filldir = item->oii_filldir;
2505 dentry = item->oii_dentry;
2508 while (!list_empty(&dev->od_ios_list)) {
2509 item = list_entry(dev->od_ios_list.next,
2510 struct osd_ios_item, oii_list);
2511 list_del_init(&item->oii_list);
2512 dput(item->oii_dentry);
2519 /* There maybe the case that the object has been removed, but its OI
2520 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2521 * file-level backup/restore. So here cleanup the stale OI mappings. */
2522 while (map->olm_name != NULL) {
2523 struct dentry *child;
2525 if (fid_is_zero(&map->olm_fid)) {
2530 child = osd_ios_lookup_one_len(map->olm_name,
2531 osd_sb(dev)->s_root,
2535 else if (PTR_ERR(child) == -ENOENT)
2536 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2537 NULL, DTO_INDEX_DELETE,
2545 char *osd_lf_fid2name(const struct lu_fid *fid)
2547 const struct osd_lf_map *map = osd_lf_maps;
2549 while (map->olm_name != NULL) {
2550 if (!lu_fid_eq(fid, &map->olm_fid)) {
2555 if (map->olm_flags & OLF_SHOW_NAME)
2556 return map->olm_name;
2564 /* OI scrub start/stop */
2566 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2568 struct osd_scrub *scrub = &dev->od_scrub;
2569 struct ptlrpc_thread *thread = &scrub->os_thread;
2570 struct l_wait_info lwi = { 0 };
2571 struct task_struct *task;
2575 /* os_lock: sync status between stop and scrub thread */
2576 spin_lock(&scrub->os_lock);
2579 if (thread_is_running(thread)) {
2580 spin_unlock(&scrub->os_lock);
2581 if (!(scrub->os_file.sf_flags & SF_AUTO ||
2582 scrub->os_partial_scan) ||
2583 (flags & SS_AUTO_PARTIAL))
2586 osd_scrub_join(dev, flags, false);
2587 spin_lock(&scrub->os_lock);
2588 if (!thread_is_running(thread))
2591 spin_unlock(&scrub->os_lock);
2595 if (unlikely(thread_is_stopping(thread))) {
2596 spin_unlock(&scrub->os_lock);
2597 l_wait_event(thread->t_ctl_waitq,
2598 thread_is_stopped(thread),
2600 spin_lock(&scrub->os_lock);
2603 spin_unlock(&scrub->os_lock);
2605 if (scrub->os_file.sf_status == SS_COMPLETED) {
2606 if (!(flags & SS_SET_FAILOUT))
2607 flags |= SS_CLEAR_FAILOUT;
2609 if (!(flags & SS_SET_DRYRUN))
2610 flags |= SS_CLEAR_DRYRUN;
2615 scrub->os_start_flags = flags;
2616 thread_set_flags(thread, 0);
2617 task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2620 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2621 osd_scrub2name(scrub), rc);
2625 l_wait_event(thread->t_ctl_waitq,
2626 thread_is_running(thread) || thread_is_stopped(thread),
2632 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2637 /* od_otable_mutex: prevent curcurrent start/stop */
2638 mutex_lock(&dev->od_otable_mutex);
2639 rc = do_osd_scrub_start(dev, flags);
2640 mutex_unlock(&dev->od_otable_mutex);
2642 RETURN(rc == -EALREADY ? 0 : rc);
2645 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2647 struct ptlrpc_thread *thread = &scrub->os_thread;
2648 struct l_wait_info lwi = { 0 };
2650 /* os_lock: sync status between stop and scrub thread */
2651 spin_lock(&scrub->os_lock);
2652 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2653 thread_set_flags(thread, SVC_STOPPING);
2654 spin_unlock(&scrub->os_lock);
2655 wake_up_all(&thread->t_ctl_waitq);
2656 l_wait_event(thread->t_ctl_waitq,
2657 thread_is_stopped(thread),
2659 /* Do not skip the last lock/unlock, which can guarantee that
2660 * the caller cannot return until the OI scrub thread exit. */
2661 spin_lock(&scrub->os_lock);
2663 spin_unlock(&scrub->os_lock);
2666 static void osd_scrub_stop(struct osd_device *dev)
2668 /* od_otable_mutex: prevent curcurrent start/stop */
2669 mutex_lock(&dev->od_otable_mutex);
2670 dev->od_scrub.os_paused = 1;
2671 do_osd_scrub_stop(&dev->od_scrub);
2672 mutex_unlock(&dev->od_otable_mutex);
2675 /* OI scrub setup/cleanup */
2677 static const char osd_scrub_name[] = "OI_scrub";
2679 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2681 struct osd_thread_info *info = osd_oti_get(env);
2682 struct osd_scrub *scrub = &dev->od_scrub;
2683 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
2684 struct scrub_file *sf = &scrub->os_file;
2685 struct super_block *sb = osd_sb(dev);
2686 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2687 struct lvfs_run_ctxt saved;
2689 struct inode *inode;
2690 struct lu_fid *fid = &info->oti_fid;
2692 bool restored = false;
2696 memset(scrub, 0, sizeof(*scrub));
2697 OBD_SET_CTXT_MAGIC(ctxt);
2698 ctxt->pwdmnt = dev->od_mnt;
2699 ctxt->pwd = dev->od_mnt->mnt_root;
2700 ctxt->fs = get_ds();
2702 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2703 init_rwsem(&scrub->os_rwsem);
2704 spin_lock_init(&scrub->os_lock);
2705 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2707 push_ctxt(&saved, ctxt);
2708 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2710 pop_ctxt(&saved, ctxt);
2711 RETURN(PTR_ERR(filp));
2714 inode = file_inode(filp);
2715 /* 'What the @fid is' is not imporatant, because the object
2716 * has no OI mapping, and only is visible inside the OSD.*/
2717 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2718 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2720 filp_close(filp, NULL);
2721 pop_ctxt(&saved, ctxt);
2725 scrub->os_inode = igrab(inode);
2726 filp_close(filp, NULL);
2727 pop_ctxt(&saved, ctxt);
2729 rc = osd_scrub_file_load(scrub);
2730 if (rc == -ENOENT) {
2731 osd_scrub_file_init(scrub, es->s_uuid);
2732 /* If the "/O" dir does not exist when mount (indicated by
2733 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2734 * then it is quite probably that the device is a new one,
2735 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2737 * For the rare case that "/O" and "OI_scrub" both lost on
2738 * an old device, it can be found and cleared later.
2740 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2741 * need to check "filter_fid_old" and to convert it to
2742 * "filter_fid" for each object, and all the IGIF should
2743 * have their FID mapping in OI files already. */
2744 if (dev->od_maybe_new)
2745 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2747 } else if (rc != 0) {
2748 GOTO(cleanup_inode, rc);
2750 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2751 struct obd_uuid *old_uuid;
2752 struct obd_uuid *new_uuid;
2754 OBD_ALLOC_PTR(old_uuid);
2755 OBD_ALLOC_PTR(new_uuid);
2756 if (old_uuid == NULL || new_uuid == NULL) {
2757 CERROR("%.16s: UUID has been changed, but"
2758 "failed to allocate RAM for report\n",
2759 LDISKFS_SB(sb)->s_es->s_volume_name);
2761 class_uuid_unparse(sf->sf_uuid, old_uuid);
2762 class_uuid_unparse(es->s_uuid, new_uuid);
2763 CERROR("%.16s: UUID has been changed from "
2765 LDISKFS_SB(sb)->s_es->s_volume_name,
2766 old_uuid->uuid, new_uuid->uuid);
2768 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2771 if (old_uuid != NULL)
2772 OBD_FREE_PTR(old_uuid);
2773 if (new_uuid != NULL)
2774 OBD_FREE_PTR(new_uuid);
2775 } else if (sf->sf_status == SS_SCANNING) {
2776 sf->sf_status = SS_CRASHED;
2781 if (sf->sf_pos_last_checkpoint != 0)
2782 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2784 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2787 rc = osd_scrub_file_store(scrub);
2789 GOTO(cleanup_inode, rc);
2792 /* Initialize OI files. */
2793 rc = osd_oi_init(info, dev, restored);
2795 GOTO(cleanup_inode, rc);
2797 rc = osd_initial_OI_scrub(info, dev);
2799 GOTO(cleanup_oi, rc);
2801 if (sf->sf_flags & SF_UPGRADE ||
2802 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2803 sf->sf_success_count > 0)) {
2804 dev->od_igif_inoi = 0;
2805 dev->od_check_ff = dev->od_is_ost;
2807 dev->od_igif_inoi = 1;
2808 dev->od_check_ff = 0;
2811 if (sf->sf_flags & SF_INCONSISTENT)
2812 /* The 'od_igif_inoi' will be set under the
2814 * 1) new created system, or
2815 * 2) restored from file-level backup, or
2816 * 3) the upgrading completed.
2818 * The 'od_igif_inoi' may be cleared by OI scrub
2819 * later if found that the system is upgrading. */
2820 dev->od_igif_inoi = 1;
2822 if (!dev->od_noscrub &&
2823 ((sf->sf_status == SS_PAUSED) ||
2824 (sf->sf_status == SS_CRASHED &&
2825 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2826 SF_UPGRADE | SF_AUTO)) ||
2827 (sf->sf_status == SS_INIT &&
2828 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2830 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2833 GOTO(cleanup_oi, rc);
2835 /* it is possible that dcache entries may keep objects after they are
2836 * deleted by OSD. While it looks safe this can cause object data to
2837 * stay until umount causing failures in tests calculating free space,
2838 * e.g. replay-ost-single. Since those dcache entries are not used
2839 * anymore let's just free them after use here */
2840 shrink_dcache_sb(sb);
2844 osd_oi_fini(info, dev);
2846 iput(scrub->os_inode);
2847 scrub->os_inode = NULL;
2852 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2854 struct osd_scrub *scrub = &dev->od_scrub;
2856 LASSERT(dev->od_otable_it == NULL);
2858 if (scrub->os_inode != NULL) {
2859 osd_scrub_stop(dev);
2860 iput(scrub->os_inode);
2861 scrub->os_inode = NULL;
2863 if (dev->od_oi_table != NULL)
2864 osd_oi_fini(osd_oti_get(env), dev);
2867 /* object table based iteration APIs */
2869 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2870 struct dt_object *dt, __u32 attr)
2872 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2873 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2874 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2875 struct osd_scrub *scrub = &dev->od_scrub;
2876 struct osd_otable_it *it;
2881 /* od_otable_mutex: prevent curcurrent init/fini */
2882 mutex_lock(&dev->od_otable_mutex);
2883 if (dev->od_otable_it != NULL)
2884 GOTO(out, it = ERR_PTR(-EALREADY));
2888 GOTO(out, it = ERR_PTR(-ENOMEM));
2890 dev->od_otable_it = it;
2892 it->ooi_cache.ooc_consumer_idx = -1;
2893 if (flags & DOIF_OUTUSED)
2894 it->ooi_used_outside = 1;
2896 if (flags & DOIF_RESET)
2899 if (valid & DOIV_ERROR_HANDLE) {
2900 if (flags & DOIF_FAILOUT)
2901 start |= SS_SET_FAILOUT;
2903 start |= SS_CLEAR_FAILOUT;
2906 if (valid & DOIV_DRYRUN) {
2907 if (flags & DOIF_DRYRUN)
2908 start |= SS_SET_DRYRUN;
2910 start |= SS_CLEAR_DRYRUN;
2913 rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2914 if (rc < 0 && rc != -EALREADY) {
2915 dev->od_otable_it = NULL;
2917 GOTO(out, it = ERR_PTR(rc));
2920 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2925 mutex_unlock(&dev->od_otable_mutex);
2926 return (struct dt_it *)it;
2929 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2931 struct osd_otable_it *it = (struct osd_otable_it *)di;
2932 struct osd_device *dev = it->ooi_dev;
2934 /* od_otable_mutex: prevent curcurrent init/fini */
2935 mutex_lock(&dev->od_otable_mutex);
2936 do_osd_scrub_stop(&dev->od_scrub);
2937 LASSERT(dev->od_otable_it == it);
2939 dev->od_otable_it = NULL;
2940 mutex_unlock(&dev->od_otable_mutex);
2944 static int osd_otable_it_get(const struct lu_env *env,
2945 struct dt_it *di, const struct dt_key *key)
2950 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2955 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2957 spin_lock(&scrub->os_lock);
2958 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2959 scrub->os_waiting ||
2960 !thread_is_running(&scrub->os_thread))
2961 it->ooi_waiting = 0;
2963 it->ooi_waiting = 1;
2964 spin_unlock(&scrub->os_lock);
2966 return !it->ooi_waiting;
2969 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2971 struct osd_otable_it *it = (struct osd_otable_it *)di;
2972 struct osd_device *dev = it->ooi_dev;
2973 struct osd_scrub *scrub = &dev->od_scrub;
2974 struct osd_otable_cache *ooc = &it->ooi_cache;
2975 struct ptlrpc_thread *thread = &scrub->os_thread;
2976 struct l_wait_info lwi = { 0 };
2980 LASSERT(it->ooi_user_ready);
2983 if (!thread_is_running(thread) && !it->ooi_used_outside)
2986 if (ooc->ooc_cached_items > 0) {
2987 ooc->ooc_cached_items--;
2988 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2989 ~OSD_OTABLE_IT_CACHE_MASK;
2993 if (it->ooi_all_cached) {
2994 l_wait_event(thread->t_ctl_waitq,
2995 !thread_is_running(thread),
3000 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
3001 spin_lock(&scrub->os_lock);
3002 scrub->os_waiting = 0;
3003 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3004 spin_unlock(&scrub->os_lock);
3007 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
3008 l_wait_event(thread->t_ctl_waitq,
3009 osd_otable_it_wakeup(scrub, it),
3012 if (!thread_is_running(thread) && !it->ooi_used_outside)
3015 rc = osd_otable_it_preload(env, it);
3022 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
3023 const struct dt_it *di)
3028 static int osd_otable_it_key_size(const struct lu_env *env,
3029 const struct dt_it *di)
3031 return sizeof(__u64);
3034 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
3035 struct dt_rec *rec, __u32 attr)
3037 struct osd_otable_it *it = (struct osd_otable_it *)di;
3038 struct osd_otable_cache *ooc = &it->ooi_cache;
3040 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
3042 /* Filter out Invald FID already. */
3043 LASSERTF(fid_is_sane((struct lu_fid *)rec),
3044 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
3045 PFID((struct lu_fid *)rec),
3046 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
3051 static __u64 osd_otable_it_store(const struct lu_env *env,
3052 const struct dt_it *di)
3054 struct osd_otable_it *it = (struct osd_otable_it *)di;
3055 struct osd_otable_cache *ooc = &it->ooi_cache;
3058 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
3059 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
3061 hash = ooc->ooc_pos_preload;
3066 * Set the OSD layer iteration start position as the specified hash.
3068 static int osd_otable_it_load(const struct lu_env *env,
3069 const struct dt_it *di, __u64 hash)
3071 struct osd_otable_it *it = (struct osd_otable_it *)di;
3072 struct osd_device *dev = it->ooi_dev;
3073 struct osd_otable_cache *ooc = &it->ooi_cache;
3074 struct osd_scrub *scrub = &dev->od_scrub;
3078 /* Forbid to set iteration position after iteration started. */
3079 if (it->ooi_user_ready)
3082 LASSERT(!scrub->os_partial_scan);
3084 if (hash > OSD_OTABLE_MAX_HASH)
3085 hash = OSD_OTABLE_MAX_HASH;
3087 ooc->ooc_pos_preload = hash;
3088 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
3089 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
3091 it->ooi_user_ready = 1;
3092 if (!scrub->os_full_speed)
3093 wake_up_all(&scrub->os_thread.t_ctl_waitq);
3095 /* Unplug OSD layer iteration by the first next() call. */
3096 rc = osd_otable_it_next(env, (struct dt_it *)it);
3101 static int osd_otable_it_key_rec(const struct lu_env *env,
3102 const struct dt_it *di, void *key_rec)
3107 const struct dt_index_operations osd_otable_ops = {
3109 .init = osd_otable_it_init,
3110 .fini = osd_otable_it_fini,
3111 .get = osd_otable_it_get,
3112 .put = osd_otable_it_put,
3113 .next = osd_otable_it_next,
3114 .key = osd_otable_it_key,
3115 .key_size = osd_otable_it_key_size,
3116 .rec = osd_otable_it_rec,
3117 .store = osd_otable_it_store,
3118 .load = osd_otable_it_load,
3119 .key_rec = osd_otable_it_key_rec,
3123 /* high priority inconsistent items list APIs */
3125 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
3127 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
3130 struct osd_inconsistent_item *oii;
3131 struct osd_scrub *scrub = &dev->od_scrub;
3132 struct ptlrpc_thread *thread = &scrub->os_thread;
3137 if (unlikely(oii == NULL))
3140 INIT_LIST_HEAD(&oii->oii_list);
3141 oii->oii_cache = *oic;
3142 oii->oii_insert = insert;
3144 if (scrub->os_partial_scan) {
3145 __u64 now = cfs_time_current_sec();
3147 /* If there haven't been errors in a long time,
3148 * decay old count until either the errors are
3149 * gone or we reach the current interval. */
3150 while (unlikely(scrub->os_bad_oimap_count > 0 &&
3151 scrub->os_bad_oimap_time +
3152 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
3153 scrub->os_bad_oimap_count >>= 1;
3154 scrub->os_bad_oimap_time +=
3155 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
3158 scrub->os_bad_oimap_time = now;
3159 if (++scrub->os_bad_oimap_count >
3160 dev->od_full_scrub_threshold_rate)
3161 scrub->os_full_scrub = 1;
3164 spin_lock(&scrub->os_lock);
3165 if (unlikely(!thread_is_running(thread))) {
3166 spin_unlock(&scrub->os_lock);
3171 if (list_empty(&scrub->os_inconsistent_items))
3173 list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
3174 spin_unlock(&scrub->os_lock);
3177 wake_up_all(&thread->t_ctl_waitq);
3182 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
3183 struct osd_inode_id *id)
3185 struct osd_scrub *scrub = &dev->od_scrub;
3186 struct osd_inconsistent_item *oii;
3189 spin_lock(&scrub->os_lock);
3190 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
3191 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
3192 *id = oii->oii_cache.oic_lid;
3193 spin_unlock(&scrub->os_lock);
3197 spin_unlock(&scrub->os_lock);
3204 static const char *scrub_status_names[] = {
3215 static const char *scrub_flags_names[] = {
3223 static const char *scrub_param_names[] = {
3229 static void scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
3235 seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
3237 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
3240 seq_printf(m, "%s%c", names[i],
3241 bits != 0 ? ',' : '\n');
3246 static void scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3249 seq_printf(m, "%s: %llu seconds\n", prefix,
3250 cfs_time_current_sec() - time);
3252 seq_printf(m, "%s: N/A\n", prefix);
3255 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3258 seq_printf(m, "%s: %llu\n", prefix, pos);
3260 seq_printf(m, "%s: N/A\n", prefix);
3263 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3265 struct osd_scrub *scrub = &dev->od_scrub;
3266 struct scrub_file *sf = &scrub->os_file;
3270 down_read(&scrub->os_rwsem);
3271 seq_printf(m, "name: OI_scrub\n"
3275 sf->sf_magic, (int)sf->sf_oi_count,
3276 scrub_status_names[sf->sf_status]);
3278 scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
3280 scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
3282 scrub_time_dump(m, sf->sf_time_last_complete,
3283 "time_since_last_completed");
3285 scrub_time_dump(m, sf->sf_time_latest_start,
3286 "time_since_latest_start");
3288 scrub_time_dump(m, sf->sf_time_last_checkpoint,
3289 "time_since_last_checkpoint");
3291 scrub_pos_dump(m, sf->sf_pos_latest_start,
3292 "latest_start_position");
3294 scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3295 "last_checkpoint_position");
3297 scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3298 "first_failure_position");
3300 checked = sf->sf_items_checked + scrub->os_new_checked;
3301 seq_printf(m, "checked: %llu\n"
3304 "prior_updated: %llu\n"
3307 "success_count: %u\n",
3308 checked, sf->sf_items_updated, sf->sf_items_failed,
3309 sf->sf_items_updated_prior, sf->sf_items_noscrub,
3310 sf->sf_items_igif, sf->sf_success_count);
3313 if (thread_is_running(&scrub->os_thread)) {
3314 cfs_duration_t duration = cfs_time_current() -
3315 scrub->os_time_last_checkpoint;
3316 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3318 __u32 rtime = sf->sf_run_time +
3319 cfs_duration_sec(duration + HALF_SEC);
3322 do_div(new_checked, duration);
3324 do_div(speed, rtime);
3325 seq_printf(m, "run_time: %u seconds\n"
3326 "average_speed: %llu objects/sec\n"
3327 "real-time_speed: %llu objects/sec\n"
3328 "current_position: %u\n"
3329 "lf_scanned: %llu\n"
3330 "lf_repaired: %llu\n"
3331 "lf_failed: %llu\n",
3332 rtime, speed, new_checked, scrub->os_pos_current,
3333 scrub->os_lf_scanned, scrub->os_lf_repaired,
3334 scrub->os_lf_failed);
3336 if (sf->sf_run_time != 0)
3337 do_div(speed, sf->sf_run_time);
3338 seq_printf(m, "run_time: %u seconds\n"
3339 "average_speed: %llu objects/sec\n"
3340 "real-time_speed: N/A\n"
3341 "current_position: N/A\n"
3342 "lf_scanned: %llu\n"
3343 "lf_repaired: %llu\n"
3344 "lf_failed: %llu\n",
3345 sf->sf_run_time, speed, scrub->os_lf_scanned,
3346 scrub->os_lf_repaired, scrub->os_lf_failed);
3349 up_read(&scrub->os_rwsem);