4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2012, 2014, Intel Corporation.
26 * lustre/osd-ldiskfs/osd_scrub.c
28 * Top-level entry points into osd module
30 * The OI scrub is used for rebuilding Object Index files when restores MDT from
33 * The otable based iterator scans ldiskfs inode table to feed up layer LFSCK.
35 * Author: Fan Yong <yong.fan@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_LFSCK
40 #include <linux/kthread.h>
41 #include <lustre/lustre_idl.h>
42 #include <lustre_disk.h>
43 #include <dt_object.h>
44 #include <linux/xattr.h>
46 #include "osd_internal.h"
48 #include "osd_scrub.h"
50 #define HALF_SEC msecs_to_jiffies(MSEC_PER_SEC >> 1)
52 #define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL
54 #define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */
55 #define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */
56 #define SCRUB_NEXT_EXIT 3 /* exit all the loops */
57 #define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */
58 #define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */
59 #define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */
60 #define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */
61 #define SCRUB_NEXT_NOLMA 8 /* the inode has no FID-in-LMA */
62 #define SCRUB_NEXT_OSTOBJ 9 /* for OST-object */
63 #define SCRUB_NEXT_OSTOBJ_OLD 10 /* old OST-object, no LMA or no FID-on-OST
68 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
70 return container_of0(scrub, struct osd_device, od_scrub);
73 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
75 return osd_sb(osd_scrub2dev(scrub));
78 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
79 struct osd_otable_cache *ooc)
81 return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
84 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
86 return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
90 * update/insert/delete the specified OI mapping (@fid @id) according to the ops
92 * \retval 1, changed nothing
93 * \retval 0, changed successfully
94 * \retval -ve, on error
96 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
97 struct osd_device *dev,
98 const struct lu_fid *fid,
99 const struct osd_inode_id *id,
101 enum oi_check_flags flags)
107 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
110 /* DTO_INDEX_INSERT is enough for other two ops:
111 * delete/update, but save stack. */
112 th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
113 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
116 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
117 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
118 PFID(fid), id->oii_ino, id->oii_gen, rc);
123 case DTO_INDEX_UPDATE:
124 rc = osd_oi_update(info, dev, fid, id, th, flags);
125 if (unlikely(rc == -ENOENT)) {
126 /* Some unlink thread may removed the OI mapping. */
130 case DTO_INDEX_INSERT:
131 rc = osd_oi_insert(info, dev, fid, id, th, flags);
132 if (unlikely(rc == -EEXIST)) {
134 /* XXX: There are trouble things when adding OI
135 * mapping for IGIF object, which may cause
136 * multiple objects to be mapped to the same
137 * IGIF formatted FID. Consider the following
140 * 1) The MDT is upgrading from 1.8 device.
141 * The OI scrub generates IGIF FID1 for the
142 * OBJ1 and adds the OI mapping.
144 * 2) For some reason, the OI scrub does not
145 * process all the IGIF objects completely.
147 * 3) The MDT is backuped and restored against
150 * 4) When the MDT mounts up, the OI scrub will
151 * try to rebuild the OI files. For some IGIF
152 * object, OBJ2, which was not processed by the
153 * OI scrub before the backup/restore, and the
154 * new generated IGIF formatted FID may be just
155 * the FID1, the same as OBJ1.
157 * Under such case, the OI scrub cannot know how
158 * to generate new FID for the OBJ2.
160 * Currently, we do nothing for that. One possible
161 * solution is to generate new normal FID for the
164 * Anyway, it is rare, only exists in theory. */
167 case DTO_INDEX_DELETE:
168 rc = osd_oi_delete(info, dev, fid, th, flags);
170 /* It is normal that the unlink thread has removed the
171 * OI mapping already. */
176 LASSERTF(0, "Unexpected ops %d\n", ops);
180 ldiskfs_journal_stop(th);
182 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
183 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
184 PFID(fid), id ? 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(inode->i_nlink == 0))
619 osd_scrub_refresh_mapping(info, dev, fid, lid,
620 DTO_INDEX_DELETE, false,
621 (val == SCRUB_NEXT_OSTOBJ ||
622 val == SCRUB_NEXT_OSTOBJ_OLD) ?
623 OI_KNOWN_ON_OST : 0);
624 up_write(&scrub->os_rwsem);
626 if (inode != NULL && !IS_ERR(inode))
630 LASSERT(!list_empty(&oii->oii_list));
632 spin_lock(&scrub->os_lock);
633 list_del_init(&oii->oii_list);
634 spin_unlock(&scrub->os_lock);
637 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
642 static int osd_scrub_prep(struct osd_device *dev)
644 struct osd_scrub *scrub = &dev->od_scrub;
645 struct ptlrpc_thread *thread = &scrub->os_thread;
646 struct scrub_file *sf = &scrub->os_file;
647 __u32 flags = scrub->os_start_flags;
649 bool drop_dryrun = false;
652 CDEBUG(D_LFSCK, "%.16s: OI scrub prep, flags = 0x%x\n",
653 osd_scrub2name(scrub), flags);
655 down_write(&scrub->os_rwsem);
656 if (flags & SS_SET_FAILOUT)
657 sf->sf_param |= SP_FAILOUT;
658 else if (flags & SS_CLEAR_FAILOUT)
659 sf->sf_param &= ~SP_FAILOUT;
661 if (flags & SS_SET_DRYRUN) {
662 sf->sf_param |= SP_DRYRUN;
663 } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
664 sf->sf_param &= ~SP_DRYRUN;
668 if (flags & SS_RESET)
669 osd_scrub_file_reset(scrub,
670 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
672 if (flags & SS_AUTO_FULL) {
673 scrub->os_full_speed = 1;
674 scrub->os_partial_scan = 0;
675 sf->sf_flags |= SF_AUTO;
676 } else if (flags & SS_AUTO_PARTIAL) {
677 scrub->os_full_speed = 0;
678 scrub->os_partial_scan = 1;
679 sf->sf_flags |= SF_AUTO;
680 } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
682 scrub->os_full_speed = 1;
683 scrub->os_partial_scan = 0;
685 scrub->os_full_speed = 0;
686 scrub->os_partial_scan = 0;
689 spin_lock(&scrub->os_lock);
690 scrub->os_in_prior = 0;
691 scrub->os_waiting = 0;
692 scrub->os_paused = 0;
693 scrub->os_in_join = 0;
694 scrub->os_full_scrub = 0;
695 spin_unlock(&scrub->os_lock);
696 scrub->os_new_checked = 0;
697 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
698 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
699 else if (sf->sf_pos_last_checkpoint != 0)
700 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
702 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
704 scrub->os_pos_current = sf->sf_pos_latest_start;
705 sf->sf_status = SS_SCANNING;
706 sf->sf_time_latest_start = cfs_time_current_sec();
707 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
708 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
709 rc = osd_scrub_file_store(scrub);
711 spin_lock(&scrub->os_lock);
712 thread_set_flags(thread, SVC_RUNNING);
713 spin_unlock(&scrub->os_lock);
714 wake_up_all(&thread->t_ctl_waitq);
716 up_write(&scrub->os_rwsem);
721 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
723 struct scrub_file *sf = &scrub->os_file;
726 if (likely(cfs_time_before(cfs_time_current(),
727 scrub->os_time_next_checkpoint) ||
728 scrub->os_new_checked == 0))
731 down_write(&scrub->os_rwsem);
732 sf->sf_items_checked += scrub->os_new_checked;
733 scrub->os_new_checked = 0;
734 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
735 sf->sf_time_last_checkpoint = cfs_time_current_sec();
736 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
737 scrub->os_time_last_checkpoint);
738 rc = osd_scrub_file_store(scrub);
739 up_write(&scrub->os_rwsem);
744 static void osd_scrub_post(struct osd_scrub *scrub, int result)
746 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 result = osd_scrub_file_store(scrub);
787 up_write(&scrub->os_rwsem);
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 struct osd_inconsistent_item *oii;
1014 oii = list_entry(scrub->os_inconsistent_items.next,
1015 struct osd_inconsistent_item, oii_list);
1016 *oic = &oii->oii_cache;
1017 scrub->os_in_prior = 1;
1022 return SCRUB_NEXT_WAIT;
1024 rc = osd_iit_next(param, &scrub->os_pos_current);
1028 *oic = &scrub->os_oic;
1029 fid = &(*oic)->oic_fid;
1030 lid = &(*oic)->oic_lid;
1031 rc = osd_iit_iget(info, dev, fid, lid,
1032 scrub->os_pos_current, param->sb, true);
1036 static int osd_preload_next(struct osd_thread_info *info,
1037 struct osd_device *dev, struct osd_iit_param *param,
1038 struct osd_idmap_cache **oic, const bool noslot)
1040 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1041 struct osd_scrub *scrub;
1042 struct ptlrpc_thread *thread;
1045 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1049 scrub = &dev->od_scrub;
1050 thread = &scrub->os_thread;
1051 if (thread_is_running(thread) &&
1052 ooc->ooc_pos_preload >= scrub->os_pos_current)
1053 return SCRUB_NEXT_EXIT;
1055 rc = osd_iit_iget(info, dev,
1056 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1057 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1058 ooc->ooc_pos_preload, param->sb, false);
1059 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1060 * ignore the failure, so it still need to skip the inode next time. */
1061 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1066 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1068 spin_lock(&scrub->os_lock);
1069 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1070 !list_empty(&scrub->os_inconsistent_items) ||
1071 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1072 scrub->os_waiting = 0;
1074 scrub->os_waiting = 1;
1075 spin_unlock(&scrub->os_lock);
1077 return !scrub->os_waiting;
1080 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1081 struct osd_iit_param *param,
1082 struct osd_idmap_cache *oic, bool *noslot, int rc)
1084 struct l_wait_info lwi = { 0 };
1085 struct osd_scrub *scrub = &dev->od_scrub;
1086 struct scrub_file *sf = &scrub->os_file;
1087 struct ptlrpc_thread *thread = &scrub->os_thread;
1088 struct osd_otable_it *it = dev->od_otable_it;
1089 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
1092 case SCRUB_NEXT_CONTINUE:
1094 case SCRUB_NEXT_WAIT:
1096 case SCRUB_NEXT_NOSCRUB:
1097 down_write(&scrub->os_rwsem);
1098 scrub->os_new_checked++;
1099 sf->sf_items_noscrub++;
1100 up_write(&scrub->os_rwsem);
1104 rc = osd_scrub_check_update(info, dev, oic, rc);
1108 rc = osd_scrub_checkpoint(scrub);
1110 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1111 "rc = %d\n", osd_scrub2name(scrub),
1112 scrub->os_pos_current, rc);
1113 /* Continue, as long as the scrub itself can go ahead. */
1116 if (scrub->os_in_prior) {
1117 scrub->os_in_prior = 0;
1122 scrub->os_pos_current = param->gbase + ++(param->offset);
1125 if (it != NULL && it->ooi_waiting && ooc != NULL &&
1126 ooc->ooc_pos_preload < scrub->os_pos_current) {
1127 spin_lock(&scrub->os_lock);
1128 it->ooi_waiting = 0;
1129 wake_up_all(&thread->t_ctl_waitq);
1130 spin_unlock(&scrub->os_lock);
1133 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1136 if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1142 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1145 if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1152 static int osd_preload_exec(struct osd_thread_info *info,
1153 struct osd_device *dev, struct osd_iit_param *param,
1154 struct osd_idmap_cache *oic, bool *noslot, int rc)
1156 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1159 ooc->ooc_cached_items++;
1160 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1161 ~OSD_OTABLE_IT_CACHE_MASK;
1163 return rc > 0 ? 0 : rc;
1166 #define SCRUB_IT_ALL 1
1167 #define SCRUB_IT_CRASH 2
1169 static void osd_scrub_join(struct osd_device *dev, __u32 flags,
1172 struct osd_scrub *scrub = &dev->od_scrub;
1173 struct ptlrpc_thread *thread = &scrub->os_thread;
1174 struct scrub_file *sf = &scrub->os_file;
1178 LASSERT(!(flags & SS_AUTO_PARTIAL));
1180 down_write(&scrub->os_rwsem);
1181 scrub->os_in_join = 1;
1182 if (flags & SS_SET_FAILOUT)
1183 sf->sf_param |= SP_FAILOUT;
1184 else if (flags & SS_CLEAR_FAILOUT)
1185 sf->sf_param &= ~SP_FAILOUT;
1187 if (flags & SS_SET_DRYRUN)
1188 sf->sf_param |= SP_DRYRUN;
1189 else if (flags & SS_CLEAR_DRYRUN)
1190 sf->sf_param &= ~SP_DRYRUN;
1192 if (flags & SS_RESET) {
1193 osd_scrub_file_reset(scrub,
1194 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1195 inconsistent ? SF_INCONSISTENT : 0);
1196 sf->sf_status = SS_SCANNING;
1199 if (flags & SS_AUTO_FULL) {
1200 sf->sf_flags |= SF_AUTO;
1201 scrub->os_full_speed = 1;
1204 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
1205 scrub->os_full_speed = 1;
1207 scrub->os_full_speed = 0;
1209 scrub->os_new_checked = 0;
1210 if (sf->sf_pos_last_checkpoint != 0)
1211 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
1213 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1215 scrub->os_pos_current = sf->sf_pos_latest_start;
1216 sf->sf_time_latest_start = cfs_time_current_sec();
1217 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
1218 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
1219 rc = osd_scrub_file_store(scrub);
1221 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file when join "
1222 "the OI scrub: rc = %d\n", osd_scrub2name(scrub), rc);
1224 spin_lock(&scrub->os_lock);
1225 scrub->os_waiting = 0;
1226 scrub->os_paused = 0;
1227 scrub->os_partial_scan = 0;
1228 scrub->os_in_join = 0;
1229 scrub->os_full_scrub = 0;
1230 spin_unlock(&scrub->os_lock);
1231 wake_up_all(&thread->t_ctl_waitq);
1232 up_write(&scrub->os_rwsem);
1237 static int osd_inode_iteration(struct osd_thread_info *info,
1238 struct osd_device *dev, __u32 max, bool preload)
1240 struct osd_scrub *scrub = &dev->od_scrub;
1241 struct ptlrpc_thread *thread = &scrub->os_thread;
1242 struct scrub_file *sf = &scrub->os_file;
1243 osd_iit_next_policy next;
1244 osd_iit_exec_policy exec;
1247 struct osd_iit_param param = { NULL };
1248 struct l_wait_info lwi = { 0 };
1254 param.sb = osd_sb(dev);
1258 while (scrub->os_partial_scan && !scrub->os_in_join) {
1259 struct osd_idmap_cache *oic = NULL;
1261 rc = osd_scrub_next(info, dev, ¶m, &oic, noslot);
1263 case SCRUB_NEXT_EXIT:
1265 case SCRUB_NEXT_CRASH:
1266 RETURN(SCRUB_IT_CRASH);
1267 case SCRUB_NEXT_FATAL:
1269 case SCRUB_NEXT_WAIT: {
1270 struct kstatfs *ksfs = &info->oti_ksfs;
1273 if (dev->od_full_scrub_ratio == OFSR_NEVER ||
1274 unlikely(sf->sf_items_updated_prior == 0))
1277 if (dev->od_full_scrub_ratio == OFSR_DIRECTLY ||
1278 scrub->os_full_scrub) {
1279 osd_scrub_join(dev, SS_AUTO_FULL | SS_RESET,
1284 rc = param.sb->s_op->statfs(param.sb->s_root, ksfs);
1286 __u64 used = ksfs->f_files - ksfs->f_ffree;
1288 do_div(used, sf->sf_items_updated_prior);
1289 /* If we hit too much inconsistent OI
1290 * mappings during the partial scan,
1291 * then scan the device completely. */
1292 if (used < dev->od_full_scrub_ratio) {
1294 SS_AUTO_FULL | SS_RESET, true);
1300 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) &&
1304 saved_flags = sf->sf_flags;
1305 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
1306 SF_UPGRADE | SF_AUTO);
1307 sf->sf_status = SS_COMPLETED;
1308 l_wait_event(thread->t_ctl_waitq,
1309 !thread_is_running(thread) ||
1310 !scrub->os_partial_scan ||
1311 scrub->os_in_join ||
1312 !list_empty(&scrub->os_inconsistent_items),
1314 sf->sf_flags = saved_flags;
1315 sf->sf_status = SS_SCANNING;
1317 if (unlikely(!thread_is_running(thread)))
1320 if (!scrub->os_partial_scan || scrub->os_in_join)
1326 LASSERTF(rc == 0, "rc = %d\n", rc);
1328 osd_scrub_exec(info, dev, ¶m, oic, &noslot, rc);
1335 l_wait_event(thread->t_ctl_waitq,
1336 !thread_is_running(thread) || !scrub->os_in_join,
1339 if (unlikely(!thread_is_running(thread)))
1345 next = osd_scrub_next;
1346 exec = osd_scrub_exec;
1347 pos = &scrub->os_pos_current;
1348 count = &scrub->os_new_checked;
1350 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1352 next = osd_preload_next;
1353 exec = osd_preload_exec;
1354 pos = &ooc->ooc_pos_preload;
1355 count = &ooc->ooc_cached_items;
1357 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1359 while (*pos <= limit && *count < max) {
1360 struct osd_idmap_cache *oic = NULL;
1361 struct ldiskfs_group_desc *desc;
1363 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1364 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1368 ldiskfs_lock_group(param.sb, param.bg);
1369 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1370 ldiskfs_unlock_group(param.sb, param.bg);
1371 *pos = 1 + (param.bg + 1) *
1372 LDISKFS_INODES_PER_GROUP(param.sb);
1375 ldiskfs_unlock_group(param.sb, param.bg);
1377 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1378 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1379 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1380 if (param.bitmap == NULL) {
1381 CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1382 "scrub will stop, urgent mode\n",
1383 osd_scrub2name(scrub), (__u32)param.bg);
1387 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1390 ldiskfs_itable_unused_count(param.sb, desc) >
1391 LDISKFS_INODES_PER_GROUP(param.sb))
1394 rc = next(info, dev, ¶m, &oic, noslot);
1396 case SCRUB_NEXT_BREAK:
1398 case SCRUB_NEXT_EXIT:
1399 brelse(param.bitmap);
1401 case SCRUB_NEXT_CRASH:
1402 brelse(param.bitmap);
1403 RETURN(SCRUB_IT_CRASH);
1404 case SCRUB_NEXT_FATAL:
1405 brelse(param.bitmap);
1409 rc = exec(info, dev, ¶m, oic, &noslot, rc);
1411 brelse(param.bitmap);
1417 brelse(param.bitmap);
1421 RETURN(SCRUB_IT_ALL);
1425 static int osd_otable_it_preload(const struct lu_env *env,
1426 struct osd_otable_it *it)
1428 struct osd_device *dev = it->ooi_dev;
1429 struct osd_scrub *scrub = &dev->od_scrub;
1430 struct osd_otable_cache *ooc = &it->ooi_cache;
1434 rc = osd_inode_iteration(osd_oti_get(env), dev,
1435 OSD_OTABLE_IT_CACHE_SIZE, true);
1436 if (rc == SCRUB_IT_ALL)
1437 it->ooi_all_cached = 1;
1439 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1440 spin_lock(&scrub->os_lock);
1441 scrub->os_waiting = 0;
1442 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1443 spin_unlock(&scrub->os_lock);
1446 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1449 static int osd_scrub_main(void *args)
1452 struct osd_device *dev = (struct osd_device *)args;
1453 struct osd_scrub *scrub = &dev->od_scrub;
1454 struct ptlrpc_thread *thread = &scrub->os_thread;
1458 rc = lu_env_init(&env, LCT_LOCAL);
1460 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1461 osd_scrub2name(scrub), rc);
1465 rc = osd_scrub_prep(dev);
1467 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1468 osd_scrub2name(scrub), rc);
1472 if (!scrub->os_full_speed && !scrub->os_partial_scan) {
1473 struct l_wait_info lwi = { 0 };
1474 struct osd_otable_it *it = dev->od_otable_it;
1475 struct osd_otable_cache *ooc = &it->ooi_cache;
1477 l_wait_event(thread->t_ctl_waitq,
1478 it->ooi_user_ready || !thread_is_running(thread),
1480 if (unlikely(!thread_is_running(thread)))
1483 scrub->os_pos_current = ooc->ooc_pos_preload;
1486 CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1487 osd_scrub2name(scrub), scrub->os_start_flags,
1488 scrub->os_pos_current);
1490 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1491 if (unlikely(rc == SCRUB_IT_CRASH))
1492 GOTO(out, rc = -EINVAL);
1496 osd_scrub_post(scrub, rc);
1497 CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1498 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1501 while (!list_empty(&scrub->os_inconsistent_items)) {
1502 struct osd_inconsistent_item *oii;
1504 oii = list_entry(scrub->os_inconsistent_items.next,
1505 struct osd_inconsistent_item, oii_list);
1506 list_del_init(&oii->oii_list);
1512 spin_lock(&scrub->os_lock);
1513 thread_set_flags(thread, SVC_STOPPED);
1514 wake_up_all(&thread->t_ctl_waitq);
1515 spin_unlock(&scrub->os_lock);
1519 /* initial OI scrub */
1521 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1522 struct dentry *, filldir_t filldir);
1524 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1525 loff_t offset, __u64 ino, unsigned d_type);
1526 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1527 loff_t offset, __u64 ino, unsigned d_type);
1528 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1529 loff_t offset, __u64 ino, unsigned d_type);
1530 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1531 loff_t offset, __u64 ino, unsigned d_type);
1534 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1535 struct dentry *dentry, filldir_t filldir);
1537 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1538 struct dentry *dentry, filldir_t filldir);
1541 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1542 struct dentry *dentry, filldir_t filldir);
1545 OLF_SCAN_SUBITEMS = 0x0001,
1546 OLF_HIDE_FID = 0x0002,
1547 OLF_SHOW_NAME = 0x0004,
1549 OLF_IDX_IN_FID = 0x0010,
1554 struct lu_fid olm_fid;
1557 scandir_t olm_scandir;
1558 filldir_t olm_filldir;
1561 /* Add the new introduced local files in the list in the future. */
1562 static const struct osd_lf_map osd_lf_maps[] = {
1564 { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1565 sizeof(CATLIST) - 1, NULL, NULL },
1568 { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1569 OLF_SCAN_SUBITEMS, sizeof(MOUNT_CONFIGS_DIR) - 1,
1570 osd_ios_general_scan, osd_ios_varfid_fill },
1572 /* NIDTBL_VERSIONS */
1573 { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1574 sizeof(MGS_NIDTBL_DIR) - 1, osd_ios_general_scan,
1575 osd_ios_varfid_fill },
1578 { "PENDING", { 0, 0, 0 }, 0, sizeof("PENDING") - 1, NULL, NULL },
1581 { "ROOT", { FID_SEQ_ROOT, FID_OID_ROOT, 0 },
1582 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, sizeof("ROOT") - 1,
1583 osd_ios_ROOT_scan, NULL },
1585 /* changelog_catalog */
1586 { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, sizeof(CHANGELOG_CATALOG) - 1,
1589 /* changelog_users */
1590 { CHANGELOG_USERS, { 0, 0, 0 }, 0, sizeof(CHANGELOG_USERS) - 1,
1594 { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1595 sizeof("fld") - 1, NULL, NULL },
1598 { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1599 sizeof(LAST_RCVD) - 1, NULL, NULL },
1602 { REPLY_DATA, { FID_SEQ_LOCAL_FILE, REPLY_DATA_OID, 0 }, OLF_SHOW_NAME,
1603 sizeof(REPLY_DATA) - 1, NULL, NULL },
1606 { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1607 sizeof(LOV_OBJID) - 1, NULL, NULL },
1610 { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1611 OLF_SHOW_NAME, sizeof(LOV_OBJSEQ) - 1, NULL, NULL },
1614 { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QMT_DIR) - 1,
1615 osd_ios_general_scan, osd_ios_varfid_fill },
1618 { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(QSD_DIR) - 1,
1619 osd_ios_general_scan, osd_ios_varfid_fill },
1622 { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1623 OLF_SHOW_NAME, sizeof("seq_ctl") - 1, NULL, NULL },
1626 { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1627 OLF_SHOW_NAME, sizeof("seq_srv") - 1, NULL, NULL },
1630 { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1631 OLF_SHOW_NAME, sizeof(HEALTH_CHECK) - 1, NULL, NULL },
1634 { LFSCK_DIR, { 0, 0, 0 }, 0, sizeof(LFSCK_DIR) - 1,
1635 osd_ios_general_scan, osd_ios_varfid_fill },
1637 /* lfsck_bookmark */
1638 { LFSCK_BOOKMARK, { 0, 0, 0 }, 0, sizeof(LFSCK_BOOKMARK) - 1,
1642 { LFSCK_LAYOUT, { 0, 0, 0 }, 0, sizeof(LFSCK_LAYOUT) - 1,
1645 /* lfsck_namespace */
1646 { LFSCK_NAMESPACE, { 0, 0, 0 }, 0, sizeof(LFSCK_NAMESPACE) - 1,
1649 /* OBJECTS, upgrade from old device */
1650 { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, sizeof(OBJECTS) - 1,
1651 osd_ios_OBJECTS_scan, NULL },
1653 /* lquota_v2.user, upgrade from old device */
1654 { "lquota_v2.user", { 0, 0, 0 }, 0, sizeof("lquota_v2.user") - 1,
1657 /* lquota_v2.group, upgrade from old device */
1658 { "lquota_v2.group", { 0, 0, 0 }, 0, sizeof("lquota_v2.group") - 1,
1661 /* LAST_GROUP, upgrade from old device */
1662 { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1663 OLF_SHOW_NAME, sizeof("LAST_GROUP") - 1, NULL, NULL },
1665 /* committed batchid for cross-MDT operation */
1666 { "BATCHID", { FID_SEQ_LOCAL_FILE, BATCHID_COMMITTED_OID, 0 },
1667 OLF_SHOW_NAME, sizeof("BATCHID") - 1, NULL, NULL },
1669 /* OSP update logs update_log{_dir} use f_seq = FID_SEQ_UPDATE_LOG{_DIR}
1670 * and f_oid = index for their log files. See lu_update_log{_dir}_fid()
1671 * for more details. */
1674 { "update_log", { FID_SEQ_UPDATE_LOG, 0, 0 },
1675 OLF_SHOW_NAME | OLF_IDX_IN_FID, sizeof("update_log") - 1,
1678 /* update_log_dir */
1679 { "update_log_dir", { FID_SEQ_UPDATE_LOG_DIR, 0, 0 },
1680 OLF_SHOW_NAME | OLF_SCAN_SUBITEMS | OLF_IDX_IN_FID,
1681 sizeof("update_log_dir") - 1,
1682 osd_ios_general_scan, osd_ios_uld_fill },
1685 { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1686 OLF_SCAN_SUBITEMS, sizeof("lost+found") - 1,
1687 osd_ios_general_scan, osd_ios_lf_fill },
1689 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1692 /* Add the new introduced files under .lustre/ in the list in the future. */
1693 static const struct osd_lf_map osd_dl_maps[] = {
1695 { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1696 sizeof("fid") - 1, NULL, NULL },
1698 /* .lustre/lost+found */
1699 { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1700 sizeof("lost+found") - 1, NULL, NULL },
1702 { NULL, { 0, 0, 0 }, 0, 0, NULL, NULL }
1705 struct osd_ios_item {
1706 struct list_head oii_list;
1707 struct dentry *oii_dentry;
1708 scandir_t oii_scandir;
1709 filldir_t oii_filldir;
1712 struct osd_ios_filldir_buf {
1713 #ifdef HAVE_DIR_CONTEXT
1714 /* please keep it as first member */
1715 struct dir_context ctx;
1717 struct osd_thread_info *oifb_info;
1718 struct osd_device *oifb_dev;
1719 struct dentry *oifb_dentry;
1722 static inline struct dentry *
1723 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1725 struct dentry *dentry;
1727 dentry = ll_lookup_one_len(name, parent, namelen);
1728 if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1730 return ERR_PTR(-ENOENT);
1737 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1738 scandir_t scandir, filldir_t filldir)
1740 struct osd_ios_item *item;
1743 OBD_ALLOC_PTR(item);
1747 INIT_LIST_HEAD(&item->oii_list);
1748 item->oii_dentry = dget(dentry);
1749 item->oii_scandir = scandir;
1750 item->oii_filldir = filldir;
1751 list_add_tail(&item->oii_list, &dev->od_ios_list);
1757 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1759 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1760 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1761 * reference the inode, or fixed if it is missing or references another inode.
1764 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1765 struct inode *inode, const struct lu_fid *fid, int flags)
1767 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1768 struct osd_inode_id *id = &info->oti_id;
1769 struct osd_inode_id *id2 = &info->oti_id2;
1770 struct osd_scrub *scrub = &dev->od_scrub;
1771 struct scrub_file *sf = &scrub->os_file;
1776 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1777 if (rc != 0 && rc != -ENODATA) {
1778 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1779 "rc = %d\n", osd_name(dev), rc);
1784 osd_id_gen(id, inode->i_ino, inode->i_generation);
1785 if (rc == -ENODATA) {
1786 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID) {
1787 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1790 if (flags & OLF_IDX_IN_FID) {
1791 LASSERT(dev->od_index >= 0);
1793 tfid.f_oid = dev->od_index;
1796 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1798 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1799 "scrub: rc = %d\n", osd_name(dev), rc);
1804 if (lma->lma_compat & LMAC_NOT_IN_OI)
1807 tfid = lma->lma_self_fid;
1810 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1815 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1816 DTO_INDEX_INSERT, true, 0);
1823 if (osd_id_eq_strict(id, id2))
1826 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1827 osd_scrub_file_reset(scrub,
1828 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1830 rc = osd_scrub_file_store(scrub);
1835 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1836 DTO_INDEX_UPDATE, true, 0);
1844 * It scans the /lost+found, and for the OST-object (with filter_fid
1845 * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1847 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1848 loff_t offset, __u64 ino, unsigned d_type)
1850 struct osd_ios_filldir_buf *fill_buf = buf;
1851 struct osd_thread_info *info = fill_buf->oifb_info;
1852 struct osd_device *dev = fill_buf->oifb_dev;
1853 struct lu_fid *fid = &info->oti_fid;
1854 struct osd_scrub *scrub = &dev->od_scrub;
1855 struct dentry *parent = fill_buf->oifb_dentry;
1856 struct dentry *child;
1857 struct inode *dir = parent->d_inode;
1858 struct inode *inode;
1862 /* skip any '.' started names */
1866 scrub->os_lf_scanned++;
1867 child = osd_ios_lookup_one_len(name, parent, namelen);
1868 if (IS_ERR(child)) {
1869 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1870 osd_name(dev), namelen, name, (int)PTR_ERR(child));
1874 inode = child->d_inode;
1875 if (S_ISDIR(inode->i_mode)) {
1876 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1879 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1880 "rc = %d\n", osd_name(dev), namelen, name, rc);
1884 if (!S_ISREG(inode->i_mode))
1887 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1888 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1889 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1891 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1892 "/lost+found.\n", namelen, name, PFID(fid));
1893 scrub->os_lf_repaired++;
1895 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1897 osd_name(dev), namelen, name, PFID(fid), rc);
1901 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1902 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1903 * can process them in furtuer. */
1909 scrub->os_lf_failed++;
1911 /* skip the failure to make the scanning to continue. */
1915 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1916 loff_t offset, __u64 ino, unsigned d_type)
1918 struct osd_ios_filldir_buf *fill_buf = buf;
1919 struct osd_device *dev = fill_buf->oifb_dev;
1920 struct dentry *child;
1924 /* skip any '.' started names */
1928 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1930 RETURN(PTR_ERR(child));
1932 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1934 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1935 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1936 osd_ios_varfid_fill);
1942 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1943 loff_t offset, __u64 ino, unsigned d_type)
1945 struct osd_ios_filldir_buf *fill_buf = buf;
1946 struct osd_device *dev = fill_buf->oifb_dev;
1947 const struct osd_lf_map *map;
1948 struct dentry *child;
1952 /* skip any '.' started names */
1956 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1957 if (map->olm_namelen != namelen)
1960 if (strncmp(map->olm_name, name, namelen) == 0)
1964 if (map->olm_name == NULL)
1967 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1969 RETURN(PTR_ERR(child));
1971 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1972 &map->olm_fid, map->olm_flags);
1978 static int osd_ios_uld_fill(void *buf, const char *name, int namelen,
1979 loff_t offset, __u64 ino, unsigned d_type)
1981 struct osd_ios_filldir_buf *fill_buf = buf;
1982 struct dentry *child;
1987 /* skip any non-DFID format name */
1991 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1993 RETURN(PTR_ERR(child));
1995 /* skip the start '[' */
1996 sscanf(&name[1], SFID, RFID(&tfid));
1997 if (fid_is_sane(&tfid))
1998 rc = osd_ios_scan_one(fill_buf->oifb_info, fill_buf->oifb_dev,
1999 child->d_inode, &tfid, 0);
2007 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
2008 loff_t offset, __u64 ino, unsigned d_type)
2010 struct osd_ios_filldir_buf *fill_buf = buf;
2011 struct osd_device *dev = fill_buf->oifb_dev;
2012 const struct osd_lf_map *map;
2013 struct dentry *child;
2017 /* skip any '.' started names */
2021 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
2022 if (map->olm_namelen != namelen)
2025 if (strncmp(map->olm_name, name, namelen) == 0)
2029 if (map->olm_name == NULL)
2032 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
2034 RETURN(PTR_ERR(child));
2036 if (!(map->olm_flags & OLF_NO_OI))
2037 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
2038 &map->olm_fid, map->olm_flags);
2039 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
2040 rc = osd_ios_new_item(dev, child, map->olm_scandir,
2048 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
2049 struct dentry *dentry, filldir_t filldir)
2051 struct osd_ios_filldir_buf buf = {
2052 #ifdef HAVE_DIR_CONTEXT
2053 .ctx.actor = filldir,
2057 .oifb_dentry = dentry };
2058 struct file *filp = &info->oti_file;
2059 struct inode *inode = dentry->d_inode;
2060 const struct file_operations *fops = inode->i_fop;
2064 LASSERT(filldir != NULL);
2067 filp->f_path.dentry = dentry;
2068 filp->f_mode = FMODE_64BITHASH;
2069 filp->f_mapping = inode->i_mapping;
2071 filp->private_data = NULL;
2072 set_file_inode(filp, inode);
2074 #ifdef HAVE_DIR_CONTEXT
2075 buf.ctx.pos = filp->f_pos;
2076 rc = fops->iterate(filp, &buf.ctx);
2077 filp->f_pos = buf.ctx.pos;
2079 rc = fops->readdir(filp, &buf, filldir);
2081 fops->release(inode, filp);
2087 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
2088 struct dentry *dentry, filldir_t filldir)
2090 struct osd_scrub *scrub = &dev->od_scrub;
2091 struct scrub_file *sf = &scrub->os_file;
2092 struct dentry *child;
2096 /* It is existing MDT0 device. We only allow the case of object without
2097 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
2098 * can generate IGIF mode FID for the object and related OI mapping. If
2099 * it is on other MDTs, then becuase file-level backup/restore, related
2100 * OI mapping may be invalid already, we do not know which is the right
2101 * FID for the object. We only allow IGIF objects to reside on the MDT0.
2103 * XXX: For the case of object on non-MDT0 device with neither LMA nor
2104 * "fid" xattr, then something crashed. We cannot re-generate the
2105 * FID directly, instead, the OI scrub will scan the OI structure
2106 * and try to re-generate the LMA from the OI mapping. But if the
2107 * OI mapping crashed or lost also, then we have to give up under
2108 * double failure cases. */
2109 scrub->os_convert_igif = 1;
2110 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
2111 strlen(dot_lustre_name));
2112 if (IS_ERR(child)) {
2113 rc = PTR_ERR(child);
2114 if (rc == -ENOENT) {
2115 /* It is 1.8 MDT device. */
2116 if (!(sf->sf_flags & SF_UPGRADE)) {
2117 osd_scrub_file_reset(scrub,
2118 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
2120 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2121 rc = osd_scrub_file_store(scrub);
2127 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
2128 * so the client will get IGIF for the ".lustre" object when
2131 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
2132 * it does not know whether there are some old clients cached
2133 * the ".lustre" IGIF during the upgrading. Two choices:
2135 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
2136 * It will allow the old connected clients to access the
2137 * ".lustre" with cached IGIF. But it will cause others
2138 * on the MDT failed to check "fid_is_dot_lustre()".
2140 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
2141 * for ".lustre" in spite of whether there are some clients
2142 * cached the ".lustre" IGIF or not. It enables the check
2143 * "fid_is_dot_lustre()" on the MDT, although it will cause
2144 * that the old connected clients cannot access the ".lustre"
2145 * with the cached IGIF.
2147 * Usually, it is rare case for the old connected clients
2148 * to access the ".lustre" with cached IGIF. So we prefer
2149 * to the solution 2). */
2150 rc = osd_ios_scan_one(info, dev, child->d_inode,
2151 &LU_DOT_LUSTRE_FID, 0);
2153 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
2162 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
2163 struct dentry *dentry, filldir_t filldir)
2165 struct osd_scrub *scrub = &dev->od_scrub;
2166 struct scrub_file *sf = &scrub->os_file;
2167 struct dentry *child;
2171 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
2172 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
2173 rc = osd_scrub_file_store(scrub);
2178 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
2179 if (!IS_ERR(child)) {
2180 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2183 rc = PTR_ERR(child);
2186 if (rc != 0 && rc != -ENOENT)
2189 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
2190 if (!IS_ERR(child)) {
2191 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
2194 rc = PTR_ERR(child);
2203 static int osd_initial_OI_scrub(struct osd_thread_info *info,
2204 struct osd_device *dev)
2206 struct osd_ios_item *item = NULL;
2207 scandir_t scandir = osd_ios_general_scan;
2208 filldir_t filldir = osd_ios_root_fill;
2209 struct dentry *dentry = osd_sb(dev)->s_root;
2210 const struct osd_lf_map *map = osd_lf_maps;
2214 /* Lookup IGIF in OI by force for initial OI scrub. */
2215 dev->od_igif_inoi = 1;
2218 rc = scandir(info, dev, dentry, filldir);
2220 dput(item->oii_dentry);
2227 if (list_empty(&dev->od_ios_list))
2230 item = list_entry(dev->od_ios_list.next,
2231 struct osd_ios_item, oii_list);
2232 list_del_init(&item->oii_list);
2234 LASSERT(item->oii_scandir != NULL);
2235 scandir = item->oii_scandir;
2236 filldir = item->oii_filldir;
2237 dentry = item->oii_dentry;
2240 while (!list_empty(&dev->od_ios_list)) {
2241 item = list_entry(dev->od_ios_list.next,
2242 struct osd_ios_item, oii_list);
2243 list_del_init(&item->oii_list);
2244 dput(item->oii_dentry);
2251 /* There maybe the case that the object has been removed, but its OI
2252 * mapping is still in the OI file, such as the "CATALOGS" after MDT
2253 * file-level backup/restore. So here cleanup the stale OI mappings. */
2254 while (map->olm_name != NULL) {
2255 struct dentry *child;
2257 if (fid_is_zero(&map->olm_fid)) {
2262 child = osd_ios_lookup_one_len(map->olm_name,
2263 osd_sb(dev)->s_root,
2267 else if (PTR_ERR(child) == -ENOENT)
2268 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2269 NULL, DTO_INDEX_DELETE,
2277 char *osd_lf_fid2name(const struct lu_fid *fid)
2279 const struct osd_lf_map *map = osd_lf_maps;
2281 while (map->olm_name != NULL) {
2282 if (!lu_fid_eq(fid, &map->olm_fid)) {
2287 if (map->olm_flags & OLF_SHOW_NAME)
2288 return map->olm_name;
2296 /* OI scrub start/stop */
2298 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2300 struct osd_scrub *scrub = &dev->od_scrub;
2301 struct ptlrpc_thread *thread = &scrub->os_thread;
2302 struct l_wait_info lwi = { 0 };
2303 struct task_struct *task;
2307 /* os_lock: sync status between stop and scrub thread */
2308 spin_lock(&scrub->os_lock);
2311 if (thread_is_running(thread)) {
2312 spin_unlock(&scrub->os_lock);
2313 if (!(scrub->os_file.sf_flags & SF_AUTO) ||
2314 (flags & (SS_AUTO_FULL | SS_AUTO_PARTIAL)))
2317 osd_scrub_join(dev, flags, false);
2318 spin_lock(&scrub->os_lock);
2319 if (!thread_is_running(thread))
2322 spin_unlock(&scrub->os_lock);
2326 if (unlikely(thread_is_stopping(thread))) {
2327 spin_unlock(&scrub->os_lock);
2328 l_wait_event(thread->t_ctl_waitq,
2329 thread_is_stopped(thread),
2331 spin_lock(&scrub->os_lock);
2334 spin_unlock(&scrub->os_lock);
2336 if (scrub->os_file.sf_status == SS_COMPLETED) {
2337 if (!(flags & SS_SET_FAILOUT))
2338 flags |= SS_CLEAR_FAILOUT;
2340 if (!(flags & SS_SET_DRYRUN))
2341 flags |= SS_CLEAR_DRYRUN;
2346 scrub->os_start_flags = flags;
2347 thread_set_flags(thread, 0);
2348 task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2351 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2352 osd_scrub2name(scrub), rc);
2356 l_wait_event(thread->t_ctl_waitq,
2357 thread_is_running(thread) || thread_is_stopped(thread),
2363 int osd_scrub_start(struct osd_device *dev, __u32 flags)
2368 /* od_otable_mutex: prevent curcurrent start/stop */
2369 mutex_lock(&dev->od_otable_mutex);
2370 rc = do_osd_scrub_start(dev, flags);
2371 mutex_unlock(&dev->od_otable_mutex);
2373 RETURN(rc == -EALREADY ? 0 : rc);
2376 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2378 struct ptlrpc_thread *thread = &scrub->os_thread;
2379 struct l_wait_info lwi = { 0 };
2381 /* os_lock: sync status between stop and scrub thread */
2382 spin_lock(&scrub->os_lock);
2383 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2384 thread_set_flags(thread, SVC_STOPPING);
2385 spin_unlock(&scrub->os_lock);
2386 wake_up_all(&thread->t_ctl_waitq);
2387 l_wait_event(thread->t_ctl_waitq,
2388 thread_is_stopped(thread),
2390 /* Do not skip the last lock/unlock, which can guarantee that
2391 * the caller cannot return until the OI scrub thread exit. */
2392 spin_lock(&scrub->os_lock);
2394 spin_unlock(&scrub->os_lock);
2397 static void osd_scrub_stop(struct osd_device *dev)
2399 /* od_otable_mutex: prevent curcurrent start/stop */
2400 mutex_lock(&dev->od_otable_mutex);
2401 dev->od_scrub.os_paused = 1;
2402 do_osd_scrub_stop(&dev->od_scrub);
2403 mutex_unlock(&dev->od_otable_mutex);
2406 /* OI scrub setup/cleanup */
2408 static const char osd_scrub_name[] = "OI_scrub";
2410 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2412 struct osd_thread_info *info = osd_oti_get(env);
2413 struct osd_scrub *scrub = &dev->od_scrub;
2414 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
2415 struct scrub_file *sf = &scrub->os_file;
2416 struct super_block *sb = osd_sb(dev);
2417 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2418 struct lvfs_run_ctxt saved;
2420 struct inode *inode;
2421 struct lu_fid *fid = &info->oti_fid;
2426 memset(scrub, 0, sizeof(*scrub));
2427 OBD_SET_CTXT_MAGIC(ctxt);
2428 ctxt->pwdmnt = dev->od_mnt;
2429 ctxt->pwd = dev->od_mnt->mnt_root;
2430 ctxt->fs = get_ds();
2432 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2433 init_rwsem(&scrub->os_rwsem);
2434 spin_lock_init(&scrub->os_lock);
2435 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2437 push_ctxt(&saved, ctxt);
2438 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2440 pop_ctxt(&saved, ctxt);
2441 RETURN(PTR_ERR(filp));
2444 inode = filp->f_path.dentry->d_inode;
2445 /* 'What the @fid is' is not imporatant, because the object
2446 * has no OI mapping, and only is visible inside the OSD.*/
2447 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2448 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2450 filp_close(filp, NULL);
2451 pop_ctxt(&saved, ctxt);
2455 scrub->os_inode = igrab(inode);
2456 filp_close(filp, NULL);
2457 pop_ctxt(&saved, ctxt);
2459 rc = osd_scrub_file_load(scrub);
2460 if (rc == -ENOENT) {
2461 osd_scrub_file_init(scrub, es->s_uuid);
2462 /* If the "/O" dir does not exist when mount (indicated by
2463 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2464 * then it is quite probably that the device is a new one,
2465 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2467 * For the rare case that "/O" and "OI_scrub" both lost on
2468 * an old device, it can be found and cleared later.
2470 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2471 * need to check "filter_fid_old" and to convert it to
2472 * "filter_fid" for each object, and all the IGIF should
2473 * have their FID mapping in OI files already. */
2474 if (dev->od_maybe_new)
2475 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2477 } else if (rc != 0) {
2478 GOTO(cleanup_inode, rc);
2480 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2481 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2483 } else if (sf->sf_status == SS_SCANNING) {
2484 sf->sf_status = SS_CRASHED;
2489 if (sf->sf_pos_last_checkpoint != 0)
2490 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2492 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2495 rc = osd_scrub_file_store(scrub);
2497 GOTO(cleanup_inode, rc);
2500 /* Initialize OI files. */
2501 rc = osd_oi_init(info, dev);
2503 GOTO(cleanup_inode, rc);
2505 rc = osd_initial_OI_scrub(info, dev);
2507 GOTO(cleanup_oi, rc);
2509 if (sf->sf_flags & SF_UPGRADE ||
2510 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2511 sf->sf_success_count > 0)) {
2512 dev->od_igif_inoi = 0;
2513 dev->od_check_ff = dev->od_is_ost;
2515 dev->od_igif_inoi = 1;
2516 dev->od_check_ff = 0;
2519 if (sf->sf_flags & SF_INCONSISTENT)
2520 /* The 'od_igif_inoi' will be set under the
2522 * 1) new created system, or
2523 * 2) restored from file-level backup, or
2524 * 3) the upgrading completed.
2526 * The 'od_igif_inoi' may be cleared by OI scrub
2527 * later if found that the system is upgrading. */
2528 dev->od_igif_inoi = 1;
2530 if (!dev->od_noscrub &&
2531 ((sf->sf_status == SS_PAUSED) ||
2532 (sf->sf_status == SS_CRASHED &&
2533 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2534 SF_UPGRADE | SF_AUTO)) ||
2535 (sf->sf_status == SS_INIT &&
2536 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2538 rc = osd_scrub_start(dev, SS_AUTO_FULL);
2541 GOTO(cleanup_oi, rc);
2543 /* it is possible that dcache entries may keep objects after they are
2544 * deleted by OSD. While it looks safe this can cause object data to
2545 * stay until umount causing failures in tests calculating free space,
2546 * e.g. replay-ost-single. Since those dcache entries are not used
2547 * anymore let's just free them after use here */
2548 shrink_dcache_sb(sb);
2552 osd_oi_fini(info, dev);
2554 iput(scrub->os_inode);
2555 scrub->os_inode = NULL;
2560 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2562 struct osd_scrub *scrub = &dev->od_scrub;
2564 LASSERT(dev->od_otable_it == NULL);
2566 if (scrub->os_inode != NULL) {
2567 osd_scrub_stop(dev);
2568 iput(scrub->os_inode);
2569 scrub->os_inode = NULL;
2571 if (dev->od_oi_table != NULL)
2572 osd_oi_fini(osd_oti_get(env), dev);
2575 /* object table based iteration APIs */
2577 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2578 struct dt_object *dt, __u32 attr)
2580 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2581 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2582 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2583 struct osd_scrub *scrub = &dev->od_scrub;
2584 struct osd_otable_it *it;
2589 /* od_otable_mutex: prevent curcurrent init/fini */
2590 mutex_lock(&dev->od_otable_mutex);
2591 if (dev->od_otable_it != NULL)
2592 GOTO(out, it = ERR_PTR(-EALREADY));
2596 GOTO(out, it = ERR_PTR(-ENOMEM));
2598 dev->od_otable_it = it;
2600 it->ooi_cache.ooc_consumer_idx = -1;
2601 if (flags & DOIF_OUTUSED)
2602 it->ooi_used_outside = 1;
2604 if (flags & DOIF_RESET)
2607 if (valid & DOIV_ERROR_HANDLE) {
2608 if (flags & DOIF_FAILOUT)
2609 start |= SS_SET_FAILOUT;
2611 start |= SS_CLEAR_FAILOUT;
2614 if (valid & DOIV_DRYRUN) {
2615 if (flags & DOIF_DRYRUN)
2616 start |= SS_SET_DRYRUN;
2618 start |= SS_CLEAR_DRYRUN;
2621 rc = do_osd_scrub_start(dev, start & ~SS_AUTO_PARTIAL);
2622 if (rc < 0 && rc != -EALREADY) {
2623 dev->od_otable_it = NULL;
2625 GOTO(out, it = ERR_PTR(rc));
2628 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2633 mutex_unlock(&dev->od_otable_mutex);
2634 return (struct dt_it *)it;
2637 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2639 struct osd_otable_it *it = (struct osd_otable_it *)di;
2640 struct osd_device *dev = it->ooi_dev;
2642 /* od_otable_mutex: prevent curcurrent init/fini */
2643 mutex_lock(&dev->od_otable_mutex);
2644 do_osd_scrub_stop(&dev->od_scrub);
2645 LASSERT(dev->od_otable_it == it);
2647 dev->od_otable_it = NULL;
2648 mutex_unlock(&dev->od_otable_mutex);
2652 static int osd_otable_it_get(const struct lu_env *env,
2653 struct dt_it *di, const struct dt_key *key)
2658 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2663 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2665 spin_lock(&scrub->os_lock);
2666 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2667 scrub->os_waiting ||
2668 !thread_is_running(&scrub->os_thread))
2669 it->ooi_waiting = 0;
2671 it->ooi_waiting = 1;
2672 spin_unlock(&scrub->os_lock);
2674 return !it->ooi_waiting;
2677 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2679 struct osd_otable_it *it = (struct osd_otable_it *)di;
2680 struct osd_device *dev = it->ooi_dev;
2681 struct osd_scrub *scrub = &dev->od_scrub;
2682 struct osd_otable_cache *ooc = &it->ooi_cache;
2683 struct ptlrpc_thread *thread = &scrub->os_thread;
2684 struct l_wait_info lwi = { 0 };
2688 LASSERT(it->ooi_user_ready);
2691 if (!thread_is_running(thread) && !it->ooi_used_outside)
2694 if (ooc->ooc_cached_items > 0) {
2695 ooc->ooc_cached_items--;
2696 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2697 ~OSD_OTABLE_IT_CACHE_MASK;
2701 if (it->ooi_all_cached) {
2702 l_wait_event(thread->t_ctl_waitq,
2703 !thread_is_running(thread),
2708 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2709 spin_lock(&scrub->os_lock);
2710 scrub->os_waiting = 0;
2711 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2712 spin_unlock(&scrub->os_lock);
2715 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2716 l_wait_event(thread->t_ctl_waitq,
2717 osd_otable_it_wakeup(scrub, it),
2720 if (!thread_is_running(thread) && !it->ooi_used_outside)
2723 rc = osd_otable_it_preload(env, it);
2730 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2731 const struct dt_it *di)
2736 static int osd_otable_it_key_size(const struct lu_env *env,
2737 const struct dt_it *di)
2739 return sizeof(__u64);
2742 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2743 struct dt_rec *rec, __u32 attr)
2745 struct osd_otable_it *it = (struct osd_otable_it *)di;
2746 struct osd_otable_cache *ooc = &it->ooi_cache;
2748 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2750 /* Filter out Invald FID already. */
2751 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2752 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2753 PFID((struct lu_fid *)rec),
2754 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2759 static __u64 osd_otable_it_store(const struct lu_env *env,
2760 const struct dt_it *di)
2762 struct osd_otable_it *it = (struct osd_otable_it *)di;
2763 struct osd_otable_cache *ooc = &it->ooi_cache;
2766 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2767 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2769 hash = ooc->ooc_pos_preload;
2774 * Set the OSD layer iteration start position as the specified hash.
2776 static int osd_otable_it_load(const struct lu_env *env,
2777 const struct dt_it *di, __u64 hash)
2779 struct osd_otable_it *it = (struct osd_otable_it *)di;
2780 struct osd_device *dev = it->ooi_dev;
2781 struct osd_otable_cache *ooc = &it->ooi_cache;
2782 struct osd_scrub *scrub = &dev->od_scrub;
2786 /* Forbid to set iteration position after iteration started. */
2787 if (it->ooi_user_ready)
2790 if (hash > OSD_OTABLE_MAX_HASH)
2791 hash = OSD_OTABLE_MAX_HASH;
2793 ooc->ooc_pos_preload = hash;
2794 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2795 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2797 it->ooi_user_ready = 1;
2798 if (!scrub->os_full_speed)
2799 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2801 /* Unplug OSD layer iteration by the first next() call. */
2802 rc = osd_otable_it_next(env, (struct dt_it *)it);
2807 static int osd_otable_it_key_rec(const struct lu_env *env,
2808 const struct dt_it *di, void *key_rec)
2813 const struct dt_index_operations osd_otable_ops = {
2815 .init = osd_otable_it_init,
2816 .fini = osd_otable_it_fini,
2817 .get = osd_otable_it_get,
2818 .put = osd_otable_it_put,
2819 .next = osd_otable_it_next,
2820 .key = osd_otable_it_key,
2821 .key_size = osd_otable_it_key_size,
2822 .rec = osd_otable_it_rec,
2823 .store = osd_otable_it_store,
2824 .load = osd_otable_it_load,
2825 .key_rec = osd_otable_it_key_rec,
2829 /* high priority inconsistent items list APIs */
2831 #define SCRUB_BAD_OIMAP_DECAY_INTERVAL 60
2833 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2836 struct osd_inconsistent_item *oii;
2837 struct osd_scrub *scrub = &dev->od_scrub;
2838 struct ptlrpc_thread *thread = &scrub->os_thread;
2843 if (unlikely(oii == NULL))
2846 INIT_LIST_HEAD(&oii->oii_list);
2847 oii->oii_cache = *oic;
2848 oii->oii_insert = insert;
2850 if (scrub->os_partial_scan) {
2851 __u64 now = cfs_time_current_sec();
2853 /* If there haven't been errors in a long time,
2854 * decay old count until either the errors are
2855 * gone or we reach the current interval. */
2856 while (unlikely(scrub->os_bad_oimap_count > 0 &&
2857 scrub->os_bad_oimap_time +
2858 SCRUB_BAD_OIMAP_DECAY_INTERVAL < now)) {
2859 scrub->os_bad_oimap_count >>= 1;
2860 scrub->os_bad_oimap_time +=
2861 SCRUB_BAD_OIMAP_DECAY_INTERVAL;
2864 scrub->os_bad_oimap_time = now;
2865 if (++scrub->os_bad_oimap_count >
2866 dev->od_full_scrub_threshold_rate)
2867 scrub->os_full_scrub = 1;
2870 spin_lock(&scrub->os_lock);
2871 if (unlikely(!thread_is_running(thread))) {
2872 spin_unlock(&scrub->os_lock);
2877 if (list_empty(&scrub->os_inconsistent_items))
2879 list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2880 spin_unlock(&scrub->os_lock);
2883 wake_up_all(&thread->t_ctl_waitq);
2888 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2889 struct osd_inode_id *id)
2891 struct osd_scrub *scrub = &dev->od_scrub;
2892 struct osd_inconsistent_item *oii;
2895 spin_lock(&scrub->os_lock);
2896 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2897 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2898 *id = oii->oii_cache.oic_lid;
2899 spin_unlock(&scrub->os_lock);
2903 spin_unlock(&scrub->os_lock);
2910 static const char *scrub_status_names[] = {
2921 static const char *scrub_flags_names[] = {
2929 static const char *scrub_param_names[] = {
2935 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2942 rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2946 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2949 rc = seq_printf(m, "%s%c", names[i],
2950 bits != 0 ? ',' : '\n');
2958 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2963 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2964 cfs_time_current_sec() - time);
2966 rc = seq_printf(m, "%s: N/A\n", prefix);
2970 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2975 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2977 rc = seq_printf(m, "%s: N/A\n", prefix);
2981 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2983 struct osd_scrub *scrub = &dev->od_scrub;
2984 struct scrub_file *sf = &scrub->os_file;
2989 down_read(&scrub->os_rwsem);
2990 rc = seq_printf(m, "name: OI_scrub\n"
2994 sf->sf_magic, (int)sf->sf_oi_count,
2995 scrub_status_names[sf->sf_status]);
2999 rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
3004 rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
3009 rc = scrub_time_dump(m, sf->sf_time_last_complete,
3010 "time_since_last_completed");
3014 rc = scrub_time_dump(m, sf->sf_time_latest_start,
3015 "time_since_latest_start");
3019 rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
3020 "time_since_last_checkpoint");
3024 rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
3025 "latest_start_position");
3029 rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
3030 "last_checkpoint_position");
3034 rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
3035 "first_failure_position");
3039 checked = sf->sf_items_checked + scrub->os_new_checked;
3040 rc = seq_printf(m, "checked: "LPU64"\n"
3041 "updated: "LPU64"\n"
3043 "prior_updated: "LPU64"\n"
3044 "noscrub: "LPU64"\n"
3046 "success_count: %u\n",
3047 checked, sf->sf_items_updated, sf->sf_items_failed,
3048 sf->sf_items_updated_prior, sf->sf_items_noscrub,
3049 sf->sf_items_igif, sf->sf_success_count);
3054 if (thread_is_running(&scrub->os_thread)) {
3055 cfs_duration_t duration = cfs_time_current() -
3056 scrub->os_time_last_checkpoint;
3057 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
3059 __u32 rtime = sf->sf_run_time +
3060 cfs_duration_sec(duration + HALF_SEC);
3063 do_div(new_checked, duration);
3065 do_div(speed, rtime);
3066 rc = seq_printf(m, "run_time: %u seconds\n"
3067 "average_speed: "LPU64" objects/sec\n"
3068 "real-time_speed: "LPU64" objects/sec\n"
3069 "current_position: %u\n"
3070 "lf_scanned: "LPU64"\n"
3071 "lf_repaired: "LPU64"\n"
3072 "lf_failed: "LPU64"\n",
3073 rtime, speed, new_checked, scrub->os_pos_current,
3074 scrub->os_lf_scanned, scrub->os_lf_repaired,
3075 scrub->os_lf_failed);
3077 if (sf->sf_run_time != 0)
3078 do_div(speed, sf->sf_run_time);
3079 rc = seq_printf(m, "run_time: %u seconds\n"
3080 "average_speed: "LPU64" objects/sec\n"
3081 "real-time_speed: N/A\n"
3082 "current_position: N/A\n"
3083 "lf_scanned: "LPU64"\n"
3084 "lf_repaired: "LPU64"\n"
3085 "lf_failed: "LPU64"\n",
3086 sf->sf_run_time, speed, scrub->os_lf_scanned,
3087 scrub->os_lf_repaired, scrub->os_lf_failed);
3091 up_read(&scrub->os_rwsem);
3092 return (rc < 0 ? -ENOSPC : 0);