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, 2013, 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 <lustre/lustre_idl.h>
41 #include <lustre_disk.h>
42 #include <dt_object.h>
43 #include <linux/xattr.h>
45 #include "osd_internal.h"
47 #include "osd_scrub.h"
49 #define HALF_SEC msecs_to_jiffies(MSEC_PER_SEC >> 1)
51 #define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL
53 #define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */
54 #define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */
55 #define SCRUB_NEXT_EXIT 3 /* exit all the loops */
56 #define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */
57 #define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */
58 #define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */
59 #define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */
60 #define SCRUB_NEXT_NOLMA 8 /* the inode has no FID-in-LMA */
61 #define SCRUB_NEXT_OSTOBJ 9 /* for OST-object */
62 #define SCRUB_NEXT_OSTOBJ_OLD 10 /* old OST-object, no LMA or no FID-on-OST
67 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
69 return container_of0(scrub, struct osd_device, od_scrub);
72 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
74 return osd_sb(osd_scrub2dev(scrub));
77 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
78 struct osd_otable_cache *ooc)
80 return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
83 static inline const char *osd_scrub2name(struct osd_scrub *scrub)
85 return LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
89 * update/insert/delete the specified OI mapping (@fid @id) according to the ops
91 * \retval 1, changed nothing
92 * \retval 0, changed successfully
93 * \retval -ve, on error
95 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
96 struct osd_device *dev,
97 const struct lu_fid *fid,
98 const struct osd_inode_id *id,
100 enum oi_check_flags flags)
106 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN && !force)
109 /* DTO_INDEX_INSERT is enough for other two ops:
110 * delete/update, but save stack. */
111 th = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
112 osd_dto_credits_noquota[DTO_INDEX_INSERT]);
115 CDEBUG(D_LFSCK, "%s: fail to start trans for scrub op %d "
116 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
117 PFID(fid), id->oii_ino, id->oii_gen, rc);
122 case DTO_INDEX_UPDATE:
123 rc = osd_oi_update(info, dev, fid, id, th, flags);
124 if (unlikely(rc == -ENOENT)) {
125 /* Some unlink thread may removed the OI mapping. */
129 case DTO_INDEX_INSERT:
130 rc = osd_oi_insert(info, dev, fid, id, th, flags);
131 if (unlikely(rc == -EEXIST)) {
133 /* XXX: There are trouble things when adding OI
134 * mapping for IGIF object, which may cause
135 * multiple objects to be mapped to the same
136 * IGIF formatted FID. Consider the following
139 * 1) The MDT is upgrading from 1.8 device.
140 * The OI scrub generates IGIF FID1 for the
141 * OBJ1 and adds the OI mapping.
143 * 2) For some reason, the OI scrub does not
144 * process all the IGIF objects completely.
146 * 3) The MDT is backuped and restored against
149 * 4) When the MDT mounts up, the OI scrub will
150 * try to rebuild the OI files. For some IGIF
151 * object, OBJ2, which was not processed by the
152 * OI scrub before the backup/restore, and the
153 * new generated IGIF formatted FID may be just
154 * the FID1, the same as OBJ1.
156 * Under such case, the OI scrub cannot know how
157 * to generate new FID for the OBJ2.
159 * Currently, we do nothing for that. One possible
160 * solution is to generate new normal FID for the
163 * Anyway, it is rare, only exists in theory. */
166 case DTO_INDEX_DELETE:
167 rc = osd_oi_delete(info, dev, fid, th, flags);
169 /* It is normal that the unlink thread has removed the
170 * OI mapping already. */
175 LASSERTF(0, "Unexpected ops %d\n", ops);
179 ldiskfs_journal_stop(th);
181 CDEBUG(D_LFSCK, "%s: fail to refresh OI map for scrub op %d "
182 DFID" => %u/%u: rc = %d\n", osd_name(dev), ops,
183 PFID(fid), id->oii_ino, id->oii_gen, rc);
188 /* OI_scrub file ops */
190 static void osd_scrub_file_to_cpu(struct scrub_file *des,
191 struct scrub_file *src)
193 memcpy(des->sf_uuid, src->sf_uuid, 16);
194 des->sf_flags = le64_to_cpu(src->sf_flags);
195 des->sf_magic = le32_to_cpu(src->sf_magic);
196 des->sf_status = le16_to_cpu(src->sf_status);
197 des->sf_param = le16_to_cpu(src->sf_param);
198 des->sf_time_last_complete =
199 le64_to_cpu(src->sf_time_last_complete);
200 des->sf_time_latest_start =
201 le64_to_cpu(src->sf_time_latest_start);
202 des->sf_time_last_checkpoint =
203 le64_to_cpu(src->sf_time_last_checkpoint);
204 des->sf_pos_latest_start =
205 le64_to_cpu(src->sf_pos_latest_start);
206 des->sf_pos_last_checkpoint =
207 le64_to_cpu(src->sf_pos_last_checkpoint);
208 des->sf_pos_first_inconsistent =
209 le64_to_cpu(src->sf_pos_first_inconsistent);
210 des->sf_items_checked =
211 le64_to_cpu(src->sf_items_checked);
212 des->sf_items_updated =
213 le64_to_cpu(src->sf_items_updated);
214 des->sf_items_failed =
215 le64_to_cpu(src->sf_items_failed);
216 des->sf_items_updated_prior =
217 le64_to_cpu(src->sf_items_updated_prior);
218 des->sf_run_time = le32_to_cpu(src->sf_run_time);
219 des->sf_success_count = le32_to_cpu(src->sf_success_count);
220 des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
221 des->sf_internal_flags = le16_to_cpu(src->sf_internal_flags);
222 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
225 static void osd_scrub_file_to_le(struct scrub_file *des,
226 struct scrub_file *src)
228 memcpy(des->sf_uuid, src->sf_uuid, 16);
229 des->sf_flags = cpu_to_le64(src->sf_flags);
230 des->sf_magic = cpu_to_le32(src->sf_magic);
231 des->sf_status = cpu_to_le16(src->sf_status);
232 des->sf_param = cpu_to_le16(src->sf_param);
233 des->sf_time_last_complete =
234 cpu_to_le64(src->sf_time_last_complete);
235 des->sf_time_latest_start =
236 cpu_to_le64(src->sf_time_latest_start);
237 des->sf_time_last_checkpoint =
238 cpu_to_le64(src->sf_time_last_checkpoint);
239 des->sf_pos_latest_start =
240 cpu_to_le64(src->sf_pos_latest_start);
241 des->sf_pos_last_checkpoint =
242 cpu_to_le64(src->sf_pos_last_checkpoint);
243 des->sf_pos_first_inconsistent =
244 cpu_to_le64(src->sf_pos_first_inconsistent);
245 des->sf_items_checked =
246 cpu_to_le64(src->sf_items_checked);
247 des->sf_items_updated =
248 cpu_to_le64(src->sf_items_updated);
249 des->sf_items_failed =
250 cpu_to_le64(src->sf_items_failed);
251 des->sf_items_updated_prior =
252 cpu_to_le64(src->sf_items_updated_prior);
253 des->sf_run_time = cpu_to_le32(src->sf_run_time);
254 des->sf_success_count = cpu_to_le32(src->sf_success_count);
255 des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
256 des->sf_internal_flags = cpu_to_le16(src->sf_internal_flags);
257 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
260 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
262 struct scrub_file *sf = &scrub->os_file;
264 memset(sf, 0, sizeof(*sf));
265 memcpy(sf->sf_uuid, uuid, 16);
266 sf->sf_magic = SCRUB_MAGIC_V1;
267 sf->sf_status = SS_INIT;
270 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
272 struct scrub_file *sf = &scrub->os_file;
274 CDEBUG(D_LFSCK, "%.16s: reset OI scrub file, flags = "LPX64"\n",
275 osd_scrub2name(scrub), flags);
276 memcpy(sf->sf_uuid, uuid, 16);
277 sf->sf_status = SS_INIT;
278 sf->sf_flags |= flags;
280 sf->sf_time_latest_start = 0;
281 sf->sf_time_last_checkpoint = 0;
282 sf->sf_pos_latest_start = 0;
283 sf->sf_pos_last_checkpoint = 0;
284 sf->sf_pos_first_inconsistent = 0;
285 sf->sf_items_checked = 0;
286 sf->sf_items_updated = 0;
287 sf->sf_items_failed = 0;
288 sf->sf_items_updated_prior = 0;
289 sf->sf_items_noscrub = 0;
290 sf->sf_items_igif = 0;
293 static int osd_scrub_file_load(struct osd_scrub *scrub)
296 int len = sizeof(scrub->os_file_disk);
299 rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
301 struct scrub_file *sf = &scrub->os_file;
303 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
304 if (sf->sf_magic != SCRUB_MAGIC_V1) {
305 CDEBUG(D_LFSCK, "%.16s: invalid scrub magic "
306 "0x%x != 0x%x\n", osd_scrub2name(scrub),
307 sf->sf_magic, SCRUB_MAGIC_V1);
308 /* Process it as new scrub file. */
313 } else if (rc != 0) {
314 CDEBUG(D_LFSCK, "%.16s: fail to load scrub file, "
315 "expected = %d: rc = %d\n",
316 osd_scrub2name(scrub), len, rc);
320 /* return -ENOENT for empty scrub file case. */
327 int osd_scrub_file_store(struct osd_scrub *scrub)
329 struct osd_device *dev;
332 int len = sizeof(scrub->os_file_disk);
336 dev = container_of0(scrub, struct osd_device, od_scrub);
337 credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
338 osd_dto_credits_noquota[DTO_WRITE_BLOCK];
339 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC, credits);
342 CDEBUG(D_LFSCK, "%.16s: fail to start trans for scrub store: "
343 "rc = %d\n", osd_scrub2name(scrub), rc);
347 osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
348 rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
350 ldiskfs_journal_stop(jh);
352 CDEBUG(D_LFSCK, "%.16s: fail to store scrub file, "
353 "expected = %d: rc = %d\n",
354 osd_scrub2name(scrub), len, rc);
356 scrub->os_time_last_checkpoint = cfs_time_current();
357 scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
358 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
363 osd_scrub_convert_ff(struct osd_thread_info *info, struct osd_device *dev,
364 struct inode *inode, const struct lu_fid *fid)
366 struct filter_fid_old *ff = &info->oti_ff;
367 struct dentry *dentry = &info->oti_obj_dentry;
371 bool removed = false;
375 if (dev->od_scrub.os_file.sf_param & SP_DRYRUN)
378 /* We want the LMA to fit into the 256-byte OST inode, so operate
380 * 1) read old XATTR_NAME_FID and save the parent FID;
381 * 2) delete the old XATTR_NAME_FID;
382 * 3) make new LMA and add it;
383 * 4) generate new XATTR_NAME_FID with the saved parent FID and add it.
385 * Making the LMA to fit into the 256-byte OST inode can save time for
386 * normal osd_check_lma() and for other OI scrub scanning in future.
387 * So it is worth to make some slow conversion here. */
388 jh = osd_journal_start_sb(osd_sb(dev), LDISKFS_HT_MISC,
389 osd_dto_credits_noquota[DTO_XATTR_SET] * 3);
392 CDEBUG(D_LFSCK, "%s: fail to start trans for convert ff "
393 DFID": rc = %d\n", osd_name(dev), PFID(fid), rc);
397 /* 1) read old XATTR_NAME_FID and save the parent FID */
398 rc = __osd_xattr_get(inode, dentry, XATTR_NAME_FID, ff, sizeof(*ff));
399 if (rc == sizeof(*ff)) {
400 /* 2) delete the old XATTR_NAME_FID */
401 ll_vfs_dq_init(inode);
402 rc = inode->i_op->removexattr(dentry, XATTR_NAME_FID);
407 } else if (unlikely(rc == -ENODATA)) {
409 } else if (rc != sizeof(struct filter_fid)) {
410 GOTO(stop, rc = -EINVAL);
413 /* 3) make new LMA and add it */
414 rc = osd_ea_fid_set(info, inode, fid, LMAC_FID_ON_OST, 0);
415 if (rc == 0 && reset)
416 size = sizeof(struct filter_fid);
417 else if (rc != 0 && removed)
418 /* If failed, we should try to add the old back. */
419 size = sizeof(struct filter_fid_old);
421 /* 4) generate new XATTR_NAME_FID with the saved parent FID and add it*/
425 rc1 = __osd_xattr_set(info, inode, XATTR_NAME_FID, ff, size,
427 if (rc1 != 0 && rc == 0)
434 ldiskfs_journal_stop(jh);
436 CDEBUG(D_LFSCK, "%s: fail to convert ff "DFID": rc = %d\n",
437 osd_name(dev), PFID(fid), rc);
442 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
443 struct osd_idmap_cache *oic, int val)
445 struct osd_scrub *scrub = &dev->od_scrub;
446 struct scrub_file *sf = &scrub->os_file;
447 struct lu_fid *fid = &oic->oic_fid;
448 struct osd_inode_id *lid = &oic->oic_lid;
449 struct osd_inode_id *lid2 = &info->oti_id;
450 struct osd_inconsistent_item *oii = NULL;
451 struct inode *inode = NULL;
452 int ops = DTO_INDEX_UPDATE;
455 bool converted = false;
458 down_write(&scrub->os_rwsem);
459 scrub->os_new_checked++;
463 if (scrub->os_in_prior)
464 oii = list_entry(oic, struct osd_inconsistent_item,
467 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
470 if (fid_is_igif(fid))
473 if (val == SCRUB_NEXT_OSTOBJ_OLD) {
474 inode = osd_iget(info, dev, lid);
477 /* Someone removed the inode. */
478 if (rc == -ENOENT || rc == -ESTALE)
483 sf->sf_flags |= SF_UPGRADE;
484 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
485 dev->od_check_ff = 1;
486 rc = osd_scrub_convert_ff(info, dev, inode, fid);
493 if ((val == SCRUB_NEXT_NOLMA) &&
494 (!scrub->os_convert_igif || OBD_FAIL_CHECK(OBD_FAIL_FID_NOLMA)))
497 if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
500 rc = osd_oi_lookup(info, dev, fid, lid2,
501 (val == SCRUB_NEXT_OSTOBJ ||
502 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
504 if (rc != -ENOENT && rc != -ESTALE)
509 inode = osd_iget(info, dev, lid);
512 /* Someone removed the inode. */
513 if (rc == -ENOENT || rc == -ESTALE)
519 scrub->os_full_speed = 1;
520 ops = DTO_INDEX_INSERT;
521 idx = osd_oi_fid2idx(dev, fid);
523 case SCRUB_NEXT_NOLMA:
524 sf->sf_flags |= SF_UPGRADE;
525 if (!(sf->sf_param & SP_DRYRUN)) {
526 rc = osd_ea_fid_set(info, inode, fid, 0, 0);
531 if (!(sf->sf_flags & SF_INCONSISTENT))
532 dev->od_igif_inoi = 0;
534 case SCRUB_NEXT_OSTOBJ:
535 sf->sf_flags |= SF_INCONSISTENT;
536 case SCRUB_NEXT_OSTOBJ_OLD:
539 sf->sf_flags |= SF_RECREATED;
540 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
541 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
544 } else if (osd_id_eq(lid, lid2)) {
546 sf->sf_items_updated++;
550 scrub->os_full_speed = 1;
551 sf->sf_flags |= SF_INCONSISTENT;
553 /* XXX: If the device is restored from file-level backup, then
554 * some IGIFs may have been already in OI files, and some
555 * may be not yet. Means upgrading from 1.8 may be partly
556 * processed, but some clients may hold some immobilized
557 * IGIFs, and use them to access related objects. Under
558 * such case, OSD does not know whether an given IGIF has
559 * been processed or to be processed, and it also cannot
560 * generate local ino#/gen# directly from the immobilized
561 * IGIF because of the backup/restore. Then force OSD to
562 * lookup the given IGIF in OI files, and if no entry,
563 * then ask the client to retry after upgrading completed.
564 * No better choice. */
565 dev->od_igif_inoi = 1;
568 rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops, false,
569 (val == SCRUB_NEXT_OSTOBJ ||
570 val == SCRUB_NEXT_OSTOBJ_OLD) ? OI_KNOWN_ON_OST : 0);
572 if (scrub->os_in_prior)
573 sf->sf_items_updated_prior++;
575 sf->sf_items_updated++;
577 /* The target has been changed, need to be re-loaded. */
578 lu_object_purge(info->oti_env, osd2lu_dev(dev), fid);
585 sf->sf_items_failed++;
586 if (sf->sf_pos_first_inconsistent == 0 ||
587 sf->sf_pos_first_inconsistent > lid->oii_ino)
588 sf->sf_pos_first_inconsistent = lid->oii_ino;
593 /* There may be conflict unlink during the OI scrub,
594 * if happend, then remove the new added OI mapping. */
595 if (ops == DTO_INDEX_INSERT && inode != NULL && !IS_ERR(inode) &&
596 unlikely(inode->i_nlink == 0))
597 osd_scrub_refresh_mapping(info, dev, fid, lid,
598 DTO_INDEX_DELETE, false,
599 (val == SCRUB_NEXT_OSTOBJ ||
600 val == SCRUB_NEXT_OSTOBJ_OLD) ?
601 OI_KNOWN_ON_OST : 0);
602 up_write(&scrub->os_rwsem);
604 if (inode != NULL && !IS_ERR(inode))
608 LASSERT(!list_empty(&oii->oii_list));
610 spin_lock(&scrub->os_lock);
611 list_del_init(&oii->oii_list);
612 spin_unlock(&scrub->os_lock);
615 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
620 static int osd_scrub_prep(struct osd_device *dev)
622 struct osd_scrub *scrub = &dev->od_scrub;
623 struct ptlrpc_thread *thread = &scrub->os_thread;
624 struct scrub_file *sf = &scrub->os_file;
625 __u32 flags = scrub->os_start_flags;
627 bool drop_dryrun = false;
630 down_write(&scrub->os_rwsem);
631 if (flags & SS_SET_FAILOUT)
632 sf->sf_param |= SP_FAILOUT;
634 if (flags & SS_CLEAR_FAILOUT)
635 sf->sf_param &= ~SP_FAILOUT;
637 if (flags & SS_SET_DRYRUN)
638 sf->sf_param |= SP_DRYRUN;
640 if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
641 sf->sf_param &= ~SP_DRYRUN;
645 if (flags & SS_RESET)
646 osd_scrub_file_reset(scrub,
647 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
649 if (flags & SS_AUTO) {
650 scrub->os_full_speed = 1;
651 sf->sf_flags |= SF_AUTO;
652 /* For the case of OI scrub auto triggered, NOT dryrun. */
653 sf->sf_param &= ~SP_FAILOUT;
655 scrub->os_full_speed = 0;
658 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
659 scrub->os_full_speed = 1;
661 scrub->os_in_prior = 0;
662 spin_lock(&scrub->os_lock);
663 scrub->os_waiting = 0;
664 scrub->os_paused = 0;
665 spin_unlock(&scrub->os_lock);
666 scrub->os_new_checked = 0;
667 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
668 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
669 else if (sf->sf_pos_last_checkpoint != 0)
670 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
672 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
674 scrub->os_pos_current = sf->sf_pos_latest_start;
675 sf->sf_status = SS_SCANNING;
676 sf->sf_time_latest_start = cfs_time_current_sec();
677 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
678 rc = osd_scrub_file_store(scrub);
680 spin_lock(&scrub->os_lock);
681 thread_set_flags(thread, SVC_RUNNING);
682 spin_unlock(&scrub->os_lock);
683 wake_up_all(&thread->t_ctl_waitq);
685 up_write(&scrub->os_rwsem);
690 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
692 struct scrub_file *sf = &scrub->os_file;
695 if (likely(cfs_time_before(cfs_time_current(),
696 scrub->os_time_next_checkpoint) ||
697 scrub->os_new_checked == 0))
700 down_write(&scrub->os_rwsem);
701 sf->sf_items_checked += scrub->os_new_checked;
702 scrub->os_new_checked = 0;
703 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
704 sf->sf_time_last_checkpoint = cfs_time_current_sec();
705 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
706 scrub->os_time_last_checkpoint);
707 rc = osd_scrub_file_store(scrub);
708 up_write(&scrub->os_rwsem);
713 static void osd_scrub_post(struct osd_scrub *scrub, int result)
715 struct scrub_file *sf = &scrub->os_file;
718 down_write(&scrub->os_rwsem);
719 spin_lock(&scrub->os_lock);
720 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
721 spin_unlock(&scrub->os_lock);
722 if (scrub->os_new_checked > 0) {
723 sf->sf_items_checked += scrub->os_new_checked;
724 scrub->os_new_checked = 0;
725 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
727 sf->sf_time_last_checkpoint = cfs_time_current_sec();
729 struct osd_device *dev =
730 container_of0(scrub, struct osd_device, od_scrub);
732 dev->od_igif_inoi = 1;
733 dev->od_check_ff = 0;
734 sf->sf_status = SS_COMPLETED;
735 if (!(sf->sf_param & SP_DRYRUN)) {
736 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
737 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
738 SF_UPGRADE | SF_AUTO);
740 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
741 sf->sf_success_count++;
742 } else if (result == 0) {
743 if (scrub->os_paused)
744 sf->sf_status = SS_PAUSED;
746 sf->sf_status = SS_STOPPED;
748 sf->sf_status = SS_FAILED;
750 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
751 scrub->os_time_last_checkpoint);
752 result = osd_scrub_file_store(scrub);
753 up_write(&scrub->os_rwsem);
758 /* iteration engine */
760 struct osd_iit_param {
761 struct super_block *sb;
762 struct buffer_head *bitmap;
768 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
769 struct osd_device *dev,
770 struct osd_iit_param *param,
771 struct osd_idmap_cache **oic,
774 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
775 struct osd_device *dev,
776 struct osd_iit_param *param,
777 struct osd_idmap_cache *oic,
778 int *noslot, int rc);
780 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
782 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
783 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
784 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
785 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
786 return SCRUB_NEXT_BREAK;
788 *pos = param->gbase + param->offset;
794 * \retval SCRUB_NEXT_OSTOBJ_OLD: FID-on-OST
795 * \retval 0: FID-on-MDT
797 static int osd_scrub_check_local_fldb(struct osd_thread_info *info,
798 struct osd_device *dev,
801 /* XXX: The initial OI scrub will scan the top level /O to generate
802 * a small local FLDB according to the <seq>. If the given FID
803 * is in the local FLDB, then it is FID-on-OST; otherwise it's
804 * quite possible for FID-on-MDT. */
806 return SCRUB_NEXT_OSTOBJ_OLD;
811 static int osd_scrub_get_fid(struct osd_thread_info *info,
812 struct osd_device *dev, struct inode *inode,
813 struct lu_fid *fid, bool scrub)
815 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
817 bool has_lma = false;
819 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
822 if (lma->lma_compat & LMAC_NOT_IN_OI ||
823 lma->lma_incompat & LMAI_AGENT)
824 return SCRUB_NEXT_CONTINUE;
826 *fid = lma->lma_self_fid;
830 if (lma->lma_compat & LMAC_FID_ON_OST)
831 return SCRUB_NEXT_OSTOBJ;
833 if (fid_is_idif(fid))
834 return SCRUB_NEXT_OSTOBJ_OLD;
836 /* For local object. */
837 if (fid_is_internal(fid))
840 /* For external visible MDT-object with non-normal FID. */
841 if (fid_is_namespace_visible(fid) && !fid_is_norm(fid))
844 /* For the object with normal FID, it may be MDT-object,
845 * or may be 2.4 OST-object, need further distinguish.
846 * Fall through to next section. */
849 if (rc == -ENODATA || rc == 0) {
850 rc = osd_get_idif(info, inode, &info->oti_obj_dentry, fid);
853 /* It is old 2.x (x <= 3) or 1.8 OST-object. */
854 rc = SCRUB_NEXT_OSTOBJ_OLD;
860 /* It is FID-on-OST, but we do not know how
861 * to generate its FID, ignore it directly. */
862 rc = SCRUB_NEXT_CONTINUE;
864 /* It is 2.4 OST-object. */
865 rc = SCRUB_NEXT_OSTOBJ_OLD;
873 if (dev->od_scrub.os_convert_igif) {
874 lu_igif_build(fid, inode->i_ino,
875 inode->i_generation);
877 rc = SCRUB_NEXT_NOLMA;
881 /* It may be FID-on-OST, or may be FID for
882 * non-MDT0, anyway, we do not know how to
883 * generate its FID, ignore it directly. */
884 rc = SCRUB_NEXT_CONTINUE;
889 /* For OI scrub case only: the object has LMA but has no ff
890 * (or ff crashed). It may be MDT-object, may be OST-object
891 * with crashed ff. The last check is local FLDB. */
892 rc = osd_scrub_check_local_fldb(info, dev, fid);
898 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
899 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
900 struct super_block *sb, bool scrub)
906 /* Not handle the backend root object and agent parent object.
907 * They are neither visible to namespace nor have OI mappings. */
908 if (unlikely(pos == osd_sb(dev)->s_root->d_inode->i_ino ||
909 pos == osd_remote_parent_ino(dev)))
910 RETURN(SCRUB_NEXT_CONTINUE);
912 osd_id_gen(lid, pos, OSD_OII_NOGEN);
913 inode = osd_iget(info, dev, lid);
916 /* The inode may be removed after bitmap searching, or the
917 * file is new created without inode initialized yet. */
918 if (rc == -ENOENT || rc == -ESTALE)
919 RETURN(SCRUB_NEXT_CONTINUE);
921 CDEBUG(D_LFSCK, "%.16s: fail to read inode, ino# = %u: "
922 "rc = %d\n", LDISKFS_SB(sb)->s_es->s_volume_name,
928 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
929 /* Only skip it for the first OI scrub accessing. */
930 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
931 GOTO(put, rc = SCRUB_NEXT_NOSCRUB);
934 rc = osd_scrub_get_fid(info, dev, inode, fid, scrub);
943 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
944 struct osd_iit_param *param,
945 struct osd_idmap_cache **oic, int noslot)
947 struct osd_scrub *scrub = &dev->od_scrub;
948 struct ptlrpc_thread *thread = &scrub->os_thread;
950 struct osd_inode_id *lid;
953 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
954 struct l_wait_info lwi;
956 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
957 l_wait_event(thread->t_ctl_waitq,
958 !list_empty(&scrub->os_inconsistent_items) ||
959 !thread_is_running(thread),
963 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
964 spin_lock(&scrub->os_lock);
965 thread_set_flags(thread, SVC_STOPPING);
966 spin_unlock(&scrub->os_lock);
967 return SCRUB_NEXT_CRASH;
970 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
971 return SCRUB_NEXT_FATAL;
973 if (unlikely(!thread_is_running(thread)))
974 return SCRUB_NEXT_EXIT;
976 if (!list_empty(&scrub->os_inconsistent_items)) {
977 struct osd_inconsistent_item *oii;
979 oii = list_entry(scrub->os_inconsistent_items.next,
980 struct osd_inconsistent_item, oii_list);
981 *oic = &oii->oii_cache;
982 scrub->os_in_prior = 1;
987 return SCRUB_NEXT_WAIT;
989 rc = osd_iit_next(param, &scrub->os_pos_current);
993 *oic = &scrub->os_oic;
994 fid = &(*oic)->oic_fid;
995 lid = &(*oic)->oic_lid;
996 rc = osd_iit_iget(info, dev, fid, lid,
997 scrub->os_pos_current, param->sb, true);
1001 static int osd_preload_next(struct osd_thread_info *info,
1002 struct osd_device *dev, struct osd_iit_param *param,
1003 struct osd_idmap_cache **oic, int noslot)
1005 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1006 struct osd_scrub *scrub;
1007 struct ptlrpc_thread *thread;
1010 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
1014 scrub = &dev->od_scrub;
1015 thread = &scrub->os_thread;
1016 if (thread_is_running(thread) &&
1017 ooc->ooc_pos_preload >= scrub->os_pos_current)
1018 return SCRUB_NEXT_EXIT;
1020 rc = osd_iit_iget(info, dev,
1021 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
1022 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
1023 ooc->ooc_pos_preload, param->sb, false);
1024 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
1025 * ignore the failure, so it still need to skip the inode next time. */
1026 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
1031 osd_scrub_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
1033 spin_lock(&scrub->os_lock);
1034 if (osd_scrub_has_window(scrub, &it->ooi_cache) ||
1035 !list_empty(&scrub->os_inconsistent_items) ||
1036 it->ooi_waiting || !thread_is_running(&scrub->os_thread))
1037 scrub->os_waiting = 0;
1039 scrub->os_waiting = 1;
1040 spin_unlock(&scrub->os_lock);
1042 return !scrub->os_waiting;
1045 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
1046 struct osd_iit_param *param,
1047 struct osd_idmap_cache *oic, int *noslot, int rc)
1049 struct l_wait_info lwi = { 0 };
1050 struct osd_scrub *scrub = &dev->od_scrub;
1051 struct scrub_file *sf = &scrub->os_file;
1052 struct ptlrpc_thread *thread = &scrub->os_thread;
1053 struct osd_otable_it *it = dev->od_otable_it;
1054 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
1057 case SCRUB_NEXT_CONTINUE:
1059 case SCRUB_NEXT_WAIT:
1061 case SCRUB_NEXT_NOSCRUB:
1062 down_write(&scrub->os_rwsem);
1063 scrub->os_new_checked++;
1064 sf->sf_items_noscrub++;
1065 up_write(&scrub->os_rwsem);
1069 rc = osd_scrub_check_update(info, dev, oic, rc);
1073 rc = osd_scrub_checkpoint(scrub);
1075 CDEBUG(D_LFSCK, "%.16s: fail to checkpoint, pos = %u: "
1076 "rc = %d\n", osd_scrub2name(scrub),
1077 scrub->os_pos_current, rc);
1078 /* Continue, as long as the scrub itself can go ahead. */
1081 if (scrub->os_in_prior) {
1082 scrub->os_in_prior = 0;
1087 scrub->os_pos_current = param->gbase + ++(param->offset);
1090 if (it != NULL && it->ooi_waiting && ooc != NULL &&
1091 ooc->ooc_pos_preload < scrub->os_pos_current) {
1092 spin_lock(&scrub->os_lock);
1093 it->ooi_waiting = 0;
1094 wake_up_all(&thread->t_ctl_waitq);
1095 spin_unlock(&scrub->os_lock);
1098 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
1101 if (ooc != NULL && osd_scrub_has_window(scrub, ooc)) {
1107 l_wait_event(thread->t_ctl_waitq, osd_scrub_wakeup(scrub, it),
1110 if (ooc != NULL && osd_scrub_has_window(scrub, ooc))
1117 static int osd_preload_exec(struct osd_thread_info *info,
1118 struct osd_device *dev, struct osd_iit_param *param,
1119 struct osd_idmap_cache *oic, int *noslot, int rc)
1121 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1124 ooc->ooc_cached_items++;
1125 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
1126 ~OSD_OTABLE_IT_CACHE_MASK;
1128 return rc > 0 ? 0 : rc;
1131 #define SCRUB_IT_ALL 1
1132 #define SCRUB_IT_CRASH 2
1134 static int osd_inode_iteration(struct osd_thread_info *info,
1135 struct osd_device *dev, __u32 max, bool preload)
1137 osd_iit_next_policy next;
1138 osd_iit_exec_policy exec;
1141 struct osd_iit_param param;
1148 struct osd_scrub *scrub = &dev->od_scrub;
1150 next = osd_scrub_next;
1151 exec = osd_scrub_exec;
1152 pos = &scrub->os_pos_current;
1153 count = &scrub->os_new_checked;
1155 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
1157 next = osd_preload_next;
1158 exec = osd_preload_exec;
1159 pos = &ooc->ooc_pos_preload;
1160 count = &ooc->ooc_cached_items;
1162 param.sb = osd_sb(dev);
1163 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
1165 while (*pos <= limit && *count < max) {
1166 struct osd_idmap_cache *oic = NULL;
1167 struct ldiskfs_group_desc *desc;
1169 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
1170 desc = ldiskfs_get_group_desc(param.sb, param.bg, NULL);
1174 ldiskfs_lock_group(param.sb, param.bg);
1175 if (desc->bg_flags & cpu_to_le16(LDISKFS_BG_INODE_UNINIT)) {
1176 ldiskfs_unlock_group(param.sb, param.bg);
1177 *pos = 1 + (param.bg + 1) *
1178 LDISKFS_INODES_PER_GROUP(param.sb);
1181 ldiskfs_unlock_group(param.sb, param.bg);
1183 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
1184 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
1185 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
1186 if (param.bitmap == NULL) {
1187 CDEBUG(D_LFSCK, "%.16s: fail to read bitmap for %u, "
1188 "scrub will stop, urgent mode\n",
1189 LDISKFS_SB(param.sb)->s_es->s_volume_name,
1194 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
1197 ldiskfs_itable_unused_count(param.sb, desc) >
1198 LDISKFS_INODES_PER_GROUP(param.sb))
1201 rc = next(info, dev, ¶m, &oic, noslot);
1203 case SCRUB_NEXT_BREAK:
1205 case SCRUB_NEXT_EXIT:
1206 brelse(param.bitmap);
1208 case SCRUB_NEXT_CRASH:
1209 brelse(param.bitmap);
1210 RETURN(SCRUB_IT_CRASH);
1211 case SCRUB_NEXT_FATAL:
1212 brelse(param.bitmap);
1216 rc = exec(info, dev, ¶m, oic, &noslot, rc);
1218 brelse(param.bitmap);
1224 brelse(param.bitmap);
1228 RETURN(SCRUB_IT_ALL);
1232 static int osd_otable_it_preload(const struct lu_env *env,
1233 struct osd_otable_it *it)
1235 struct osd_device *dev = it->ooi_dev;
1236 struct osd_scrub *scrub = &dev->od_scrub;
1237 struct osd_otable_cache *ooc = &it->ooi_cache;
1241 rc = osd_inode_iteration(osd_oti_get(env), dev,
1242 OSD_OTABLE_IT_CACHE_SIZE, true);
1243 if (rc == SCRUB_IT_ALL)
1244 it->ooi_all_cached = 1;
1246 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1247 spin_lock(&scrub->os_lock);
1248 scrub->os_waiting = 0;
1249 wake_up_all(&scrub->os_thread.t_ctl_waitq);
1250 spin_unlock(&scrub->os_lock);
1253 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1256 static int osd_scrub_main(void *args)
1259 struct osd_device *dev = (struct osd_device *)args;
1260 struct osd_scrub *scrub = &dev->od_scrub;
1261 struct ptlrpc_thread *thread = &scrub->os_thread;
1265 rc = lu_env_init(&env, LCT_LOCAL);
1267 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to init env: rc = %d\n",
1268 osd_scrub2name(scrub), rc);
1272 rc = osd_scrub_prep(dev);
1274 CDEBUG(D_LFSCK, "%.16s: OI scrub fail to scrub prep: rc = %d\n",
1275 osd_scrub2name(scrub), rc);
1279 if (!scrub->os_full_speed) {
1280 struct l_wait_info lwi = { 0 };
1281 struct osd_otable_it *it = dev->od_otable_it;
1282 struct osd_otable_cache *ooc = &it->ooi_cache;
1284 l_wait_event(thread->t_ctl_waitq,
1285 it->ooi_user_ready || !thread_is_running(thread),
1287 if (unlikely(!thread_is_running(thread)))
1290 scrub->os_pos_current = ooc->ooc_pos_preload;
1293 CDEBUG(D_LFSCK, "%.16s: OI scrub start, flags = 0x%x, pos = %u\n",
1294 osd_scrub2name(scrub), scrub->os_start_flags,
1295 scrub->os_pos_current);
1297 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
1298 if (unlikely(rc == SCRUB_IT_CRASH))
1299 GOTO(out, rc = -EINVAL);
1303 osd_scrub_post(scrub, rc);
1304 CDEBUG(D_LFSCK, "%.16s: OI scrub: stop, pos = %u: rc = %d\n",
1305 osd_scrub2name(scrub), scrub->os_pos_current, rc);
1308 while (!list_empty(&scrub->os_inconsistent_items)) {
1309 struct osd_inconsistent_item *oii;
1311 oii = list_entry(scrub->os_inconsistent_items.next,
1312 struct osd_inconsistent_item, oii_list);
1313 list_del_init(&oii->oii_list);
1319 spin_lock(&scrub->os_lock);
1320 thread_set_flags(thread, SVC_STOPPED);
1321 wake_up_all(&thread->t_ctl_waitq);
1322 spin_unlock(&scrub->os_lock);
1326 /* initial OI scrub */
1328 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1329 struct dentry *, filldir_t filldir);
1331 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1332 loff_t offset, __u64 ino, unsigned d_type);
1333 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1334 loff_t offset, __u64 ino, unsigned d_type);
1335 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1336 loff_t offset, __u64 ino, unsigned d_type);
1339 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1340 struct dentry *dentry, filldir_t filldir);
1342 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1343 struct dentry *dentry, filldir_t filldir);
1346 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1347 struct dentry *dentry, filldir_t filldir);
1350 OLF_SCAN_SUBITEMS = 0x0001,
1351 OLF_HIDE_FID = 0x0002,
1352 OLF_SHOW_NAME = 0x0004,
1358 struct lu_fid olm_fid;
1360 scandir_t olm_scandir;
1361 filldir_t olm_filldir;
1364 /* Add the new introduced local files in the list in the future. */
1365 static const struct osd_lf_map osd_lf_maps[] = {
1367 { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1371 { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1372 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1373 osd_ios_varfid_fill },
1375 /* NIDTBL_VERSIONS */
1376 { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1377 osd_ios_general_scan, osd_ios_varfid_fill },
1380 { "PENDING", { 0, 0, 0 }, 0, NULL, NULL },
1383 { "ROOT", { FID_SEQ_ROOT, 1, 0 },
1384 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1386 /* changelog_catalog */
1387 { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1389 /* changelog_users */
1390 { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1393 { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1397 { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1400 /* lfsck_bookmark */
1401 { "lfsck_bookmark", { 0, 0, 0 }, 0, NULL, NULL },
1404 { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1408 { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1409 OLF_SHOW_NAME, NULL, NULL },
1412 { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1413 osd_ios_general_scan, osd_ios_varfid_fill },
1416 { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1417 osd_ios_general_scan, osd_ios_varfid_fill },
1420 { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1421 OLF_SHOW_NAME, NULL, NULL },
1424 { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1425 OLF_SHOW_NAME, NULL, NULL },
1428 { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1429 OLF_SHOW_NAME, NULL, NULL },
1431 /* lfsck_namespace */
1432 { "lfsck_namespace", { 0, 0, 0 }, 0, NULL, NULL },
1434 /* OBJECTS, upgrade from old device */
1435 { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1437 /* lquota_v2.user, upgrade from old device */
1438 { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1440 /* lquota_v2.group, upgrade from old device */
1441 { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1443 /* LAST_GROUP, upgrade from old device */
1444 { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1445 OLF_SHOW_NAME, NULL, NULL },
1447 /* SLAVE_LOG, llog for destroy slave stripes of striped dir */
1448 { "SLAVE_LOG", { FID_SEQ_LOCAL_FILE, SLAVE_LLOG_CATALOGS_OID, 0 },
1449 OLF_SHOW_NAME, NULL, NULL },
1452 { "lost+found", { FID_SEQ_LOCAL_FILE, OSD_LPF_OID, 0 },
1453 OLF_SCAN_SUBITEMS, osd_ios_general_scan, osd_ios_lf_fill },
1455 { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1458 /* Add the new introduced files under .lustre/ in the list in the future. */
1459 static const struct osd_lf_map osd_dl_maps[] = {
1461 { "fid", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_OBF, 0 }, 0,
1464 /* .lustre/lost+found */
1465 { "lost+found", { FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE_LPF, 0 }, 0,
1468 { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1471 struct osd_ios_item {
1472 struct list_head oii_list;
1473 struct dentry *oii_dentry;
1474 scandir_t oii_scandir;
1475 filldir_t oii_filldir;
1478 struct osd_ios_filldir_buf {
1479 #ifdef HAVE_DIR_CONTEXT
1480 /* please keep it as first member */
1481 struct dir_context ctx;
1483 struct osd_thread_info *oifb_info;
1484 struct osd_device *oifb_dev;
1485 struct dentry *oifb_dentry;
1488 static inline struct dentry *
1489 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1491 struct dentry *dentry;
1493 dentry = ll_lookup_one_len(name, parent, namelen);
1494 if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1496 return ERR_PTR(-ENOENT);
1503 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1504 scandir_t scandir, filldir_t filldir)
1506 struct osd_ios_item *item;
1509 OBD_ALLOC_PTR(item);
1513 INIT_LIST_HEAD(&item->oii_list);
1514 item->oii_dentry = dget(dentry);
1515 item->oii_scandir = scandir;
1516 item->oii_filldir = filldir;
1517 list_add_tail(&item->oii_list, &dev->od_ios_list);
1523 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1525 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1526 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1527 * reference the inode, or fixed if it is missing or references another inode.
1530 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1531 struct inode *inode, const struct lu_fid *fid, int flags)
1533 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1534 struct osd_inode_id *id = &info->oti_id;
1535 struct osd_inode_id *id2 = &info->oti_id2;
1536 struct osd_scrub *scrub = &dev->od_scrub;
1537 struct scrub_file *sf = &scrub->os_file;
1542 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1543 if (rc != 0 && rc != -ENODATA) {
1544 CDEBUG(D_LFSCK, "%s: fail to get lma for init OI scrub: "
1545 "rc = %d\n", osd_name(dev), rc);
1550 osd_id_gen(id, inode->i_ino, inode->i_generation);
1551 if (rc == -ENODATA) {
1552 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1553 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1556 rc = osd_ea_fid_set(info, inode, &tfid, 0, 0);
1558 CDEBUG(D_LFSCK, "%s: fail to set LMA for init OI "
1559 "scrub: rc = %d\n", osd_name(dev), rc);
1564 if (lma->lma_compat & LMAC_NOT_IN_OI)
1567 tfid = lma->lma_self_fid;
1570 rc = osd_oi_lookup(info, dev, &tfid, id2, 0);
1575 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1576 DTO_INDEX_INSERT, true, 0);
1583 if (osd_id_eq_strict(id, id2))
1586 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1587 osd_scrub_file_reset(scrub,
1588 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1590 rc = osd_scrub_file_store(scrub);
1595 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1596 DTO_INDEX_UPDATE, true, 0);
1604 * It scans the /lost+found, and for the OST-object (with filter_fid
1605 * or filter_fid_old), move them back to its proper /O/<seq>/d<x>.
1607 static int osd_ios_lf_fill(void *buf, const char *name, int namelen,
1608 loff_t offset, __u64 ino, unsigned d_type)
1610 struct osd_ios_filldir_buf *fill_buf = buf;
1611 struct osd_thread_info *info = fill_buf->oifb_info;
1612 struct osd_device *dev = fill_buf->oifb_dev;
1613 struct lu_fid *fid = &info->oti_fid;
1614 struct osd_scrub *scrub = &dev->od_scrub;
1615 struct dentry *parent = fill_buf->oifb_dentry;
1616 struct dentry *child;
1617 struct inode *dir = parent->d_inode;
1618 struct inode *inode;
1622 /* skip any '.' started names */
1626 scrub->os_lf_scanned++;
1627 child = osd_ios_lookup_one_len(name, parent, namelen);
1628 if (IS_ERR(child)) {
1629 CDEBUG(D_LFSCK, "%s: cannot lookup child '%.*s': rc = %d\n",
1630 osd_name(dev), namelen, name, (int)PTR_ERR(child));
1634 inode = child->d_inode;
1635 if (S_ISDIR(inode->i_mode)) {
1636 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1639 CDEBUG(D_LFSCK, "%s: cannot add child '%.*s': "
1640 "rc = %d\n", osd_name(dev), namelen, name, rc);
1644 if (!S_ISREG(inode->i_mode))
1647 rc = osd_scrub_get_fid(info, dev, inode, fid, true);
1648 if (rc == SCRUB_NEXT_OSTOBJ || rc == SCRUB_NEXT_OSTOBJ_OLD) {
1649 rc = osd_obj_map_recover(info, dev, dir, child, fid);
1651 CDEBUG(D_LFSCK, "recovered '%.*s' ["DFID"] from "
1652 "/lost+found.\n", namelen, name, PFID(fid));
1653 scrub->os_lf_repaired++;
1655 CDEBUG(D_LFSCK, "%s: cannot rename for '%.*s' "
1657 osd_name(dev), namelen, name, PFID(fid), rc);
1661 /* XXX: For MDT-objects, we can move them from /lost+found to namespace
1662 * visible place, such as the /ROOT/.lustre/lost+found, then LFSCK
1663 * can process them in furtuer. */
1669 scrub->os_lf_failed++;
1671 /* skip the failure to make the scanning to continue. */
1675 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1676 loff_t offset, __u64 ino, unsigned d_type)
1678 struct osd_ios_filldir_buf *fill_buf = buf;
1679 struct osd_device *dev = fill_buf->oifb_dev;
1680 struct dentry *child;
1684 /* skip any '.' started names */
1688 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1690 RETURN(PTR_ERR(child));
1692 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1694 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1695 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1696 osd_ios_varfid_fill);
1702 static int osd_ios_dl_fill(void *buf, const char *name, int namelen,
1703 loff_t offset, __u64 ino, unsigned d_type)
1705 struct osd_ios_filldir_buf *fill_buf = buf;
1706 struct osd_device *dev = fill_buf->oifb_dev;
1707 const struct osd_lf_map *map;
1708 struct dentry *child;
1712 /* skip any '.' started names */
1716 for (map = osd_dl_maps; map->olm_name != NULL; map++) {
1717 if (strlen(map->olm_name) != namelen)
1720 if (strncmp(map->olm_name, name, namelen) == 0)
1724 if (map->olm_name == NULL)
1727 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1729 RETURN(PTR_ERR(child));
1731 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1732 &map->olm_fid, map->olm_flags);
1738 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1739 loff_t offset, __u64 ino, unsigned d_type)
1741 struct osd_ios_filldir_buf *fill_buf = buf;
1742 struct osd_device *dev = fill_buf->oifb_dev;
1743 const struct osd_lf_map *map;
1744 struct dentry *child;
1748 /* skip any '.' started names */
1752 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1753 if (strlen(map->olm_name) != namelen)
1756 if (strncmp(map->olm_name, name, namelen) == 0)
1760 if (map->olm_name == NULL)
1763 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1765 RETURN(PTR_ERR(child));
1767 if (!(map->olm_flags & OLF_NO_OI))
1768 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1769 &map->olm_fid, map->olm_flags);
1770 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1771 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1779 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1780 struct dentry *dentry, filldir_t filldir)
1782 struct osd_ios_filldir_buf buf = {
1783 #ifdef HAVE_DIR_CONTEXT
1784 .ctx.actor = filldir,
1788 .oifb_dentry = dentry };
1789 struct file *filp = &info->oti_it_ea.oie_file;
1790 struct inode *inode = dentry->d_inode;
1791 const struct file_operations *fops = inode->i_fop;
1795 LASSERT(filldir != NULL);
1798 filp->f_dentry = dentry;
1799 filp->f_mode = FMODE_64BITHASH;
1800 filp->f_mapping = inode->i_mapping;
1802 filp->private_data = NULL;
1803 set_file_inode(filp, inode);
1805 #ifdef HAVE_DIR_CONTEXT
1806 buf.ctx.pos = filp->f_pos;
1807 rc = fops->iterate(filp, &buf.ctx);
1808 filp->f_pos = buf.ctx.pos;
1810 rc = fops->readdir(filp, &buf, filldir);
1812 fops->release(inode, filp);
1818 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1819 struct dentry *dentry, filldir_t filldir)
1821 struct osd_scrub *scrub = &dev->od_scrub;
1822 struct scrub_file *sf = &scrub->os_file;
1823 struct dentry *child;
1827 /* It is existing MDT0 device. We only allow the case of object without
1828 * LMA to happen on the MDT0, which is usually for old 1.8 MDT. Then we
1829 * can generate IGIF mode FID for the object and related OI mapping. If
1830 * it is on other MDTs, then becuase file-level backup/restore, related
1831 * OI mapping may be invalid already, we do not know which is the right
1832 * FID for the object. We only allow IGIF objects to reside on the MDT0.
1834 * XXX: For the case of object on non-MDT0 device with neither LMA nor
1835 * "fid" xattr, then something crashed. We cannot re-generate the
1836 * FID directly, instead, the OI scrub will scan the OI structure
1837 * and try to re-generate the LMA from the OI mapping. But if the
1838 * OI mapping crashed or lost also, then we have to give up under
1839 * double failure cases. */
1840 scrub->os_convert_igif = 1;
1841 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1842 strlen(dot_lustre_name));
1843 if (IS_ERR(child)) {
1844 rc = PTR_ERR(child);
1845 if (rc == -ENOENT) {
1846 /* It is 1.8 MDT device. */
1847 if (!(sf->sf_flags & SF_UPGRADE)) {
1848 osd_scrub_file_reset(scrub,
1849 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1851 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
1852 rc = osd_scrub_file_store(scrub);
1858 /* For lustre-2.x (x <= 3), the ".lustre" has NO FID-in-LMA,
1859 * so the client will get IGIF for the ".lustre" object when
1862 * From the OI scrub view, when the MDT upgrade to Lustre-2.4,
1863 * it does not know whether there are some old clients cached
1864 * the ".lustre" IGIF during the upgrading. Two choices:
1866 * 1) Generate IGIF-in-LMA and IGIF-in-OI for the ".lustre".
1867 * It will allow the old connected clients to access the
1868 * ".lustre" with cached IGIF. But it will cause others
1869 * on the MDT failed to check "fid_is_dot_lustre()".
1871 * 2) Use fixed FID {FID_SEQ_DOT_LUSTRE, FID_OID_DOT_LUSTRE, 0}
1872 * for ".lustre" in spite of whether there are some clients
1873 * cached the ".lustre" IGIF or not. It enables the check
1874 * "fid_is_dot_lustre()" on the MDT, although it will cause
1875 * that the old connected clients cannot access the ".lustre"
1876 * with the cached IGIF.
1878 * Usually, it is rare case for the old connected clients
1879 * to access the ".lustre" with cached IGIF. So we prefer
1880 * to the solution 2). */
1881 rc = osd_ios_scan_one(info, dev, child->d_inode,
1882 &LU_DOT_LUSTRE_FID, 0);
1884 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1893 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1894 struct dentry *dentry, filldir_t filldir)
1896 struct osd_scrub *scrub = &dev->od_scrub;
1897 struct scrub_file *sf = &scrub->os_file;
1898 struct dentry *child;
1902 if (unlikely(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID)) {
1903 sf->sf_internal_flags &= ~SIF_NO_HANDLE_OLD_FID;
1904 rc = osd_scrub_file_store(scrub);
1909 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1910 if (!IS_ERR(child)) {
1911 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1914 rc = PTR_ERR(child);
1917 if (rc != 0 && rc != -ENOENT)
1920 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1921 if (!IS_ERR(child)) {
1922 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1925 rc = PTR_ERR(child);
1934 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1935 struct osd_device *dev)
1937 struct osd_ios_item *item = NULL;
1938 scandir_t scandir = osd_ios_general_scan;
1939 filldir_t filldir = osd_ios_root_fill;
1940 struct dentry *dentry = osd_sb(dev)->s_root;
1941 const struct osd_lf_map *map = osd_lf_maps;
1945 /* Lookup IGIF in OI by force for initial OI scrub. */
1946 dev->od_igif_inoi = 1;
1949 rc = scandir(info, dev, dentry, filldir);
1951 dput(item->oii_dentry);
1958 if (list_empty(&dev->od_ios_list))
1961 item = list_entry(dev->od_ios_list.next,
1962 struct osd_ios_item, oii_list);
1963 list_del_init(&item->oii_list);
1965 LASSERT(item->oii_scandir != NULL);
1966 scandir = item->oii_scandir;
1967 filldir = item->oii_filldir;
1968 dentry = item->oii_dentry;
1971 while (!list_empty(&dev->od_ios_list)) {
1972 item = list_entry(dev->od_ios_list.next,
1973 struct osd_ios_item, oii_list);
1974 list_del_init(&item->oii_list);
1975 dput(item->oii_dentry);
1982 /* There maybe the case that the object has been removed, but its OI
1983 * mapping is still in the OI file, such as the "CATALOGS" after MDT
1984 * file-level backup/restore. So here cleanup the stale OI mappings. */
1985 while (map->olm_name != NULL) {
1986 struct dentry *child;
1988 if (fid_is_zero(&map->olm_fid)) {
1993 child = osd_ios_lookup_one_len(map->olm_name,
1994 osd_sb(dev)->s_root,
1995 strlen(map->olm_name));
1998 else if (PTR_ERR(child) == -ENOENT)
1999 osd_scrub_refresh_mapping(info, dev, &map->olm_fid,
2000 NULL, DTO_INDEX_DELETE,
2008 char *osd_lf_fid2name(const struct lu_fid *fid)
2010 const struct osd_lf_map *map = osd_lf_maps;
2012 while (map->olm_name != NULL) {
2013 if (!lu_fid_eq(fid, &map->olm_fid)) {
2018 if (map->olm_flags & OLF_SHOW_NAME)
2019 return map->olm_name;
2027 /* OI scrub start/stop */
2029 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
2031 struct osd_scrub *scrub = &dev->od_scrub;
2032 struct ptlrpc_thread *thread = &scrub->os_thread;
2033 struct l_wait_info lwi = { 0 };
2034 struct task_struct *task;
2039 /* os_lock: sync status between stop and scrub thread */
2040 spin_lock(&scrub->os_lock);
2041 if (thread_is_running(thread)) {
2042 spin_unlock(&scrub->os_lock);
2044 } else if (unlikely(thread_is_stopping(thread))) {
2045 spin_unlock(&scrub->os_lock);
2046 l_wait_event(thread->t_ctl_waitq,
2047 thread_is_stopped(thread),
2051 spin_unlock(&scrub->os_lock);
2053 if (scrub->os_file.sf_status == SS_COMPLETED) {
2054 if (!(flags & SS_SET_FAILOUT))
2055 flags |= SS_CLEAR_FAILOUT;
2057 if (!(flags & SS_SET_DRYRUN))
2058 flags |= SS_CLEAR_DRYRUN;
2063 scrub->os_start_flags = flags;
2064 thread_set_flags(thread, 0);
2065 task = kthread_run(osd_scrub_main, dev, "OI_scrub");
2068 CERROR("%.16s: cannot start iteration thread: rc = %d\n",
2069 osd_scrub2name(scrub), rc);
2073 l_wait_event(thread->t_ctl_waitq,
2074 thread_is_running(thread) || thread_is_stopped(thread),
2080 int osd_scrub_start(struct osd_device *dev)
2085 /* od_otable_mutex: prevent curcurrent start/stop */
2086 mutex_lock(&dev->od_otable_mutex);
2087 rc = do_osd_scrub_start(dev, SS_AUTO);
2088 mutex_unlock(&dev->od_otable_mutex);
2090 RETURN(rc == -EALREADY ? 0 : rc);
2093 static void do_osd_scrub_stop(struct osd_scrub *scrub)
2095 struct ptlrpc_thread *thread = &scrub->os_thread;
2096 struct l_wait_info lwi = { 0 };
2098 /* os_lock: sync status between stop and scrub thread */
2099 spin_lock(&scrub->os_lock);
2100 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2101 thread_set_flags(thread, SVC_STOPPING);
2102 spin_unlock(&scrub->os_lock);
2103 wake_up_all(&thread->t_ctl_waitq);
2104 l_wait_event(thread->t_ctl_waitq,
2105 thread_is_stopped(thread),
2107 /* Do not skip the last lock/unlock, which can guarantee that
2108 * the caller cannot return until the OI scrub thread exit. */
2109 spin_lock(&scrub->os_lock);
2111 spin_unlock(&scrub->os_lock);
2114 static void osd_scrub_stop(struct osd_device *dev)
2116 /* od_otable_mutex: prevent curcurrent start/stop */
2117 mutex_lock(&dev->od_otable_mutex);
2118 dev->od_scrub.os_paused = 1;
2119 do_osd_scrub_stop(&dev->od_scrub);
2120 mutex_unlock(&dev->od_otable_mutex);
2123 /* OI scrub setup/cleanup */
2125 static const char osd_scrub_name[] = "OI_scrub";
2127 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
2129 struct osd_thread_info *info = osd_oti_get(env);
2130 struct osd_scrub *scrub = &dev->od_scrub;
2131 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
2132 struct scrub_file *sf = &scrub->os_file;
2133 struct super_block *sb = osd_sb(dev);
2134 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
2135 struct lvfs_run_ctxt saved;
2137 struct inode *inode;
2138 struct lu_fid *fid = &info->oti_fid;
2143 memset(scrub, 0, sizeof(*scrub));
2144 OBD_SET_CTXT_MAGIC(ctxt);
2145 ctxt->pwdmnt = dev->od_mnt;
2146 ctxt->pwd = dev->od_mnt->mnt_root;
2147 ctxt->fs = get_ds();
2149 init_waitqueue_head(&scrub->os_thread.t_ctl_waitq);
2150 init_rwsem(&scrub->os_rwsem);
2151 spin_lock_init(&scrub->os_lock);
2152 INIT_LIST_HEAD(&scrub->os_inconsistent_items);
2154 push_ctxt(&saved, ctxt);
2155 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
2157 pop_ctxt(&saved, ctxt);
2158 RETURN(PTR_ERR(filp));
2161 inode = filp->f_dentry->d_inode;
2162 /* 'What the @fid is' is not imporatant, because the object
2163 * has no OI mapping, and only is visible inside the OSD.*/
2164 lu_igif_build(fid, inode->i_ino, inode->i_generation);
2165 rc = osd_ea_fid_set(info, inode, fid, LMAC_NOT_IN_OI, 0);
2167 filp_close(filp, 0);
2168 pop_ctxt(&saved, ctxt);
2172 scrub->os_inode = igrab(inode);
2173 filp_close(filp, 0);
2174 pop_ctxt(&saved, ctxt);
2176 rc = osd_scrub_file_load(scrub);
2177 if (rc == -ENOENT) {
2178 osd_scrub_file_init(scrub, es->s_uuid);
2179 /* If the "/O" dir does not exist when mount (indicated by
2180 * osd_device::od_maybe_new), neither for the "/OI_scrub",
2181 * then it is quite probably that the device is a new one,
2182 * under such case, mark it as SIF_NO_HANDLE_OLD_FID.
2184 * For the rare case that "/O" and "OI_scrub" both lost on
2185 * an old device, it can be found and cleared later.
2187 * For the system with "SIF_NO_HANDLE_OLD_FID", we do not
2188 * need to check "filter_fid_old" and to convert it to
2189 * "filter_fid" for each object, and all the IGIF should
2190 * have their FID mapping in OI files already. */
2191 if (dev->od_maybe_new)
2192 sf->sf_internal_flags = SIF_NO_HANDLE_OLD_FID;
2194 } else if (rc != 0) {
2195 GOTO(cleanup_inode, rc);
2197 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
2198 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
2200 } else if (sf->sf_status == SS_SCANNING) {
2201 sf->sf_status = SS_CRASHED;
2206 if (sf->sf_pos_last_checkpoint != 0)
2207 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
2209 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
2212 rc = osd_scrub_file_store(scrub);
2214 GOTO(cleanup_inode, rc);
2217 /* Initialize OI files. */
2218 rc = osd_oi_init(info, dev);
2220 GOTO(cleanup_inode, rc);
2222 rc = osd_initial_OI_scrub(info, dev);
2224 GOTO(cleanup_oi, rc);
2226 if (sf->sf_flags & SF_UPGRADE ||
2227 !(sf->sf_internal_flags & SIF_NO_HANDLE_OLD_FID ||
2228 sf->sf_success_count > 0)) {
2229 dev->od_igif_inoi = 0;
2230 dev->od_check_ff = dev->od_is_ost;
2232 dev->od_igif_inoi = 1;
2233 dev->od_check_ff = 0;
2236 if (sf->sf_flags & SF_INCONSISTENT)
2237 /* The 'od_igif_inoi' will be set under the
2239 * 1) new created system, or
2240 * 2) restored from file-level backup, or
2241 * 3) the upgrading completed.
2243 * The 'od_igif_inoi' may be cleared by OI scrub
2244 * later if found that the system is upgrading. */
2245 dev->od_igif_inoi = 1;
2247 if (!dev->od_noscrub &&
2248 ((sf->sf_status == SS_PAUSED) ||
2249 (sf->sf_status == SS_CRASHED &&
2250 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2251 SF_UPGRADE | SF_AUTO)) ||
2252 (sf->sf_status == SS_INIT &&
2253 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
2255 rc = osd_scrub_start(dev);
2258 GOTO(cleanup_oi, rc);
2260 /* it is possible that dcache entries may keep objects after they are
2261 * deleted by OSD. While it looks safe this can cause object data to
2262 * stay until umount causing failures in tests calculating free space,
2263 * e.g. replay-ost-single. Since those dcache entries are not used
2264 * anymore let's just free them after use here */
2265 shrink_dcache_sb(sb);
2269 osd_oi_fini(info, dev);
2271 iput(scrub->os_inode);
2272 scrub->os_inode = NULL;
2277 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
2279 struct osd_scrub *scrub = &dev->od_scrub;
2281 LASSERT(dev->od_otable_it == NULL);
2283 if (scrub->os_inode != NULL) {
2284 osd_scrub_stop(dev);
2285 iput(scrub->os_inode);
2286 scrub->os_inode = NULL;
2288 if (dev->od_oi_table != NULL)
2289 osd_oi_fini(osd_oti_get(env), dev);
2292 /* object table based iteration APIs */
2294 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
2295 struct dt_object *dt, __u32 attr,
2296 struct lustre_capa *capa)
2298 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
2299 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
2300 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
2301 struct osd_scrub *scrub = &dev->od_scrub;
2302 struct osd_otable_it *it;
2307 /* od_otable_mutex: prevent curcurrent init/fini */
2308 mutex_lock(&dev->od_otable_mutex);
2309 if (dev->od_otable_it != NULL)
2310 GOTO(out, it = ERR_PTR(-EALREADY));
2314 GOTO(out, it = ERR_PTR(-ENOMEM));
2316 dev->od_otable_it = it;
2318 it->ooi_cache.ooc_consumer_idx = -1;
2319 if (flags & DOIF_OUTUSED)
2320 it->ooi_used_outside = 1;
2322 if (flags & DOIF_RESET)
2325 if (valid & DOIV_ERROR_HANDLE) {
2326 if (flags & DOIF_FAILOUT)
2327 start |= SS_SET_FAILOUT;
2329 start |= SS_CLEAR_FAILOUT;
2332 if (valid & DOIV_DRYRUN) {
2333 if (flags & DOIF_DRYRUN)
2334 start |= SS_SET_DRYRUN;
2336 start |= SS_CLEAR_DRYRUN;
2339 rc = do_osd_scrub_start(dev, start);
2340 if (rc < 0 && rc != -EALREADY) {
2341 dev->od_otable_it = NULL;
2343 GOTO(out, it = ERR_PTR(rc));
2346 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
2351 mutex_unlock(&dev->od_otable_mutex);
2352 return (struct dt_it *)it;
2355 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
2357 struct osd_otable_it *it = (struct osd_otable_it *)di;
2358 struct osd_device *dev = it->ooi_dev;
2360 /* od_otable_mutex: prevent curcurrent init/fini */
2361 mutex_lock(&dev->od_otable_mutex);
2362 do_osd_scrub_stop(&dev->od_scrub);
2363 LASSERT(dev->od_otable_it == it);
2365 dev->od_otable_it = NULL;
2366 mutex_unlock(&dev->od_otable_mutex);
2370 static int osd_otable_it_get(const struct lu_env *env,
2371 struct dt_it *di, const struct dt_key *key)
2376 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
2381 osd_otable_it_wakeup(struct osd_scrub *scrub, struct osd_otable_it *it)
2383 spin_lock(&scrub->os_lock);
2384 if (it->ooi_cache.ooc_pos_preload < scrub->os_pos_current ||
2385 scrub->os_waiting ||
2386 !thread_is_running(&scrub->os_thread))
2387 it->ooi_waiting = 0;
2389 it->ooi_waiting = 1;
2390 spin_unlock(&scrub->os_lock);
2392 return !it->ooi_waiting;
2395 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
2397 struct osd_otable_it *it = (struct osd_otable_it *)di;
2398 struct osd_device *dev = it->ooi_dev;
2399 struct osd_scrub *scrub = &dev->od_scrub;
2400 struct osd_otable_cache *ooc = &it->ooi_cache;
2401 struct ptlrpc_thread *thread = &scrub->os_thread;
2402 struct l_wait_info lwi = { 0 };
2406 LASSERT(it->ooi_user_ready);
2409 if (!thread_is_running(thread) && !it->ooi_used_outside)
2412 if (ooc->ooc_cached_items > 0) {
2413 ooc->ooc_cached_items--;
2414 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
2415 ~OSD_OTABLE_IT_CACHE_MASK;
2419 if (it->ooi_all_cached) {
2420 l_wait_event(thread->t_ctl_waitq,
2421 !thread_is_running(thread),
2426 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
2427 spin_lock(&scrub->os_lock);
2428 scrub->os_waiting = 0;
2429 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2430 spin_unlock(&scrub->os_lock);
2433 if (it->ooi_cache.ooc_pos_preload >= scrub->os_pos_current)
2434 l_wait_event(thread->t_ctl_waitq,
2435 osd_otable_it_wakeup(scrub, it),
2438 if (!thread_is_running(thread) && !it->ooi_used_outside)
2441 rc = osd_otable_it_preload(env, it);
2448 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
2449 const struct dt_it *di)
2454 static int osd_otable_it_key_size(const struct lu_env *env,
2455 const struct dt_it *di)
2457 return sizeof(__u64);
2460 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
2461 struct dt_rec *rec, __u32 attr)
2463 struct osd_otable_it *it = (struct osd_otable_it *)di;
2464 struct osd_otable_cache *ooc = &it->ooi_cache;
2466 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
2468 /* Filter out Invald FID already. */
2469 LASSERTF(fid_is_sane((struct lu_fid *)rec),
2470 "Invalid FID "DFID", p_idx = %d, c_idx = %d\n",
2471 PFID((struct lu_fid *)rec),
2472 ooc->ooc_producer_idx, ooc->ooc_consumer_idx);
2477 static __u64 osd_otable_it_store(const struct lu_env *env,
2478 const struct dt_it *di)
2480 struct osd_otable_it *it = (struct osd_otable_it *)di;
2481 struct osd_otable_cache *ooc = &it->ooi_cache;
2484 if (it->ooi_user_ready && ooc->ooc_consumer_idx != -1)
2485 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
2487 hash = ooc->ooc_pos_preload;
2492 * Set the OSD layer iteration start position as the specified hash.
2494 static int osd_otable_it_load(const struct lu_env *env,
2495 const struct dt_it *di, __u64 hash)
2497 struct osd_otable_it *it = (struct osd_otable_it *)di;
2498 struct osd_device *dev = it->ooi_dev;
2499 struct osd_otable_cache *ooc = &it->ooi_cache;
2500 struct osd_scrub *scrub = &dev->od_scrub;
2504 /* Forbid to set iteration position after iteration started. */
2505 if (it->ooi_user_ready)
2508 if (hash > OSD_OTABLE_MAX_HASH)
2509 hash = OSD_OTABLE_MAX_HASH;
2511 ooc->ooc_pos_preload = hash;
2512 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
2513 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
2515 it->ooi_user_ready = 1;
2516 if (!scrub->os_full_speed)
2517 wake_up_all(&scrub->os_thread.t_ctl_waitq);
2519 /* Unplug OSD layer iteration by the first next() call. */
2520 rc = osd_otable_it_next(env, (struct dt_it *)it);
2525 static int osd_otable_it_key_rec(const struct lu_env *env,
2526 const struct dt_it *di, void *key_rec)
2531 const struct dt_index_operations osd_otable_ops = {
2533 .init = osd_otable_it_init,
2534 .fini = osd_otable_it_fini,
2535 .get = osd_otable_it_get,
2536 .put = osd_otable_it_put,
2537 .next = osd_otable_it_next,
2538 .key = osd_otable_it_key,
2539 .key_size = osd_otable_it_key_size,
2540 .rec = osd_otable_it_rec,
2541 .store = osd_otable_it_store,
2542 .load = osd_otable_it_load,
2543 .key_rec = osd_otable_it_key_rec,
2547 /* high priority inconsistent items list APIs */
2549 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
2552 struct osd_inconsistent_item *oii;
2553 struct osd_scrub *scrub = &dev->od_scrub;
2554 struct ptlrpc_thread *thread = &scrub->os_thread;
2559 if (unlikely(oii == NULL))
2562 INIT_LIST_HEAD(&oii->oii_list);
2563 oii->oii_cache = *oic;
2564 oii->oii_insert = insert;
2566 spin_lock(&scrub->os_lock);
2567 if (unlikely(!thread_is_running(thread))) {
2568 spin_unlock(&scrub->os_lock);
2573 if (list_empty(&scrub->os_inconsistent_items))
2575 list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
2576 spin_unlock(&scrub->os_lock);
2579 wake_up_all(&thread->t_ctl_waitq);
2584 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
2585 struct osd_inode_id *id)
2587 struct osd_scrub *scrub = &dev->od_scrub;
2588 struct osd_inconsistent_item *oii;
2591 spin_lock(&scrub->os_lock);
2592 list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
2593 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
2594 *id = oii->oii_cache.oic_lid;
2595 spin_unlock(&scrub->os_lock);
2599 spin_unlock(&scrub->os_lock);
2606 static const char *scrub_status_names[] = {
2617 static const char *scrub_flags_names[] = {
2625 static const char *scrub_param_names[] = {
2631 static int scrub_bits_dump(struct seq_file *m, int bits, const char *names[],
2638 rc = seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2642 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2645 rc = seq_printf(m, "%s%c", names[i],
2646 bits != 0 ? ',' : '\n');
2654 static int scrub_time_dump(struct seq_file *m, __u64 time, const char *prefix)
2659 rc = seq_printf(m, "%s: "LPU64" seconds\n", prefix,
2660 cfs_time_current_sec() - time);
2662 rc = seq_printf(m, "%s: N/A\n", prefix);
2666 static int scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
2671 rc = seq_printf(m, "%s: "LPU64"\n", prefix, pos);
2673 rc = seq_printf(m, "%s: N/A\n", prefix);
2677 int osd_scrub_dump(struct seq_file *m, struct osd_device *dev)
2679 struct osd_scrub *scrub = &dev->od_scrub;
2680 struct scrub_file *sf = &scrub->os_file;
2685 down_read(&scrub->os_rwsem);
2686 rc = seq_printf(m, "name: OI_scrub\n"
2690 sf->sf_magic, (int)sf->sf_oi_count,
2691 scrub_status_names[sf->sf_status]);
2695 rc = scrub_bits_dump(m, sf->sf_flags, scrub_flags_names,
2700 rc = scrub_bits_dump(m, sf->sf_param, scrub_param_names,
2705 rc = scrub_time_dump(m, sf->sf_time_last_complete,
2706 "time_since_last_completed");
2710 rc = scrub_time_dump(m, sf->sf_time_latest_start,
2711 "time_since_latest_start");
2715 rc = scrub_time_dump(m, sf->sf_time_last_checkpoint,
2716 "time_since_last_checkpoint");
2720 rc = scrub_pos_dump(m, sf->sf_pos_latest_start,
2721 "latest_start_position");
2725 rc = scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
2726 "last_checkpoint_position");
2730 rc = scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
2731 "first_failure_position");
2735 checked = sf->sf_items_checked + scrub->os_new_checked;
2736 rc = seq_printf(m, "checked: "LPU64"\n"
2737 "updated: "LPU64"\n"
2739 "prior_updated: "LPU64"\n"
2740 "noscrub: "LPU64"\n"
2742 "success_count: %u\n",
2743 checked, sf->sf_items_updated, sf->sf_items_failed,
2744 sf->sf_items_updated_prior, sf->sf_items_noscrub,
2745 sf->sf_items_igif, sf->sf_success_count);
2750 if (thread_is_running(&scrub->os_thread)) {
2751 cfs_duration_t duration = cfs_time_current() -
2752 scrub->os_time_last_checkpoint;
2753 __u64 new_checked = msecs_to_jiffies(scrub->os_new_checked *
2755 __u32 rtime = sf->sf_run_time +
2756 cfs_duration_sec(duration + HALF_SEC);
2759 do_div(new_checked, duration);
2761 do_div(speed, rtime);
2762 rc = seq_printf(m, "run_time: %u seconds\n"
2763 "average_speed: "LPU64" objects/sec\n"
2764 "real-time_speed: "LPU64" objects/sec\n"
2765 "current_position: %u\n"
2766 "lf_scanned: "LPU64"\n"
2767 "lf_reparied: "LPU64"\n"
2768 "lf_failed: "LPU64"\n",
2769 rtime, speed, new_checked, scrub->os_pos_current,
2770 scrub->os_lf_scanned, scrub->os_lf_repaired,
2771 scrub->os_lf_failed);
2773 if (sf->sf_run_time != 0)
2774 do_div(speed, sf->sf_run_time);
2775 rc = seq_printf(m, "run_time: %u seconds\n"
2776 "average_speed: "LPU64" objects/sec\n"
2777 "real-time_speed: N/A\n"
2778 "current_position: N/A\n"
2779 "lf_scanned: "LPU64"\n"
2780 "lf_reparied: "LPU64"\n"
2781 "lf_failed: "LPU64"\n",
2782 sf->sf_run_time, speed, scrub->os_lf_scanned,
2783 scrub->os_lf_repaired, scrub->os_lf_failed);
2787 up_read(&scrub->os_rwsem);
2788 return (rc < 0 ? -ENOSPC : 0);