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)
107 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
110 /* DTO_INDEX_INSERT is enough for other two ops:
111 * delete/update, but save stack. */
112 th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
113 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
116 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
117 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
118 PFID(fid), id->oii_ino, id->oii_gen, rc);
123 case DTO_INDEX_UPDATE:
124 rc = osd_oi_update(info, dev, fid, id, th, flags);
125 if (unlikely(rc == -ENOENT)) {
126 /* Some unlink thread may removed the OI mapping. */
130 case DTO_INDEX_INSERT:
131 rc = osd_oi_insert(info, dev, fid, id, th, flags);
132 if (unlikely(rc == -EEXIST)) {
134 /* XXX: There are trouble things when adding OI
135 * mapping for IGIF object, which may cause
136 * multiple objects to be mapped to the same
137 * IGIF formatted FID. Consider the following
140 * 1) The MDT is upgrading from 1.8 device.
141 * The OI scrub generates IGIF FID1 for the
142 * OBJ1 and adds the OI mapping.
144 * 2) For some reason, the OI scrub does not
145 * process all the IGIF objects completely.
147 * 3) The MDT is backuped and restored against
150 * 4) When the MDT mounts up, the OI scrub will
151 * try to rebuild the OI files. For some IGIF
152 * object, OBJ2, which was not processed by the
153 * OI scrub before the backup/restore, and the
154 * new generated IGIF formatted FID may be just
155 * the FID1, the same as OBJ1.
157 * Under such case, the OI scrub cannot know how
158 * to generate new FID for the OBJ2.
160 * Currently, we do nothing for that. One possible
161 * solution is to generate new normal FID for the
164 * Anyway, it is rare, only exists in theory. */
167 case DTO_INDEX_DELETE:
168 rc = osd_oi_delete(info, dev, fid, th, flags);
170 /* It is normal that the unlink thread has removed the
171 * OI mapping already. */
176 LASSERTF(0, "Unexpected ops %d\n", ops);
180 ldiskfs_journal_stop(th);
182 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
183 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
184 PFID(fid), id ? id->oii_ino : -1, id ? id->oii_gen : -1,
190 /* OI_scrub file ops */
192 static void osd_scrub_file_to_cpu(struct scrub_file *des,
193 struct scrub_file *src)
195 memcpy(des->sf_uuid, src->sf_uuid, 16);
196 des->sf_flags = le64_to_cpu(src->sf_flags);
197 des->sf_magic = le32_to_cpu(src->sf_magic);
198 des->sf_status = le16_to_cpu(src->sf_status);
199 des->sf_param = le16_to_cpu(src->sf_param);
200 des->sf_time_last_complete =
201 le64_to_cpu(src->sf_time_last_complete);
202 des->sf_time_latest_start =
203 le64_to_cpu(src->sf_time_latest_start);
204 des->sf_time_last_checkpoint =
205 le64_to_cpu(src->sf_time_last_checkpoint);
206 des->sf_pos_latest_start =
207 le64_to_cpu(src->sf_pos_latest_start);
208 des->sf_pos_last_checkpoint =
209 le64_to_cpu(src->sf_pos_last_checkpoint);
210 des->sf_pos_first_inconsistent =
211 le64_to_cpu(src->sf_pos_first_inconsistent);
212 des->sf_items_checked =
213 le64_to_cpu(src->sf_items_checked);
214 des->sf_items_updated =
215 le64_to_cpu(src->sf_items_updated);
216 des->sf_items_failed =
217 le64_to_cpu(src->sf_items_failed);
218 des->sf_items_updated_prior =
219 le64_to_cpu(src->sf_items_updated_prior);
220 des->sf_run_time = le32_to_cpu(src->sf_run_time);
221 des->sf_success_count = le32_to_cpu(src->sf_success_count);
222 des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
223 des->sf_internal_flags = le16_to_cpu(src->sf_internal_flags);
224 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
227 static void osd_scrub_file_to_le(struct scrub_file *des,
228 struct scrub_file *src)
230 memcpy(des->sf_uuid, src->sf_uuid, 16);
231 des->sf_flags = cpu_to_le64(src->sf_flags);
232 des->sf_magic = cpu_to_le32(src->sf_magic);
233 des->sf_status = cpu_to_le16(src->sf_status);
234 des->sf_param = cpu_to_le16(src->sf_param);
235 des->sf_time_last_complete =
236 cpu_to_le64(src->sf_time_last_complete);
237 des->sf_time_latest_start =
238 cpu_to_le64(src->sf_time_latest_start);
239 des->sf_time_last_checkpoint =
240 cpu_to_le64(src->sf_time_last_checkpoint);
241 des->sf_pos_latest_start =
242 cpu_to_le64(src->sf_pos_latest_start);
243 des->sf_pos_last_checkpoint =
244 cpu_to_le64(src->sf_pos_last_checkpoint);
245 des->sf_pos_first_inconsistent =
246 cpu_to_le64(src->sf_pos_first_inconsistent);
247 des->sf_items_checked =
248 cpu_to_le64(src->sf_items_checked);
249 des->sf_items_updated =
250 cpu_to_le64(src->sf_items_updated);
251 des->sf_items_failed =
252 cpu_to_le64(src->sf_items_failed);
253 des->sf_items_updated_prior =
254 cpu_to_le64(src->sf_items_updated_prior);
255 des->sf_run_time = cpu_to_le32(src->sf_run_time);
256 des->sf_success_count = cpu_to_le32(src->sf_success_count);
257 des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
258 des->sf_internal_flags = cpu_to_le16(src->sf_internal_flags);
259 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
262 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
264 struct scrub_file *sf = &scrub->os_file;
266 memset(sf, 0, sizeof(*sf));
267 memcpy(sf->sf_uuid, uuid, 16);
268 sf->sf_magic = SCRUB_MAGIC_V1;
269 sf->sf_status = SS_INIT;
272 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
274 struct scrub_file *sf = &scrub->os_file;
276 CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, old flags = "
277 LPX64", add flags = "LPX64"\n",
278 osd_scrub2name(scrub), sf->sf_flags, flags);
280 memcpy(sf->sf_uuid, uuid, 16);
281 sf->sf_status = SS_INIT;
282 sf->sf_flags |= flags;
283 sf->sf_flags &= ~SF_AUTO;
285 sf->sf_time_latest_start = 0;
286 sf->sf_time_last_checkpoint = 0;
287 sf->sf_pos_latest_start = 0;
288 sf->sf_pos_last_checkpoint = 0;
289 sf->sf_pos_first_inconsistent = 0;
290 sf->sf_items_checked = 0;
291 sf->sf_items_updated = 0;
292 sf->sf_items_failed = 0;
293 if (!scrub->os_in_join)
294 sf->sf_items_updated_prior = 0;
296 sf->sf_items_noscrub = 0;
297 sf->sf_items_igif = 0;
300 static int osd_scrub_file_load(struct osd_scrub *scrub)
303 int len = sizeof(scrub->os_file_disk);
306 rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
308 struct scrub_file *sf = &scrub->os_file;
310 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
311 if (sf->sf_magic != SCRUB_MAGIC_V1) {
312 CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
313 "0x%x != 0x%x\n", osd_scrub2name(scrub),
314 sf->sf_magic, SCRUB_MAGIC_V1);
315 /* Process it as new scrub file. */
320 } else if (rc != 0) {
321 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
322 "expected = %d: rc = %d\n",
323 osd_scrub2name(scrub), len, rc);
327 /* return -ENOENT for empty scrub file case. */
334 int osd_scrub_file_store(struct osd_scrub *scrub)
336 struct osd_device *dev;
339 int len = sizeof(scrub->os_file_disk);
343 dev = container_of0(scrub, struct osd_device, od_scrub);
344 credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
345 osd_dto_credits_noquota[DTO_WRITE_BLOCK];
346 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
349 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
350 "rc = %d\n", osd_scrub2name(scrub), rc);
354 osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
355 rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
357 ldiskfs_journal_stop(jh);
359 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
360 "expected = %d: rc = %d\n",
361 osd_scrub2name(scrub), len, rc);
363 scrub->os_time_last_checkpoint = cfs_time_current();
364 scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
365 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
370 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
371 struct inode *inode, const struct lu_fid *fid)
373 struct filter_fid_old *ff = &info->oti_ff;
374 struct dentry *dentry = &info->oti_obj_dentry;
375 struct lu_fid *tfid = &info->oti_fid;
379 bool removed = false;
383 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
386 if (fid_is_idif(fid) && dev->od_index_in_idif == 0) {
387 struct ost_id *oi = &info->oti_ostid;
389 fid_to_ostid(fid, oi);
390 ostid_to_fid(tfid, oi, 0);
395 /* We want the LMA to fit into the 256-byte OST inode, so operate
397 * 1) read old XATTR_NAME_FID and save the parent FID;
398 * 2) delete the old XATTR_NAME_FID;
399 * 3) make new LMA and add it;
400 * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
402 * Making the LMA to fit into the 256-byte OST inode can save time for
403 * normal osd_check_lma() and for other OI scrub scanning in future.
404 * So it is worth to make some slow conversion here. */
405 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
406 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
409 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
410 DFID": rc = %d\n", osd_name(dev), PFID(tfid), rc);
414 /* 1) read old XATTR_NAME_FID and save the parent FID */
415 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
416 if (rc == sizeof(*ff)) {
417 /* 2) delete the old XATTR_NAME_FID */
418 ll_vfs_dq_init(inode);
419 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
424 } else if (unlikely(rc == -ENODATA)) {
426 } else if (rc != sizeof(struct filter_fid)) {
427 GOTO(stop, rc = -EINVAL);
430 /* 3) make new LMA and add it */
431 rc = osd_ea_fid_set(info, inode, tfid, LMAC_FID_ON_OST, 0);
432 if (rc == 0 && reset)
433 size = sizeof(struct filter_fid);
434 else if (rc != 0 && removed)
435 /* If failed, we should try to add the old back. */
436 size = sizeof(struct filter_fid_old);
438 /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
442 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
444 if (rc1 != 0 && rc == 0)
451 ldiskfs_journal_stop(jh);
453 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
454 osd_name(dev), PFID(tfid), rc);
459 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
460 struct osd_idmap_cache *oic, int val)
462 struct osd_scrub *scrub = &dev->od_scrub;
463 struct scrub_file *sf = &scrub->os_file;
464 struct lu_fid *fid = &oic->oic_fid;
465 struct osd_inode_id *lid = &oic->oic_lid;
466 struct osd_inode_id *lid2 = &info->oti_id;
467 struct osd_inconsistent_item *oii = NULL;
468 struct inode *inode = NULL;
469 int ops = DTO_INDEX_UPDATE;
472 bool converted = false;
475 down_write(&scrub->os_rwsem);
476 scrub->os_new_checked++;
480 if (scrub->os_in_prior)
481 oii = list_entry(oic, struct osd_inconsistent_item,
484 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
487 if (fid_is_igif(fid))
490 if (val == SCRUB_NEXT_OSTOBJ_OLD) {
491 inode = osd_iget(info, dev, lid);
494 /* Someone removed the inode. */
495 if (rc == -ENOENT || rc == -ESTALE)
500 sf->sf_flags |= SF_UPGRADE;
501 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
502 dev->od_check_ff = 1;
503 rc = osd_scrub_convert_ff(info, dev, inode, fid);
510 if ((val == SCRUB_NEXT_NOLMA) &&
511 (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
514 if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA)) {
515 ops = DTO_INDEX_INSERT;
520 rc = osd_oi_lookup(info, dev, fid, lid2,
521 (val == SCRUB_NEXT_OSTOBJ ||
522 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
525 ops = DTO_INDEX_INSERT;
526 else if (rc != -ESTALE)
531 inode = osd_iget(info, dev, lid);
534 /* Someone removed the inode. */
535 if (rc == -ENOENT || rc == -ESTALE)
541 if (!scrub->os_partial_scan)
542 scrub->os_full_speed = 1;
544 idx = osd_oi_fid2idx(dev, fid);
546 case SCRUB_NEXT_NOLMA:
547 sf->sf_flags |= SF_UPGRADE;
548 if (!(sf->sf_param & SP_DRYRUN)) {
549 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
554 if (!(sf->sf_flags & SF_INCONSISTENT))
555 dev->od_igif_inoi = 0;
557 case SCRUB_NEXT_OSTOBJ:
558 sf->sf_flags |= SF_INCONSISTENT;
559 case SCRUB_NEXT_OSTOBJ_OLD:
562 sf->sf_flags |= SF_RECREATED;
563 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
564 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
567 } else if (osd_id_eq(lid, lid2)) {
569 sf->sf_items_updated++;
573 if (!scrub->os_partial_scan)
574 scrub->os_full_speed = 1;
576 sf->sf_flags |= SF_INCONSISTENT;
578 /* XXX: If the device is restored from file-level backup, then
579 * some IGIFs may have been already in OI files, and some
580 * may be not yet. Means upgrading from 1.8 may be partly
581 * processed, but some clients may hold some immobilized
582 * IGIFs, and use them to access related objects. Under
583 * such case, OSD does not know whether an given IGIF has
584 * been processed or to be processed, and it also cannot
585 * generate local ino#/gen# directly from the immobilized
586 * IGIF because of the backup/restore. Then force OSD to
587 * lookup the given IGIF in OI files, and if no entry,
588 * then ask the client to retry after upgrading completed.
589 * No better choice. */
590 dev->od_igif_inoi = 1;
593 rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
594 (val == SCRUB_NEXT_OSTOBJ ||
595 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
597 if (scrub->os_in_prior)
598 sf->sf_items_updated_prior++;
600 sf->sf_items_updated++;
607 sf->sf_items_failed++;
608 if (sf->sf_pos_first_inconsistent == 0 ||
609 sf->sf_pos_first_inconsistent > lid->oii_ino)
610 sf->sf_pos_first_inconsistent = lid->oii_ino;
615 /* There may be conflict unlink during the OI scrub,
616 * if happend, then remove the new added OI mapping. */
617 if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
618 unlikely(ldiskfs_test_inode_state(inode,
619 LDISKFS_STATE_LUSTRE_DESTROY)))
620 osd_scrub_refresh_mapping(info, dev, fid, lid,
621 DTO_INDEX_DELETE, false,
622 (val == SCRUB_NEXT_OSTOBJ ||
623 val == SCRUB_NEXT_OSTOBJ_OLD) ?
624 OI_KNOWN_ON_OST : 0);
625 up_write(&scrub->os_rwsem);
627 if (inode != NULL && !IS_ERR(inode))
631 LASSERT(list_empty(&oii->oii_list));
636 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
641 static int osd_scrub_prep(struct osd_device *dev)
643 struct osd_scrub *scrub = &dev->od_scrub;
644 struct ptlrpc_thread *thread = &scrub->os_thread;
645 struct scrub_file *sf = &scrub->os_file;
646 __u32 flags = scrub->os_start_flags;
648 bool drop_dryrun = false;
651 CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
652 osd_scrub2name(scrub), flags);
654 down_write(&scrub->os_rwsem);
655 if (flags & SS_SET_FAILOUT)
656 sf->sf_param |= SP_FAILOUT;
657 else if (flags & SS_CLEAR_FAILOUT)
658 sf->sf_param &= ~SP_FAILOUT;
660 if (flags & SS_SET_DRYRUN) {
661 sf->sf_param |= SP_DRYRUN;
662 } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
663 sf->sf_param &= ~SP_DRYRUN;
667 if (flags & SS_RESET)
668 osd_scrub_file_reset(scrub,
669 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
671 if (flags & SS_AUTO_FULL) {
672 scrub->os_full_speed = 1;
673 scrub->os_partial_scan = 0;
674 sf->sf_flags |= SF_AUTO;
675 } else if (flags & SS_AUTO_PARTIAL) {
676 scrub->os_full_speed = 0;
677 scrub->os_partial_scan = 1;
678 sf->sf_flags |= SF_AUTO;
679 } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
681 scrub->os_full_speed = 1;
682 scrub->os_partial_scan = 0;
684 scrub->os_full_speed = 0;
685 scrub->os_partial_scan = 0;
688 spin_lock(&scrub->os_lock);
689 scrub->os_in_prior = 0;
690 scrub->os_waiting = 0;
691 scrub->os_paused = 0;
692 scrub->os_in_join = 0;
693 scrub->os_full_scrub = 0;
694 spin_unlock(&scrub->os_lock);
695 scrub->os_new_checked = 0;
696 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
697 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
698 else if (sf->sf_pos_last_checkpoint != 0)
699 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
701 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
703 scrub->os_pos_current = sf->sf_pos_latest_start;
704 sf->sf_status = SS_SCANNING;
705 sf->sf_time_latest_start = cfs_time_current_sec();
706 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
707 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
708 rc = osd_scrub_file_store(scrub);
710 spin_lock(&scrub->os_lock);
711 thread_set_flags(thread, SVC_RUNNING);
712 spin_unlock(&scrub->os_lock);
713 wake_up_all(&thread->t_ctl_waitq);
715 up_write(&scrub->os_rwsem);
720 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
722 struct scrub_file *sf = &scrub->os_file;
725 if (likely(cfs_time_before(cfs_time_current(),
726 scrub->os_time_next_checkpoint) ||
727 scrub->os_new_checked == 0))
730 down_write(&scrub->os_rwsem);
731 sf->sf_items_checked += scrub->os_new_checked;
732 scrub->os_new_checked = 0;
733 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
734 sf->sf_time_last_checkpoint = cfs_time_current_sec();
735 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
736 scrub->os_time_last_checkpoint);
737 rc = osd_scrub_file_store(scrub);
738 up_write(&scrub->os_rwsem);
743 static int osd_scrub_post(struct osd_scrub *scrub, int result)
745 struct scrub_file *sf = &scrub->os_file;
749 CDEBUG(D_LFSCK, "%.16s: OI scrub post, result = %d\n",
750 osd_scrub2name(scrub), result);
752 down_write(&scrub->os_rwsem);
753 spin_lock(&scrub->os_lock);
754 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
755 spin_unlock(&scrub->os_lock);
756 if (scrub->os_new_checked > 0) {
757 sf->sf_items_checked += scrub->os_new_checked;
758 scrub->os_new_checked = 0;
759 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
761 sf->sf_time_last_checkpoint = cfs_time_current_sec();
763 struct osd_device *dev =
764 container_of0(scrub, struct osd_device, od_scrub);
766 dev->od_igif_inoi = 1;
767 dev->od_check_ff = 0;
768 sf->sf_status = SS_COMPLETED;
769 if (!(sf->sf_param & SP_DRYRUN)) {
770 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
771 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
772 SF_UPGRADE | SF_AUTO);
774 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
775 sf->sf_success_count++;
776 } else if (result == 0) {
777 if (scrub->os_paused)
778 sf->sf_status = SS_PAUSED;
780 sf->sf_status = SS_STOPPED;
782 sf->sf_status = SS_FAILED;
784 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
785 scrub->os_time_last_checkpoint);
786 rc = osd_scrub_file_store(scrub);
787 up_write(&scrub->os_rwsem);
789 RETURN(rc < 0 ? rc : result);
792 /* iteration engine */
794 struct osd_iit_param {
795 struct super_block *sb;
796 struct buffer_head *bitmap;
802 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
803 struct osd_device *dev,
804 struct osd_iit_param *param,
805 struct osd_idmap_cache **oic,
808 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
809 struct osd_device *dev,
810 struct osd_iit_param *param,
811 struct osd_idmap_cache *oic,
812 bool *noslot, int rc);
814 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
816 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
817 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
818 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
819 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
820 return SCRUB_NEXT_BREAK;
822 *pos = param->gbase + param->offset;
828 * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
829 * \retval 0: FID-on-MDT
831 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
832 struct osd_device *dev,
835 /* XXX: The initial OI scrub will scan the top level /O to generate
836 * a small local FLDB according to the <seq>. If the given FID
837 * is in the local FLDB, then it is FID-on-OST; otherwise it's
838 * quite possible for FID-on-MDT. */
840 return SCRUB_NEXT_OSTOBJ_OLD;
845 static int osd_scrub_get_fid(struct osd_thread_info *info,
846 struct osd_device *dev, struct inode *inode,
847 struct lu_fid *fid, bool scrub)
849 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
851 bool has_lma = false;
853 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
856 if (lma->lma_compat & LMAC_NOT_IN_OI ||
857 lma->lma_incompat & LMAI_AGENT)
858 return SCRUB_NEXT_CONTINUE;
860 *fid = lma->lma_self_fid;
864 if (lma->lma_compat & LMAC_FID_ON_OST)
865 return SCRUB_NEXT_OSTOBJ;
867 if (fid_is_idif(fid))
868 return SCRUB_NEXT_OSTOBJ_OLD;
870 /* For local object. */
871 if (fid_is_internal(fid))
874 /* For external visible MDT-object with non-normal FID. */
875 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
878 /* For the object with normal FID, it may be MDT-object,
879 * or may be 2.4 OST-object, need further distinguish.
880 * Fall through to next section. */
883 if (rc == -ENODATA || rc == 0) {
884 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
887 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
888 rc = SCRUB_NEXT_OSTOBJ_OLD;
894 /* It is FID-on-OST, but we do not know how
895 * to generate its FID, ignore it directly. */
896 rc = SCRUB_NEXT_CONTINUE;
898 /* It is 2.4 OST-object. */
899 rc = SCRUB_NEXT_OSTOBJ_OLD;
907 if (dev->od_scrub.os_convert_igif) {
908 lu_igif_build(fid, inode->i_ino,
909 inode->i_generation);
911 rc = SCRUB_NEXT_NOLMA;
915 /* It may be FID-on-OST, or may be FID for
916 * non-MDT0, anyway, we do not know how to
917 * generate its FID, ignore it directly. */
918 rc = SCRUB_NEXT_CONTINUE;
923 /* For OI scrub case only: the object has LMA but has no ff
924 * (or ff crashed). It may be MDT-object, may be OST-object
925 * with crashed ff. The last check is local FLDB. */
926 rc = osd_scrub_check_local_fldb(info, dev, fid);
932 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
933 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
934 struct super_block *sb, bool scrub)
940 /* Not handle the backend root object and agent parent object.
941 * They are neither visible to namespace nor have OI mappings. */
942 if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
943 pos == osd_remote_parent_ino(dev)))
944 RETURN(SCRUB_NEXT_CONTINUE);
946 osd_id_gen(lid, pos, OSD_OII_NOGEN);
947 inode = osd_iget(info, dev, lid);
950 /* The inode may be removed after bitmap searching, or the
951 * file is new created without inode initialized yet. */
952 if (rc == -ENOENT || rc == -ESTALE)
953 RETURN(SCRUB_NEXT_CONTINUE);
955 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
956 "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
962 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
963 /* Only skip it for the first OI scrub accessing. */
964 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
965 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
968 rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
977 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
978 struct osd_iit_param *param,
979 struct osd_idmap_cache **oic, const bool noslot)
981 struct osd_scrub *scrub = &dev->od_scrub;
982 struct ptlrpc_thread *thread = &scrub->os_thread;
984 struct osd_inode_id *lid;
987 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
988 struct l_wait_info lwi;
990 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
991 if (likely(lwi.lwi_timeout > 0))
992 l_wait_event(thread->t_ctl_waitq,
993 !list_empty(&scrub->os_inconsistent_items) ||
994 !thread_is_running(thread),
998 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
999 spin_lock(&scrub->os_lock);
1000 thread_set_flags(thread, SVC_STOPPING);
1001 spin_unlock(&scrub->os_lock);
1002 return SCRUB_NEXT_CRASH;
1005 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
1006 return SCRUB_NEXT_FATAL;
1008 if (unlikely(!thread_is_running(thread)))
1009 return SCRUB_NEXT_EXIT;
1011 if (!list_empty(&scrub->os_inconsistent_items)) {
1012 spin_lock(&scrub->os_lock);
1013 if (likely(!list_empty(&scrub->os_inconsistent_items))) {
1014 struct osd_inconsistent_item *oii;
1016 oii = list_entry(scrub->os_inconsistent_items.next,
1017 struct osd_inconsistent_item, oii_list);
1018 list_del_init(&oii->oii_list);
1019 spin_unlock(&scrub->os_lock);
1021 *oic = &oii->oii_cache;
1022 scrub->os_in_prior = 1;
1026 spin_unlock(&scrub->os_lock);
1030 return SCRUB_NEXT_WAIT;
1032 rc = osd_iit_next(param, &scrub->os_pos_current);
1036 *oic = &scrub->os_oic;
1037 fid = &(*oic)->oic_fid;
1038 lid = &(*oic)->oic_lid;
1039 rc = osd_iit_iget(info, dev, fid, lid,
1040 scrub->os_pos_current, param->sb, true);
1044 static int osd_preload_next(struct osd_thread_info *info,
1045 struct osd_device *dev, struct osd_iit_param *param,
1046 struct osd_idmap_cache **oic, const bool noslot)
1048 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1049 struct osd_scrub *scrub;
1050 struct ptlrpc_thread *thread;
1053 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1057 scrub = &dev->od_scrub;
1058 thread = &scrub->os_thread;
1059 if (thread_is_running(thread) &&
1060 ooc->ooc_pos_preload >= scrub->os_pos_current)
1061 return SCRUB_NEXT_EXIT;
1063 rc = osd_iit_iget(info, dev,
1064 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1065 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1066 ooc->ooc_pos_preload, param->sb, false);
1067 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1068 * ignore the failure, so it still need to skip the inode next time. */
1069 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1074 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1076 spin_lock(&scrub->os_lock);
1077 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1078 !list_empty(&scrub->os_inconsistent_items) ||
1079 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1080 scrub->os_waiting = 0;
1082 scrub->os_waiting = 1;
1083 spin_unlock(&scrub->os_lock);
1085 return !scrub->os_waiting;
1088 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1089 struct osd_iit_param *param,
1090 struct osd_idmap_cache *oic, bool *noslot, int rc)
1092 struct l_wait_info lwi = { 0 };
1093 struct osd_scrub *scrub = &dev->od_scrub;
1094 struct scrub_file *sf = &scrub->os_file;
1095 struct ptlrpc_thread *thread = &scrub->os_thread;
1096 struct osd_otable_it *it = dev->od_otable_it;
1097 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
1100 case SCRUB_NEXT_CONTINUE:
1102 case SCRUB_NEXT_WAIT:
1104 case SCRUB_NEXT_NOSCRUB:
1105 down_write(&scrub->os_rwsem);
1106 scrub->os_new_checked++;
1107 sf->sf_items_noscrub++;
1108 up_write(&scrub->os_rwsem);
1112 rc = osd_scrub_check_update(info, dev, oic, rc);
1114 scrub->os_in_prior = 0;
1118 rc = osd_scrub_checkpoint(scrub);
1120 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1121 "rc = %d\n", osd_scrub2name(scrub),
1122 scrub->os_pos_current, rc);
1123 /* Continue, as long as the scrub itself can go ahead. */
1126 if (scrub->os_in_prior) {
1127 scrub->os_in_prior = 0;
1132 scrub->os_pos_current = param->gbase + ++(param->offset);
1135 if (it != NULL && it->ooi_waiting && ooc != NULL &&
1136 ooc->ooc_pos_preload < scrub->os_pos_current) {
1137 spin_lock(&scrub->os_lock);
1138 it->ooi_waiting = 0;
1139 wake_up_all(&thread->t_ctl_waitq);
1140 spin_unlock(&scrub->os_lock);
1143 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1146 if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1152 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1155 if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1162 static int osd_preload_exec(struct osd_thread_info *info,
1163 struct osd_device *dev, struct osd_iit_param *param,
1164 struct osd_idmap_cache *oic, bool *noslot, int rc)
1166 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1169 ooc->ooc_cached_items++;
1170 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1171 ~OSD_OTABLE_IT_CACHE_MASK;
1173 return rc > 0 ? 0 : rc;
1176 #define SCRUB_IT_ALL 1
1177 #define SCRUB_IT_CRASH 2
1179 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1182 struct osd_scrub *scrub = &dev->od_scrub;
1183 struct ptlrpc_thread *thread = &scrub->os_thread;
1184 struct scrub_file *sf = &scrub->os_file;
1188 LASSERT(!(flags & SS_AUTO_PARTIAL));
1190 down_write(&scrub->os_rwsem);
1191 scrub->os_in_join = 1;
1192 if (flags & SS_SET_FAILOUT)
1193 sf->sf_param |= SP_FAILOUT;
1194 else if (flags & SS_CLEAR_FAILOUT)
1195 sf->sf_param &= ~SP_FAILOUT;
1197 if (flags & SS_SET_DRYRUN)
1198 sf->sf_param |= SP_DRYRUN;
1199 else if (flags & SS_CLEAR_DRYRUN)
1200 sf->sf_param &= ~SP_DRYRUN;
1202 if (flags & SS_RESET) {
1203 osd_scrub_file_reset(scrub,
1204 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1205 inconsistent ? SF_INCONSISTENT : 0);
1206 sf->sf_status = SS_SCANNING;
1209 if (flags & SS_AUTO_FULL) {
1210 sf->sf_flags |= SF_AUTO;
1211 scrub->os_full_speed = 1;
1214 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1215 scrub->os_full_speed = 1;
1217 scrub->os_full_speed = 0;
1219 scrub->os_new_checked = 0;
1220 if (sf->sf_pos_last_checkpoint != 0)
1221 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1223 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1225 scrub->os_pos_current = sf->sf_pos_latest_start;
1226 sf->sf_time_latest_start = cfs_time_current_sec();
1227 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1228 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1229 rc = osd_scrub_file_store(scrub);
1231 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1232 "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1234 spin_lock(&scrub->os_lock);
1235 scrub->os_waiting = 0;
1236 scrub->os_paused = 0;
1237 scrub->os_partial_scan = 0;
1238 scrub->os_in_join = 0;
1239 scrub->os_full_scrub = 0;
1240 spin_unlock(&scrub->os_lock);
1241 wake_up_all(&thread->t_ctl_waitq);
1242 up_write(&scrub->os_rwsem);
1247 static int osd_inode_iteration(struct osd_thread_info *info,
1248 struct osd_device *dev, __u32 max, bool preload)
1250 struct osd_scrub *scrub = &dev->od_scrub;
1251 struct ptlrpc_thread *thread = &scrub->os_thread;
1252 struct scrub_file *sf = &scrub->os_file;
1253 osd_iit_next_policy next;
1254 osd_iit_exec_policy exec;
1257 struct osd_iit_param param = { NULL };
1258 struct l_wait_info lwi = { 0 };
1264 param.sb = osd_sb(dev);
1268 while (scrub->os_partial_scan && !scrub->os_in_join) {
1269 struct osd_idmap_cache *oic = NULL;
1271 rc = osd_scrub_next(info, dev, ¶m, &oic, noslot);
1273 case SCRUB_NEXT_EXIT:
1275 case SCRUB_NEXT_CRASH:
1276 RETURN(SCRUB_IT_CRASH);
1277 case SCRUB_NEXT_FATAL:
1279 case SCRUB_NEXT_WAIT: {
1280 struct kstatfs *ksfs = &info->oti_ksfs;
1283 if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1284 unlikely(sf->sf_items_updated_prior == 0))
1287 if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1288 scrub->os_full_scrub) {
1289 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1294 rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1296 __u64 used = ksfs->f_files - ksfs->f_ffree;
1298 do_div(used, sf->sf_items_updated_prior);
1299 /* If we hit too much inconsistent OI
1300 * mappings during the partial scan,
1301 * then scan the device completely. */
1302 if (used < dev->od_full_scrub_ratio) {
1304 SS_AUTO_FULL | SS_RESET, true);
1310 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1314 saved_flags = sf->sf_flags;
1315 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1316 SF_UPGRADE | SF_AUTO);
1317 sf->sf_status = SS_COMPLETED;
1318 l_wait_event(thread->t_ctl_waitq,
1319 !thread_is_running(thread) ||
1320 !scrub->os_partial_scan ||
1321 scrub->os_in_join ||
1322 !list_empty(&scrub->os_inconsistent_items),
1324 sf->sf_flags = saved_flags;
1325 sf->sf_status = SS_SCANNING;
1327 if (unlikely(!thread_is_running(thread)))
1330 if (!scrub->os_partial_scan || scrub->os_in_join)
1336 LASSERTF(rc == 0, "rc = %d\n", rc);
1338 osd_scrub_exec(info, dev, ¶m, oic, &noslot, rc);
1345 l_wait_event(thread->t_ctl_waitq,
1346 !thread_is_running(thread) || !scrub->os_in_join,
1349 if (unlikely(!thread_is_running(thread)))
1355 next = osd_scrub_next;
1356 exec = osd_scrub_exec;
1357 pos = &scrub->os_pos_current;
1358 count = &scrub->os_new_checked;
1360 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1362 next = osd_preload_next;
1363 exec = osd_preload_exec;
1364 pos = &ooc->ooc_pos_preload;
1365 count = &ooc->ooc_cached_items;
1367 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1369 while (*pos <= limit && *count < max) {
1370 struct osd_idmap_cache *oic = NULL;
1371 struct ldiskfs_group_desc *desc;
1373 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1374 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1378 ldiskfs_lock_group(param.sb, param.bg);
1379 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1380 ldiskfs_unlock_group(param.sb, param.bg);
1381 *pos = 1 + (param.bg + 1) *
1382 LDISKFS_INODES_PER_GROUP(param.sb);
1385 ldiskfs_unlock_group(param.sb, param.bg);
1387 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1388 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1389 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1390 if (param.bitmap == NULL) {
1391 CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1392 "scrub will stop, urgent mode\n",
1393 osd_scrub2name(scrub), (__u32)param.bg);
1397 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1400 ldiskfs_itable_unused_count(param.sb, desc) >
1401 LDISKFS_INODES_PER_GROUP(param.sb))
1404 rc = next(info, dev, ¶m, &oic, noslot);
1406 case SCRUB_NEXT_BREAK:
1408 case SCRUB_NEXT_EXIT:
1409 brelse(param.bitmap);
1411 case SCRUB_NEXT_CRASH:
1412 brelse(param.bitmap);
1413 RETURN(SCRUB_IT_CRASH);
1414 case SCRUB_NEXT_FATAL:
1415 brelse(param.bitmap);
1419 rc = exec(info, dev, ¶m, oic, &noslot, rc);
1421 brelse(param.bitmap);
1427 brelse(param.bitmap);
1431 RETURN(SCRUB_IT_ALL);
1435 static int osd_otable_it_preload(const struct lu_env *env,
1436 struct osd_otable_it *it)
1438 struct osd_device *dev = it->ooi_dev;
1439 struct osd_scrub *scrub = &dev->od_scrub;
1440 struct osd_otable_cache *ooc = &it->ooi_cache;
1444 rc = osd_inode_iteration(osd_oti_get(env), dev,
1445 OSD_OTABLE_IT_CACHE_SIZE, true);
1446 if (rc == SCRUB_IT_ALL)
1447 it->ooi_all_cached = 1;
1449 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1450 spin_lock(&scrub->os_lock);
1451 scrub->os_waiting = 0;
1452 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1453 spin_unlock(&scrub->os_lock);
1456 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1459 static int osd_scrub_main(void *args)
1462 struct osd_device *dev = (struct osd_device *)args;
1463 struct osd_scrub *scrub = &dev->od_scrub;
1464 struct ptlrpc_thread *thread = &scrub->os_thread;
1468 rc = lu_env_init(&env, LCT_LOCAL);
1470 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1471 osd_scrub2name(scrub), rc);
1475 rc = osd_scrub_prep(dev);
1477 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1478 osd_scrub2name(scrub), rc);
1482 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1483 struct l_wait_info lwi = { 0 };
1484 struct osd_otable_it *it = dev->od_otable_it;
1485 struct osd_otable_cache *ooc = &it->ooi_cache;
1487 l_wait_event(thread->t_ctl_waitq,
1488 it->ooi_user_ready || !thread_is_running(thread),
1490 if (unlikely(!thread_is_running(thread)))
1493 scrub->os_pos_current = ooc->ooc_pos_preload;
1496 CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1497 osd_scrub2name(scrub), scrub->os_start_flags,
1498 scrub->os_pos_current);
1500 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1501 if (unlikely(rc == SCRUB_IT_CRASH))
1502 GOTO(out, rc = -EINVAL);
1506 rc = osd_scrub_post(scrub, rc);
1507 CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1508 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1511 while (!list_empty(&scrub->os_inconsistent_items)) {
1512 struct osd_inconsistent_item *oii;
1514 oii = list_entry(scrub->os_inconsistent_items.next,
1515 struct osd_inconsistent_item, oii_list);
1516 list_del_init(&oii->oii_list);
1522 spin_lock(&scrub->os_lock);
1523 thread_set_flags(thread, SVC_STOPPED);
1524 wake_up_all(&thread->t_ctl_waitq);
1525 spin_unlock(&scrub->os_lock);
1529 /* initial OI scrub */
1531 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1532 struct dentry *, filldir_t filldir);
1534 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1535 loff_t offset, __u64 ino, unsigned d_type);
1536 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1537 loff_t offset, __u64 ino, unsigned d_type);
1538 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1539 loff_t offset, __u64 ino, unsigned d_type);
1540 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1541 loff_t offset, __u64 ino, unsigned d_type);
1544 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1545 struct dentry *dentry, filldir_t filldir);
1547 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1548 struct dentry *dentry, filldir_t filldir);
1551 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1552 struct dentry *dentry, filldir_t filldir);
1555 OLF_SCAN_SUBITEMS = 0x0001,
1556 OLF_HIDE_FID = 0x0002,
1557 OLF_SHOW_NAME = 0x0004,
1559 OLF_IDX_IN_FID = 0x0010,
1564 struct lu_fid olm_fid;
1567 scandir_t olm_scandir;
1568 filldir_t olm_filldir;
1571 /* Add the new introduced local files in the list in the future. */
1572 static const struct osd_lf_map osd_lf_maps[] = {
1574 { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1575 sizeof(CATLIST) - 1, NULL, NULL },
1578 { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1579 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1580 osd_ios_general_scan, osd_ios_varfid_fill },
1582 /* NIDTBL_VERSIONS */
1583 { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1584 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1585 osd_ios_varfid_fill },
1588 { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1591 { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1592 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1593 osd_ios_ROOT_scan, NULL },
1595 /* changelog_catalog */
1596 { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1599 /* changelog_users */
1600 { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1604 { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1605 sizeof("fld") - 1, NULL, NULL },
1608 { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1609 sizeof(LAST_RCVD) - 1, NULL, NULL },
1612 { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1613 sizeof(REPLY_DATA) - 1, NULL, NULL },
1616 { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1617 sizeof(LOV_OBJID) - 1, NULL, NULL },
1620 { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1621 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1624 { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1625 osd_ios_general_scan, osd_ios_varfid_fill },
1628 { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1629 osd_ios_general_scan, osd_ios_varfid_fill },
1632 { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1633 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1636 { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1637 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1640 { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1641 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1644 { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1645 osd_ios_general_scan, osd_ios_varfid_fill },
1647 /* lfsck_bookmark */
1648 { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1652 { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1655 /* lfsck_namespace */
1656 { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1659 /* OBJECTS, upgrade from old device */
1660 { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1661 osd_ios_OBJECTS_scan, NULL },
1663 /* lquota_v2.user, upgrade from old device */
1664 { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1667 /* lquota_v2.group, upgrade from old device */
1668 { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1671 /* LAST_GROUP, upgrade from old device */
1672 { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1673 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1675 /* committed batchid for cross-MDT operation */
1676 { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1677 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1679 /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1680 * and f_oid = index for their log files. See lu_update_log{_dir}_fid()
1681 * for more details. */
1684 { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1685 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1688 /* update_log_dir */
1689 { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1690 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1691 sizeof("update_log_dir") - 1,
1692 osd_ios_general_scan, osd_ios_uld_fill },
1695 { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1696 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1697 osd_ios_general_scan, osd_ios_lf_fill },
1699 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1702 /* Add the new introduced files under .lustre/ in the list in the future. */
1703 static const struct osd_lf_map osd_dl_maps[] = {
1705 { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1706 sizeof("fid") - 1, NULL, NULL },
1708 /* .lustre/lost+found */
1709 { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1710 sizeof("lost+found") - 1, NULL, NULL },
1712 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1715 struct osd_ios_item {
1716 struct list_head oii_list;
1717 struct dentry *oii_dentry;
1718 scandir_t oii_scandir;
1719 filldir_t oii_filldir;
1722 struct osd_ios_filldir_buf {
1723 #ifdef HAVE_DIR_CONTEXT
1724 /* please keep it as first member */
1725 struct dir_context ctx;
1727 struct osd_thread_info *oifb_info;
1728 struct osd_device *oifb_dev;
1729 struct dentry *oifb_dentry;
1732 static inline struct dentry *
1733 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1735 struct dentry *dentry;
1737 dentry = ll_lookup_one_len(name, parent, namelen);
1738 if (IS_ERR(dentry)) {
1739 int rc = PTR_ERR(dentry);
1742 CERROR("Fail to find %.*s in %.*s (%lu/%u): rc = %d\n",
1743 namelen, name, parent->d_name.len,
1744 parent->d_name.name, parent->d_inode->i_ino,
1745 parent->d_inode->i_generation, rc);
1750 if (dentry->d_inode == NULL) {
1752 return ERR_PTR(-ENOENT);
1759 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1760 scandir_t scandir, filldir_t filldir)
1762 struct osd_ios_item *item;
1765 OBD_ALLOC_PTR(item);
1769 INIT_LIST_HEAD(&item->oii_list);
1770 item->oii_dentry = dget(dentry);
1771 item->oii_scandir = scandir;
1772 item->oii_filldir = filldir;
1773 list_add_tail(&item->oii_list, &dev->od_ios_list);
1779 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1781 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1782 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1783 * reference the inode, or fixed if it is missing or references another inode.
1786 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1787 struct inode *inode, const struct lu_fid *fid, int flags)
1789 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1790 struct osd_inode_id *id = &info->oti_id;
1791 struct osd_inode_id *id2 = &info->oti_id2;
1792 struct osd_scrub *scrub = &dev->od_scrub;
1793 struct scrub_file *sf = &scrub->os_file;
1798 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1799 if (rc != 0 && rc != -ENODATA) {
1800 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1801 "rc = %d\n", osd_name(dev), rc);
1806 osd_id_gen(id, inode->i_ino, inode->i_generation);
1807 if (rc == -ENODATA) {
1808 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1809 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1812 if (flags & OLF_IDX_IN_FID) {
1813 LASSERT(dev->od_index >= 0);
1815 tfid.f_oid = dev->od_index;
1818 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1820 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1821 "scrub: rc = %d\n", osd_name(dev), rc);
1826 if (lma->lma_compat & LMAC_NOT_IN_OI)
1829 tfid = lma->lma_self_fid;
1832 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1837 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1838 DTO_INDEX_INSERT, true, 0);
1845 if (osd_id_eq_strict(id, id2))
1848 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1849 osd_scrub_file_reset(scrub,
1850 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1852 rc = osd_scrub_file_store(scrub);
1857 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1858 DTO_INDEX_UPDATE, true, 0);
1866 * It scans the /lost+found, and for the OST-object (with filter_fid
1867 * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1869 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1870 loff_t offset, __u64 ino, unsigned d_type)
1872 struct osd_ios_filldir_buf *fill_buf = buf;
1873 struct osd_thread_info *info = fill_buf->oifb_info;
1874 struct osd_device *dev = fill_buf->oifb_dev;
1875 struct lu_fid *fid = &info->oti_fid;
1876 struct osd_scrub *scrub = &dev->od_scrub;
1877 struct dentry *parent = fill_buf->oifb_dentry;
1878 struct dentry *child;
1879 struct inode *dir = parent->d_inode;
1880 struct inode *inode;
1884 /* skip any '.' started names */
1888 scrub->os_lf_scanned++;
1889 child = osd_ios_lookup_one_len(name, parent, namelen);
1890 if (IS_ERR(child)) {
1891 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1892 osd_name(dev), namelen, name, (int)PTR_ERR(child));
1896 inode = child->d_inode;
1897 if (S_ISDIR(inode->i_mode)) {
1898 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1901 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1902 "rc = %d\n", osd_name(dev), namelen, name, rc);
1906 if (!S_ISREG(inode->i_mode))
1909 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1910 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1911 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1913 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1914 "/lost+found.\n", namelen, name, PFID(fid));
1915 scrub->os_lf_repaired++;
1917 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1919 osd_name(dev), namelen, name, PFID(fid), rc);
1923 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1924 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1925 * can process them in furtuer. */
1931 scrub->os_lf_failed++;
1933 /* skip the failure to make the scanning to continue. */
1937 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1938 loff_t offset, __u64 ino, unsigned d_type)
1940 struct osd_ios_filldir_buf *fill_buf = buf;
1941 struct osd_device *dev = fill_buf->oifb_dev;
1942 struct dentry *child;
1946 /* skip any '.' started names */
1950 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1952 RETURN(PTR_ERR(child));
1954 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1956 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1957 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1958 osd_ios_varfid_fill);
1964 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1965 loff_t offset, __u64 ino, unsigned d_type)
1967 struct osd_ios_filldir_buf *fill_buf = buf;
1968 struct osd_device *dev = fill_buf->oifb_dev;
1969 const struct osd_lf_map *map;
1970 struct dentry *child;
1974 /* skip any '.' started names */
1978 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1979 if (map->olm_namelen != namelen)
1982 if (strncmp(map->olm_name, name, namelen) == 0)
1986 if (map->olm_name == NULL)
1989 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1991 RETURN(PTR_ERR(child));
1993 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1994 &map->olm_fid, map->olm_flags);
2000 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
2001 loff_t offset, __u64 ino, unsigned d_type)
2003 struct osd_ios_filldir_buf *fill_buf = buf;
2004 struct dentry *child;
2009 /* skip any non-DFID format name */
2013 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2015 RETURN(PTR_ERR(child));
2017 /* skip the start '[' */
2018 sscanf(&name[1], SFID, RFID(&tfid));
2019 if (fid_is_sane(&tfid))
2020 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
2021 child->d_inode, &tfid, 0);
2029 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2030 loff_t offset, __u64 ino, unsigned d_type)
2032 struct osd_ios_filldir_buf *fill_buf = buf;
2033 struct osd_device *dev = fill_buf->oifb_dev;
2034 const struct osd_lf_map *map;
2035 struct dentry *child;
2039 /* skip any '.' started names */
2043 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2044 if (map->olm_namelen != namelen)
2047 if (strncmp(map->olm_name, name, namelen) == 0)
2051 if (map->olm_name == NULL)
2054 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2056 RETURN(PTR_ERR(child));
2058 if (!(map->olm_flags & OLF_NO_OI))
2059 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2060 &map->olm_fid, map->olm_flags);
2061 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2062 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2070 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2071 struct dentry *dentry, filldir_t filldir)
2073 struct osd_ios_filldir_buf buf = {
2074 #ifdef HAVE_DIR_CONTEXT
2075 .ctx.actor = filldir,
2079 .oifb_dentry = dentry };
2080 struct file *filp = &info->oti_file;
2081 struct inode *inode = dentry->d_inode;
2082 const struct file_operations *fops = inode->i_fop;
2086 LASSERT(filldir != NULL);
2089 filp->f_path.dentry = dentry;
2090 filp->f_mode = FMODE_64BITHASH;
2091 filp->f_mapping = inode->i_mapping;
2093 filp->private_data = NULL;
2094 set_file_inode(filp, inode);
2096 #ifdef HAVE_DIR_CONTEXT
2097 buf.ctx.pos = filp->f_pos;
2098 rc = fops->iterate(filp, &buf.ctx);
2099 filp->f_pos = buf.ctx.pos;
2101 rc = fops->readdir(filp, &buf, filldir);
2103 fops->release(inode, filp);
2109 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2110 struct dentry *dentry, filldir_t filldir)
2112 struct osd_scrub *scrub = &dev->od_scrub;
2113 struct scrub_file *sf = &scrub->os_file;
2114 struct dentry *child;
2118 /* It is existing MDT0 device. We only allow the case of object without
2119 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2120 * can generate IGIF mode FID for the object and related OI mapping. If
2121 * it is on other MDTs, then becuase file-level backup/restore, related
2122 * OI mapping may be invalid already, we do not know which is the right
2123 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2125 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2126 * "fid" xattr, then something crashed. We cannot re-generate the
2127 * FID directly, instead, the OI scrub will scan the OI structure
2128 * and try to re-generate the LMA from the OI mapping. But if the
2129 * OI mapping crashed or lost also, then we have to give up under
2130 * double failure cases. */
2131 scrub->os_convert_igif = 1;
2132 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2133 strlen(dot_lustre_name));
2134 if (IS_ERR(child)) {
2135 rc = PTR_ERR(child);
2136 if (rc == -ENOENT) {
2137 /* It is 1.8 MDT device. */
2138 if (!(sf->sf_flags & SF_UPGRADE)) {
2139 osd_scrub_file_reset(scrub,
2140 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2142 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2143 rc = osd_scrub_file_store(scrub);
2149 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2150 * so the client will get IGIF for the ".lustre" object when
2153 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2154 * it does not know whether there are some old clients cached
2155 * the ".lustre" IGIF during the upgrading. Two choices:
2157 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2158 * It will allow the old connected clients to access the
2159 * ".lustre" with cached IGIF. But it will cause others
2160 * on the MDT failed to check "fid_is_dot_lustre()".
2162 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2163 * for ".lustre" in spite of whether there are some clients
2164 * cached the ".lustre" IGIF or not. It enables the check
2165 * "fid_is_dot_lustre()" on the MDT, although it will cause
2166 * that the old connected clients cannot access the ".lustre"
2167 * with the cached IGIF.
2169 * Usually, it is rare case for the old connected clients
2170 * to access the ".lustre" with cached IGIF. So we prefer
2171 * to the solution 2). */
2172 rc = osd_ios_scan_one(info, dev, child->d_inode,
2173 &LU_DOT_LUSTRE_FID, 0);
2175 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2184 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2185 struct dentry *dentry, filldir_t filldir)
2187 struct osd_scrub *scrub = &dev->od_scrub;
2188 struct scrub_file *sf = &scrub->os_file;
2189 struct dentry *child;
2193 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2194 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2195 rc = osd_scrub_file_store(scrub);
2200 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2201 if (!IS_ERR(child)) {
2202 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2205 rc = PTR_ERR(child);
2208 if (rc != 0 && rc != -ENOENT)
2211 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2212 if (!IS_ERR(child)) {
2213 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2216 rc = PTR_ERR(child);
2225 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2226 struct osd_device *dev)
2228 struct osd_ios_item *item = NULL;
2229 scandir_t scandir = osd_ios_general_scan;
2230 filldir_t filldir = osd_ios_root_fill;
2231 struct dentry *dentry = osd_sb(dev)->s_root;
2232 const struct osd_lf_map *map = osd_lf_maps;
2236 /* Lookup IGIF in OI by force for initial OI scrub. */
2237 dev->od_igif_inoi = 1;
2240 rc = scandir(info, dev, dentry, filldir);
2242 dput(item->oii_dentry);
2249 if (list_empty(&dev->od_ios_list))
2252 item = list_entry(dev->od_ios_list.next,
2253 struct osd_ios_item, oii_list);
2254 list_del_init(&item->oii_list);
2256 LASSERT(item->oii_scandir != NULL);
2257 scandir = item->oii_scandir;
2258 filldir = item->oii_filldir;
2259 dentry = item->oii_dentry;
2262 while (!list_empty(&dev->od_ios_list)) {
2263 item = list_entry(dev->od_ios_list.next,
2264 struct osd_ios_item, oii_list);
2265 list_del_init(&item->oii_list);
2266 dput(item->oii_dentry);
2273 /* There maybe the case that the object has been removed, but its OI
2274 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2275 * file-level backup/restore. So here cleanup the stale OI mappings. */
2276 while (map->olm_name != NULL) {
2277 struct dentry *child;
2279 if (fid_is_zero(&map->olm_fid)) {
2284 child = osd_ios_lookup_one_len(map->olm_name,
2285 osd_sb(dev)->s_root,
2289 else if (PTR_ERR(child) == -ENOENT)
2290 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2291 NULL, DTO_INDEX_DELETE,
2299 char *osd_lf_fid2name(const struct lu_fid *fid)
2301 const struct osd_lf_map *map = osd_lf_maps;
2303 while (map->olm_name != NULL) {
2304 if (!lu_fid_eq(fid, &map->olm_fid)) {
2309 if (map->olm_flags & OLF_SHOW_NAME)
2310 return map->olm_name;
2318 /* OI scrub start/stop */
2320 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2322 struct osd_scrub *scrub = &dev->od_scrub;
2323 struct ptlrpc_thread *thread = &scrub->os_thread;
2324 struct l_wait_info lwi = { 0 };
2325 struct task_struct *task;
2329 /* os_lock: sync status between stop and scrub thread */
2330 spin_lock(&scrub->os_lock);
2333 if (thread_is_running(thread)) {
2334 spin_unlock(&scrub->os_lock);
2335 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2336 (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2339 osd_scrub_join(dev, flags, false);
2340 spin_lock(&scrub->os_lock);
2341 if (!thread_is_running(thread))
2344 spin_unlock(&scrub->os_lock);
2348 if (unlikely(thread_is_stopping(thread))) {
2349 spin_unlock(&scrub->os_lock);
2350 l_wait_event(thread->t_ctl_waitq,
2351 thread_is_stopped(thread),
2353 spin_lock(&scrub->os_lock);
2356 spin_unlock(&scrub->os_lock);
2358 if (scrub->os_file.sf_status == SS_COMPLETED) {
2359 if (!(flags & SS_SET_FAILOUT))
2360 flags |= SS_CLEAR_FAILOUT;
2362 if (!(flags & SS_SET_DRYRUN))
2363 flags |= SS_CLEAR_DRYRUN;
2368 scrub->os_start_flags = flags;
2369 thread_set_flags(thread, 0);
2370 task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2373 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2374 osd_scrub2name(scrub), rc);
2378 l_wait_event(thread->t_ctl_waitq,
2379 thread_is_running(thread) || thread_is_stopped(thread),
2385 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2390 /* od_otable_mutex: prevent curcurrent start/stop */
2391 mutex_lock(&dev->od_otable_mutex);
2392 rc = do_osd_scrub_start(dev, flags);
2393 mutex_unlock(&dev->od_otable_mutex);
2395 RETURN(rc == -EALREADY ? 0 : rc);
2398 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2400 struct ptlrpc_thread *thread = &scrub->os_thread;
2401 struct l_wait_info lwi = { 0 };
2403 /* os_lock: sync status between stop and scrub thread */
2404 spin_lock(&scrub->os_lock);
2405 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2406 thread_set_flags(thread, SVC_STOPPING);
2407 spin_unlock(&scrub->os_lock);
2408 wake_up_all(&thread->t_ctl_waitq);
2409 l_wait_event(thread->t_ctl_waitq,
2410 thread_is_stopped(thread),
2412 /* Do not skip the last lock/unlock, which can guarantee that
2413 * the caller cannot return until the OI scrub thread exit. */
2414 spin_lock(&scrub->os_lock);
2416 spin_unlock(&scrub->os_lock);
2419 static void osd_scrub_stop(struct osd_device *dev)
2421 /* od_otable_mutex: prevent curcurrent start/stop */
2422 mutex_lock(&dev->od_otable_mutex);
2423 dev->od_scrub.os_paused = 1;
2424 do_osd_scrub_stop(&dev->od_scrub);
2425 mutex_unlock(&dev->od_otable_mutex);
2428 /* OI scrub setup/cleanup */
2430 static const char osd_scrub_name[] = "OI_scrub";
2432 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2434 struct osd_thread_info *info = osd_oti_get(env);
2435 struct osd_scrub *scrub = &dev->od_scrub;
2436 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
2437 struct scrub_file *sf = &scrub->os_file;
2438 struct super_block *sb = osd_sb(dev);
2439 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2440 struct lvfs_run_ctxt saved;
2442 struct inode *inode;
2443 struct lu_fid *fid = &info->oti_fid;
2448 memset(scrub, 0, sizeof(*scrub));
2449 OBD_SET_CTXT_MAGIC(ctxt);
2450 ctxt->pwdmnt = dev->od_mnt;
2451 ctxt->pwd = dev->od_mnt->mnt_root;
2452 ctxt->fs = get_ds();
2454 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2455 init_rwsem(&scrub->os_rwsem);
2456 spin_lock_init(&scrub->os_lock);
2457 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2459 push_ctxt(&saved, ctxt);
2460 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2462 pop_ctxt(&saved, ctxt);
2463 RETURN(PTR_ERR(filp));
2466 inode = filp->f_path.dentry->d_inode;
2467 /* 'What the @fid is' is not imporatant, because the object
2468 * has no OI mapping, and only is visible inside the OSD.*/
2469 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2470 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2472 filp_close(filp, NULL);
2473 pop_ctxt(&saved, ctxt);
2477 scrub->os_inode = igrab(inode);
2478 filp_close(filp, NULL);
2479 pop_ctxt(&saved, ctxt);
2481 rc = osd_scrub_file_load(scrub);
2482 if (rc == -ENOENT) {
2483 osd_scrub_file_init(scrub, es->s_uuid);
2484 /* If the "/O" dir does not exist when mount (indicated by
2485 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2486 * then it is quite probably that the device is a new one,
2487 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2489 * For the rare case that "/O" and "OI_scrub" both lost on
2490 * an old device, it can be found and cleared later.
2492 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2493 * need to check "filter_fid_old" and to convert it to
2494 * "filter_fid" for each object, and all the IGIF should
2495 * have their FID mapping in OI files already. */
2496 if (dev->od_maybe_new)
2497 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2499 } else if (rc != 0) {
2500 GOTO(cleanup_inode, rc);
2502 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2503 struct obd_uuid *old_uuid;
2504 struct obd_uuid *new_uuid;
2506 OBD_ALLOC_PTR(old_uuid);
2507 OBD_ALLOC_PTR(new_uuid);
2508 if (old_uuid == NULL || new_uuid == NULL) {
2509 CERROR("%.16s: UUID has been changed, but"
2510 "failed to allocate RAM for report\n",
2511 LDISKFS_SB(sb)->s_es->s_volume_name);
2513 class_uuid_unparse(sf->sf_uuid, old_uuid);
2514 class_uuid_unparse(es->s_uuid, new_uuid);
2515 CERROR("%.16s: UUID has been changed from "
2517 LDISKFS_SB(sb)->s_es->s_volume_name,
2518 old_uuid->uuid, new_uuid->uuid);
2520 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2522 if (old_uuid != NULL)
2523 OBD_FREE_PTR(old_uuid);
2524 if (new_uuid != NULL)
2525 OBD_FREE_PTR(new_uuid);
2526 } else if (sf->sf_status == SS_SCANNING) {
2527 sf->sf_status = SS_CRASHED;
2532 if (sf->sf_pos_last_checkpoint != 0)
2533 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2535 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2538 rc = osd_scrub_file_store(scrub);
2540 GOTO(cleanup_inode, rc);
2543 /* Initialize OI files. */
2544 rc = osd_oi_init(info, dev);
2546 GOTO(cleanup_inode, rc);
2548 rc = osd_initial_OI_scrub(info, dev);
2550 GOTO(cleanup_oi, rc);
2552 if (sf->sf_flags & SF_UPGRADE ||
2553 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2554 sf->sf_success_count > 0)) {
2555 dev->od_igif_inoi = 0;
2556 dev->od_check_ff = dev->od_is_ost;
2558 dev->od_igif_inoi = 1;
2559 dev->od_check_ff = 0;
2562 if (sf->sf_flags & SF_INCONSISTENT)
2563 /* The 'od_igif_inoi' will be set under the
2565 * 1) new created system, or
2566 * 2) restored from file-level backup, or
2567 * 3) the upgrading completed.
2569 * The 'od_igif_inoi' may be cleared by OI scrub
2570 * later if found that the system is upgrading. */
2571 dev->od_igif_inoi = 1;
2573 if (!dev->od_noscrub &&
2574 ((sf->sf_status == SS_PAUSED) ||
2575 (sf->sf_status == SS_CRASHED &&
2576 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2577 SF_UPGRADE | SF_AUTO)) ||
2578 (sf->sf_status == SS_INIT &&
2579 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2581 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2584 GOTO(cleanup_oi, rc);
2586 /* it is possible that dcache entries may keep objects after they are
2587 * deleted by OSD. While it looks safe this can cause object data to
2588 * stay until umount causing failures in tests calculating free space,
2589 * e.g. replay-ost-single. Since those dcache entries are not used
2590 * anymore let's just free them after use here */
2591 shrink_dcache_sb(sb);
2595 osd_oi_fini(info, dev);
2597 iput(scrub->os_inode);
2598 scrub->os_inode = NULL;
2603 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2605 struct osd_scrub *scrub = &dev->od_scrub;
2607 LASSERT(dev->od_otable_it == NULL);
2609 if (scrub->os_inode != NULL) {
2610 osd_scrub_stop(dev);
2611 iput(scrub->os_inode);
2612 scrub->os_inode = NULL;
2614 if (dev->od_oi_table != NULL)
2615 osd_oi_fini(osd_oti_get(env), dev);
2618 /* object table based iteration APIs */
2620 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2621 struct dt_object *dt, __u32 attr)
2623 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2624 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2625 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2626 struct osd_scrub *scrub = &dev->od_scrub;
2627 struct osd_otable_it *it;
2632 /* od_otable_mutex: prevent curcurrent init/fini */
2633 mutex_lock(&dev->od_otable_mutex);
2634 if (dev->od_otable_it != NULL)
2635 GOTO(out, it = ERR_PTR(-EALREADY));
2639 GOTO(out, it = ERR_PTR(-ENOMEM));
2641 dev->od_otable_it = it;
2643 it->ooi_cache.ooc_consumer_idx = -1;
2644 if (flags & DOIF_OUTUSED)
2645 it->ooi_used_outside = 1;
2647 if (flags & DOIF_RESET)
2650 if (valid & DOIV_ERROR_HANDLE) {
2651 if (flags & DOIF_FAILOUT)
2652 start |= SS_SET_FAILOUT;
2654 start |= SS_CLEAR_FAILOUT;
2657 if (valid & DOIV_DRYRUN) {
2658 if (flags & DOIF_DRYRUN)
2659 start |= SS_SET_DRYRUN;
2661 start |= SS_CLEAR_DRYRUN;
2664 rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2665 if (rc < 0 && rc != -EALREADY) {
2666 dev->od_otable_it = NULL;
2668 GOTO(out, it = ERR_PTR(rc));
2671 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2676 mutex_unlock(&dev->od_otable_mutex);
2677 return (struct dt_it *)it;
2680 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2682 struct osd_otable_it *it = (struct osd_otable_it *)di;
2683 struct osd_device *dev = it->ooi_dev;
2685 /* od_otable_mutex: prevent curcurrent init/fini */
2686 mutex_lock(&dev->od_otable_mutex);
2687 do_osd_scrub_stop(&dev->od_scrub);
2688 LASSERT(dev->od_otable_it == it);
2690 dev->od_otable_it = NULL;
2691 mutex_unlock(&dev->od_otable_mutex);
2695 static int osd_otable_it_get(const struct lu_env *env,
2696 struct dt_it *di, const struct dt_key *key)
2701 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2706 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2708 spin_lock(&scrub->os_lock);
2709 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2710 scrub->os_waiting ||
2711 !thread_is_running(&scrub->os_thread))
2712 it->ooi_waiting = 0;
2714 it->ooi_waiting = 1;
2715 spin_unlock(&scrub->os_lock);
2717 return !it->ooi_waiting;
2720 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2722 struct osd_otable_it *it = (struct osd_otable_it *)di;
2723 struct osd_device *dev = it->ooi_dev;
2724 struct osd_scrub *scrub = &dev->od_scrub;
2725 struct osd_otable_cache *ooc = &it->ooi_cache;
2726 struct ptlrpc_thread *thread = &scrub->os_thread;
2727 struct l_wait_info lwi = { 0 };
2731 LASSERT(it->ooi_user_ready);
2734 if (!thread_is_running(thread) && !it->ooi_used_outside)
2737 if (ooc->ooc_cached_items > 0) {
2738 ooc->ooc_cached_items--;
2739 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2740 ~OSD_OTABLE_IT_CACHE_MASK;
2744 if (it->ooi_all_cached) {
2745 l_wait_event(thread->t_ctl_waitq,
2746 !thread_is_running(thread),
2751 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2752 spin_lock(&scrub->os_lock);
2753 scrub->os_waiting = 0;
2754 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2755 spin_unlock(&scrub->os_lock);
2758 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2759 l_wait_event(thread->t_ctl_waitq,
2760 osd_otable_it_wakeup(scrub, it),
2763 if (!thread_is_running(thread) && !it->ooi_used_outside)
2766 rc = osd_otable_it_preload(env, it);
2773 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2774 const struct dt_it *di)
2779 static int osd_otable_it_key_size(const struct lu_env *env,
2780 const struct dt_it *di)
2782 return sizeof(__u64);
2785 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2786 struct dt_rec *rec, __u32 attr)
2788 struct osd_otable_it *it = (struct osd_otable_it *)di;
2789 struct osd_otable_cache *ooc = &it->ooi_cache;
2791 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2793 /* Filter out Invald FID already. */
2794 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2795 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2796 PFID((struct lu_fid *)rec),
2797 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2802 static __u64 osd_otable_it_store(const struct lu_env *env,
2803 const struct dt_it *di)
2805 struct osd_otable_it *it = (struct osd_otable_it *)di;
2806 struct osd_otable_cache *ooc = &it->ooi_cache;
2809 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2810 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2812 hash = ooc->ooc_pos_preload;
2817 * Set the OSD layer iteration start position as the specified hash.
2819 static int osd_otable_it_load(const struct lu_env *env,
2820 const struct dt_it *di, __u64 hash)
2822 struct osd_otable_it *it = (struct osd_otable_it *)di;
2823 struct osd_device *dev = it->ooi_dev;
2824 struct osd_otable_cache *ooc = &it->ooi_cache;
2825 struct osd_scrub *scrub = &dev->od_scrub;
2829 /* Forbid to set iteration position after iteration started. */
2830 if (it->ooi_user_ready)
2833 if (hash > OSD_OTABLE_MAX_HASH)
2834 hash = OSD_OTABLE_MAX_HASH;
2836 ooc->ooc_pos_preload = hash;
2837 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2838 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2840 it->ooi_user_ready = 1;
2841 if (!scrub->os_full_speed)
2842 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2844 /* Unplug OSD layer iteration by the first next() call. */
2845 rc = osd_otable_it_next(env, (struct dt_it *)it);
2850 static int osd_otable_it_key_rec(const struct lu_env *env,
2851 const struct dt_it *di, void *key_rec)
2856 const struct dt_index_operations osd_otable_ops = {
2858 .init = osd_otable_it_init,
2859 .fini = osd_otable_it_fini,
2860 .get = osd_otable_it_get,
2861 .put = osd_otable_it_put,
2862 .next = osd_otable_it_next,
2863 .key = osd_otable_it_key,
2864 .key_size = osd_otable_it_key_size,
2865 .rec = osd_otable_it_rec,
2866 .store = osd_otable_it_store,
2867 .load = osd_otable_it_load,
2868 .key_rec = osd_otable_it_key_rec,
2872 /* high priority inconsistent items list APIs */
2874 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
2876 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2879 struct osd_inconsistent_item *oii;
2880 struct osd_scrub *scrub = &dev->od_scrub;
2881 struct ptlrpc_thread *thread = &scrub->os_thread;
2886 if (unlikely(oii == NULL))
2889 INIT_LIST_HEAD(&oii->oii_list);
2890 oii->oii_cache = *oic;
2891 oii->oii_insert = insert;
2893 if (scrub->os_partial_scan) {
2894 __u64 now = cfs_time_current_sec();
2896 /* If there haven't been errors in a long time,
2897 * decay old count until either the errors are
2898 * gone or we reach the current interval. */
2899 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2900 scrub->os_bad_oimap_time +
2901 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2902 scrub->os_bad_oimap_count >>= 1;
2903 scrub->os_bad_oimap_time +=
2904 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2907 scrub->os_bad_oimap_time = now;
2908 if (++scrub->os_bad_oimap_count >
2909 dev->od_full_scrub_threshold_rate)
2910 scrub->os_full_scrub = 1;
2913 spin_lock(&scrub->os_lock);
2914 if (unlikely(!thread_is_running(thread))) {
2915 spin_unlock(&scrub->os_lock);
2920 if (list_empty(&scrub->os_inconsistent_items))
2922 list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2923 spin_unlock(&scrub->os_lock);
2926 wake_up_all(&thread->t_ctl_waitq);
2931 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2932 struct osd_inode_id *id)
2934 struct osd_scrub *scrub = &dev->od_scrub;
2935 struct osd_inconsistent_item *oii;
2938 spin_lock(&scrub->os_lock);
2939 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2940 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2941 *id = oii->oii_cache.oic_lid;
2942 spin_unlock(&scrub->os_lock);
2946 spin_unlock(&scrub->os_lock);
2953 static const char *scrub_status_names[] = {
2964 static const char *scrub_flags_names[] = {
2972 static const char *scrub_param_names[] = {
2978 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2985 rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2989 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2992 rc = seq_printf(m, "%s%c", names[i],
2993 bits != 0 ? ',' : '\n');
3001 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
3006 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
3007 cfs_time_current_sec() - time);
3009 rc = seq_printf(m, "%s: N/A\n", prefix);
3013 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
3018 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
3020 rc = seq_printf(m, "%s: N/A\n", prefix);
3024 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
3026 struct osd_scrub *scrub = &dev->od_scrub;
3027 struct scrub_file *sf = &scrub->os_file;
3032 down_read(&scrub->os_rwsem);
3033 rc = seq_printf(m, "name: OI_scrub\n"
3037 sf->sf_magic, (int)sf->sf_oi_count,
3038 scrub_status_names[sf->sf_status]);
3042 rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
3047 rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
3052 rc = scrub_time_dump(m, sf->sf_time_last_complete,
3053 "time_since_last_completed");
3057 rc = scrub_time_dump(m, sf->sf_time_latest_start,
3058 "time_since_latest_start");
3062 rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
3063 "time_since_last_checkpoint");
3067 rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
3068 "latest_start_position");
3072 rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3073 "last_checkpoint_position");
3077 rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3078 "first_failure_position");
3082 checked = sf->sf_items_checked + scrub->os_new_checked;
3083 rc = seq_printf(m, "checked: "LPU64"\n"
3084 "updated: "LPU64"\n"
3086 "prior_updated: "LPU64"\n"
3087 "noscrub: "LPU64"\n"
3089 "success_count: %u\n",
3090 checked, sf->sf_items_updated, sf->sf_items_failed,
3091 sf->sf_items_updated_prior, sf->sf_items_noscrub,
3092 sf->sf_items_igif, sf->sf_success_count);
3097 if (thread_is_running(&scrub->os_thread)) {
3098 cfs_duration_t duration = cfs_time_current() -
3099 scrub->os_time_last_checkpoint;
3100 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3102 __u32 rtime = sf->sf_run_time +
3103 cfs_duration_sec(duration + HALF_SEC);
3106 do_div(new_checked, duration);
3108 do_div(speed, rtime);
3109 rc = seq_printf(m, "run_time: %u seconds\n"
3110 "average_speed: "LPU64" objects/sec\n"
3111 "real-time_speed: "LPU64" objects/sec\n"
3112 "current_position: %u\n"
3113 "lf_scanned: "LPU64"\n"
3114 "lf_repaired: "LPU64"\n"
3115 "lf_failed: "LPU64"\n",
3116 rtime, speed, new_checked, scrub->os_pos_current,
3117 scrub->os_lf_scanned, scrub->os_lf_repaired,
3118 scrub->os_lf_failed);
3120 if (sf->sf_run_time != 0)
3121 do_div(speed, sf->sf_run_time);
3122 rc = seq_printf(m, "run_time: %u seconds\n"
3123 "average_speed: "LPU64" objects/sec\n"
3124 "real-time_speed: N/A\n"
3125 "current_position: N/A\n"
3126 "lf_scanned: "LPU64"\n"
3127 "lf_repaired: "LPU64"\n"
3128 "lf_failed: "LPU64"\n",
3129 sf->sf_run_time, speed, scrub->os_lf_scanned,
3130 scrub->os_lf_repaired, scrub->os_lf_failed);
3134 up_read(&scrub->os_rwsem);
3135 return (rc < 0 ? -ENOSPC : 0);