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, 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>
39 # define EXPORT_SYMTAB
41 #define DEBUG_SUBSYSTEM S_MDS
43 #include <lustre/lustre_idl.h>
44 #include <lustre_disk.h>
45 #include <dt_object.h>
47 #include "osd_internal.h"
49 #include "osd_scrub.h"
51 #define HALF_SEC (CFS_HZ >> 1)
53 #define OSD_OTABLE_MAX_HASH 0x00000000ffffffffULL
55 #define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */
56 #define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */
57 #define SCRUB_NEXT_EXIT 3 /* exit all the loops */
58 #define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */
59 #define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */
60 #define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */
61 #define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */
62 #define SCRUB_NEXT_NOLMA 8 /* the inode has no FID-in-LMA */
66 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
68 return container_of0(scrub, struct osd_device, od_scrub);
71 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
73 return osd_sb(osd_scrub2dev(scrub));
76 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
77 struct osd_otable_cache *ooc)
79 return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
82 static int osd_scrub_refresh_mapping(struct osd_thread_info *info,
83 struct osd_device *dev,
84 const struct lu_fid *fid,
85 const struct osd_inode_id *id, int ops)
87 struct lu_fid *oi_fid = &info->oti_fid2;
88 struct osd_inode_id *oi_id = &info->oti_id2;
89 struct iam_container *bag;
90 struct iam_path_descr *ipd;
95 fid_cpu_to_be(oi_fid, fid);
96 osd_id_pack(oi_id, id);
97 jh = ldiskfs_journal_start_sb(osd_sb(dev),
98 osd_dto_credits_noquota[ops]);
101 CERROR("%.16s: fail to start trans for scrub store: rc = %d\n",
102 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
106 bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
107 ipd = osd_idx_ipd_get(info->oti_env, bag);
108 if (unlikely(ipd == NULL)) {
109 ldiskfs_journal_stop(jh);
110 CERROR("%.16s: fail to get ipd for scrub store\n",
111 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
115 if (ops == DTO_INDEX_UPDATE) {
116 rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
117 (struct iam_rec *)oi_id, ipd);
119 rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
120 (struct iam_rec *)oi_id, ipd);
123 /* XXX: There are trouble things when adding OI
124 * mapping for IGIF object, which may cause
125 * multiple objects to be mapped to the same
126 * IGIF formatted FID. Consider the following
129 * 1) The MDT is upgrading from 1.8 device.
130 * The OI scrub generates IGIF FID1 for the
131 * OBJ1 and adds the OI mapping.
133 * 2) For some reason, the OI scrub does not
134 * process all the IGIF objects completely.
136 * 3) The MDT is backuped and restored against
139 * 4) When the MDT mounts up, the OI scrub will
140 * try to rebuild the OI files. For some IGIF
141 * object, OBJ2, which was not processed by the
142 * OI scrub before the backup/restore, and the
143 * new generated IGIF formatted FID may be just
144 * the FID1, the same as OBJ1.
146 * Under such case, the OI scrub cannot know how
147 * to generate new FID for the OBJ2.
149 * Currently, we do nothing for that. One possible
150 * solution is to generate new normal FID for the
153 * Anyway, it is rare, only exists in theory. */
156 osd_ipd_put(info->oti_env, bag, ipd);
157 ldiskfs_journal_stop(jh);
161 /* OI_scrub file ops */
163 static void osd_scrub_file_to_cpu(struct scrub_file *des,
164 struct scrub_file *src)
166 memcpy(des->sf_uuid, src->sf_uuid, 16);
167 des->sf_flags = le64_to_cpu(src->sf_flags);
168 des->sf_magic = le32_to_cpu(src->sf_magic);
169 des->sf_status = le16_to_cpu(src->sf_status);
170 des->sf_param = le16_to_cpu(src->sf_param);
171 des->sf_time_last_complete =
172 le64_to_cpu(src->sf_time_last_complete);
173 des->sf_time_latest_start =
174 le64_to_cpu(src->sf_time_latest_start);
175 des->sf_time_last_checkpoint =
176 le64_to_cpu(src->sf_time_last_checkpoint);
177 des->sf_pos_latest_start =
178 le64_to_cpu(src->sf_pos_latest_start);
179 des->sf_pos_last_checkpoint =
180 le64_to_cpu(src->sf_pos_last_checkpoint);
181 des->sf_pos_first_inconsistent =
182 le64_to_cpu(src->sf_pos_first_inconsistent);
183 des->sf_items_checked =
184 le64_to_cpu(src->sf_items_checked);
185 des->sf_items_updated =
186 le64_to_cpu(src->sf_items_updated);
187 des->sf_items_failed =
188 le64_to_cpu(src->sf_items_failed);
189 des->sf_items_updated_prior =
190 le64_to_cpu(src->sf_items_updated_prior);
191 des->sf_run_time = le32_to_cpu(src->sf_run_time);
192 des->sf_success_count = le32_to_cpu(src->sf_success_count);
193 des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
194 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
197 static void osd_scrub_file_to_le(struct scrub_file *des,
198 struct scrub_file *src)
200 memcpy(des->sf_uuid, src->sf_uuid, 16);
201 des->sf_flags = cpu_to_le64(src->sf_flags);
202 des->sf_magic = cpu_to_le32(src->sf_magic);
203 des->sf_status = cpu_to_le16(src->sf_status);
204 des->sf_param = cpu_to_le16(src->sf_param);
205 des->sf_time_last_complete =
206 cpu_to_le64(src->sf_time_last_complete);
207 des->sf_time_latest_start =
208 cpu_to_le64(src->sf_time_latest_start);
209 des->sf_time_last_checkpoint =
210 cpu_to_le64(src->sf_time_last_checkpoint);
211 des->sf_pos_latest_start =
212 cpu_to_le64(src->sf_pos_latest_start);
213 des->sf_pos_last_checkpoint =
214 cpu_to_le64(src->sf_pos_last_checkpoint);
215 des->sf_pos_first_inconsistent =
216 cpu_to_le64(src->sf_pos_first_inconsistent);
217 des->sf_items_checked =
218 cpu_to_le64(src->sf_items_checked);
219 des->sf_items_updated =
220 cpu_to_le64(src->sf_items_updated);
221 des->sf_items_failed =
222 cpu_to_le64(src->sf_items_failed);
223 des->sf_items_updated_prior =
224 cpu_to_le64(src->sf_items_updated_prior);
225 des->sf_run_time = cpu_to_le32(src->sf_run_time);
226 des->sf_success_count = cpu_to_le32(src->sf_success_count);
227 des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
228 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
231 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
233 struct scrub_file *sf = &scrub->os_file;
235 memset(sf, 0, sizeof(*sf));
236 memcpy(sf->sf_uuid, uuid, 16);
237 sf->sf_magic = SCRUB_MAGIC_V1;
238 sf->sf_status = SS_INIT;
241 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
243 struct scrub_file *sf = &scrub->os_file;
245 CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
246 memcpy(sf->sf_uuid, uuid, 16);
247 sf->sf_status = SS_INIT;
248 sf->sf_flags |= flags;
251 sf->sf_time_latest_start = 0;
252 sf->sf_time_last_checkpoint = 0;
253 sf->sf_pos_latest_start = 0;
254 sf->sf_pos_last_checkpoint = 0;
255 sf->sf_pos_first_inconsistent = 0;
256 sf->sf_items_checked = 0;
257 sf->sf_items_updated = 0;
258 sf->sf_items_failed = 0;
259 sf->sf_items_updated_prior = 0;
260 sf->sf_items_noscrub = 0;
261 sf->sf_items_igif = 0;
264 static int osd_scrub_file_load(struct osd_scrub *scrub)
267 char *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
268 int len = sizeof(scrub->os_file_disk);
271 rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
273 struct scrub_file *sf = &scrub->os_file;
275 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
276 if (sf->sf_magic != SCRUB_MAGIC_V1) {
277 CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
278 name, sf->sf_magic, SCRUB_MAGIC_V1);
279 /* Process it as new scrub file. */
284 } else if (rc != 0) {
285 CERROR("%.16s: fail to load scrub file, expected = %d, "
286 "rc = %d\n", name, len, rc);
290 /* return -ENOENT for empty scrub file case. */
297 int osd_scrub_file_store(struct osd_scrub *scrub)
299 struct osd_device *dev;
302 int len = sizeof(scrub->os_file_disk);
306 dev = container_of0(scrub, struct osd_device, od_scrub);
307 credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
308 osd_dto_credits_noquota[DTO_WRITE_BLOCK];
309 jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
312 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
313 LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
317 osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
318 rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
320 ldiskfs_journal_stop(jh);
322 CERROR("%.16s: fail to store scrub file, expected = %d, "
324 LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
326 scrub->os_time_last_checkpoint = cfs_time_current();
327 scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
328 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
334 static int osd_scrub_prep(struct osd_device *dev)
336 struct osd_scrub *scrub = &dev->od_scrub;
337 struct ptlrpc_thread *thread = &scrub->os_thread;
338 struct scrub_file *sf = &scrub->os_file;
339 __u32 flags = scrub->os_start_flags;
343 down_write(&scrub->os_rwsem);
344 if (flags & SS_SET_FAILOUT)
345 sf->sf_param |= SP_FAILOUT;
347 if (flags & SS_CLEAR_FAILOUT)
348 sf->sf_param &= ~SP_FAILOUT;
350 if (flags & SS_RESET)
351 osd_scrub_file_reset(scrub,
352 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, 0);
354 if (flags & SS_AUTO) {
355 scrub->os_full_speed = 1;
356 sf->sf_flags |= SF_AUTO;
358 scrub->os_full_speed = 0;
361 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))
362 scrub->os_full_speed = 1;
364 scrub->os_in_prior = 0;
365 scrub->os_waiting = 0;
366 scrub->os_paused = 0;
367 scrub->os_new_checked = 0;
368 if (sf->sf_pos_last_checkpoint != 0)
369 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
371 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
373 scrub->os_pos_current = sf->sf_pos_latest_start;
374 sf->sf_status = SS_SCANNING;
375 sf->sf_time_latest_start = cfs_time_current_sec();
376 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
377 rc = osd_scrub_file_store(scrub);
379 spin_lock(&scrub->os_lock);
380 thread_set_flags(thread, SVC_RUNNING);
381 spin_unlock(&scrub->os_lock);
382 cfs_waitq_broadcast(&thread->t_ctl_waitq);
384 up_write(&scrub->os_rwsem);
390 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
391 struct osd_idmap_cache *oic, int val)
393 struct osd_scrub *scrub = &dev->od_scrub;
394 struct scrub_file *sf = &scrub->os_file;
395 struct lu_fid *fid = &oic->oic_fid;
396 struct osd_inode_id *lid = &oic->oic_lid;
397 struct osd_inode_id *lid2 = &info->oti_id;
398 struct osd_inconsistent_item *oii = NULL;
399 struct inode *inode = NULL;
400 int ops = DTO_INDEX_UPDATE;
405 down_write(&scrub->os_rwsem);
406 scrub->os_new_checked++;
410 if (scrub->os_in_prior)
411 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
414 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
417 if (fid_is_igif(fid))
420 if (val == SCRUB_NEXT_NOLMA && !dev->od_handle_nolma)
423 if ((oii != NULL && oii->oii_insert) || (val == SCRUB_NEXT_NOLMA))
426 /* XXX: Currently, no FID-in-LMA for OST object, so osd_oi_lookup()
427 * without checking FLD is enough.
429 * It should be updated if FID-in-LMA for OSD object introduced
431 rc = osd_oi_lookup(info, dev, fid, lid2, false);
437 inode = osd_iget(info, dev, lid);
440 /* Someone removed the inode. */
441 if (rc == -ENOENT || rc == -ESTALE)
446 /* Prevent the inode to be unlinked during OI scrub. */
447 mutex_lock(&inode->i_mutex);
448 if (unlikely(inode->i_nlink == 0)) {
449 mutex_unlock(&inode->i_mutex);
454 ops = DTO_INDEX_INSERT;
455 idx = osd_oi_fid2idx(dev, fid);
456 if (val == SCRUB_NEXT_NOLMA) {
457 rc = osd_ea_fid_set(info, inode, fid);
461 sf->sf_flags |= SF_RECREATED | SF_INCONSISTENT;
462 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
463 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
465 } else if (osd_id_eq(lid, lid2)) {
468 sf->sf_flags |= SF_INCONSISTENT;
471 rc = osd_scrub_refresh_mapping(info, dev, fid, lid, ops);
473 if (scrub->os_in_prior)
474 sf->sf_items_updated_prior++;
476 sf->sf_items_updated++;
483 sf->sf_items_failed++;
484 if (sf->sf_pos_first_inconsistent == 0 ||
485 sf->sf_pos_first_inconsistent > lid->oii_ino)
486 sf->sf_pos_first_inconsistent = lid->oii_ino;
491 if (ops == DTO_INDEX_INSERT) {
492 mutex_unlock(&inode->i_mutex);
495 up_write(&scrub->os_rwsem);
498 LASSERT(!cfs_list_empty(&oii->oii_list));
500 spin_lock(&scrub->os_lock);
501 cfs_list_del_init(&oii->oii_list);
502 spin_unlock(&scrub->os_lock);
505 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
508 static int osd_scrub_checkpoint(struct osd_scrub *scrub)
510 struct scrub_file *sf = &scrub->os_file;
513 if (likely(cfs_time_before(cfs_time_current(),
514 scrub->os_time_next_checkpoint) ||
515 scrub->os_new_checked == 0))
518 down_write(&scrub->os_rwsem);
519 sf->sf_items_checked += scrub->os_new_checked;
520 scrub->os_new_checked = 0;
521 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
522 sf->sf_time_last_checkpoint = cfs_time_current_sec();
523 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
524 scrub->os_time_last_checkpoint);
525 rc = osd_scrub_file_store(scrub);
526 up_write(&scrub->os_rwsem);
531 static void osd_scrub_post(struct osd_scrub *scrub, int result)
533 struct scrub_file *sf = &scrub->os_file;
536 down_write(&scrub->os_rwsem);
537 spin_lock(&scrub->os_lock);
538 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
539 spin_unlock(&scrub->os_lock);
540 if (scrub->os_new_checked > 0) {
541 sf->sf_items_checked += scrub->os_new_checked;
542 scrub->os_new_checked = 0;
543 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
545 sf->sf_time_last_checkpoint = cfs_time_current_sec();
547 sf->sf_status = SS_COMPLETED;
548 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
549 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
550 SF_UPGRADE | SF_AUTO);
551 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
552 sf->sf_success_count++;
553 } else if (result == 0) {
554 if (scrub->os_paused)
555 sf->sf_status = SS_PAUSED;
557 sf->sf_status = SS_STOPPED;
559 sf->sf_status = SS_FAILED;
561 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
562 scrub->os_time_last_checkpoint);
563 result = osd_scrub_file_store(scrub);
565 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
566 LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
568 up_write(&scrub->os_rwsem);
573 /* iteration engine */
575 struct osd_iit_param {
576 struct super_block *sb;
577 struct buffer_head *bitmap;
583 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
584 struct osd_device *dev,
585 struct osd_iit_param *param,
586 struct osd_idmap_cache **oic,
589 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
590 struct osd_device *dev,
591 struct osd_iit_param *param,
592 struct osd_idmap_cache *oic,
593 int *noslot, int rc);
595 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
597 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
598 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
599 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
600 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
601 return SCRUB_NEXT_BREAK;
603 *pos = param->gbase + param->offset;
608 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
609 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
610 struct super_block *sb, bool scrub)
612 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
616 osd_id_gen(lid, pos, OSD_OII_NOGEN);
617 inode = osd_iget(info, dev, lid);
620 /* The inode may be removed after bitmap searching, or the
621 * file is new created without inode initialized yet. */
622 if (rc == -ENOENT || rc == -ESTALE)
623 return SCRUB_NEXT_CONTINUE;
625 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
626 LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
630 /* If the inode has no OI mapping, then it is special locally used,
631 * should be invisible to OI scrub or up layer LFSCK. */
632 if (ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NO_OI)) {
634 return SCRUB_NEXT_CONTINUE;
638 ldiskfs_test_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB)) {
639 /* Only skip it for the first OI scrub accessing. */
640 ldiskfs_clear_inode_state(inode, LDISKFS_STATE_LUSTRE_NOSCRUB);
642 return SCRUB_NEXT_NOSCRUB;
645 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
648 if (!fid_is_client_visible(&lma->lma_self_fid))
649 rc = SCRUB_NEXT_CONTINUE;
651 *fid = lma->lma_self_fid;
653 } else if (rc == -ENODATA) {
654 lu_igif_build(fid, inode->i_ino, inode->i_generation);
656 rc = SCRUB_NEXT_NOLMA;
664 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
665 struct osd_iit_param *param,
666 struct osd_idmap_cache **oic, int noslot)
668 struct osd_scrub *scrub = &dev->od_scrub;
669 struct ptlrpc_thread *thread = &scrub->os_thread;
671 struct osd_inode_id *lid;
674 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
675 struct l_wait_info lwi;
677 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
678 l_wait_event(thread->t_ctl_waitq,
679 !cfs_list_empty(&scrub->os_inconsistent_items) ||
680 !thread_is_running(thread),
684 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
685 spin_lock(&scrub->os_lock);
686 thread_set_flags(thread, SVC_STOPPING);
687 spin_unlock(&scrub->os_lock);
688 return SCRUB_NEXT_CRASH;
691 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
692 return SCRUB_NEXT_FATAL;
694 if (unlikely(!thread_is_running(thread)))
695 return SCRUB_NEXT_EXIT;
697 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
698 struct osd_inconsistent_item *oii;
700 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
701 struct osd_inconsistent_item, oii_list);
702 *oic = &oii->oii_cache;
703 scrub->os_in_prior = 1;
708 return SCRUB_NEXT_WAIT;
710 rc = osd_iit_next(param, &scrub->os_pos_current);
714 *oic = &scrub->os_oic;
715 fid = &(*oic)->oic_fid;
716 lid = &(*oic)->oic_lid;
717 rc = osd_iit_iget(info, dev, fid, lid,
718 scrub->os_pos_current, param->sb, true);
722 static int osd_preload_next(struct osd_thread_info *info,
723 struct osd_device *dev, struct osd_iit_param *param,
724 struct osd_idmap_cache **oic, int noslot)
726 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
727 struct osd_scrub *scrub;
728 struct ptlrpc_thread *thread;
731 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
735 scrub = &dev->od_scrub;
736 thread = &scrub->os_thread;
737 if (thread_is_running(thread) &&
738 ooc->ooc_pos_preload >= scrub->os_pos_current)
739 return SCRUB_NEXT_EXIT;
741 rc = osd_iit_iget(info, dev,
742 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
743 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
744 ooc->ooc_pos_preload, param->sb, false);
745 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
746 * ignore the failure, so it still need to skip the inode next time. */
747 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
751 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
752 struct osd_iit_param *param,
753 struct osd_idmap_cache *oic, int *noslot, int rc)
755 struct l_wait_info lwi = { 0 };
756 struct osd_scrub *scrub = &dev->od_scrub;
757 struct scrub_file *sf = &scrub->os_file;
758 struct ptlrpc_thread *thread = &scrub->os_thread;
759 struct osd_otable_it *it = dev->od_otable_it;
760 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
763 case SCRUB_NEXT_CONTINUE:
765 case SCRUB_NEXT_WAIT:
767 case SCRUB_NEXT_NOSCRUB:
768 down_write(&scrub->os_rwsem);
769 scrub->os_new_checked++;
770 sf->sf_items_noscrub++;
771 up_write(&scrub->os_rwsem);
775 rc = osd_scrub_check_update(info, dev, oic, rc);
779 rc = osd_scrub_checkpoint(scrub);
781 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
782 LDISKFS_SB(param->sb)->s_es->s_volume_name,
783 scrub->os_pos_current, rc);
784 /* Continue, as long as the scrub itself can go ahead. */
787 if (scrub->os_in_prior) {
788 scrub->os_in_prior = 0;
793 scrub->os_pos_current = param->gbase + ++(param->offset);
794 if (it != NULL && it->ooi_waiting &&
795 ooc->ooc_pos_preload < scrub->os_pos_current) {
797 cfs_waitq_broadcast(&thread->t_ctl_waitq);
800 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
804 if (osd_scrub_has_window(scrub, ooc)) {
809 scrub->os_waiting = 1;
810 l_wait_event(thread->t_ctl_waitq,
811 osd_scrub_has_window(scrub, ooc) ||
812 !cfs_list_empty(&scrub->os_inconsistent_items) ||
813 !thread_is_running(thread),
815 scrub->os_waiting = 0;
817 if (osd_scrub_has_window(scrub, ooc))
824 static int osd_preload_exec(struct osd_thread_info *info,
825 struct osd_device *dev, struct osd_iit_param *param,
826 struct osd_idmap_cache *oic, int *noslot, int rc)
828 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
831 ooc->ooc_cached_items++;
832 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
833 ~OSD_OTABLE_IT_CACHE_MASK;
835 return rc > 0 ? 0 : rc;
838 #define SCRUB_IT_ALL 1
839 #define SCRUB_IT_CRASH 2
841 static int osd_inode_iteration(struct osd_thread_info *info,
842 struct osd_device *dev, __u32 max, bool preload)
844 osd_iit_next_policy next;
845 osd_iit_exec_policy exec;
848 struct osd_iit_param param;
855 struct osd_scrub *scrub = &dev->od_scrub;
857 next = osd_scrub_next;
858 exec = osd_scrub_exec;
859 pos = &scrub->os_pos_current;
860 count = &scrub->os_new_checked;
862 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
864 next = osd_preload_next;
865 exec = osd_preload_exec;
866 pos = &ooc->ooc_pos_preload;
867 count = &ooc->ooc_cached_items;
869 param.sb = osd_sb(dev);
870 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
872 while (*pos <= limit && *count < max) {
873 struct osd_idmap_cache *oic = NULL;
875 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
876 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
877 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
878 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
879 if (param.bitmap == NULL) {
880 CERROR("%.16s: fail to read bitmap for %u, "
881 "scrub will stop, urgent mode\n",
882 LDISKFS_SB(param.sb)->s_es->s_volume_name,
887 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
889 rc = next(info, dev, ¶m, &oic, noslot);
891 case SCRUB_NEXT_BREAK:
893 case SCRUB_NEXT_EXIT:
894 brelse(param.bitmap);
896 case SCRUB_NEXT_CRASH:
897 brelse(param.bitmap);
898 RETURN(SCRUB_IT_CRASH);
899 case SCRUB_NEXT_FATAL:
900 brelse(param.bitmap);
904 rc = exec(info, dev, ¶m, oic, &noslot, rc);
906 brelse(param.bitmap);
910 if (preload && dev->od_otable_it->ooi_stopping) {
911 brelse(param.bitmap);
917 brelse(param.bitmap);
919 if (preload && dev->od_otable_it->ooi_stopping)
924 RETURN(SCRUB_IT_ALL);
928 static int osd_otable_it_preload(const struct lu_env *env,
929 struct osd_otable_it *it)
931 struct osd_device *dev = it->ooi_dev;
932 struct osd_scrub *scrub = &dev->od_scrub;
933 struct osd_otable_cache *ooc = &it->ooi_cache;
937 rc = osd_inode_iteration(osd_oti_get(env), dev,
938 OSD_OTABLE_IT_CACHE_SIZE, true);
939 if (rc == SCRUB_IT_ALL)
940 it->ooi_all_cached = 1;
942 CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
943 le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
944 ooc->ooc_pos_preload, rc);
946 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
947 scrub->os_waiting = 0;
948 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
951 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
954 static int osd_scrub_main(void *args)
957 struct osd_device *dev = (struct osd_device *)args;
958 struct osd_scrub *scrub = &dev->od_scrub;
959 struct ptlrpc_thread *thread = &scrub->os_thread;
960 struct super_block *sb = osd_sb(dev);
964 cfs_daemonize("OI_scrub");
965 rc = lu_env_init(&env, LCT_DT_THREAD);
967 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
968 LDISKFS_SB(sb)->s_es->s_volume_name, rc);
972 rc = osd_scrub_prep(dev);
974 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
975 LDISKFS_SB(sb)->s_es->s_volume_name, rc);
979 if (!scrub->os_full_speed) {
980 struct l_wait_info lwi = { 0 };
981 struct osd_otable_it *it = dev->od_otable_it;
982 struct osd_otable_cache *ooc = &it->ooi_cache;
984 l_wait_event(thread->t_ctl_waitq,
985 it->ooi_user_ready || !thread_is_running(thread),
987 if (unlikely(!thread_is_running(thread)))
990 LASSERT(scrub->os_pos_current >= ooc->ooc_pos_preload);
991 scrub->os_pos_current = ooc->ooc_pos_preload;
994 CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
995 scrub->os_start_flags, scrub->os_pos_current);
997 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, false);
998 if (unlikely(rc == SCRUB_IT_CRASH))
999 GOTO(out, rc = -EINVAL);
1003 osd_scrub_post(scrub, rc);
1004 CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
1005 rc, scrub->os_pos_current);
1008 while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
1009 struct osd_inconsistent_item *oii;
1011 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
1012 struct osd_inconsistent_item, oii_list);
1013 cfs_list_del_init(&oii->oii_list);
1019 spin_lock(&scrub->os_lock);
1020 thread_set_flags(thread, SVC_STOPPED);
1021 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1022 spin_unlock(&scrub->os_lock);
1026 /* initial OI scrub */
1028 typedef int (*scandir_t)(struct osd_thread_info *, struct osd_device *,
1029 struct dentry *, filldir_t filldir);
1031 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1032 loff_t offset, __u64 ino, unsigned d_type);
1035 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1036 struct dentry *dentry, filldir_t filldir);
1038 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1039 struct dentry *dentry, filldir_t filldir);
1042 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1043 struct dentry *dentry, filldir_t filldir);
1046 OLF_SCAN_SUBITEMS = 0x0001,
1047 OLF_HIDE_FID = 0x0002,
1048 OLF_SHOW_NAME = 0x0004,
1053 struct lu_fid olm_fid;
1055 scandir_t olm_scandir;
1056 filldir_t olm_filldir;
1059 /* Add the new introduced local files in the list in the future. */
1060 static const struct osd_lf_map osd_lf_maps[] = {
1062 { CATLIST, { FID_SEQ_LOCAL_FILE, LLOG_CATALOGS_OID, 0 }, OLF_SHOW_NAME,
1066 { MOUNT_CONFIGS_DIR, { FID_SEQ_LOCAL_FILE, MGS_CONFIGS_OID, 0 },
1067 OLF_SCAN_SUBITEMS, osd_ios_general_scan,
1068 osd_ios_varfid_fill },
1070 /* NIDTBL_VERSIONS */
1071 { MGS_NIDTBL_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1072 osd_ios_general_scan, osd_ios_varfid_fill },
1075 { "PENDING", { FID_SEQ_LOCAL_FILE, MDD_ORPHAN_OID, 0 }, 0, NULL, NULL },
1078 { "ROOT", { FID_SEQ_LOCAL_FILE, MDD_ROOT_INDEX_OID, 0 },
1079 OLF_SCAN_SUBITEMS | OLF_HIDE_FID, osd_ios_ROOT_scan, NULL },
1082 { CAPA_KEYS, { FID_SEQ_LOCAL_FILE, MDD_CAPA_KEYS_OID, 0 }, 0,
1085 /* changelog_catalog */
1086 { CHANGELOG_CATALOG, { 0, 0, 0 }, 0, NULL, NULL },
1088 /* changelog_users */
1089 { CHANGELOG_USERS, { 0, 0, 0 }, 0, NULL, NULL },
1092 { "fld", { FID_SEQ_LOCAL_FILE, FLD_INDEX_OID, 0 }, OLF_SHOW_NAME,
1096 { LAST_RCVD, { FID_SEQ_LOCAL_FILE, LAST_RECV_OID, 0 }, OLF_SHOW_NAME,
1099 /* lfsck_bookmark */
1100 { "lfsck_bookmark", { FID_SEQ_LOCAL_FILE, LFSCK_BOOKMARK_OID, 0 }, 0,
1104 { LOV_OBJID, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OID, 0 }, OLF_SHOW_NAME,
1108 { LOV_OBJSEQ, { FID_SEQ_LOCAL_FILE, MDD_LOV_OBJ_OSEQ, 0 },
1109 OLF_SHOW_NAME, NULL, NULL },
1112 { QMT_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1113 osd_ios_general_scan, osd_ios_varfid_fill },
1116 { QSD_DIR, { 0, 0, 0 }, OLF_SCAN_SUBITEMS,
1117 osd_ios_general_scan, osd_ios_varfid_fill },
1119 /* seq-200000003-lastid */
1120 { "seq-200000003-lastid", { FID_SEQ_LOCAL_NAME, 1, 0 }, 0,
1124 { "seq_ctl", { FID_SEQ_LOCAL_FILE, FID_SEQ_CTL_OID, 0 },
1125 OLF_SHOW_NAME, NULL, NULL },
1128 { "seq_srv", { FID_SEQ_LOCAL_FILE, FID_SEQ_SRV_OID, 0 },
1129 OLF_SHOW_NAME, NULL, NULL },
1132 { "LAST_GROUP", { FID_SEQ_LOCAL_FILE, OFD_LAST_GROUP_OID, 0 },
1133 OLF_SHOW_NAME, NULL, NULL },
1136 { HEALTH_CHECK, { FID_SEQ_LOCAL_FILE, OFD_HEALTH_CHECK_OID, 0 },
1137 OLF_SHOW_NAME, NULL, NULL },
1139 /* lfsck_namespace */
1140 { "lfsck_namespace", { FID_SEQ_LOCAL_FILE, LFSCK_NAMESPACE_OID, 0 }, 0,
1143 /* OBJECTS, upgrade from old device */
1144 { OBJECTS, { 0, 0, 0 }, OLF_SCAN_SUBITEMS, osd_ios_OBJECTS_scan, NULL },
1146 /* lquota_v2.user, upgrade from old device */
1147 { "lquota_v2.user", { 0, 0, 0 }, 0, NULL, NULL },
1149 /* lquota_v2.group, upgrade from old device */
1150 { "lquota_v2.group", { 0, 0, 0 }, 0, NULL, NULL },
1152 { NULL, { 0, 0, 0 }, 0, NULL, NULL }
1155 struct osd_ios_item {
1156 cfs_list_t oii_list;
1157 struct dentry *oii_dentry;
1158 scandir_t oii_scandir;
1159 filldir_t oii_filldir;
1162 struct osd_ios_filldir_buf {
1163 struct osd_thread_info *oifb_info;
1164 struct osd_device *oifb_dev;
1165 struct dentry *oifb_dentry;
1168 static inline struct dentry *
1169 osd_ios_lookup_one_len(const char *name, struct dentry *parent, int namelen)
1171 struct dentry *dentry;
1173 dentry = ll_lookup_one_len(name, parent, namelen);
1174 if (!IS_ERR(dentry) && dentry->d_inode == NULL) {
1176 return ERR_PTR(-ENOENT);
1183 osd_ios_llogname2fid(struct lu_fid *fid, const char *name, int namelen)
1188 fid->f_seq = FID_SEQ_LLOG;
1190 id = id * 10 + name[i++] - '0';
1192 fid->f_oid = id & 0x00000000ffffffffULL;
1193 fid->f_ver = id >> 32;
1197 osd_ios_Oname2fid(struct lu_fid *fid, const char *name, int namelen)
1203 seq = seq * 10 + name[i++] - '0';
1205 lu_last_id_fid(fid, seq);
1209 osd_ios_new_item(struct osd_device *dev, struct dentry *dentry,
1210 scandir_t scandir, filldir_t filldir)
1212 struct osd_ios_item *item;
1214 OBD_ALLOC_PTR(item);
1218 CFS_INIT_LIST_HEAD(&item->oii_list);
1219 item->oii_dentry = dget(dentry);
1220 item->oii_scandir = scandir;
1221 item->oii_filldir = filldir;
1222 cfs_list_add_tail(&item->oii_list, &dev->od_ios_list);
1227 * osd_ios_scan_one() - check/fix LMA FID and OI entry for one inode
1229 * The passed \a inode's \a fid is verified against the LMA FID. If the \a fid
1230 * is NULL or is empty the IGIF FID is used. The FID is verified in the OI to
1231 * reference the inode, or fixed if it is missing or references another inode.
1234 osd_ios_scan_one(struct osd_thread_info *info, struct osd_device *dev,
1235 struct inode *inode, const struct lu_fid *fid, int flags)
1237 struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
1238 struct osd_inode_id *id = &info->oti_id;
1239 struct osd_inode_id *id2 = &info->oti_id2;
1240 struct osd_scrub *scrub = &dev->od_scrub;
1241 struct scrub_file *sf = &scrub->os_file;
1246 rc = osd_get_lma(info, inode, &info->oti_obj_dentry, lma);
1247 if (rc != 0 && rc != -ENODATA)
1250 osd_id_gen(id, inode->i_ino, inode->i_generation);
1251 if (rc == -ENODATA) {
1252 if (fid == NULL || fid_is_zero(fid) || flags & OLF_HIDE_FID)
1253 lu_igif_build(&tfid, inode->i_ino, inode->i_generation);
1256 rc = osd_ea_fid_set(info, inode, &tfid);
1260 tfid = lma->lma_self_fid;
1263 rc = __osd_oi_lookup(info, dev, &tfid, id2);
1268 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id,
1273 if (osd_id_eq_strict(id, id2))
1276 if (!(sf->sf_flags & SF_INCONSISTENT)) {
1277 osd_scrub_file_reset(scrub,
1278 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1280 rc = osd_scrub_file_store(scrub);
1285 rc = osd_scrub_refresh_mapping(info, dev, &tfid, id, DTO_INDEX_UPDATE);
1290 static int osd_ios_varfid_fill(void *buf, const char *name, int namelen,
1291 loff_t offset, __u64 ino, unsigned d_type)
1293 struct osd_ios_filldir_buf *fill_buf = buf;
1294 struct osd_device *dev = fill_buf->oifb_dev;
1295 struct dentry *child;
1299 /* skip any '.' started names */
1303 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1305 RETURN(PTR_ERR(child));
1307 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1309 if (rc == 0 && S_ISDIR(child->d_inode->i_mode))
1310 rc = osd_ios_new_item(dev, child, osd_ios_general_scan,
1311 osd_ios_varfid_fill);
1317 static int osd_ios_root_fill(void *buf, const char *name, int namelen,
1318 loff_t offset, __u64 ino, unsigned d_type)
1320 struct osd_ios_filldir_buf *fill_buf = buf;
1321 struct osd_device *dev = fill_buf->oifb_dev;
1322 const struct osd_lf_map *map;
1323 struct dentry *child;
1327 /* skip any '.' started names */
1331 for (map = osd_lf_maps; map->olm_name != NULL; map++) {
1332 if (strlen(map->olm_name) != namelen)
1335 if (strncmp(map->olm_name, name, namelen) == 0)
1339 if (map->olm_name == NULL)
1342 child = osd_ios_lookup_one_len(name, fill_buf->oifb_dentry, namelen);
1344 RETURN(PTR_ERR(child));
1346 rc = osd_ios_scan_one(fill_buf->oifb_info, dev, child->d_inode,
1347 &map->olm_fid, map->olm_flags);
1348 if (rc == 0 && map->olm_flags & OLF_SCAN_SUBITEMS)
1349 rc = osd_ios_new_item(dev, child, map->olm_scandir,
1357 osd_ios_general_scan(struct osd_thread_info *info, struct osd_device *dev,
1358 struct dentry *dentry, filldir_t filldir)
1360 struct osd_ios_filldir_buf buf = { info, dev, dentry };
1361 struct file *filp = &info->oti_it_ea.oie_file;
1362 struct inode *inode = dentry->d_inode;
1363 const struct file_operations *fops = inode->i_fop;
1367 LASSERT(filldir != NULL);
1370 filp->f_dentry = dentry;
1371 filp->f_mode = FMODE_64BITHASH;
1372 filp->f_mapping = inode->i_mapping;
1374 filp->private_data = NULL;
1376 rc = fops->readdir(filp, &buf, filldir);
1377 fops->release(inode, filp);
1383 osd_ios_ROOT_scan(struct osd_thread_info *info, struct osd_device *dev,
1384 struct dentry *dentry, filldir_t filldir)
1386 struct osd_scrub *scrub = &dev->od_scrub;
1387 struct scrub_file *sf = &scrub->os_file;
1388 struct dentry *child;
1392 /* It is existing MDT device. */
1393 dev->od_handle_nolma = 1;
1394 child = osd_ios_lookup_one_len(dot_lustre_name, dentry,
1395 strlen(dot_lustre_name));
1396 if (IS_ERR(child)) {
1397 rc = PTR_ERR(child);
1398 if (rc == -ENOENT) {
1399 /* It is 1.8 MDT device. */
1400 if (!(sf->sf_flags & SF_UPGRADE)) {
1401 osd_scrub_file_reset(scrub,
1402 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid,
1404 rc = osd_scrub_file_store(scrub);
1410 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1418 osd_ios_OBJECTS_scan(struct osd_thread_info *info, struct osd_device *dev,
1419 struct dentry *dentry, filldir_t filldir)
1421 struct dentry *child;
1425 child = osd_ios_lookup_one_len(ADMIN_USR, dentry, strlen(ADMIN_USR));
1426 if (!IS_ERR(child)) {
1427 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1430 rc = PTR_ERR(child);
1433 if (rc != 0 && rc != -ENOENT)
1436 child = osd_ios_lookup_one_len(ADMIN_GRP, dentry, strlen(ADMIN_GRP));
1437 if (!IS_ERR(child)) {
1438 rc = osd_ios_scan_one(info, dev, child->d_inode, NULL, 0);
1441 rc = PTR_ERR(child);
1450 static int osd_initial_OI_scrub(struct osd_thread_info *info,
1451 struct osd_device *dev)
1453 struct osd_ios_item *item = NULL;
1454 scandir_t scandir = osd_ios_general_scan;
1455 filldir_t filldir = osd_ios_root_fill;
1456 struct dentry *dentry = osd_sb(dev)->s_root;
1461 rc = scandir(info, dev, dentry, filldir);
1463 dput(item->oii_dentry);
1470 if (cfs_list_empty(&dev->od_ios_list))
1473 item = cfs_list_entry(dev->od_ios_list.next,
1474 struct osd_ios_item, oii_list);
1475 cfs_list_del_init(&item->oii_list);
1477 LASSERT(item->oii_scandir != NULL);
1478 scandir = item->oii_scandir;
1479 filldir = item->oii_filldir;
1480 dentry = item->oii_dentry;
1483 while (!cfs_list_empty(&dev->od_ios_list)) {
1484 item = cfs_list_entry(dev->od_ios_list.next,
1485 struct osd_ios_item, oii_list);
1486 cfs_list_del_init(&item->oii_list);
1487 dput(item->oii_dentry);
1494 char *osd_lf_fid2name(const struct lu_fid *fid)
1496 const struct osd_lf_map *map = osd_lf_maps;
1498 while (map->olm_name != NULL) {
1499 if (!lu_fid_eq(fid, &map->olm_fid)) {
1504 if (map->olm_flags & OLF_SHOW_NAME)
1505 return map->olm_name;
1513 /* OI scrub start/stop */
1515 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
1517 struct osd_scrub *scrub = &dev->od_scrub;
1518 struct ptlrpc_thread *thread = &scrub->os_thread;
1519 struct l_wait_info lwi = { 0 };
1524 /* os_lock: sync status between stop and scrub thread */
1525 spin_lock(&scrub->os_lock);
1526 if (thread_is_running(thread)) {
1527 spin_unlock(&scrub->os_lock);
1529 } else if (unlikely(thread_is_stopping(thread))) {
1530 spin_unlock(&scrub->os_lock);
1531 l_wait_event(thread->t_ctl_waitq,
1532 thread_is_stopped(thread),
1536 spin_unlock(&scrub->os_lock);
1538 if (scrub->os_file.sf_status == SS_COMPLETED)
1541 scrub->os_start_flags = flags;
1542 thread_set_flags(thread, 0);
1543 rc = cfs_create_thread(osd_scrub_main, dev, 0);
1545 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
1546 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
1550 l_wait_event(thread->t_ctl_waitq,
1551 thread_is_running(thread) || thread_is_stopped(thread),
1557 int osd_scrub_start(struct osd_device *dev)
1562 /* od_otable_mutex: prevent curcurrent start/stop */
1563 mutex_lock(&dev->od_otable_mutex);
1564 rc = do_osd_scrub_start(dev, SS_AUTO);
1565 mutex_unlock(&dev->od_otable_mutex);
1567 RETURN(rc == -EALREADY ? 0 : rc);
1570 static void do_osd_scrub_stop(struct osd_scrub *scrub)
1572 struct ptlrpc_thread *thread = &scrub->os_thread;
1573 struct l_wait_info lwi = { 0 };
1575 /* os_lock: sync status between stop and scrub thread */
1576 spin_lock(&scrub->os_lock);
1577 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
1578 thread_set_flags(thread, SVC_STOPPING);
1579 spin_unlock(&scrub->os_lock);
1580 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1581 l_wait_event(thread->t_ctl_waitq,
1582 thread_is_stopped(thread),
1584 /* Do not skip the last lock/unlock, which can guarantee that
1585 * the caller cannot return until the OI scrub thread exit. */
1586 spin_lock(&scrub->os_lock);
1588 spin_unlock(&scrub->os_lock);
1591 static void osd_scrub_stop(struct osd_device *dev)
1593 /* od_otable_mutex: prevent curcurrent start/stop */
1594 mutex_lock(&dev->od_otable_mutex);
1595 dev->od_scrub.os_paused = 1;
1596 do_osd_scrub_stop(&dev->od_scrub);
1597 mutex_unlock(&dev->od_otable_mutex);
1600 /* OI scrub setup/cleanup */
1602 static const char osd_scrub_name[] = "OI_scrub";
1604 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1606 struct osd_thread_info *info = osd_oti_get(env);
1607 struct osd_scrub *scrub = &dev->od_scrub;
1608 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
1609 struct scrub_file *sf = &scrub->os_file;
1610 struct super_block *sb = osd_sb(dev);
1611 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
1612 struct lvfs_run_ctxt saved;
1618 memset(scrub, 0, sizeof(*scrub));
1619 OBD_SET_CTXT_MAGIC(ctxt);
1620 ctxt->pwdmnt = dev->od_mnt;
1621 ctxt->pwd = dev->od_mnt->mnt_root;
1622 ctxt->fs = get_ds();
1624 cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1625 init_rwsem(&scrub->os_rwsem);
1626 spin_lock_init(&scrub->os_lock);
1627 CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1629 push_ctxt(&saved, ctxt, NULL);
1630 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1632 RETURN(PTR_ERR(filp));
1634 scrub->os_inode = igrab(filp->f_dentry->d_inode);
1635 filp_close(filp, 0);
1636 pop_ctxt(&saved, ctxt, NULL);
1637 ldiskfs_set_inode_state(scrub->os_inode,
1638 LDISKFS_STATE_LUSTRE_NO_OI);
1640 rc = osd_scrub_file_load(scrub);
1641 if (rc == -ENOENT) {
1642 osd_scrub_file_init(scrub, es->s_uuid);
1644 } else if (rc != 0) {
1647 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1648 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1650 } else if (sf->sf_status == SS_SCANNING) {
1651 sf->sf_status = SS_CRASHED;
1656 if (sf->sf_pos_last_checkpoint != 0)
1657 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1659 scrub->os_pos_current = LDISKFS_FIRST_INO(sb) + 1;
1662 rc = osd_scrub_file_store(scrub);
1667 /* Initialize OI files. */
1668 rc = osd_oi_init(info, dev);
1672 rc = osd_initial_OI_scrub(info, dev);
1673 if (rc == 0 && !dev->od_noscrub &&
1674 ((sf->sf_status == SS_PAUSED) ||
1675 (sf->sf_status == SS_CRASHED &&
1676 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE |
1678 (sf->sf_status == SS_INIT &&
1679 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_UPGRADE))))
1680 rc = osd_scrub_start(dev);
1685 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1687 struct osd_scrub *scrub = &dev->od_scrub;
1689 LASSERT(dev->od_otable_it == NULL);
1691 if (scrub->os_inode != NULL) {
1692 osd_scrub_stop(dev);
1693 iput(scrub->os_inode);
1694 scrub->os_inode = NULL;
1696 if (dev->od_oi_table != NULL)
1697 osd_oi_fini(osd_oti_get(env), dev);
1700 /* object table based iteration APIs */
1702 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1703 struct dt_object *dt, __u32 attr,
1704 struct lustre_capa *capa)
1706 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1707 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1708 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1709 struct osd_scrub *scrub = &dev->od_scrub;
1710 struct osd_otable_it *it;
1715 /* od_otable_mutex: prevent curcurrent init/fini */
1716 mutex_lock(&dev->od_otable_mutex);
1717 if (dev->od_otable_it != NULL)
1718 GOTO(out, it = ERR_PTR(-EALREADY));
1722 GOTO(out, it = ERR_PTR(-ENOMEM));
1724 dev->od_otable_it = it;
1726 it->ooi_pid = cfs_curproc_pid();
1727 it->ooi_cache.ooc_consumer_idx = -1;
1728 if (flags & DOIF_OUTUSED)
1729 it->ooi_used_outside = 1;
1731 if (flags & DOIF_RESET)
1734 if (valid & DOIV_ERROR_HANDLE) {
1735 if (flags & DOIF_FAILOUT)
1736 start |= SS_SET_FAILOUT;
1738 start |= SS_CLEAR_FAILOUT;
1741 rc = do_osd_scrub_start(dev, start);
1742 if (rc < 0 && rc != -EALREADY) {
1743 dev->od_otable_it = NULL;
1745 GOTO(out, it = ERR_PTR(rc));
1748 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
1753 mutex_unlock(&dev->od_otable_mutex);
1754 return (struct dt_it *)it;
1757 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1759 struct osd_otable_it *it = (struct osd_otable_it *)di;
1760 struct osd_device *dev = it->ooi_dev;
1762 /* od_otable_mutex: prevent curcurrent init/fini */
1763 mutex_lock(&dev->od_otable_mutex);
1764 do_osd_scrub_stop(&dev->od_scrub);
1765 LASSERT(dev->od_otable_it == it);
1767 dev->od_otable_it = NULL;
1768 mutex_unlock(&dev->od_otable_mutex);
1772 static int osd_otable_it_get(const struct lu_env *env,
1773 struct dt_it *di, const struct dt_key *key)
1781 * Sometimes the otable-based iteration driver (LFSCK) may be blocked in OSD
1782 * layer when someone wants to stop/pause the iteration. Under such case, we
1783 * need some mechanism to notify the event and wakeup the blocker.
1785 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1787 struct osd_otable_it *it = (struct osd_otable_it *)di;
1788 struct osd_device *dev = it->ooi_dev;
1790 /* od_otable_mutex: prevent curcurrent init/fini */
1791 mutex_lock(&dev->od_otable_mutex);
1792 if (it->ooi_pid == cfs_curproc_pid()) {
1793 dev->od_scrub.os_paused = 1;
1795 struct ptlrpc_thread *thread = &dev->od_scrub.os_thread;
1797 it->ooi_stopping = 1;
1798 if (it->ooi_waiting)
1799 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1801 mutex_unlock(&dev->od_otable_mutex);
1804 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1806 struct osd_otable_it *it = (struct osd_otable_it *)di;
1807 struct osd_device *dev = it->ooi_dev;
1808 struct osd_scrub *scrub = &dev->od_scrub;
1809 struct osd_otable_cache *ooc = &it->ooi_cache;
1810 struct ptlrpc_thread *thread = &scrub->os_thread;
1811 struct l_wait_info lwi = { 0 };
1815 LASSERT(it->ooi_user_ready);
1818 if (!thread_is_running(thread) && !it->ooi_used_outside)
1821 if (ooc->ooc_cached_items > 0) {
1822 ooc->ooc_cached_items--;
1823 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
1824 ~OSD_OTABLE_IT_CACHE_MASK;
1828 if (it->ooi_all_cached) {
1829 l_wait_event(thread->t_ctl_waitq,
1830 !thread_is_running(thread),
1835 it->ooi_waiting = 1;
1836 l_wait_event(thread->t_ctl_waitq,
1837 ooc->ooc_pos_preload < scrub->os_pos_current ||
1838 !thread_is_running(thread) ||
1841 it->ooi_waiting = 0;
1843 if (!thread_is_running(thread) && !it->ooi_used_outside)
1846 if (it->ooi_stopping)
1849 rc = osd_otable_it_preload(env, it);
1856 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1857 const struct dt_it *di)
1862 static int osd_otable_it_key_size(const struct lu_env *env,
1863 const struct dt_it *di)
1865 return sizeof(__u64);
1868 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1869 struct dt_rec *rec, __u32 attr)
1871 struct osd_otable_it *it = (struct osd_otable_it *)di;
1872 struct osd_otable_cache *ooc = &it->ooi_cache;
1874 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
1878 static __u64 osd_otable_it_store(const struct lu_env *env,
1879 const struct dt_it *di)
1881 struct osd_otable_it *it = (struct osd_otable_it *)di;
1882 struct osd_otable_cache *ooc = &it->ooi_cache;
1885 if (it->ooi_user_ready)
1886 hash = ooc->ooc_pos_preload;
1888 hash = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino;
1893 * Set the OSD layer iteration start position as the specified hash.
1895 static int osd_otable_it_load(const struct lu_env *env,
1896 const struct dt_it *di, __u64 hash)
1898 struct osd_otable_it *it = (struct osd_otable_it *)di;
1899 struct osd_device *dev = it->ooi_dev;
1900 struct osd_otable_cache *ooc = &it->ooi_cache;
1901 struct osd_scrub *scrub = &dev->od_scrub;
1905 /* Forbid to set iteration position after iteration started. */
1906 if (it->ooi_user_ready)
1909 if (hash > OSD_OTABLE_MAX_HASH)
1910 hash = OSD_OTABLE_MAX_HASH;
1912 /* Skip the one that has been processed last time. */
1913 if (ooc->ooc_pos_preload > hash)
1914 ooc->ooc_pos_preload = hash;
1916 if (ooc->ooc_pos_preload <= LDISKFS_FIRST_INO(osd_sb(dev)))
1917 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev)) + 1;
1919 it->ooi_user_ready = 1;
1920 if (!scrub->os_full_speed)
1921 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1923 /* Unplug OSD layer iteration by the first next() call. */
1924 rc = osd_otable_it_next(env, (struct dt_it *)it);
1929 static int osd_otable_it_key_rec(const struct lu_env *env,
1930 const struct dt_it *di, void *key_rec)
1935 const struct dt_index_operations osd_otable_ops = {
1937 .init = osd_otable_it_init,
1938 .fini = osd_otable_it_fini,
1939 .get = osd_otable_it_get,
1940 .put = osd_otable_it_put,
1941 .next = osd_otable_it_next,
1942 .key = osd_otable_it_key,
1943 .key_size = osd_otable_it_key_size,
1944 .rec = osd_otable_it_rec,
1945 .store = osd_otable_it_store,
1946 .load = osd_otable_it_load,
1947 .key_rec = osd_otable_it_key_rec,
1951 /* high priority inconsistent items list APIs */
1953 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
1956 struct osd_inconsistent_item *oii;
1957 struct osd_scrub *scrub = &dev->od_scrub;
1958 struct ptlrpc_thread *thread = &scrub->os_thread;
1963 if (unlikely(oii == NULL))
1966 CFS_INIT_LIST_HEAD(&oii->oii_list);
1967 oii->oii_cache = *oic;
1968 oii->oii_insert = insert;
1970 spin_lock(&scrub->os_lock);
1971 if (unlikely(!thread_is_running(thread))) {
1972 spin_unlock(&scrub->os_lock);
1977 if (cfs_list_empty(&scrub->os_inconsistent_items))
1979 cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
1980 spin_unlock(&scrub->os_lock);
1983 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1988 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
1989 struct osd_inode_id *id)
1991 struct osd_scrub *scrub = &dev->od_scrub;
1992 struct osd_inconsistent_item *oii;
1995 spin_lock(&scrub->os_lock);
1996 cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
1997 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
1998 *id = oii->oii_cache.oic_lid;
1999 spin_unlock(&scrub->os_lock);
2003 spin_unlock(&scrub->os_lock);
2010 static const char *scrub_status_names[] = {
2021 static const char *scrub_flags_names[] = {
2029 static const char *scrub_param_names[] = {
2034 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
2042 rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
2048 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
2051 rc = snprintf(*buf, *len, "%s%c", names[i],
2052 bits != 0 ? ',' : '\n');
2063 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
2068 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
2069 cfs_time_current_sec() - time);
2071 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2080 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
2085 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
2087 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
2096 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
2098 struct osd_scrub *scrub = &dev->od_scrub;
2099 struct scrub_file *sf = &scrub->os_file;
2106 down_read(&scrub->os_rwsem);
2107 rc = snprintf(buf, len,
2112 sf->sf_magic, (int)sf->sf_oi_count,
2113 scrub_status_names[sf->sf_status]);
2119 rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
2124 rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
2129 rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
2130 "time_since_last_completed");
2134 rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
2135 "time_since_latest_start");
2139 rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
2140 "time_since_last_checkpoint");
2144 rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
2145 "latest_start_position");
2149 rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
2150 "last_checkpoint_position");
2154 rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
2155 "first_failure_position");
2159 checked = sf->sf_items_checked + scrub->os_new_checked;
2160 rc = snprintf(buf, len,
2161 "checked: "LPU64"\n"
2162 "updated: "LPU64"\n"
2164 "prior_updated: "LPU64"\n"
2165 "noscrub: "LPU64"\n"
2167 "success_count: %u\n",
2168 checked, sf->sf_items_updated, sf->sf_items_failed,
2169 sf->sf_items_updated_prior, sf->sf_items_noscrub,
2170 sf->sf_items_igif, sf->sf_success_count);
2177 if (thread_is_running(&scrub->os_thread)) {
2178 cfs_duration_t duration = cfs_time_current() -
2179 scrub->os_time_last_checkpoint;
2180 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
2181 __u32 rtime = sf->sf_run_time +
2182 cfs_duration_sec(duration + HALF_SEC);
2185 do_div(new_checked, duration);
2187 do_div(speed, rtime);
2188 rc = snprintf(buf, len,
2189 "run_time: %u seconds\n"
2190 "average_speed: "LPU64" objects/sec\n"
2191 "real-time_speed: "LPU64" objects/sec\n"
2192 "current_position: %u\n",
2193 rtime, speed, new_checked, scrub->os_pos_current);
2195 if (sf->sf_run_time != 0)
2196 do_div(speed, sf->sf_run_time);
2197 rc = snprintf(buf, len,
2198 "run_time: %u seconds\n"
2199 "average_speed: "LPU64" objects/sec\n"
2200 "real-time_speed: N/A\n"
2201 "current_position: N/A\n",
2202 sf->sf_run_time, speed);
2212 up_read(&scrub->os_rwsem);