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 Whamcloud, Inc.
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 static inline struct osd_device *osd_scrub2dev(struct osd_scrub *scrub)
55 return container_of0(scrub, struct osd_device, od_scrub);
58 static inline struct super_block *osd_scrub2sb(struct osd_scrub *scrub)
60 return osd_sb(osd_scrub2dev(scrub));
63 static void osd_scrub_file_to_cpu(struct scrub_file *des,
64 struct scrub_file *src)
66 memcpy(des->sf_uuid, src->sf_uuid, 16);
67 des->sf_flags = le64_to_cpu(src->sf_flags);
68 des->sf_magic = le32_to_cpu(src->sf_magic);
69 des->sf_status = le16_to_cpu(src->sf_status);
70 des->sf_param = le16_to_cpu(src->sf_param);
71 des->sf_time_last_complete =
72 le64_to_cpu(src->sf_time_last_complete);
73 des->sf_time_latest_start =
74 le64_to_cpu(src->sf_time_latest_start);
75 des->sf_time_last_checkpoint =
76 le64_to_cpu(src->sf_time_last_checkpoint);
77 des->sf_pos_latest_start =
78 le64_to_cpu(src->sf_pos_latest_start);
79 des->sf_pos_last_checkpoint =
80 le64_to_cpu(src->sf_pos_last_checkpoint);
81 des->sf_pos_first_inconsistent =
82 le64_to_cpu(src->sf_pos_first_inconsistent);
83 des->sf_items_checked =
84 le64_to_cpu(src->sf_items_checked);
85 des->sf_items_updated =
86 le64_to_cpu(src->sf_items_updated);
87 des->sf_items_failed =
88 le64_to_cpu(src->sf_items_failed);
89 des->sf_items_updated_prior =
90 le64_to_cpu(src->sf_items_updated_prior);
91 des->sf_run_time = le32_to_cpu(src->sf_run_time);
92 des->sf_success_count = le32_to_cpu(src->sf_success_count);
93 des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
94 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
97 static void osd_scrub_file_to_le(struct scrub_file *des,
98 struct scrub_file *src)
100 memcpy(des->sf_uuid, src->sf_uuid, 16);
101 des->sf_flags = cpu_to_le64(src->sf_flags);
102 des->sf_magic = cpu_to_le32(src->sf_magic);
103 des->sf_status = cpu_to_le16(src->sf_status);
104 des->sf_param = cpu_to_le16(src->sf_param);
105 des->sf_time_last_complete =
106 cpu_to_le64(src->sf_time_last_complete);
107 des->sf_time_latest_start =
108 cpu_to_le64(src->sf_time_latest_start);
109 des->sf_time_last_checkpoint =
110 cpu_to_le64(src->sf_time_last_checkpoint);
111 des->sf_pos_latest_start =
112 cpu_to_le64(src->sf_pos_latest_start);
113 des->sf_pos_last_checkpoint =
114 cpu_to_le64(src->sf_pos_last_checkpoint);
115 des->sf_pos_first_inconsistent =
116 cpu_to_le64(src->sf_pos_first_inconsistent);
117 des->sf_items_checked =
118 cpu_to_le64(src->sf_items_checked);
119 des->sf_items_updated =
120 cpu_to_le64(src->sf_items_updated);
121 des->sf_items_failed =
122 cpu_to_le64(src->sf_items_failed);
123 des->sf_items_updated_prior =
124 cpu_to_le64(src->sf_items_updated_prior);
125 des->sf_run_time = cpu_to_le32(src->sf_run_time);
126 des->sf_success_count = cpu_to_le32(src->sf_success_count);
127 des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
128 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
131 static void osd_scrub_file_init(struct osd_scrub *scrub, __u8 *uuid)
133 struct scrub_file *sf = &scrub->os_file;
135 memset(sf, 0, sizeof(*sf));
136 memcpy(sf->sf_uuid, uuid, 16);
137 sf->sf_magic = SCRUB_MAGIC_V1;
138 sf->sf_status = SS_INIT;
141 void osd_scrub_file_reset(struct osd_scrub *scrub, __u8 *uuid, __u64 flags)
143 struct scrub_file *sf = &scrub->os_file;
145 CDEBUG(D_LFSCK, "Reset OI scrub file, flags = "LPX64"\n", flags);
146 memcpy(sf->sf_uuid, uuid, 16);
147 sf->sf_status = SS_INIT;
148 sf->sf_flags |= flags;
151 sf->sf_time_latest_start = 0;
152 sf->sf_time_last_checkpoint = 0;
153 sf->sf_pos_latest_start = 0;
154 sf->sf_pos_last_checkpoint = 0;
155 sf->sf_pos_first_inconsistent = 0;
156 sf->sf_items_checked = 0;
157 sf->sf_items_updated = 0;
158 sf->sf_items_failed = 0;
159 sf->sf_items_updated_prior = 0;
160 sf->sf_items_noscrub = 0;
161 sf->sf_items_igif = 0;
164 static int osd_scrub_file_load(struct osd_scrub *scrub)
167 char *name = LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name;
168 int len = sizeof(scrub->os_file_disk);
171 rc = osd_ldiskfs_read(scrub->os_inode, &scrub->os_file_disk, len, &pos);
173 struct scrub_file *sf = &scrub->os_file;
175 osd_scrub_file_to_cpu(sf, &scrub->os_file_disk);
176 if (sf->sf_magic != SCRUB_MAGIC_V1) {
177 CWARN("%.16s: invalid scrub magic 0x%x != 0x%x\n,",
178 name, sf->sf_magic, SCRUB_MAGIC_V1);
179 /* Process it as new scrub file. */
184 } else if (rc != 0) {
185 CERROR("%.16s: fail to load scrub file, expected = %d, "
186 "rc = %d\n", name, len, rc);
190 /* return -ENOENT for empty scrub file case. */
197 int osd_scrub_file_store(struct osd_scrub *scrub)
199 struct osd_device *dev;
202 int len = sizeof(scrub->os_file_disk);
206 dev = container_of0(scrub, struct osd_device, od_scrub);
207 credits = osd_dto_credits_noquota[DTO_WRITE_BASE] +
208 osd_dto_credits_noquota[DTO_WRITE_BLOCK];
209 jh = ldiskfs_journal_start_sb(osd_sb(dev), credits);
212 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
213 LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,rc);
217 osd_scrub_file_to_le(&scrub->os_file_disk, &scrub->os_file);
218 rc = osd_ldiskfs_write_record(scrub->os_inode, &scrub->os_file_disk,
220 ldiskfs_journal_stop(jh);
222 CERROR("%.16s: fail to store scrub file, expected = %d, "
224 LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
226 scrub->os_time_last_checkpoint = cfs_time_current();
227 scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
228 cfs_time_seconds(SCRUB_CHECKPOINT_INTERVAL);
232 static int osd_scrub_prep(struct osd_device *dev)
234 struct osd_scrub *scrub = &dev->od_scrub;
235 struct ptlrpc_thread *thread = &scrub->os_thread;
236 struct scrub_file *sf = &scrub->os_file;
237 __u32 flags = scrub->os_start_flags;
241 cfs_down_write(&scrub->os_rwsem);
242 if (flags & SS_SET_FAILOUT)
243 sf->sf_param |= SP_FAILOUT;
245 if (flags & SS_CLEAR_FAILOUT)
246 sf->sf_param &= ~SP_FAILOUT;
248 if (flags & SS_RESET)
249 osd_scrub_file_reset(scrub,
250 LDISKFS_SB(osd_sb(dev))->s_es->s_uuid, sf->sf_flags);
252 if (flags & SS_AUTO) {
253 scrub->os_full_speed = 1;
254 sf->sf_flags |= SF_AUTO;
256 scrub->os_full_speed = 0;
259 if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT))
260 scrub->os_full_speed = 1;
262 scrub->os_in_prior = 0;
263 scrub->os_waiting = 0;
264 scrub->os_new_checked = 0;
265 if (sf->sf_pos_last_checkpoint != 0)
266 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
268 sf->sf_pos_latest_start = LDISKFS_FIRST_INO(osd_sb(dev));
270 scrub->os_pos_current = sf->sf_pos_latest_start;
271 sf->sf_status = SS_SCANNING;
272 sf->sf_time_latest_start = cfs_time_current_sec();
273 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
274 rc = osd_scrub_file_store(scrub);
276 cfs_spin_lock(&scrub->os_lock);
277 thread_set_flags(thread, SVC_RUNNING);
278 cfs_spin_unlock(&scrub->os_lock);
279 cfs_waitq_broadcast(&thread->t_ctl_waitq);
281 cfs_up_write(&scrub->os_rwsem);
287 osd_scrub_error(struct osd_device *dev, struct osd_inode_id *lid, int rc)
289 struct osd_scrub *scrub = &dev->od_scrub;
290 struct scrub_file *sf = &scrub->os_file;
292 cfs_down_write(&scrub->os_rwsem);
293 scrub->os_new_checked++;
294 sf->sf_items_failed++;
295 if (sf->sf_pos_first_inconsistent == 0 ||
296 sf->sf_pos_first_inconsistent > lid->oii_ino)
297 sf->sf_pos_first_inconsistent = lid->oii_ino;
298 cfs_up_write(&scrub->os_rwsem);
299 return sf->sf_param & SP_FAILOUT ? rc : 0;
303 osd_scrub_check_update(struct osd_thread_info *info, struct osd_device *dev,
304 struct osd_idmap_cache *oic)
306 struct osd_scrub *scrub = &dev->od_scrub;
307 struct scrub_file *sf = &scrub->os_file;
308 struct osd_inode_id *lid2 = &info->oti_id;
309 struct lu_fid *oi_fid = &info->oti_fid;
310 struct osd_inode_id *oi_id = &info->oti_id2;
312 struct osd_inconsistent_item *oii = NULL;
313 struct inode *inode = NULL;
314 struct lu_fid *fid = &oic->oic_fid;
315 struct osd_inode_id *lid = &oic->oic_lid;
316 struct iam_container *bag;
317 struct iam_path_descr *ipd;
318 int ops = DTO_INDEX_UPDATE;
323 if (scrub->os_in_prior)
324 oii = cfs_list_entry(oic, struct osd_inconsistent_item,
327 cfs_down_write(&scrub->os_rwsem);
328 scrub->os_new_checked++;
329 if (lid->oii_ino < sf->sf_pos_latest_start && oii == NULL)
332 if (oii != NULL && oii->oii_insert)
335 rc = osd_oi_lookup(info, dev, fid, lid2);
341 inode = osd_iget(info, dev, lid);
344 /* Someone removed the inode. */
345 if (rc == -ENOENT || rc == -ESTALE)
350 /* Prevent the inode to be unlinked during OI scrub. */
351 cfs_mutex_lock(&inode->i_mutex);
352 if (unlikely(inode->i_nlink == 0)) {
353 cfs_mutex_unlock(&inode->i_mutex);
358 ops = DTO_INDEX_INSERT;
359 idx = osd_oi_fid2idx(dev, fid);
360 if (unlikely(!ldiskfs_test_bit(idx, sf->sf_oi_bitmap)))
361 ldiskfs_set_bit(idx, sf->sf_oi_bitmap);
362 sf->sf_flags |= SF_RECREATED;
363 } else if (osd_id_eq(lid, lid2)) {
367 sf->sf_flags |= SF_INCONSISTENT;
368 fid_cpu_to_be(oi_fid, fid);
369 osd_id_pack(oi_id, &oic->oic_lid);
370 jh = ldiskfs_journal_start_sb(osd_sb(dev),
371 osd_dto_credits_noquota[ops]);
374 CERROR("%.16s: fail to start trans for scrub store, rc = %d\n",
375 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
379 bag = &osd_fid2oi(dev, fid)->oi_dir.od_container;
380 ipd = osd_idx_ipd_get(info->oti_env, bag);
381 if (unlikely(ipd == NULL)) {
382 ldiskfs_journal_stop(jh);
383 CERROR("%.16s: fail to get ipd for scrub store\n",
384 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name);
385 GOTO(out, rc = -ENOMEM);
388 if (ops == DTO_INDEX_UPDATE)
389 rc = iam_update(jh, bag, (const struct iam_key *)oi_fid,
390 (struct iam_rec *)oi_id, ipd);
392 rc = iam_insert(jh, bag, (const struct iam_key *)oi_fid,
393 (struct iam_rec *)oi_id, ipd);
394 osd_ipd_put(info->oti_env, bag, ipd);
395 ldiskfs_journal_stop(jh);
397 if (scrub->os_in_prior)
398 sf->sf_items_updated_prior++;
400 sf->sf_items_updated++;
407 sf->sf_items_failed++;
408 if (sf->sf_pos_first_inconsistent == 0 ||
409 sf->sf_pos_first_inconsistent > lid->oii_ino)
410 sf->sf_pos_first_inconsistent = lid->oii_ino;
413 if (ops == DTO_INDEX_INSERT) {
414 cfs_mutex_unlock(&inode->i_mutex);
417 cfs_up_write(&scrub->os_rwsem);
420 LASSERT(!cfs_list_empty(&oii->oii_list));
422 cfs_spin_lock(&scrub->os_lock);
423 cfs_list_del_init(&oii->oii_list);
424 cfs_spin_unlock(&scrub->os_lock);
427 RETURN(sf->sf_param & SP_FAILOUT ? rc : 0);
430 static int do_osd_scrub_checkpoint(struct osd_scrub *scrub)
432 struct scrub_file *sf = &scrub->os_file;
436 cfs_down_write(&scrub->os_rwsem);
437 sf->sf_items_checked += scrub->os_new_checked;
438 scrub->os_new_checked = 0;
439 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
440 sf->sf_time_last_checkpoint = cfs_time_current_sec();
441 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
442 scrub->os_time_last_checkpoint);
443 rc = osd_scrub_file_store(scrub);
444 cfs_up_write(&scrub->os_rwsem);
449 static inline int osd_scrub_checkpoint(struct osd_scrub *scrub)
451 if (unlikely(cfs_time_beforeq(scrub->os_time_next_checkpoint,
452 cfs_time_current()) &&
453 scrub->os_new_checked > 0))
454 return do_osd_scrub_checkpoint(scrub);
458 static void osd_scrub_post(struct osd_scrub *scrub, int result)
460 struct scrub_file *sf = &scrub->os_file;
463 cfs_down_write(&scrub->os_rwsem);
464 cfs_spin_lock(&scrub->os_lock);
465 thread_set_flags(&scrub->os_thread, SVC_STOPPING);
466 cfs_spin_unlock(&scrub->os_lock);
467 if (scrub->os_new_checked > 0) {
468 sf->sf_items_checked += scrub->os_new_checked;
469 scrub->os_new_checked = 0;
470 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
472 sf->sf_time_last_checkpoint = cfs_time_current_sec();
474 sf->sf_status = SS_COMPLETED;
475 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
476 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT | SF_AUTO);
477 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
478 sf->sf_success_count++;
479 } else if (result == 0) {
480 if (scrub->os_paused)
481 sf->sf_status = SS_PAUSED;
483 sf->sf_status = SS_STOPPED;
485 sf->sf_status = SS_FAILED;
487 sf->sf_run_time += cfs_duration_sec(cfs_time_current() + HALF_SEC -
488 scrub->os_time_last_checkpoint);
489 result = osd_scrub_file_store(scrub);
491 CERROR("%.16s: fail to osd_scrub_post, rc = %d\n",
492 LDISKFS_SB(osd_scrub2sb(scrub))->s_es->s_volume_name,
494 cfs_up_write(&scrub->os_rwsem);
499 #define SCRUB_NEXT_BREAK 1 /* exit current loop and process next group */
500 #define SCRUB_NEXT_CONTINUE 2 /* skip current object and process next bit */
501 #define SCRUB_NEXT_EXIT 3 /* exit all the loops */
502 #define SCRUB_NEXT_WAIT 4 /* wait for free cache slot */
503 #define SCRUB_NEXT_CRASH 5 /* simulate system crash during OI scrub */
504 #define SCRUB_NEXT_FATAL 6 /* simulate failure during OI scrub */
505 #define SCRUB_NEXT_NOSCRUB 7 /* new created object, no scrub on it */
506 #define SCRUB_NEXT_IGIF 8 /* IGIF object */
508 struct osd_iit_param {
509 struct super_block *sb;
510 struct buffer_head *bitmap;
516 typedef int (*osd_iit_next_policy)(struct osd_thread_info *info,
517 struct osd_device *dev,
518 struct osd_iit_param *param,
519 struct osd_idmap_cache **oic,
522 typedef int (*osd_iit_exec_policy)(struct osd_thread_info *info,
523 struct osd_device *dev,
524 struct osd_iit_param *param,
525 struct osd_idmap_cache *oic,
526 int *noslot, int rc);
528 static inline int osd_scrub_has_window(struct osd_scrub *scrub,
529 struct osd_otable_cache *ooc)
531 return scrub->os_pos_current < ooc->ooc_pos_preload + SCRUB_WINDOW_SIZE;
534 static int osd_iit_next(struct osd_iit_param *param, __u32 *pos)
536 param->offset = ldiskfs_find_next_bit(param->bitmap->b_data,
537 LDISKFS_INODES_PER_GROUP(param->sb), param->offset);
538 if (param->offset >= LDISKFS_INODES_PER_GROUP(param->sb)) {
539 *pos = 1 + (param->bg+1) * LDISKFS_INODES_PER_GROUP(param->sb);
540 return SCRUB_NEXT_BREAK;
542 *pos = param->gbase + param->offset;
547 static int osd_iit_iget(struct osd_thread_info *info, struct osd_device *dev,
548 struct lu_fid *fid, struct osd_inode_id *lid, __u32 pos,
549 struct super_block *sb, struct inode **pinode)
554 osd_id_gen(lid, pos, OSD_OII_NOGEN);
555 inode = osd_iget_fid(info, dev, lid, fid);
558 /* The inode may be removed after bitmap searching, or the
559 * file is new created without inode initialized yet. */
560 if (rc == -ENOENT || rc == -ESTALE)
561 return SCRUB_NEXT_CONTINUE;
563 CERROR("%.16s: fail to read inode, ino# = %u, rc = %d\n",
564 LDISKFS_SB(sb)->s_es->s_volume_name, pos, rc);
572 static int osd_scrub_next(struct osd_thread_info *info, struct osd_device *dev,
573 struct osd_iit_param *param,
574 struct osd_idmap_cache **oic, int noslot)
576 struct osd_scrub *scrub = &dev->od_scrub;
577 struct ptlrpc_thread *thread = &scrub->os_thread;
579 struct osd_inode_id *lid;
583 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_DELAY) && cfs_fail_val > 0) {
584 struct l_wait_info lwi;
586 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val), NULL, NULL);
587 l_wait_event(thread->t_ctl_waitq,
588 !cfs_list_empty(&scrub->os_inconsistent_items) ||
589 !thread_is_running(thread),
593 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_CRASH)) {
594 cfs_spin_lock(&scrub->os_lock);
595 thread_set_flags(thread, SVC_STOPPING);
596 cfs_spin_unlock(&scrub->os_lock);
597 return SCRUB_NEXT_CRASH;
600 if (OBD_FAIL_CHECK(OBD_FAIL_OSD_SCRUB_FATAL))
601 return SCRUB_NEXT_FATAL;
603 if (unlikely(!thread_is_running(thread)))
604 return SCRUB_NEXT_EXIT;
606 if (!cfs_list_empty(&scrub->os_inconsistent_items)) {
607 struct osd_inconsistent_item *oii;
609 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
610 struct osd_inconsistent_item, oii_list);
611 *oic = &oii->oii_cache;
612 scrub->os_in_prior = 1;
617 return SCRUB_NEXT_WAIT;
619 rc = osd_iit_next(param, &scrub->os_pos_current);
623 *oic = &scrub->os_oic;
624 fid = &(*oic)->oic_fid;
625 lid = &(*oic)->oic_lid;
626 rc = osd_iit_iget(info, dev, fid, lid,
627 scrub->os_pos_current, param->sb, &inode);
631 if (inode->i_state & I_LUSTRE_NOSCRUB) {
632 /* Only skip it for the first OI scrub accessing. */
633 inode->i_state &= ~I_LUSTRE_NOSCRUB;
634 rc = SCRUB_NEXT_NOSCRUB;
635 } else if (!fid_is_norm(fid)) {
636 rc = SCRUB_NEXT_IGIF;
643 static int osd_preload_next(struct osd_thread_info *info,
644 struct osd_device *dev, struct osd_iit_param *param,
645 struct osd_idmap_cache **oic, int noslot)
647 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
648 struct osd_scrub *scrub;
649 struct ptlrpc_thread *thread;
653 rc = osd_iit_next(param, &ooc->ooc_pos_preload);
657 scrub = &dev->od_scrub;
658 thread = &scrub->os_thread;
659 if (thread_is_running(thread) &&
660 ooc->ooc_pos_preload >= scrub->os_pos_current)
661 return SCRUB_NEXT_EXIT;
663 rc = osd_iit_iget(info, dev,
664 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_fid,
665 &ooc->ooc_cache[ooc->ooc_producer_idx].oic_lid,
666 ooc->ooc_pos_preload, param->sb, &inode);
667 /* If succeed, it needs to move forward; otherwise up layer LFSCK may
668 * ignore the failure, so it still need to skip the inode next time. */
669 ooc->ooc_pos_preload = param->gbase + ++(param->offset);
675 static int osd_scrub_exec(struct osd_thread_info *info, struct osd_device *dev,
676 struct osd_iit_param *param,
677 struct osd_idmap_cache *oic, int *noslot, int rc)
679 struct l_wait_info lwi = { 0 };
680 struct osd_scrub *scrub = &dev->od_scrub;
681 struct scrub_file *sf = &scrub->os_file;
683 struct ptlrpc_thread *thread = &scrub->os_thread;
684 struct osd_otable_it *it = dev->od_otable_it;
685 struct osd_otable_cache *ooc = it ? &it->ooi_cache : NULL;
688 case SCRUB_NEXT_CONTINUE:
690 case SCRUB_NEXT_WAIT:
692 case SCRUB_NEXT_NOSCRUB:
693 items = &sf->sf_items_noscrub;
695 case SCRUB_NEXT_IGIF:
696 items = &sf->sf_items_igif;
701 cfs_down_write(&scrub->os_rwsem);
702 scrub->os_new_checked++;
704 cfs_up_write(&scrub->os_rwsem);
708 LASSERTF(rc <= 0, "unexpected rc = %d\n", rc);
711 rc = osd_scrub_error(dev, &oic->oic_lid, rc);
713 rc = osd_scrub_check_update(info, dev, oic);
717 rc = osd_scrub_checkpoint(scrub);
719 CERROR("%.16s: fail to checkpoint, pos = %u, rc = %d\n",
720 LDISKFS_SB(param->sb)->s_es->s_volume_name,
721 scrub->os_pos_current, rc);
722 /* Continue, as long as the scrub itself can go ahead. */
725 if (scrub->os_in_prior) {
726 scrub->os_in_prior = 0;
731 scrub->os_pos_current = param->gbase + ++(param->offset);
732 if (it != NULL && it->ooi_waiting &&
733 ooc->ooc_pos_preload < scrub->os_pos_current) {
735 cfs_waitq_broadcast(&thread->t_ctl_waitq);
738 if (scrub->os_full_speed || rc == SCRUB_NEXT_CONTINUE)
742 if (osd_scrub_has_window(scrub, ooc)) {
747 scrub->os_waiting = 1;
748 l_wait_event(thread->t_ctl_waitq,
749 osd_scrub_has_window(scrub, ooc) ||
750 !cfs_list_empty(&scrub->os_inconsistent_items) ||
751 !thread_is_running(thread),
753 scrub->os_waiting = 0;
755 if (osd_scrub_has_window(scrub, ooc))
762 static int osd_preload_exec(struct osd_thread_info *info,
763 struct osd_device *dev, struct osd_iit_param *param,
764 struct osd_idmap_cache *oic, int *noslot, int rc)
766 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
769 ooc->ooc_cached_items++;
770 ooc->ooc_producer_idx = (ooc->ooc_producer_idx + 1) &
771 ~OSD_OTABLE_IT_CACHE_MASK;
773 return rc > 0 ? 0 : rc;
776 #define SCRUB_IT_ALL 1
777 #define SCRUB_IT_CRASH 2
779 static int osd_inode_iteration(struct osd_thread_info *info,
780 struct osd_device *dev, __u32 max, int preload)
782 osd_iit_next_policy next;
783 osd_iit_exec_policy exec;
786 struct osd_iit_param param;
793 struct osd_scrub *scrub = &dev->od_scrub;
795 next = osd_scrub_next;
796 exec = osd_scrub_exec;
797 pos = &scrub->os_pos_current;
798 count = &scrub->os_new_checked;
800 struct osd_otable_cache *ooc = &dev->od_otable_it->ooi_cache;
802 next = osd_preload_next;
803 exec = osd_preload_exec;
804 pos = &ooc->ooc_pos_preload;
805 count = &ooc->ooc_cached_items;
807 param.sb = osd_sb(dev);
808 limit = le32_to_cpu(LDISKFS_SB(param.sb)->s_es->s_inodes_count);
810 while (*pos <= limit && *count < max) {
811 struct osd_idmap_cache *oic = NULL;
813 param.bg = (*pos - 1) / LDISKFS_INODES_PER_GROUP(param.sb);
814 param.offset = (*pos - 1) % LDISKFS_INODES_PER_GROUP(param.sb);
815 param.gbase = 1 + param.bg * LDISKFS_INODES_PER_GROUP(param.sb);
816 param.bitmap = ldiskfs_read_inode_bitmap(param.sb, param.bg);
817 if (param.bitmap == NULL) {
818 CERROR("%.16s: fail to read bitmap for %u, "
819 "scrub will stop, urgent mode\n",
820 LDISKFS_SB(param.sb)->s_es->s_volume_name,
825 while (param.offset < LDISKFS_INODES_PER_GROUP(param.sb) &&
827 rc = next(info, dev, ¶m, &oic, noslot);
829 case SCRUB_NEXT_BREAK:
831 case SCRUB_NEXT_EXIT:
832 brelse(param.bitmap);
834 case SCRUB_NEXT_CRASH:
835 brelse(param.bitmap);
836 RETURN(SCRUB_IT_CRASH);
837 case SCRUB_NEXT_FATAL:
838 brelse(param.bitmap);
842 rc = exec(info, dev, ¶m, oic, &noslot, rc);
844 brelse(param.bitmap);
850 brelse(param.bitmap);
854 RETURN(SCRUB_IT_ALL);
858 static int osd_scrub_main(void *args)
861 struct osd_device *dev = (struct osd_device *)args;
862 struct osd_scrub *scrub = &dev->od_scrub;
863 struct ptlrpc_thread *thread = &scrub->os_thread;
864 struct super_block *sb = osd_sb(dev);
868 cfs_daemonize("OI_scrub");
869 rc = lu_env_init(&env, LCT_DT_THREAD);
871 CERROR("%.16s: OI scrub, fail to init env, rc = %d\n",
872 LDISKFS_SB(sb)->s_es->s_volume_name, rc);
876 rc = osd_scrub_prep(dev);
878 CERROR("%.16s: OI scrub, fail to scrub prep, rc = %d\n",
879 LDISKFS_SB(sb)->s_es->s_volume_name, rc);
883 if (!scrub->os_full_speed) {
884 struct l_wait_info lwi = { 0 };
885 struct osd_otable_it *it = dev->od_otable_it;
886 struct osd_otable_cache *ooc = &it->ooi_cache;
888 l_wait_event(thread->t_ctl_waitq,
889 it->ooi_user_ready || !thread_is_running(thread),
891 if (unlikely(!thread_is_running(thread)))
894 LASSERT(scrub->os_pos_current >= ooc->ooc_pos_preload);
895 scrub->os_pos_current = ooc->ooc_pos_preload;
898 CDEBUG(D_LFSCK, "OI scrub: flags = 0x%x, pos = %u\n",
899 scrub->os_start_flags, scrub->os_pos_current);
901 rc = osd_inode_iteration(osd_oti_get(&env), dev, ~0U, 0);
902 if (unlikely(rc == SCRUB_IT_CRASH))
903 GOTO(out, rc = -EINVAL);
907 osd_scrub_post(scrub, rc);
908 CDEBUG(D_LFSCK, "OI scrub: stop, rc = %d, pos = %u\n",
909 rc, scrub->os_pos_current);
912 while (!cfs_list_empty(&scrub->os_inconsistent_items)) {
913 struct osd_inconsistent_item *oii;
915 oii = cfs_list_entry(scrub->os_inconsistent_items.next,
916 struct osd_inconsistent_item, oii_list);
917 cfs_list_del_init(&oii->oii_list);
923 cfs_spin_lock(&scrub->os_lock);
924 thread_set_flags(thread, SVC_STOPPED);
925 cfs_waitq_broadcast(&thread->t_ctl_waitq);
926 cfs_spin_unlock(&scrub->os_lock);
930 static int do_osd_scrub_start(struct osd_device *dev, __u32 flags)
932 struct osd_scrub *scrub = &dev->od_scrub;
933 struct ptlrpc_thread *thread = &scrub->os_thread;
934 struct l_wait_info lwi = { 0 };
939 /* os_lock: sync status between stop and scrub thread */
940 cfs_spin_lock(&scrub->os_lock);
941 if (thread_is_running(thread)) {
942 cfs_spin_unlock(&scrub->os_lock);
944 } else if (unlikely(thread_is_stopping(thread))) {
945 cfs_spin_unlock(&scrub->os_lock);
946 l_wait_event(thread->t_ctl_waitq,
947 thread_is_stopped(thread),
951 cfs_spin_unlock(&scrub->os_lock);
953 if (scrub->os_file.sf_status == SS_COMPLETED)
956 scrub->os_start_flags = flags;
957 thread_set_flags(thread, 0);
958 rc = cfs_create_thread(osd_scrub_main, dev, 0);
960 CERROR("%.16s: cannot start iteration thread, rc = %d\n",
961 LDISKFS_SB(osd_sb(dev))->s_es->s_volume_name, rc);
965 l_wait_event(thread->t_ctl_waitq,
966 thread_is_running(thread) || thread_is_stopped(thread),
972 int osd_scrub_start(struct osd_device *dev)
977 /* od_otable_mutex: prevent curcurrent start/stop */
978 cfs_mutex_lock(&dev->od_otable_mutex);
979 rc = do_osd_scrub_start(dev, SS_AUTO);
980 cfs_mutex_unlock(&dev->od_otable_mutex);
982 RETURN(rc == -EALREADY ? 0 : rc);
985 static void do_osd_scrub_stop(struct osd_scrub *scrub)
987 struct ptlrpc_thread *thread = &scrub->os_thread;
988 struct l_wait_info lwi = { 0 };
990 /* os_lock: sync status between stop and scrub thread */
991 cfs_spin_lock(&scrub->os_lock);
992 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
993 thread_set_flags(thread, SVC_STOPPING);
994 cfs_spin_unlock(&scrub->os_lock);
995 cfs_waitq_broadcast(&thread->t_ctl_waitq);
996 l_wait_event(thread->t_ctl_waitq,
997 thread_is_stopped(thread),
999 /* Do not skip the last lock/unlock, which can guarantee that
1000 * the caller cannot return until the OI scrub thread exit. */
1001 cfs_spin_lock(&scrub->os_lock);
1003 cfs_spin_unlock(&scrub->os_lock);
1006 static void osd_scrub_stop(struct osd_device *dev)
1008 /* od_otable_mutex: prevent curcurrent start/stop */
1009 cfs_mutex_lock(&dev->od_otable_mutex);
1010 dev->od_scrub.os_paused = 1;
1011 do_osd_scrub_stop(&dev->od_scrub);
1012 cfs_mutex_unlock(&dev->od_otable_mutex);
1015 static const char osd_scrub_name[] = "OI_scrub";
1017 int osd_scrub_setup(const struct lu_env *env, struct osd_device *dev)
1019 struct osd_thread_info *info = osd_oti_get(env);
1020 struct osd_scrub *scrub = &dev->od_scrub;
1021 struct lvfs_run_ctxt *ctxt = &scrub->os_ctxt;
1022 struct scrub_file *sf = &scrub->os_file;
1023 struct osd_inode_id *id = &scrub->os_oic.oic_lid;
1024 struct super_block *sb = osd_sb(dev);
1025 struct ldiskfs_super_block *es = LDISKFS_SB(sb)->s_es;
1026 struct inode *inode;
1027 struct lvfs_run_ctxt saved;
1034 memset(scrub, 0, sizeof(*scrub));
1035 OBD_SET_CTXT_MAGIC(ctxt);
1036 ctxt->pwdmnt = dev->od_mnt;
1037 ctxt->pwd = dev->od_mnt->mnt_root;
1038 ctxt->fs = get_ds();
1040 cfs_waitq_init(&scrub->os_thread.t_ctl_waitq);
1041 cfs_init_rwsem(&scrub->os_rwsem);
1042 cfs_spin_lock_init(&scrub->os_lock);
1043 CFS_INIT_LIST_HEAD(&scrub->os_inconsistent_items);
1045 push_ctxt(&saved, ctxt, NULL);
1046 filp = filp_open(osd_scrub_name, O_RDWR | O_CREAT, 0644);
1048 RETURN(PTR_ERR(filp));
1050 scrub->os_inode = igrab(filp->f_dentry->d_inode);
1051 filp_close(filp, 0);
1052 pop_ctxt(&saved, ctxt, NULL);
1054 rc = osd_scrub_file_load(scrub);
1055 if (rc == -ENOENT) {
1056 osd_scrub_file_init(scrub, es->s_uuid);
1059 } else if (rc != 0) {
1062 if (memcmp(sf->sf_uuid, es->s_uuid, 16) != 0) {
1063 osd_scrub_file_reset(scrub, es->s_uuid,SF_INCONSISTENT);
1065 } else if (sf->sf_status == SS_SCANNING) {
1066 sf->sf_status = SS_CRASHED;
1071 if (sf->sf_pos_last_checkpoint != 0)
1072 scrub->os_pos_current = sf->sf_pos_last_checkpoint + 1;
1074 scrub->os_pos_current = LDISKFS_FIRST_INO(sb);
1077 rc = osd_scrub_file_store(scrub);
1082 /* Initialize OI files. */
1083 rc = osd_oi_init(info, dev);
1088 rc = __osd_oi_lookup(info, dev, &LU_DOT_LUSTRE_FID, id);
1090 inode = osd_iget(info, dev, id);
1091 if (IS_ERR(inode)) {
1092 rc = PTR_ERR(inode);
1093 /* It is restored from old 2.x backup. */
1094 if (rc == -ENOENT || rc == -ESTALE) {
1095 osd_scrub_file_reset(scrub, es->s_uuid,
1097 rc = osd_scrub_file_store(scrub);
1102 } else if (rc == -ENOENT) {
1107 if (rc == 0 && !dev->od_noscrub &&
1108 ((sf->sf_status == SS_PAUSED) ||
1109 (sf->sf_status == SS_CRASHED &&
1110 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT | SF_AUTO)) ||
1111 (sf->sf_status == SS_INIT &&
1112 sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT))))
1113 rc = osd_scrub_start(dev);
1118 void osd_scrub_cleanup(const struct lu_env *env, struct osd_device *dev)
1120 struct osd_scrub *scrub = &dev->od_scrub;
1122 LASSERT(dev->od_otable_it == NULL);
1124 if (scrub->os_inode != NULL) {
1125 osd_scrub_stop(dev);
1126 iput(scrub->os_inode);
1127 scrub->os_inode = NULL;
1129 if (dev->od_oi_table != NULL)
1130 osd_oi_fini(osd_oti_get(env), dev);
1133 static struct dt_it *osd_otable_it_init(const struct lu_env *env,
1134 struct dt_object *dt, __u32 attr,
1135 struct lustre_capa *capa)
1137 enum dt_otable_it_flags flags = attr >> DT_OTABLE_IT_FLAGS_SHIFT;
1138 enum dt_otable_it_valid valid = attr & ~DT_OTABLE_IT_FLAGS_MASK;
1139 struct osd_device *dev = osd_dev(dt->do_lu.lo_dev);
1140 struct osd_scrub *scrub = &dev->od_scrub;
1141 struct osd_otable_it *it;
1146 /* od_otable_mutex: prevent curcurrent init/fini */
1147 cfs_mutex_lock(&dev->od_otable_mutex);
1148 if (dev->od_otable_it != NULL)
1149 GOTO(out, it = ERR_PTR(-EALREADY));
1153 GOTO(out, it = ERR_PTR(-ENOMEM));
1155 dev->od_otable_it = it;
1157 it->ooi_cache.ooc_consumer_idx = -1;
1158 if (flags & DOIF_OUTUSED)
1159 it->ooi_used_outside = 1;
1161 if (flags & DOIF_RESET)
1164 if (valid & DOIV_ERROR_HANDLE) {
1165 if (flags & DOIF_FAILOUT)
1166 start |= SS_SET_FAILOUT;
1168 start |= SS_CLEAR_FAILOUT;
1171 rc = do_osd_scrub_start(dev, start);
1172 if (rc == -EALREADY) {
1173 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current - 1;
1174 } else if (rc < 0) {
1175 dev->od_otable_it = NULL;
1177 GOTO(out, it = ERR_PTR(-EALREADY));
1179 it->ooi_cache.ooc_pos_preload = scrub->os_pos_current;
1185 cfs_mutex_unlock(&dev->od_otable_mutex);
1186 return (struct dt_it *)it;
1189 static void osd_otable_it_fini(const struct lu_env *env, struct dt_it *di)
1191 struct osd_otable_it *it = (struct osd_otable_it *)di;
1192 struct osd_device *dev = it->ooi_dev;
1194 /* od_otable_mutex: prevent curcurrent init/fini */
1195 cfs_mutex_lock(&dev->od_otable_mutex);
1196 do_osd_scrub_stop(&dev->od_scrub);
1197 LASSERT(dev->od_otable_it == it);
1199 dev->od_otable_it = NULL;
1200 cfs_mutex_unlock(&dev->od_otable_mutex);
1205 * XXX: Temporary used to notify otable iteration to be paused.
1207 static void osd_otable_it_put(const struct lu_env *env, struct dt_it *di)
1209 struct osd_device *dev = ((struct osd_otable_it *)di)->ooi_dev;
1211 /* od_otable_mutex: prevent curcurrent init/fini */
1212 cfs_mutex_lock(&dev->od_otable_mutex);
1213 dev->od_scrub.os_paused = 1;
1214 cfs_mutex_unlock(&dev->od_otable_mutex);
1218 * Set the OSD layer iteration start position as the specified key.
1220 * The LFSCK out of OSD layer does not know the detail of the key, so if there
1221 * are several keys, they cannot be compared out of OSD, so call "::get()" for
1222 * each key, and OSD will select the smallest one by itself.
1224 static int osd_otable_it_get(const struct lu_env *env,
1225 struct dt_it *di, const struct dt_key *key)
1227 struct osd_otable_it *it = (struct osd_otable_it *)di;
1228 struct osd_otable_cache *ooc = &it->ooi_cache;
1229 const char *str = (const char *)key;
1233 /* Forbid to set iteration position after iteration started. */
1234 if (it->ooi_user_ready)
1240 if (sscanf(str, "%u", &ino) <= 0)
1243 /* Skip the one that has been processed last time. */
1244 if (ooc->ooc_pos_preload > ++ino)
1245 ooc->ooc_pos_preload = ino;
1250 static int osd_otable_it_preload(const struct lu_env *env,
1251 struct osd_otable_it *it)
1253 struct osd_device *dev = it->ooi_dev;
1254 struct osd_scrub *scrub = &dev->od_scrub;
1255 struct osd_otable_cache *ooc = &it->ooi_cache;
1259 rc = osd_inode_iteration(osd_oti_get(env), dev,
1260 OSD_OTABLE_IT_CACHE_SIZE, 1);
1261 if (rc == SCRUB_IT_ALL)
1262 it->ooi_all_cached = 1;
1264 CDEBUG(D_LFSCK, "OSD pre-loaded: max = %u, preload = %u, rc = %d\n",
1265 le32_to_cpu(LDISKFS_SB(osd_sb(dev))->s_es->s_inodes_count),
1266 ooc->ooc_pos_preload, rc);
1268 if (scrub->os_waiting && osd_scrub_has_window(scrub, ooc)) {
1269 scrub->os_waiting = 0;
1270 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1273 RETURN(rc < 0 ? rc : ooc->ooc_cached_items);
1276 static int osd_otable_it_next(const struct lu_env *env, struct dt_it *di)
1278 struct osd_otable_it *it = (struct osd_otable_it *)di;
1279 struct osd_device *dev = it->ooi_dev;
1280 struct osd_scrub *scrub = &dev->od_scrub;
1281 struct osd_otable_cache *ooc = &it->ooi_cache;
1282 struct ptlrpc_thread *thread = &scrub->os_thread;
1283 struct l_wait_info lwi = { 0 };
1287 LASSERT(it->ooi_user_ready);
1290 if (!thread_is_running(thread) && !it->ooi_used_outside)
1293 if (ooc->ooc_cached_items > 0) {
1294 ooc->ooc_cached_items--;
1295 ooc->ooc_consumer_idx = (ooc->ooc_consumer_idx + 1) &
1296 ~OSD_OTABLE_IT_CACHE_MASK;
1300 if (it->ooi_all_cached) {
1301 l_wait_event(thread->t_ctl_waitq,
1302 !thread_is_running(thread),
1307 it->ooi_waiting = 1;
1308 l_wait_event(thread->t_ctl_waitq,
1309 ooc->ooc_pos_preload < scrub->os_pos_current ||
1310 !thread_is_running(thread),
1312 it->ooi_waiting = 0;
1314 if (!thread_is_running(thread) && !it->ooi_used_outside)
1317 rc = osd_otable_it_preload(env, it);
1324 static struct dt_key *osd_otable_it_key(const struct lu_env *env,
1325 const struct dt_it *di)
1327 struct osd_otable_it *it = (struct osd_otable_it *)di;
1328 struct osd_otable_cache *ooc = &it->ooi_cache;
1330 sprintf(it->ooi_key, "%u",
1331 ooc->ooc_cache[ooc->ooc_consumer_idx].oic_lid.oii_ino);
1332 return (struct dt_key *)it->ooi_key;
1335 static int osd_otable_it_key_size(const struct lu_env *env,
1336 const struct dt_it *di)
1338 return sizeof(((struct osd_otable_it *)di)->ooi_key);
1341 static int osd_otable_it_rec(const struct lu_env *env, const struct dt_it *di,
1342 struct dt_rec *rec, __u32 attr)
1344 struct osd_otable_it *it = (struct osd_otable_it *)di;
1345 struct osd_otable_cache *ooc = &it->ooi_cache;
1347 *(struct lu_fid *)rec = ooc->ooc_cache[ooc->ooc_consumer_idx].oic_fid;
1351 static int osd_otable_it_load(const struct lu_env *env,
1352 const struct dt_it *di, __u64 hash)
1354 struct osd_otable_it *it = (struct osd_otable_it *)di;
1355 struct osd_device *dev = it->ooi_dev;
1356 struct osd_otable_cache *ooc = &it->ooi_cache;
1357 struct osd_scrub *scrub = &dev->od_scrub;
1359 if (it->ooi_user_ready)
1362 if (ooc->ooc_pos_preload < LDISKFS_FIRST_INO(osd_sb(dev)))
1363 ooc->ooc_pos_preload = LDISKFS_FIRST_INO(osd_sb(dev));
1364 it->ooi_user_ready = 1;
1365 if (!scrub->os_full_speed)
1366 cfs_waitq_broadcast(&scrub->os_thread.t_ctl_waitq);
1368 /* Unplug OSD layer iteration by the first next() call. */
1369 return osd_otable_it_next(env, (struct dt_it *)it);
1372 const struct dt_index_operations osd_otable_ops = {
1374 .init = osd_otable_it_init,
1375 .fini = osd_otable_it_fini,
1376 .put = osd_otable_it_put,
1377 .get = osd_otable_it_get,
1378 .next = osd_otable_it_next,
1379 .key = osd_otable_it_key,
1380 .key_size = osd_otable_it_key_size,
1381 .rec = osd_otable_it_rec,
1382 .load = osd_otable_it_load,
1386 int osd_oii_insert(struct osd_device *dev, struct osd_idmap_cache *oic,
1389 struct osd_inconsistent_item *oii;
1390 struct osd_scrub *scrub = &dev->od_scrub;
1391 struct ptlrpc_thread *thread = &scrub->os_thread;
1396 if (unlikely(oii == NULL))
1399 CFS_INIT_LIST_HEAD(&oii->oii_list);
1400 oii->oii_cache = *oic;
1401 oii->oii_insert = insert;
1403 cfs_spin_lock(&scrub->os_lock);
1404 if (unlikely(!thread_is_running(thread))) {
1405 cfs_spin_unlock(&scrub->os_lock);
1410 if (cfs_list_empty(&scrub->os_inconsistent_items))
1412 cfs_list_add_tail(&oii->oii_list, &scrub->os_inconsistent_items);
1413 cfs_spin_unlock(&scrub->os_lock);
1416 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1421 int osd_oii_lookup(struct osd_device *dev, const struct lu_fid *fid,
1422 struct osd_inode_id *id)
1424 struct osd_scrub *scrub = &dev->od_scrub;
1425 struct osd_inconsistent_item *oii;
1428 cfs_spin_lock(&scrub->os_lock);
1429 cfs_list_for_each_entry(oii, &scrub->os_inconsistent_items, oii_list) {
1430 if (lu_fid_eq(fid, &oii->oii_cache.oic_fid)) {
1431 *id = oii->oii_cache.oic_lid;
1432 cfs_spin_unlock(&scrub->os_lock);
1436 cfs_spin_unlock(&scrub->os_lock);
1441 static const char *scrub_status_names[] = {
1452 static const char *scrub_flags_names[] = {
1459 static const char *scrub_param_names[] = {
1464 static int scrub_bits_dump(char **buf, int *len, int bits, const char *names[],
1472 rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1478 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
1481 rc = snprintf(*buf, *len, "%s%c", names[i],
1482 bits != 0 ? ',' : '\n');
1493 static int scrub_time_dump(char **buf, int *len, __u64 time, const char *prefix)
1498 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
1499 cfs_time_current_sec() - time);
1501 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
1510 static int scrub_pos_dump(char **buf, int *len, __u64 pos, const char *prefix)
1515 rc = snprintf(*buf, *len, "%s: "LPU64"\n", prefix, pos);
1517 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
1526 int osd_scrub_dump(struct osd_device *dev, char *buf, int len)
1528 struct osd_scrub *scrub = &dev->od_scrub;
1529 struct scrub_file *sf = &scrub->os_file;
1536 cfs_down_read(&scrub->os_rwsem);
1537 rc = snprintf(buf, len,
1542 sf->sf_magic, (int)sf->sf_oi_count,
1543 scrub_status_names[sf->sf_status]);
1549 rc = scrub_bits_dump(&buf, &len, sf->sf_flags, scrub_flags_names,
1554 rc = scrub_bits_dump(&buf, &len, sf->sf_param, scrub_param_names,
1559 rc = scrub_time_dump(&buf, &len, sf->sf_time_last_complete,
1560 "time_since_last_completed");
1564 rc = scrub_time_dump(&buf, &len, sf->sf_time_latest_start,
1565 "time_since_latest_start");
1569 rc = scrub_time_dump(&buf, &len, sf->sf_time_last_checkpoint,
1570 "time_since_last_checkpoint");
1574 rc = scrub_pos_dump(&buf, &len, sf->sf_pos_latest_start,
1575 "latest_start_position");
1579 rc = scrub_pos_dump(&buf, &len, sf->sf_pos_last_checkpoint,
1580 "last_checkpoint_position");
1584 rc = scrub_pos_dump(&buf, &len, sf->sf_pos_first_inconsistent,
1585 "first_failure_position");
1589 checked = sf->sf_items_checked + scrub->os_new_checked;
1590 rc = snprintf(buf, len,
1591 "checked: "LPU64"\n"
1592 "updated: "LPU64"\n"
1594 "prior_updated: "LPU64"\n"
1595 "noscrub: "LPU64"\n"
1597 "success_count: %u\n",
1598 checked, sf->sf_items_updated, sf->sf_items_failed,
1599 sf->sf_items_updated_prior, sf->sf_items_noscrub,
1600 sf->sf_items_igif, sf->sf_success_count);
1607 if (thread_is_running(&scrub->os_thread)) {
1608 cfs_duration_t duration = cfs_time_current() -
1609 scrub->os_time_last_checkpoint;
1610 __u64 new_checked = scrub->os_new_checked * CFS_HZ;
1611 __u32 rtime = sf->sf_run_time +
1612 cfs_duration_sec(duration + HALF_SEC);
1615 do_div(new_checked, duration);
1617 do_div(speed, rtime);
1618 rc = snprintf(buf, len,
1619 "run_time: %u seconds\n"
1620 "average_speed: "LPU64" objects/sec\n"
1621 "real-time_speed: "LPU64" objects/sec\n"
1622 "current_position: %u\n",
1623 rtime, speed, new_checked, scrub->os_pos_current);
1625 if (sf->sf_run_time != 0)
1626 do_div(speed, sf->sf_run_time);
1627 rc = snprintf(buf, len,
1628 "run_time: %u seconds\n"
1629 "average_speed: "LPU64" objects/sec\n"
1630 "real-time_speed: N/A\n"
1631 "current_position: N/A\n",
1632 sf->sf_run_time, speed);
1642 cfs_up_read(&scrub->os_rwsem);