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/mdd/mdd_lfsck.c
28 * Top-level entry points into mdd module
30 * LFSCK controller, which scans the whole device through low layer
31 * iteration APIs, drives all lfsck compeonents, controls the speed.
33 * Author: Fan Yong <yong.fan@whamcloud.com>
37 # define EXPORT_SYMTAB
39 #define DEBUG_SUBSYSTEM S_MDS
41 #include <lustre/lustre_idl.h>
42 #include <lustre_fid.h>
43 #include <obd_support.h>
45 #include "mdd_internal.h"
46 #include "mdd_lfsck.h"
48 #define HALF_SEC (CFS_HZ >> 1)
49 #define LFSCK_CHECKPOINT_INTERVAL 60
51 #define LFSCK_NAMEENTRY_DEAD 1 /* The object has been unlinked. */
52 #define LFSCK_NAMEENTRY_REMOVED 2 /* The entry has been removed. */
53 #define LFSCK_NAMEENTRY_RECREATED 3 /* The entry has been recreated. */
55 const char lfsck_bookmark_name[] = "lfsck_bookmark";
56 const char lfsck_namespace_name[] = "lfsck_namespace";
58 static const char *lfsck_status_names[] = {
70 static const char *lfsck_flags_names[] = {
77 static const char *lfsck_param_names[] = {
85 static inline struct mdd_device *mdd_lfsck2mdd(struct md_lfsck *lfsck)
87 return container_of0(lfsck, struct mdd_device, mdd_lfsck);
90 static inline char *mdd_lfsck2name(struct md_lfsck *lfsck)
92 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
94 return mdd2obd_dev(mdd)->obd_name;
97 static inline void mdd_lfsck_component_get(struct lfsck_component *com)
99 atomic_inc(&com->lc_ref);
102 static inline void mdd_lfsck_component_put(const struct lu_env *env,
103 struct lfsck_component *com)
105 if (atomic_dec_and_test(&com->lc_ref)) {
106 if (com->lc_obj != NULL)
107 lu_object_put(env, &com->lc_obj->do_lu);
108 if (com->lc_file_ram != NULL)
109 OBD_FREE(com->lc_file_ram, com->lc_file_size);
110 if (com->lc_file_disk != NULL)
111 OBD_FREE(com->lc_file_disk, com->lc_file_size);
116 static inline struct lfsck_component *
117 __mdd_lfsck_component_find(struct md_lfsck *lfsck, __u16 type, cfs_list_t *list)
119 struct lfsck_component *com;
121 cfs_list_for_each_entry(com, list, lc_link) {
122 if (com->lc_type == type)
128 static struct lfsck_component *
129 mdd_lfsck_component_find(struct md_lfsck *lfsck, __u16 type)
131 struct lfsck_component *com;
133 spin_lock(&lfsck->ml_lock);
134 com = __mdd_lfsck_component_find(lfsck, type, &lfsck->ml_list_scan);
138 com = __mdd_lfsck_component_find(lfsck, type,
139 &lfsck->ml_list_double_scan);
143 com = __mdd_lfsck_component_find(lfsck, type, &lfsck->ml_list_idle);
147 mdd_lfsck_component_get(com);
148 spin_unlock(&lfsck->ml_lock);
152 static void mdd_lfsck_component_cleanup(const struct lu_env *env,
153 struct lfsck_component *com)
155 if (!cfs_list_empty(&com->lc_link))
156 cfs_list_del_init(&com->lc_link);
157 if (!cfs_list_empty(&com->lc_link_dir))
158 cfs_list_del_init(&com->lc_link_dir);
160 mdd_lfsck_component_put(env, com);
163 static int lfsck_bits_dump(char **buf, int *len, int bits, const char *names[],
171 rc = snprintf(*buf, *len, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
177 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
180 rc = snprintf(*buf, *len, "%s%c", names[i],
181 bits != 0 ? ',' : '\n');
192 static int lfsck_time_dump(char **buf, int *len, __u64 time, const char *prefix)
197 rc = snprintf(*buf, *len, "%s: "LPU64" seconds\n", prefix,
198 cfs_time_current_sec() - time);
200 rc = snprintf(*buf, *len, "%s: N/A\n", prefix);
209 static int lfsck_pos_dump(char **buf, int *len, struct lfsck_position *pos,
214 if (fid_is_zero(&pos->lp_dir_parent)) {
215 if (pos->lp_oit_cookie == 0)
216 rc = snprintf(*buf, *len, "%s: N/A, N/A, N/A\n",
219 rc = snprintf(*buf, *len, "%s: "LPU64", N/A, N/A\n",
220 prefix, pos->lp_oit_cookie);
222 rc = snprintf(*buf, *len, "%s: "LPU64", "DFID", "LPU64"\n",
223 prefix, pos->lp_oit_cookie,
224 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
234 static void mdd_lfsck_pos_fill(const struct lu_env *env, struct md_lfsck *lfsck,
235 struct lfsck_position *pos, bool init)
237 const struct dt_it_ops *iops = &lfsck->ml_obj_oit->do_index_ops->dio_it;
239 spin_lock(&lfsck->ml_lock);
240 if (unlikely(lfsck->ml_di_oit == NULL)) {
241 spin_unlock(&lfsck->ml_lock);
242 memset(pos, 0, sizeof(*pos));
246 pos->lp_oit_cookie = iops->store(env, lfsck->ml_di_oit);
247 if (!lfsck->ml_current_oit_processed && !init)
248 pos->lp_oit_cookie--;
250 LASSERT(pos->lp_oit_cookie > 0);
252 if (lfsck->ml_di_dir != NULL) {
253 struct dt_object *dto = lfsck->ml_obj_dir;
255 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
258 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
259 fid_zero(&pos->lp_dir_parent);
260 pos->lp_dir_cookie = 0;
262 pos->lp_dir_parent = *lu_object_fid(&dto->do_lu);
265 fid_zero(&pos->lp_dir_parent);
266 pos->lp_dir_cookie = 0;
268 spin_unlock(&lfsck->ml_lock);
271 static inline void mdd_lfsck_pos_set_zero(struct lfsck_position *pos)
273 memset(pos, 0, sizeof(*pos));
276 static inline int mdd_lfsck_pos_is_zero(const struct lfsck_position *pos)
278 return pos->lp_oit_cookie == 0 && fid_is_zero(&pos->lp_dir_parent);
281 static inline int mdd_lfsck_pos_is_eq(const struct lfsck_position *pos1,
282 const struct lfsck_position *pos2)
284 if (pos1->lp_oit_cookie < pos2->lp_oit_cookie)
287 if (pos1->lp_oit_cookie > pos2->lp_oit_cookie)
290 if (fid_is_zero(&pos1->lp_dir_parent) &&
291 !fid_is_zero(&pos2->lp_dir_parent))
294 if (!fid_is_zero(&pos1->lp_dir_parent) &&
295 fid_is_zero(&pos2->lp_dir_parent))
298 if (fid_is_zero(&pos1->lp_dir_parent) &&
299 fid_is_zero(&pos2->lp_dir_parent))
302 LASSERT(lu_fid_eq(&pos1->lp_dir_parent, &pos2->lp_dir_parent));
304 if (pos1->lp_dir_cookie < pos2->lp_dir_cookie)
307 if (pos1->lp_dir_cookie > pos2->lp_dir_cookie)
313 static void mdd_lfsck_close_dir(const struct lu_env *env,
314 struct md_lfsck *lfsck)
316 struct dt_object *dir_obj = lfsck->ml_obj_dir;
317 const struct dt_it_ops *dir_iops = &dir_obj->do_index_ops->dio_it;
318 struct dt_it *dir_di = lfsck->ml_di_dir;
320 spin_lock(&lfsck->ml_lock);
321 lfsck->ml_di_dir = NULL;
322 spin_unlock(&lfsck->ml_lock);
324 dir_iops->put(env, dir_di);
325 dir_iops->fini(env, dir_di);
326 lfsck->ml_obj_dir = NULL;
327 lu_object_put(env, &dir_obj->do_lu);
330 static void __mdd_lfsck_set_speed(struct md_lfsck *lfsck, __u32 limit)
332 lfsck->ml_bookmark_ram.lb_speed_limit = limit;
333 if (limit != LFSCK_SPEED_NO_LIMIT) {
334 if (limit > CFS_HZ) {
335 lfsck->ml_sleep_rate = limit / CFS_HZ;
336 lfsck->ml_sleep_jif = 1;
338 lfsck->ml_sleep_rate = 1;
339 lfsck->ml_sleep_jif = CFS_HZ / limit;
342 lfsck->ml_sleep_jif = 0;
343 lfsck->ml_sleep_rate = 0;
347 static void mdd_lfsck_control_speed(struct md_lfsck *lfsck)
349 struct ptlrpc_thread *thread = &lfsck->ml_thread;
350 struct l_wait_info lwi;
352 if (lfsck->ml_sleep_jif > 0 &&
353 lfsck->ml_new_scanned >= lfsck->ml_sleep_rate) {
354 spin_lock(&lfsck->ml_lock);
355 if (likely(lfsck->ml_sleep_jif > 0 &&
356 lfsck->ml_new_scanned >= lfsck->ml_sleep_rate)) {
357 lwi = LWI_TIMEOUT_INTR(lfsck->ml_sleep_jif, NULL,
358 LWI_ON_SIGNAL_NOOP, NULL);
359 spin_unlock(&lfsck->ml_lock);
361 l_wait_event(thread->t_ctl_waitq,
362 !thread_is_running(thread),
364 lfsck->ml_new_scanned = 0;
366 spin_unlock(&lfsck->ml_lock);
371 /* lfsck_bookmark file ops */
373 static void inline mdd_lfsck_bookmark_to_cpu(struct lfsck_bookmark *des,
374 struct lfsck_bookmark *src)
376 des->lb_magic = le32_to_cpu(src->lb_magic);
377 des->lb_version = le16_to_cpu(src->lb_version);
378 des->lb_param = le16_to_cpu(src->lb_param);
379 des->lb_speed_limit = le32_to_cpu(src->lb_speed_limit);
382 static void inline mdd_lfsck_bookmark_to_le(struct lfsck_bookmark *des,
383 struct lfsck_bookmark *src)
385 des->lb_magic = cpu_to_le32(src->lb_magic);
386 des->lb_version = cpu_to_le16(src->lb_version);
387 des->lb_param = cpu_to_le16(src->lb_param);
388 des->lb_speed_limit = cpu_to_le32(src->lb_speed_limit);
391 static int mdd_lfsck_bookmark_load(const struct lu_env *env,
392 struct md_lfsck *lfsck)
395 int len = sizeof(struct lfsck_bookmark);
398 rc = dt_record_read(env, lfsck->ml_bookmark_obj,
399 mdd_buf_get(env, &lfsck->ml_bookmark_disk, len),
402 struct lfsck_bookmark *bm = &lfsck->ml_bookmark_ram;
404 mdd_lfsck_bookmark_to_cpu(bm, &lfsck->ml_bookmark_disk);
405 if (bm->lb_magic != LFSCK_BOOKMARK_MAGIC) {
406 CWARN("%.16s: invalid lfsck_bookmark magic "
407 "0x%x != 0x%x\n", mdd_lfsck2name(lfsck),
408 bm->lb_magic, LFSCK_BOOKMARK_MAGIC);
409 /* Process it as new lfsck_bookmark. */
413 if (rc == -EFAULT && pos == 0)
414 /* return -ENODATA for empty lfsck_bookmark. */
417 CERROR("%.16s: fail to load lfsck_bookmark, "
418 "expected = %d, rc = %d\n",
419 mdd_lfsck2name(lfsck), len, rc);
424 static int mdd_lfsck_bookmark_store(const struct lu_env *env,
425 struct md_lfsck *lfsck)
427 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
428 struct thandle *handle;
429 struct dt_object *obj = lfsck->ml_bookmark_obj;
431 int len = sizeof(struct lfsck_bookmark);
435 mdd_lfsck_bookmark_to_le(&lfsck->ml_bookmark_disk,
436 &lfsck->ml_bookmark_ram);
437 handle = dt_trans_create(env, mdd->mdd_bottom);
438 if (IS_ERR(handle)) {
439 rc = PTR_ERR(handle);
440 CERROR("%.16s: fail to create trans for storing "
441 "lfsck_bookmark: %d\n,", mdd_lfsck2name(lfsck), rc);
445 rc = dt_declare_record_write(env, obj, len, 0, handle);
447 CERROR("%.16s: fail to declare trans for storing "
448 "lfsck_bookmark: %d\n,", mdd_lfsck2name(lfsck), rc);
452 rc = dt_trans_start_local(env, mdd->mdd_bottom, handle);
454 CERROR("%.16s: fail to start trans for storing "
455 "lfsck_bookmark: %d\n,", mdd_lfsck2name(lfsck), rc);
459 rc = dt_record_write(env, obj,
460 mdd_buf_get(env, &lfsck->ml_bookmark_disk, len),
463 CERROR("%.16s: fail to store lfsck_bookmark, expected = %d, "
464 "rc = %d\n", mdd_lfsck2name(lfsck), len, rc);
469 dt_trans_stop(env, mdd->mdd_bottom, handle);
473 static int mdd_lfsck_bookmark_init(const struct lu_env *env,
474 struct md_lfsck *lfsck)
476 struct lfsck_bookmark *mb = &lfsck->ml_bookmark_ram;
479 memset(mb, 0, sizeof(*mb));
480 mb->lb_magic = LFSCK_BOOKMARK_MAGIC;
481 mb->lb_version = LFSCK_VERSION_V2;
482 mutex_lock(&lfsck->ml_mutex);
483 rc = mdd_lfsck_bookmark_store(env, lfsck);
484 mutex_unlock(&lfsck->ml_mutex);
488 /* lfsck_namespace file ops */
490 static void inline mdd_lfsck_position_to_cpu(struct lfsck_position *des,
491 struct lfsck_position *src)
493 des->lp_oit_cookie = le64_to_cpu(src->lp_oit_cookie);
494 fid_le_to_cpu(&des->lp_dir_parent, &src->lp_dir_parent);
495 des->lp_dir_cookie = le64_to_cpu(src->lp_dir_cookie);
498 static void inline mdd_lfsck_position_to_le(struct lfsck_position *des,
499 struct lfsck_position *src)
501 des->lp_oit_cookie = cpu_to_le64(src->lp_oit_cookie);
502 fid_cpu_to_le(&des->lp_dir_parent, &src->lp_dir_parent);
503 des->lp_dir_cookie = cpu_to_le64(src->lp_dir_cookie);
506 static void inline mdd_lfsck_namespace_to_cpu(struct lfsck_namespace *des,
507 struct lfsck_namespace *src)
509 des->ln_magic = le32_to_cpu(src->ln_magic);
510 des->ln_status = le32_to_cpu(src->ln_status);
511 des->ln_flags = le32_to_cpu(src->ln_flags);
512 des->ln_success_count = le32_to_cpu(src->ln_success_count);
513 des->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
514 des->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
515 des->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
516 des->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
517 des->ln_time_last_checkpoint =
518 le64_to_cpu(src->ln_time_last_checkpoint);
519 mdd_lfsck_position_to_cpu(&des->ln_pos_latest_start,
520 &src->ln_pos_latest_start);
521 mdd_lfsck_position_to_cpu(&des->ln_pos_last_checkpoint,
522 &src->ln_pos_last_checkpoint);
523 mdd_lfsck_position_to_cpu(&des->ln_pos_first_inconsistent,
524 &src->ln_pos_first_inconsistent);
525 des->ln_items_checked = le64_to_cpu(src->ln_items_checked);
526 des->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
527 des->ln_items_failed = le64_to_cpu(src->ln_items_failed);
528 des->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
529 des->ln_mlinked_checked = le64_to_cpu(src->ln_mlinked_checked);
530 des->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
531 des->ln_objs_repaired_phase2 =
532 le64_to_cpu(src->ln_objs_repaired_phase2);
533 des->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
534 des->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
535 des->ln_objs_lost_found = le64_to_cpu(src->ln_objs_lost_found);
536 fid_le_to_cpu(&des->ln_fid_latest_scanned_phase2,
537 &src->ln_fid_latest_scanned_phase2);
540 static void inline mdd_lfsck_namespace_to_le(struct lfsck_namespace *des,
541 struct lfsck_namespace *src)
543 des->ln_magic = cpu_to_le32(src->ln_magic);
544 des->ln_status = cpu_to_le32(src->ln_status);
545 des->ln_flags = cpu_to_le32(src->ln_flags);
546 des->ln_success_count = cpu_to_le32(src->ln_success_count);
547 des->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
548 des->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
549 des->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
550 des->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
551 des->ln_time_last_checkpoint =
552 cpu_to_le64(src->ln_time_last_checkpoint);
553 mdd_lfsck_position_to_le(&des->ln_pos_latest_start,
554 &src->ln_pos_latest_start);
555 mdd_lfsck_position_to_le(&des->ln_pos_last_checkpoint,
556 &src->ln_pos_last_checkpoint);
557 mdd_lfsck_position_to_le(&des->ln_pos_first_inconsistent,
558 &src->ln_pos_first_inconsistent);
559 des->ln_items_checked = cpu_to_le64(src->ln_items_checked);
560 des->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
561 des->ln_items_failed = cpu_to_le64(src->ln_items_failed);
562 des->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
563 des->ln_mlinked_checked = cpu_to_le64(src->ln_mlinked_checked);
564 des->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
565 des->ln_objs_repaired_phase2 =
566 cpu_to_le64(src->ln_objs_repaired_phase2);
567 des->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
568 des->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
569 des->ln_objs_lost_found = cpu_to_le64(src->ln_objs_lost_found);
570 fid_cpu_to_le(&des->ln_fid_latest_scanned_phase2,
571 &src->ln_fid_latest_scanned_phase2);
575 * \retval +ve: the lfsck_namespace is broken, the caller should reset it.
576 * \retval 0: succeed.
577 * \retval -ve: failed cases.
579 static int mdd_lfsck_namespace_load(const struct lu_env *env,
580 struct lfsck_component *com)
582 int len = com->lc_file_size;
585 rc = dt_xattr_get(env, com->lc_obj,
586 mdd_buf_get(env, com->lc_file_disk, len),
587 XATTR_NAME_LFSCK_NAMESPACE, BYPASS_CAPA);
589 struct lfsck_namespace *ns = com->lc_file_ram;
591 mdd_lfsck_namespace_to_cpu(ns,
592 (struct lfsck_namespace *)com->lc_file_disk);
593 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
594 CWARN("%.16s: invalid lfsck_namespace magic "
596 mdd_lfsck2name(com->lc_lfsck),
597 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
602 } else if (rc != -ENODATA) {
603 CERROR("%.16s: fail to load lfsck_namespace, expected = %d, "
604 "rc = %d\n", mdd_lfsck2name(com->lc_lfsck), len, rc);
611 static int mdd_lfsck_namespace_store(const struct lu_env *env,
612 struct lfsck_component *com, bool init)
614 struct dt_object *obj = com->lc_obj;
615 struct md_lfsck *lfsck = com->lc_lfsck;
616 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
617 struct thandle *handle;
618 int len = com->lc_file_size;
622 mdd_lfsck_namespace_to_le((struct lfsck_namespace *)com->lc_file_disk,
623 (struct lfsck_namespace *)com->lc_file_ram);
624 handle = dt_trans_create(env, mdd->mdd_bottom);
625 if (IS_ERR(handle)) {
626 rc = PTR_ERR(handle);
627 CERROR("%.16s: fail to create trans for storing "
628 "lfsck_namespace: %d\n,", mdd_lfsck2name(lfsck), rc);
632 rc = dt_declare_xattr_set(env, obj,
633 mdd_buf_get(env, com->lc_file_disk, len),
634 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
636 CERROR("%.16s: fail to declare trans for storing "
637 "lfsck_namespace: %d\n,", mdd_lfsck2name(lfsck), rc);
641 rc = dt_trans_start_local(env, mdd->mdd_bottom, handle);
643 CERROR("%.16s: fail to start trans for storing "
644 "lfsck_namespace: %d\n,", mdd_lfsck2name(lfsck), rc);
648 rc = dt_xattr_set(env, obj,
649 mdd_buf_get(env, com->lc_file_disk, len),
650 XATTR_NAME_LFSCK_NAMESPACE,
651 init ? LU_XATTR_CREATE : LU_XATTR_REPLACE,
652 handle, BYPASS_CAPA);
654 CERROR("%.16s: fail to store lfsck_namespace, len = %d, "
655 "rc = %d\n", mdd_lfsck2name(lfsck), len, rc);
660 dt_trans_stop(env, mdd->mdd_bottom, handle);
664 static int mdd_lfsck_namespace_init(const struct lu_env *env,
665 struct lfsck_component *com)
667 struct lfsck_namespace *ns = (struct lfsck_namespace *)com->lc_file_ram;
670 memset(ns, 0, sizeof(*ns));
671 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
672 ns->ln_status = LS_INIT;
673 down_write(&com->lc_sem);
674 rc = mdd_lfsck_namespace_store(env, com, true);
675 up_write(&com->lc_sem);
679 static int mdd_lfsck_namespace_lookup(const struct lu_env *env,
680 struct lfsck_component *com,
681 const struct lu_fid *fid,
684 struct lu_fid *key = &mdd_env_info(env)->mti_fid;
687 fid_cpu_to_be(key, fid);
688 rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)flags,
689 (const struct dt_key *)key, BYPASS_CAPA);
693 static int mdd_lfsck_namespace_delete(const struct lu_env *env,
694 struct lfsck_component *com,
695 const struct lu_fid *fid)
697 struct mdd_device *mdd = mdd_lfsck2mdd(com->lc_lfsck);
698 struct lu_fid *key = &mdd_env_info(env)->mti_fid;
699 struct thandle *handle;
700 struct dt_object *obj = com->lc_obj;
704 handle = dt_trans_create(env, mdd->mdd_bottom);
706 RETURN(PTR_ERR(handle));
708 rc = dt_declare_delete(env, obj, (const struct dt_key *)fid, handle);
712 rc = dt_trans_start_local(env, mdd->mdd_bottom, handle);
716 fid_cpu_to_be(key, fid);
717 rc = dt_delete(env, obj, (const struct dt_key *)key, handle,
723 dt_trans_stop(env, mdd->mdd_bottom, handle);
727 static int mdd_lfsck_namespace_update(const struct lu_env *env,
728 struct lfsck_component *com,
729 const struct lu_fid *fid,
730 __u8 flags, bool force)
732 struct mdd_device *mdd = mdd_lfsck2mdd(com->lc_lfsck);
733 struct lu_fid *key = &mdd_env_info(env)->mti_fid;
734 struct thandle *handle;
735 struct dt_object *obj = com->lc_obj;
741 rc = mdd_lfsck_namespace_lookup(env, com, fid, &tf);
742 if (rc != 0 && rc != -ENOENT)
746 if (!force || flags == tf)
750 handle = dt_trans_create(env, mdd->mdd_bottom);
752 RETURN(PTR_ERR(handle));
754 rc = dt_declare_delete(env, obj, (const struct dt_key *)fid,
759 handle = dt_trans_create(env, mdd->mdd_bottom);
761 RETURN(PTR_ERR(handle));
764 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
765 (const struct dt_key *)fid, handle);
769 rc = dt_trans_start_local(env, mdd->mdd_bottom, handle);
773 fid_cpu_to_be(key, fid);
775 rc = dt_delete(env, obj, (const struct dt_key *)key, handle,
778 CERROR("%s: fail to insert "DFID", rc = %d\n",
779 mdd_lfsck2name(com->lc_lfsck), PFID(fid), rc);
784 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
785 (const struct dt_key *)key, handle, BYPASS_CAPA, 1);
790 dt_trans_stop(env, mdd->mdd_bottom, handle);
795 * \retval +ve repaired
796 * \retval 0 no need to repair
797 * \retval -ve error cases
799 static int mdd_lfsck_namespace_double_scan_one(const struct lu_env *env,
800 struct lfsck_component *com,
801 struct mdd_object *child,
804 struct mdd_thread_info *info = mdd_env_info(env);
805 struct lu_attr *la = &info->mti_la;
806 struct lu_name *cname = &info->mti_name;
807 struct lu_fid *pfid = &info->mti_fid;
808 struct lu_fid *cfid = &info->mti_fid2;
809 struct md_lfsck *lfsck = com->lc_lfsck;
810 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
811 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
812 struct lfsck_namespace *ns =
813 (struct lfsck_namespace *)com->lc_file_ram;
814 struct linkea_data ldata = { 0 };
815 struct thandle *handle = NULL;
822 if (com->lc_journal) {
828 handle = mdd_trans_create(env, mdd);
830 RETURN(rc = PTR_ERR(handle));
832 rc = mdd_declare_links_add(env, child, handle, NULL);
836 rc = mdd_trans_start(env, mdd, handle);
840 mdd_write_lock(env, child, MOR_TGT_CHILD);
844 if (unlikely(mdd_is_dead_obj(child)))
847 rc = mdd_links_read(env, child, &ldata);
849 if ((bk->lb_param & LPF_DRYRUN) &&
850 (rc == -EINVAL || rc == -ENODATA))
856 rc = mdd_la_get(env, child, la, BYPASS_CAPA);
860 ldata.ld_lee = LINKEA_FIRST_ENTRY(ldata);
861 count = ldata.ld_leh->leh_reccount;
862 while (count-- > 0) {
863 struct mdd_object *parent = NULL;
864 struct dt_object *dir;
866 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname,
868 if (!fid_is_sane(pfid))
871 parent = mdd_object_find(env, mdd, pfid);
874 else if (IS_ERR(parent))
875 GOTO(stop, rc = PTR_ERR(parent));
877 if (!mdd_object_exists(parent))
880 /* XXX: need more processing for remote object in the future. */
881 if (mdd_object_remote(parent)) {
882 mdd_object_put(env, parent);
883 ldata.ld_lee = LINKEA_NEXT_ENTRY(ldata);
887 dir = mdd_object_child(parent);
888 if (unlikely(!dt_try_as_dir(env, dir)))
891 /* To guarantee the 'name' is terminated with '0'. */
892 memcpy(info->mti_key, cname->ln_name, cname->ln_namelen);
893 info->mti_key[cname->ln_namelen] = 0;
894 cname->ln_name = info->mti_key;
895 rc = dt_lookup(env, dir, (struct dt_rec *)cfid,
896 (const struct dt_key *)cname->ln_name,
898 if (rc != 0 && rc != -ENOENT) {
899 mdd_object_put(env, parent);
904 if (lu_fid_eq(cfid, mdo2fid(child))) {
905 mdd_object_put(env, parent);
906 ldata.ld_lee = LINKEA_NEXT_ENTRY(ldata);
913 if (ldata.ld_leh->leh_reccount > la->la_nlink)
916 /* XXX: For the case of there is linkea entry, but without name
917 * entry pointing to the object, and the object link count
918 * isn't less than the count of name entries, then add the
919 * name entry back to namespace.
921 * It is out of LFSCK 1.5 scope, will implement it in the
922 * future. Keep the linkEA entry. */
923 mdd_object_put(env, parent);
924 ldata.ld_lee = LINKEA_NEXT_ENTRY(ldata);
929 mdd_object_put(env, parent);
930 if (bk->lb_param & LPF_DRYRUN)
933 CDEBUG(D_LFSCK, "Remove linkEA: "DFID"[%.*s], "DFID"\n",
934 PFID(mdo2fid(child)), cname->ln_namelen, cname->ln_name,
936 linkea_del_buf(&ldata, cname);
941 if (!com->lc_journal) {
946 rc = mdd_links_write(env, child, &ldata, handle);
953 mdd_write_unlock(env, child);
956 mdd_trans_stop(env, mdd, rc, handle);
958 if (rc == 0 && update) {
959 ns->ln_objs_nlink_repaired++;
967 static int mdd_lfsck_namespace_reset(const struct lu_env *env,
968 struct lfsck_component *com, bool init)
970 struct lfsck_namespace *ns = (struct lfsck_namespace *)com->lc_file_ram;
971 struct mdd_device *mdd = mdd_lfsck2mdd(com->lc_lfsck);
972 struct dt_object *dto, *root;
976 down_write(&com->lc_sem);
978 memset(ns, 0, sizeof(*ns));
980 __u32 count = ns->ln_success_count;
981 __u64 last_time = ns->ln_time_last_complete;
983 memset(ns, 0, sizeof(*ns));
984 ns->ln_success_count = count;
985 ns->ln_time_last_complete = last_time;
987 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
988 ns->ln_status = LS_INIT;
990 root = dt_locate(env, mdd->mdd_bottom, &mdd->mdd_local_root_fid);
991 if (unlikely(IS_ERR(root)))
992 GOTO(out, rc = PTR_ERR(root));
994 rc = local_object_unlink(env, mdd->mdd_bottom, root,
995 lfsck_namespace_name);
999 dto = local_index_find_or_create(env, mdd->mdd_los, root,
1000 lfsck_namespace_name,
1001 S_IFREG | S_IRUGO | S_IWUSR,
1002 &dt_lfsck_features);
1004 GOTO(out, rc = PTR_ERR(dto));
1006 rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features);
1011 rc = mdd_lfsck_namespace_store(env, com, true);
1015 lu_object_put(env, &root->do_lu);
1016 up_write(&com->lc_sem);
1021 mdd_lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
1024 struct lfsck_namespace *ns = (struct lfsck_namespace *)com->lc_file_ram;
1026 down_write(&com->lc_sem);
1028 com->lc_new_checked++;
1029 ns->ln_items_failed++;
1030 if (mdd_lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
1031 mdd_lfsck_pos_fill(env, com->lc_lfsck,
1032 &ns->ln_pos_first_inconsistent, false);
1033 up_write(&com->lc_sem);
1036 static int mdd_lfsck_namespace_checkpoint(const struct lu_env *env,
1037 struct lfsck_component *com,
1040 struct md_lfsck *lfsck = com->lc_lfsck;
1041 struct lfsck_namespace *ns =
1042 (struct lfsck_namespace *)com->lc_file_ram;
1045 if (com->lc_new_checked == 0 && !init)
1048 down_write(&com->lc_sem);
1051 ns->ln_pos_latest_start = lfsck->ml_pos_current;
1053 ns->ln_pos_last_checkpoint = lfsck->ml_pos_current;
1054 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
1055 HALF_SEC - lfsck->ml_time_last_checkpoint);
1056 ns->ln_time_last_checkpoint = cfs_time_current_sec();
1057 ns->ln_items_checked += com->lc_new_checked;
1058 com->lc_new_checked = 0;
1061 rc = mdd_lfsck_namespace_store(env, com, false);
1063 up_write(&com->lc_sem);
1067 static int mdd_lfsck_namespace_prep(const struct lu_env *env,
1068 struct lfsck_component *com)
1070 struct md_lfsck *lfsck = com->lc_lfsck;
1071 struct lfsck_namespace *ns =
1072 (struct lfsck_namespace *)com->lc_file_ram;
1073 struct lfsck_position *pos = &com->lc_pos_start;
1075 if (ns->ln_status == LS_COMPLETED) {
1078 rc = mdd_lfsck_namespace_reset(env, com, false);
1083 down_write(&com->lc_sem);
1085 ns->ln_time_latest_start = cfs_time_current_sec();
1087 spin_lock(&lfsck->ml_lock);
1088 if (ns->ln_flags & LF_SCANNED_ONCE) {
1089 if (!lfsck->ml_drop_dryrun ||
1090 mdd_lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
1091 ns->ln_status = LS_SCANNING_PHASE2;
1092 cfs_list_del_init(&com->lc_link);
1093 cfs_list_add_tail(&com->lc_link,
1094 &lfsck->ml_list_double_scan);
1095 if (!cfs_list_empty(&com->lc_link_dir))
1096 cfs_list_del_init(&com->lc_link_dir);
1097 mdd_lfsck_pos_set_zero(pos);
1099 ns->ln_status = LS_SCANNING_PHASE1;
1100 ns->ln_run_time_phase1 = 0;
1101 ns->ln_run_time_phase2 = 0;
1102 ns->ln_items_checked = 0;
1103 ns->ln_items_repaired = 0;
1104 ns->ln_items_failed = 0;
1105 ns->ln_dirs_checked = 0;
1106 ns->ln_mlinked_checked = 0;
1107 ns->ln_objs_checked_phase2 = 0;
1108 ns->ln_objs_repaired_phase2 = 0;
1109 ns->ln_objs_failed_phase2 = 0;
1110 ns->ln_objs_nlink_repaired = 0;
1111 ns->ln_objs_lost_found = 0;
1112 fid_zero(&ns->ln_fid_latest_scanned_phase2);
1113 if (cfs_list_empty(&com->lc_link_dir))
1114 cfs_list_add_tail(&com->lc_link_dir,
1115 &lfsck->ml_list_dir);
1116 *pos = ns->ln_pos_first_inconsistent;
1119 ns->ln_status = LS_SCANNING_PHASE1;
1120 if (cfs_list_empty(&com->lc_link_dir))
1121 cfs_list_add_tail(&com->lc_link_dir,
1122 &lfsck->ml_list_dir);
1123 if (!lfsck->ml_drop_dryrun ||
1124 mdd_lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
1125 *pos = ns->ln_pos_last_checkpoint;
1126 pos->lp_oit_cookie++;
1128 *pos = ns->ln_pos_first_inconsistent;
1131 spin_unlock(&lfsck->ml_lock);
1133 up_write(&com->lc_sem);
1137 static int mdd_lfsck_namespace_exec_oit(const struct lu_env *env,
1138 struct lfsck_component *com,
1139 struct mdd_object *obj)
1141 down_write(&com->lc_sem);
1142 com->lc_new_checked++;
1143 if (S_ISDIR(mdd_object_type(obj)))
1144 ((struct lfsck_namespace *)com->lc_file_ram)->ln_dirs_checked++;
1145 up_write(&com->lc_sem);
1149 static int mdd_declare_lfsck_namespace_exec_dir(const struct lu_env *env,
1150 struct mdd_object *obj,
1151 struct thandle *handle)
1155 /* For destroying all invalid linkEA entries. */
1156 rc = mdo_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
1160 /* For insert new linkEA entry. */
1161 rc = mdd_declare_links_add(env, obj, handle, NULL);
1165 static int mdd_lfsck_namespace_check_exist(const struct lu_env *env,
1166 struct md_lfsck *lfsck,
1167 struct mdd_object *obj,
1170 struct dt_object *dir = lfsck->ml_obj_dir;
1171 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
1175 if (unlikely(mdd_is_dead_obj(obj)))
1176 RETURN(LFSCK_NAMEENTRY_DEAD);
1178 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
1179 (const struct dt_key *)name, BYPASS_CAPA);
1181 RETURN(LFSCK_NAMEENTRY_REMOVED);
1186 if (!lu_fid_eq(fid, mdo2fid(obj)))
1187 RETURN(LFSCK_NAMEENTRY_RECREATED);
1192 static int mdd_lfsck_namespace_exec_dir(const struct lu_env *env,
1193 struct lfsck_component *com,
1194 struct mdd_object *obj,
1195 struct lu_dirent *ent)
1197 struct mdd_thread_info *info = mdd_env_info(env);
1198 struct lu_attr *la = &info->mti_la;
1199 struct md_lfsck *lfsck = com->lc_lfsck;
1200 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
1201 struct lfsck_namespace *ns =
1202 (struct lfsck_namespace *)com->lc_file_ram;
1203 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
1204 struct linkea_data ldata = { 0 };
1205 const struct lu_fid *pfid =
1206 lu_object_fid(&lfsck->ml_obj_dir->do_lu);
1207 const struct lu_fid *cfid = mdo2fid(obj);
1208 const struct lu_name *cname;
1209 struct thandle *handle = NULL;
1210 bool repaired = false;
1211 bool locked = false;
1216 cname = mdd_name_get_const(env, ent->lde_name, ent->lde_namelen);
1217 down_write(&com->lc_sem);
1218 com->lc_new_checked++;
1220 if (ent->lde_attrs & LUDA_UPGRADE) {
1221 ns->ln_flags |= LF_UPGRADE;
1223 } else if (ent->lde_attrs & LUDA_REPAIR) {
1224 ns->ln_flags |= LF_INCONSISTENT;
1228 if (ent->lde_name[0] == '.' &&
1229 (ent->lde_namelen == 1 ||
1230 (ent->lde_namelen == 2 && ent->lde_name[1] == '.') ||
1231 fid_is_dot_lustre(&ent->lde_fid)))
1234 if (!(bk->lb_param & LPF_DRYRUN) &&
1235 (com->lc_journal || repaired)) {
1240 com->lc_journal = 1;
1241 handle = mdd_trans_create(env, mdd);
1243 GOTO(out, rc = PTR_ERR(handle));
1245 rc = mdd_declare_lfsck_namespace_exec_dir(env, obj, handle);
1249 rc = mdd_trans_start(env, mdd, handle);
1253 mdd_write_lock(env, obj, MOR_TGT_CHILD);
1257 rc = mdd_lfsck_namespace_check_exist(env, lfsck, obj, ent->lde_name);
1261 rc = mdd_links_read(env, obj, &ldata);
1263 count = ldata.ld_leh->leh_reccount;
1264 rc = linkea_links_find(&ldata, cname, pfid);
1266 /* For dir, if there are more than one linkea entries,
1267 * then remove all the other redundant linkea entries.*/
1268 if (unlikely(count > 1 &&
1269 S_ISDIR(mdd_object_type(obj))))
1276 ns->ln_flags |= LF_INCONSISTENT;
1277 if (bk->lb_param & LPF_DRYRUN) {
1282 /*For dir, remove the unmatched linkea entry directly.*/
1283 if (S_ISDIR(mdd_object_type(obj))) {
1284 if (!com->lc_journal)
1287 rc = mdo_xattr_del(env, obj, XATTR_NAME_LINK,
1288 handle, BYPASS_CAPA);
1297 } else if (unlikely(rc == -EINVAL)) {
1298 ns->ln_flags |= LF_INCONSISTENT;
1299 if (bk->lb_param & LPF_DRYRUN) {
1305 if (!com->lc_journal)
1308 /* The magic crashed, we are not sure whether there are more
1309 * corrupt data in the linkea, so remove all linkea entries. */
1310 rc = mdo_xattr_del(env, obj, XATTR_NAME_LINK, handle,
1316 } else if (rc == -ENODATA) {
1317 ns->ln_flags |= LF_UPGRADE;
1318 if (bk->lb_param & LPF_DRYRUN) {
1325 rc = linkea_data_new(&ldata, &mdd_env_info(env)->mti_link_buf);
1330 if (!com->lc_journal)
1333 rc = linkea_add_buf(&ldata, cname, pfid);
1337 rc = mdd_links_write(env, obj, &ldata, handle);
1341 count = ldata.ld_leh->leh_reccount;
1350 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
1355 (la->la_nlink == 1 || S_ISDIR(mdd_object_type(obj))))
1356 /* Usually, it is for single linked object or dir, do nothing.*/
1359 /* Following modification will be in another transaction. */
1360 if (handle != NULL) {
1361 LASSERT(mdd_write_locked(env, obj));
1363 mdd_write_unlock(env, obj);
1366 mdd_trans_stop(env, mdd, 0, handle);
1370 ns->ln_mlinked_checked++;
1371 rc = mdd_lfsck_namespace_update(env, com, cfid,
1372 count != la->la_nlink ? LLF_UNMATCH_NLINKS : 0, false);
1378 mdd_write_unlock(env, obj);
1381 mdd_trans_stop(env, mdd, rc, handle);
1385 ns->ln_items_failed++;
1386 if (mdd_lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
1387 mdd_lfsck_pos_fill(env, lfsck,
1388 &ns->ln_pos_first_inconsistent,
1390 if (!(bk->lb_param & LPF_FAILOUT))
1394 ns->ln_items_repaired++;
1396 com->lc_journal = 0;
1399 up_write(&com->lc_sem);
1403 static int mdd_lfsck_namespace_post(const struct lu_env *env,
1404 struct lfsck_component *com,
1405 int result, bool init)
1407 struct md_lfsck *lfsck = com->lc_lfsck;
1408 struct lfsck_namespace *ns =
1409 (struct lfsck_namespace *)com->lc_file_ram;
1412 down_write(&com->lc_sem);
1414 spin_lock(&lfsck->ml_lock);
1416 ns->ln_pos_last_checkpoint = lfsck->ml_pos_current;
1418 ns->ln_status = LS_SCANNING_PHASE2;
1419 ns->ln_flags |= LF_SCANNED_ONCE;
1420 ns->ln_flags &= ~LF_UPGRADE;
1421 cfs_list_del_init(&com->lc_link);
1422 cfs_list_del_init(&com->lc_link_dir);
1423 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_double_scan);
1424 } else if (result == 0) {
1425 if (lfsck->ml_paused) {
1426 ns->ln_status = LS_PAUSED;
1428 ns->ln_status = LS_STOPPED;
1429 cfs_list_del_init(&com->lc_link);
1430 cfs_list_del_init(&com->lc_link_dir);
1431 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_idle);
1434 ns->ln_status = LS_FAILED;
1435 cfs_list_del_init(&com->lc_link);
1436 cfs_list_del_init(&com->lc_link_dir);
1437 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_idle);
1439 spin_unlock(&lfsck->ml_lock);
1442 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
1443 HALF_SEC - lfsck->ml_time_last_checkpoint);
1444 ns->ln_time_last_checkpoint = cfs_time_current_sec();
1445 ns->ln_items_checked += com->lc_new_checked;
1446 com->lc_new_checked = 0;
1449 rc = mdd_lfsck_namespace_store(env, com, false);
1451 up_write(&com->lc_sem);
1456 mdd_lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
1459 struct md_lfsck *lfsck = com->lc_lfsck;
1460 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
1461 struct lfsck_namespace *ns =
1462 (struct lfsck_namespace *)com->lc_file_ram;
1467 down_read(&com->lc_sem);
1468 rc = snprintf(buf, len,
1469 "name: lfsck_namespace\n"
1475 lfsck_status_names[ns->ln_status]);
1481 rc = lfsck_bits_dump(&buf, &len, ns->ln_flags, lfsck_flags_names,
1486 rc = lfsck_bits_dump(&buf, &len, bk->lb_param, lfsck_param_names,
1491 rc = lfsck_time_dump(&buf, &len, ns->ln_time_last_complete,
1492 "time_since_last_completed");
1496 rc = lfsck_time_dump(&buf, &len, ns->ln_time_latest_start,
1497 "time_since_latest_start");
1501 rc = lfsck_time_dump(&buf, &len, ns->ln_time_last_checkpoint,
1502 "time_since_last_checkpoint");
1506 rc = lfsck_pos_dump(&buf, &len, &ns->ln_pos_latest_start,
1507 "latest_start_position");
1511 rc = lfsck_pos_dump(&buf, &len, &ns->ln_pos_last_checkpoint,
1512 "last_checkpoint_position");
1516 rc = lfsck_pos_dump(&buf, &len, &ns->ln_pos_first_inconsistent,
1517 "first_failure_position");
1521 if (ns->ln_status == LS_SCANNING_PHASE1) {
1522 struct lfsck_position pos;
1523 cfs_duration_t duration = cfs_time_current() -
1524 lfsck->ml_time_last_checkpoint;
1525 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
1526 __u64 speed = checked;
1527 __u64 new_checked = com->lc_new_checked * CFS_HZ;
1528 __u32 rtime = ns->ln_run_time_phase1 +
1529 cfs_duration_sec(duration + HALF_SEC);
1532 do_div(new_checked, duration);
1534 do_div(speed, rtime);
1535 rc = snprintf(buf, len,
1536 "checked_phase1: "LPU64"\n"
1537 "checked_phase2: "LPU64"\n"
1538 "updated_phase1: "LPU64"\n"
1539 "updated_phase2: "LPU64"\n"
1540 "failed_phase1: "LPU64"\n"
1541 "failed_phase2: "LPU64"\n"
1543 "M-linked: "LPU64"\n"
1544 "nlinks_repaired: "LPU64"\n"
1545 "lost_found: "LPU64"\n"
1546 "success_count: %u\n"
1547 "run_time_phase1: %u seconds\n"
1548 "run_time_phase2: %u seconds\n"
1549 "average_speed_phase1: "LPU64" items/sec\n"
1550 "average_speed_phase2: N/A\n"
1551 "real-time_speed_phase1: "LPU64" items/sec\n"
1552 "real-time_speed_phase2: N/A\n",
1554 ns->ln_objs_checked_phase2,
1555 ns->ln_items_repaired,
1556 ns->ln_objs_repaired_phase2,
1557 ns->ln_items_failed,
1558 ns->ln_objs_failed_phase2,
1559 ns->ln_dirs_checked,
1560 ns->ln_mlinked_checked,
1561 ns->ln_objs_nlink_repaired,
1562 ns->ln_objs_lost_found,
1563 ns->ln_success_count,
1565 ns->ln_run_time_phase2,
1573 mdd_lfsck_pos_fill(env, lfsck, &pos, false);
1574 rc = lfsck_pos_dump(&buf, &len, &pos, "current_position");
1577 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
1578 cfs_duration_t duration = cfs_time_current() -
1579 lfsck->ml_time_last_checkpoint;
1580 __u64 checked = ns->ln_objs_checked_phase2 +
1581 com->lc_new_checked;
1582 __u64 speed1 = ns->ln_items_checked;
1583 __u64 speed2 = checked;
1584 __u64 new_checked = com->lc_new_checked * CFS_HZ;
1585 __u32 rtime = ns->ln_run_time_phase2 +
1586 cfs_duration_sec(duration + HALF_SEC);
1589 do_div(new_checked, duration);
1590 if (ns->ln_run_time_phase1 != 0)
1591 do_div(speed1, ns->ln_run_time_phase1);
1593 do_div(speed2, rtime);
1594 rc = snprintf(buf, len,
1595 "checked_phase1: "LPU64"\n"
1596 "checked_phase2: "LPU64"\n"
1597 "updated_phase1: "LPU64"\n"
1598 "updated_phase2: "LPU64"\n"
1599 "failed_phase1: "LPU64"\n"
1600 "failed_phase2: "LPU64"\n"
1602 "M-linked: "LPU64"\n"
1603 "nlinks_repaired: "LPU64"\n"
1604 "lost_found: "LPU64"\n"
1605 "success_count: %u\n"
1606 "run_time_phase1: %u seconds\n"
1607 "run_time_phase2: %u seconds\n"
1608 "average_speed_phase1: "LPU64" items/sec\n"
1609 "average_speed_phase2: "LPU64" objs/sec\n"
1610 "real-time_speed_phase1: N/A\n"
1611 "real-time_speed_phase2: "LPU64" objs/sec\n"
1612 "current_position: "DFID"\n",
1613 ns->ln_items_checked,
1615 ns->ln_items_repaired,
1616 ns->ln_objs_repaired_phase2,
1617 ns->ln_items_failed,
1618 ns->ln_objs_failed_phase2,
1619 ns->ln_dirs_checked,
1620 ns->ln_mlinked_checked,
1621 ns->ln_objs_nlink_repaired,
1622 ns->ln_objs_lost_found,
1623 ns->ln_success_count,
1624 ns->ln_run_time_phase1,
1629 PFID(&ns->ln_fid_latest_scanned_phase2));
1636 __u64 speed1 = ns->ln_items_checked;
1637 __u64 speed2 = ns->ln_objs_checked_phase2;
1639 if (ns->ln_run_time_phase1 != 0)
1640 do_div(speed1, ns->ln_run_time_phase1);
1641 if (ns->ln_run_time_phase2 != 0)
1642 do_div(speed2, ns->ln_run_time_phase2);
1643 rc = snprintf(buf, len,
1644 "checked_phase1: "LPU64"\n"
1645 "checked_phase2: "LPU64"\n"
1646 "updated_phase1: "LPU64"\n"
1647 "updated_phase2: "LPU64"\n"
1648 "failed_phase1: "LPU64"\n"
1649 "failed_phase2: "LPU64"\n"
1651 "M-linked: "LPU64"\n"
1652 "nlinks_repaired: "LPU64"\n"
1653 "lost_found: "LPU64"\n"
1654 "success_count: %u\n"
1655 "run_time_phase1: %u seconds\n"
1656 "run_time_phase2: %u seconds\n"
1657 "average_speed_phase1: "LPU64" items/sec\n"
1658 "average_speed_phase2: "LPU64" objs/sec\n"
1659 "real-time_speed_phase1: N/A\n"
1660 "real-time_speed_phase2: N/A\n"
1661 "current_position: N/A\n",
1662 ns->ln_items_checked,
1663 ns->ln_objs_checked_phase2,
1664 ns->ln_items_repaired,
1665 ns->ln_objs_repaired_phase2,
1666 ns->ln_items_failed,
1667 ns->ln_objs_failed_phase2,
1668 ns->ln_dirs_checked,
1669 ns->ln_mlinked_checked,
1670 ns->ln_objs_nlink_repaired,
1671 ns->ln_objs_lost_found,
1672 ns->ln_success_count,
1673 ns->ln_run_time_phase1,
1674 ns->ln_run_time_phase2,
1686 up_read(&com->lc_sem);
1690 static int mdd_lfsck_namespace_double_scan(const struct lu_env *env,
1691 struct lfsck_component *com)
1693 struct md_lfsck *lfsck = com->lc_lfsck;
1694 struct ptlrpc_thread *thread = &lfsck->ml_thread;
1695 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
1696 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
1697 struct lfsck_namespace *ns =
1698 (struct lfsck_namespace *)com->lc_file_ram;
1699 struct dt_object *obj = com->lc_obj;
1700 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
1701 struct mdd_object *target;
1709 lfsck->ml_new_scanned = 0;
1710 lfsck->ml_time_last_checkpoint = cfs_time_current();
1711 lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
1712 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1714 di = iops->init(env, obj, 0, BYPASS_CAPA);
1716 RETURN(PTR_ERR(di));
1718 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
1719 rc = iops->get(env, di, (const struct dt_key *)&fid);
1723 /* Skip the start one, which either has been processed or non-exist. */
1724 rc = iops->next(env, di);
1728 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1732 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
1734 struct l_wait_info lwi;
1736 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
1738 l_wait_event(thread->t_ctl_waitq,
1739 !thread_is_running(thread),
1743 key = iops->key(env, di);
1744 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
1745 target = mdd_object_find(env, mdd, &fid);
1746 down_write(&com->lc_sem);
1747 if (target == NULL) {
1750 } else if (IS_ERR(target)) {
1751 rc = PTR_ERR(target);
1755 /* XXX: need more processing for remote object in the future. */
1756 if (!mdd_object_exists(target) || mdd_object_remote(target))
1759 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
1761 rc = mdd_lfsck_namespace_double_scan_one(env, com,
1765 mdd_object_put(env, target);
1768 lfsck->ml_new_scanned++;
1769 com->lc_new_checked++;
1770 ns->ln_fid_latest_scanned_phase2 = fid;
1772 ns->ln_objs_repaired_phase2++;
1774 ns->ln_objs_failed_phase2++;
1775 up_write(&com->lc_sem);
1777 if ((rc == 0) || ((rc > 0) && !(bk->lb_param & LPF_DRYRUN))) {
1778 mdd_lfsck_namespace_delete(env, com, &fid);
1779 } else if (rc < 0) {
1780 flags |= LLF_REPAIR_FAILED;
1781 mdd_lfsck_namespace_update(env, com, &fid, flags, true);
1784 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1787 if (likely(cfs_time_beforeq(cfs_time_current(),
1788 lfsck->ml_time_next_checkpoint)) ||
1789 com->lc_new_checked == 0)
1792 down_write(&com->lc_sem);
1793 ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1794 HALF_SEC - lfsck->ml_time_last_checkpoint);
1795 ns->ln_time_last_checkpoint = cfs_time_current_sec();
1796 ns->ln_objs_checked_phase2 += com->lc_new_checked;
1797 com->lc_new_checked = 0;
1798 rc = mdd_lfsck_namespace_store(env, com, false);
1799 up_write(&com->lc_sem);
1803 lfsck->ml_time_last_checkpoint = cfs_time_current();
1804 lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
1805 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1808 mdd_lfsck_control_speed(lfsck);
1809 if (unlikely(!thread_is_running(thread)))
1812 rc = iops->next(env, di);
1821 iops->fini(env, di);
1822 down_write(&com->lc_sem);
1824 ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1825 HALF_SEC - lfsck->ml_time_last_checkpoint);
1826 ns->ln_time_last_checkpoint = cfs_time_current_sec();
1827 ns->ln_objs_checked_phase2 += com->lc_new_checked;
1828 com->lc_new_checked = 0;
1831 com->lc_journal = 0;
1832 ns->ln_status = LS_COMPLETED;
1833 if (!(bk->lb_param & LPF_DRYRUN))
1835 ~(LF_SCANNED_ONCE | LF_INCONSISTENT | LF_UPGRADE);
1836 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
1837 ns->ln_success_count++;
1838 } else if (rc == 0) {
1839 if (lfsck->ml_paused)
1840 ns->ln_status = LS_PAUSED;
1842 ns->ln_status = LS_STOPPED;
1844 ns->ln_status = LS_FAILED;
1847 if (ns->ln_status != LS_PAUSED) {
1848 spin_lock(&lfsck->ml_lock);
1849 cfs_list_del_init(&com->lc_link);
1850 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_idle);
1851 spin_unlock(&lfsck->ml_lock);
1854 rc = mdd_lfsck_namespace_store(env, com, false);
1856 up_write(&com->lc_sem);
1860 static struct lfsck_operations mdd_lfsck_namespace_ops = {
1861 .lfsck_reset = mdd_lfsck_namespace_reset,
1862 .lfsck_fail = mdd_lfsck_namespace_fail,
1863 .lfsck_checkpoint = mdd_lfsck_namespace_checkpoint,
1864 .lfsck_prep = mdd_lfsck_namespace_prep,
1865 .lfsck_exec_oit = mdd_lfsck_namespace_exec_oit,
1866 .lfsck_exec_dir = mdd_lfsck_namespace_exec_dir,
1867 .lfsck_post = mdd_lfsck_namespace_post,
1868 .lfsck_dump = mdd_lfsck_namespace_dump,
1869 .lfsck_double_scan = mdd_lfsck_namespace_double_scan,
1872 /* LFSCK component setup/cleanup functions */
1874 static int mdd_lfsck_namespace_setup(const struct lu_env *env,
1875 struct md_lfsck *lfsck)
1877 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
1878 struct lfsck_component *com;
1879 struct lfsck_namespace *ns;
1880 struct dt_object *obj, *root;
1888 CFS_INIT_LIST_HEAD(&com->lc_link);
1889 CFS_INIT_LIST_HEAD(&com->lc_link_dir);
1890 init_rwsem(&com->lc_sem);
1891 atomic_set(&com->lc_ref, 1);
1892 com->lc_lfsck = lfsck;
1893 com->lc_type = LT_NAMESPACE;
1894 com->lc_ops = &mdd_lfsck_namespace_ops;
1895 com->lc_file_size = sizeof(struct lfsck_namespace);
1896 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
1897 if (com->lc_file_ram == NULL)
1898 GOTO(out, rc = -ENOMEM);
1900 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
1901 if (com->lc_file_disk == NULL)
1902 GOTO(out, rc = -ENOMEM);
1904 root = dt_locate(env, mdd->mdd_bottom, &mdd->mdd_local_root_fid);
1905 if (unlikely(IS_ERR(root)))
1906 GOTO(out, rc = PTR_ERR(root));
1908 obj = local_index_find_or_create(env, mdd->mdd_los, root,
1909 lfsck_namespace_name,
1910 S_IFREG | S_IRUGO | S_IWUSR,
1911 &dt_lfsck_features);
1912 lu_object_put(env, &root->do_lu);
1914 GOTO(out, rc = PTR_ERR(obj));
1917 rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
1921 rc = mdd_lfsck_namespace_load(env, com);
1923 rc = mdd_lfsck_namespace_reset(env, com, true);
1924 else if (rc == -ENODATA)
1925 rc = mdd_lfsck_namespace_init(env, com);
1929 ns = (struct lfsck_namespace *)com->lc_file_ram;
1930 switch (ns->ln_status) {
1935 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_idle);
1938 CERROR("%s: unknown status: %u\n",
1939 mdd_lfsck2name(lfsck), ns->ln_status);
1941 case LS_SCANNING_PHASE1:
1942 case LS_SCANNING_PHASE2:
1943 /* No need to store the status to disk right now.
1944 * If the system crashed before the status stored,
1945 * it will be loaded back when next time. */
1946 ns->ln_status = LS_CRASHED;
1950 cfs_list_add_tail(&com->lc_link, &lfsck->ml_list_scan);
1951 cfs_list_add_tail(&com->lc_link_dir, &lfsck->ml_list_dir);
1959 mdd_lfsck_component_cleanup(env, com);
1963 /* helper functions for framework */
1965 static int object_needs_lfsck(const struct lu_env *env, struct mdd_device *mdd,
1966 struct mdd_object *obj)
1968 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
1972 LASSERT(S_ISDIR(mdd_object_type(obj)));
1975 if (mdd_is_root(mdd, mdo2fid(obj))) {
1977 mdd_object_put(env, obj);
1981 /* .lustre doesn't contain "real" user objects, no need lfsck */
1982 if (fid_is_dot_lustre(mdo2fid(obj))) {
1984 mdd_object_put(env, obj);
1988 mdd_read_lock(env, obj, MOR_TGT_CHILD);
1989 if (unlikely(mdd_is_dead_obj(obj))) {
1990 mdd_read_unlock(env, obj);
1992 mdd_object_put(env, obj);
1996 rc = dt_xattr_get(env, mdd_object_child(obj),
1997 mdd_buf_get(env, NULL, 0), XATTR_NAME_LINK,
1999 mdd_read_unlock(env, obj);
2002 mdd_object_put(env, obj);
2006 if (rc < 0 && rc != -ENODATA) {
2008 mdd_object_put(env, obj);
2012 rc = mdd_parent_fid(env, obj, fid);
2014 mdd_object_put(env, obj);
2018 if (unlikely(lu_fid_eq(fid, &mdd->mdd_local_root_fid)))
2021 obj = mdd_object_find(env, mdd, fid);
2024 else if (IS_ERR(obj))
2025 return PTR_ERR(obj);
2027 if (!mdd_object_exists(obj)) {
2028 mdd_object_put(env, obj);
2032 /* Currently, only client visible directory can be remote. */
2033 if (mdd_object_remote(obj)) {
2034 mdd_object_put(env, obj);
2043 static void mdd_lfsck_unpack_ent(struct lu_dirent *ent)
2045 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
2046 ent->lde_hash = le64_to_cpu(ent->lde_hash);
2047 ent->lde_reclen = le16_to_cpu(ent->lde_reclen);
2048 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
2049 ent->lde_attrs = le32_to_cpu(ent->lde_attrs);
2051 /* Make sure the name is terminated with '0'.
2052 * The data (type) after ent::lde_name maybe
2053 * broken, but we do not care. */
2054 ent->lde_name[ent->lde_namelen] = 0;
2057 /* LFSCK wrap functions */
2059 static void mdd_lfsck_fail(const struct lu_env *env, struct md_lfsck *lfsck,
2062 struct lfsck_component *com;
2064 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2065 com->lc_ops->lfsck_fail(env, com, new_checked);
2069 static int mdd_lfsck_checkpoint(const struct lu_env *env,
2070 struct md_lfsck *lfsck)
2072 struct lfsck_component *com;
2075 if (likely(cfs_time_beforeq(cfs_time_current(),
2076 lfsck->ml_time_next_checkpoint)))
2079 mdd_lfsck_pos_fill(env, lfsck, &lfsck->ml_pos_current, false);
2080 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2081 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
2086 lfsck->ml_time_last_checkpoint = cfs_time_current();
2087 lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
2088 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
2092 static int mdd_lfsck_prep(struct lu_env *env, struct md_lfsck *lfsck)
2094 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
2095 struct mdd_object *obj = NULL;
2096 struct dt_object *dt_obj;
2097 struct lfsck_component *com;
2098 struct lfsck_component *next;
2099 struct lfsck_position *pos = NULL;
2100 const struct dt_it_ops *iops =
2101 &lfsck->ml_obj_oit->do_index_ops->dio_it;
2106 LASSERT(lfsck->ml_obj_dir == NULL);
2107 LASSERT(lfsck->ml_di_dir == NULL);
2109 lfsck->ml_current_oit_processed = 0;
2110 cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_scan, lc_link) {
2111 com->lc_new_checked = 0;
2112 if (lfsck->ml_bookmark_ram.lb_param & LPF_DRYRUN)
2113 com->lc_journal = 0;
2115 rc = com->lc_ops->lfsck_prep(env, com);
2119 if ((pos == NULL) ||
2120 (!mdd_lfsck_pos_is_zero(&com->lc_pos_start) &&
2121 mdd_lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
2122 pos = &com->lc_pos_start;
2125 /* Init otable-based iterator. */
2127 rc = iops->load(env, lfsck->ml_di_oit, 0);
2129 lfsck->ml_oit_over = 1;
2136 rc = iops->load(env, lfsck->ml_di_oit, pos->lp_oit_cookie);
2140 lfsck->ml_oit_over = 1;
2142 if (fid_is_zero(&pos->lp_dir_parent))
2145 /* Find the directory for namespace-based traverse. */
2146 obj = mdd_object_find(env, mdd, &pos->lp_dir_parent);
2149 else if (IS_ERR(obj))
2150 RETURN(PTR_ERR(obj));
2152 /* XXX: need more processing for remote object in the future. */
2153 if (!mdd_object_exists(obj) || mdd_object_remote(obj) ||
2154 unlikely(!S_ISDIR(mdd_object_type(obj))))
2157 if (unlikely(mdd_is_dead_obj(obj)))
2160 dt_obj = mdd_object_child(obj);
2161 if (unlikely(!dt_try_as_dir(env, dt_obj)))
2162 GOTO(out, rc = -ENOTDIR);
2164 /* Init the namespace-based directory traverse. */
2165 iops = &dt_obj->do_index_ops->dio_it;
2166 di = iops->init(env, dt_obj, lfsck->ml_args_dir, BYPASS_CAPA);
2168 GOTO(out, rc = PTR_ERR(di));
2170 LASSERT(pos->lp_dir_cookie < MDS_DIR_END_OFF);
2172 rc = iops->load(env, di, pos->lp_dir_cookie);
2173 if ((rc == 0) || (rc > 0 && pos->lp_dir_cookie > 0))
2174 rc = iops->next(env, di);
2180 iops->fini(env, di);
2184 lfsck->ml_obj_dir = dt_obj;
2185 spin_lock(&lfsck->ml_lock);
2186 lfsck->ml_di_dir = di;
2187 spin_unlock(&lfsck->ml_lock);
2194 mdd_object_put(env, obj);
2197 cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_scan,
2199 com->lc_ops->lfsck_post(env, com, rc, true);
2205 mdd_lfsck_pos_fill(env, lfsck, &lfsck->ml_pos_current, true);
2206 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2207 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
2212 lfsck->ml_time_last_checkpoint = cfs_time_current();
2213 lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
2214 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
2218 static int mdd_lfsck_exec_oit(const struct lu_env *env, struct md_lfsck *lfsck,
2219 struct mdd_object *obj)
2221 struct lfsck_component *com;
2222 struct dt_object *dt_obj;
2223 const struct dt_it_ops *iops;
2228 LASSERT(lfsck->ml_obj_dir == NULL);
2230 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2231 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
2236 if (!S_ISDIR(mdd_object_type(obj)) ||
2237 cfs_list_empty(&lfsck->ml_list_dir))
2240 rc = object_needs_lfsck(env, mdd_lfsck2mdd(lfsck), obj);
2244 if (unlikely(mdd_is_dead_obj(obj)))
2247 dt_obj = mdd_object_child(obj);
2248 if (unlikely(!dt_try_as_dir(env, dt_obj)))
2249 GOTO(out, rc = -ENOTDIR);
2251 iops = &dt_obj->do_index_ops->dio_it;
2252 di = iops->init(env, dt_obj, lfsck->ml_args_dir, BYPASS_CAPA);
2254 GOTO(out, rc = PTR_ERR(di));
2256 rc = iops->load(env, di, 0);
2258 rc = iops->next(env, di);
2264 iops->fini(env, di);
2268 mdd_object_get(obj);
2269 lfsck->ml_obj_dir = dt_obj;
2270 spin_lock(&lfsck->ml_lock);
2271 lfsck->ml_di_dir = di;
2272 spin_unlock(&lfsck->ml_lock);
2278 mdd_lfsck_fail(env, lfsck, false);
2279 return (rc > 0 ? 0 : rc);
2282 static int mdd_lfsck_exec_dir(const struct lu_env *env, struct md_lfsck *lfsck,
2283 struct mdd_object *obj, struct lu_dirent *ent)
2285 struct lfsck_component *com;
2288 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2289 rc = com->lc_ops->lfsck_exec_dir(env, com, obj, ent);
2296 static int mdd_lfsck_post(const struct lu_env *env, struct md_lfsck *lfsck,
2299 struct lfsck_component *com;
2300 struct lfsck_component *next;
2303 mdd_lfsck_pos_fill(env, lfsck, &lfsck->ml_pos_current, false);
2304 cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_scan, lc_link) {
2305 rc = com->lc_ops->lfsck_post(env, com, result, false);
2310 lfsck->ml_time_last_checkpoint = cfs_time_current();
2311 lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
2312 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
2316 static int mdd_lfsck_double_scan(const struct lu_env *env,
2317 struct md_lfsck *lfsck)
2319 struct lfsck_component *com;
2320 struct lfsck_component *next;
2323 cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_double_scan,
2325 if (lfsck->ml_bookmark_ram.lb_param & LPF_DRYRUN)
2326 com->lc_journal = 0;
2328 rc = com->lc_ops->lfsck_double_scan(env, com);
2337 static int mdd_lfsck_dir_engine(const struct lu_env *env,
2338 struct md_lfsck *lfsck)
2340 struct mdd_thread_info *info = mdd_env_info(env);
2341 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
2342 const struct dt_it_ops *iops =
2343 &lfsck->ml_obj_dir->do_index_ops->dio_it;
2344 struct dt_it *di = lfsck->ml_di_dir;
2345 struct lu_dirent *ent = &info->mti_ent;
2346 struct lu_fid *fid = &info->mti_fid;
2347 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
2348 struct ptlrpc_thread *thread = &lfsck->ml_thread;
2353 struct mdd_object *child;
2355 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY2) &&
2357 struct l_wait_info lwi;
2359 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
2361 l_wait_event(thread->t_ctl_waitq,
2362 !thread_is_running(thread),
2366 lfsck->ml_new_scanned++;
2367 rc = iops->rec(env, di, (struct dt_rec *)ent,
2368 lfsck->ml_args_dir);
2370 mdd_lfsck_fail(env, lfsck, true);
2371 if (bk->lb_param & LPF_FAILOUT)
2377 mdd_lfsck_unpack_ent(ent);
2378 if (ent->lde_attrs & LUDA_IGNORE)
2381 *fid = ent->lde_fid;
2382 child = mdd_object_find(env, mdd, fid);
2383 if (child == NULL) {
2385 } else if (IS_ERR(child)) {
2386 mdd_lfsck_fail(env, lfsck, true);
2387 if (bk->lb_param & LPF_FAILOUT)
2388 RETURN(PTR_ERR(child));
2393 /* XXX: need more processing for remote object in the future. */
2394 if (mdd_object_exists(child) && !mdd_object_remote(child))
2395 rc = mdd_lfsck_exec_dir(env, lfsck, child, ent);
2396 mdd_object_put(env, child);
2397 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2401 rc = mdd_lfsck_checkpoint(env, lfsck);
2402 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2406 mdd_lfsck_control_speed(lfsck);
2407 if (unlikely(!thread_is_running(thread)))
2410 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
2411 spin_lock(&lfsck->ml_lock);
2412 thread_set_flags(thread, SVC_STOPPING);
2413 spin_unlock(&lfsck->ml_lock);
2417 rc = iops->next(env, di);
2420 if (rc > 0 && !lfsck->ml_oit_over)
2421 mdd_lfsck_close_dir(env, lfsck);
2426 static int mdd_lfsck_oit_engine(const struct lu_env *env,
2427 struct md_lfsck *lfsck)
2429 struct mdd_thread_info *info = mdd_env_info(env);
2430 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
2431 const struct dt_it_ops *iops =
2432 &lfsck->ml_obj_oit->do_index_ops->dio_it;
2433 struct dt_it *di = lfsck->ml_di_oit;
2434 struct lu_fid *fid = &info->mti_fid;
2435 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
2436 struct ptlrpc_thread *thread = &lfsck->ml_thread;
2441 struct mdd_object *target;
2443 if (lfsck->ml_di_dir != NULL) {
2444 rc = mdd_lfsck_dir_engine(env, lfsck);
2449 if (unlikely(lfsck->ml_oit_over))
2452 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY1) &&
2454 struct l_wait_info lwi;
2456 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
2458 l_wait_event(thread->t_ctl_waitq,
2459 !thread_is_running(thread),
2463 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
2466 lfsck->ml_current_oit_processed = 1;
2467 lfsck->ml_new_scanned++;
2468 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
2470 mdd_lfsck_fail(env, lfsck, true);
2471 if (bk->lb_param & LPF_FAILOUT)
2477 target = mdd_object_find(env, mdd, fid);
2478 if (target == NULL) {
2480 } else if (IS_ERR(target)) {
2481 mdd_lfsck_fail(env, lfsck, true);
2482 if (bk->lb_param & LPF_FAILOUT)
2483 RETURN(PTR_ERR(target));
2488 /* XXX: In fact, low layer otable-based iteration should not
2489 * return agent object. But before LU-2646 resolved, we
2490 * need more processing for agent object. */
2491 if (mdd_object_exists(target) && !mdd_object_remote(target))
2492 rc = mdd_lfsck_exec_oit(env, lfsck, target);
2493 mdd_object_put(env, target);
2494 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2498 rc = mdd_lfsck_checkpoint(env, lfsck);
2499 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2503 mdd_lfsck_control_speed(lfsck);
2505 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
2506 spin_lock(&lfsck->ml_lock);
2507 thread_set_flags(thread, SVC_STOPPING);
2508 spin_unlock(&lfsck->ml_lock);
2512 rc = iops->next(env, di);
2513 if (unlikely(rc > 0))
2514 lfsck->ml_oit_over = 1;
2515 else if (likely(rc == 0))
2516 lfsck->ml_current_oit_processed = 0;
2518 if (unlikely(!thread_is_running(thread)))
2520 } while (rc == 0 || lfsck->ml_di_dir != NULL);
2525 static int mdd_lfsck_main(void *args)
2528 struct md_lfsck *lfsck = (struct md_lfsck *)args;
2529 struct ptlrpc_thread *thread = &lfsck->ml_thread;
2530 struct dt_object *oit_obj = lfsck->ml_obj_oit;
2531 const struct dt_it_ops *oit_iops = &oit_obj->do_index_ops->dio_it;
2532 struct dt_it *oit_di;
2536 cfs_daemonize("lfsck");
2537 rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2539 CERROR("%s: LFSCK, fail to init env, rc = %d\n",
2540 mdd_lfsck2name(lfsck), rc);
2544 oit_di = oit_iops->init(&env, oit_obj, lfsck->ml_args_oit, BYPASS_CAPA);
2545 if (IS_ERR(oit_di)) {
2546 rc = PTR_ERR(oit_di);
2547 CERROR("%s: LFSCK, fail to init iteration, rc = %d\n",
2548 mdd_lfsck2name(lfsck), rc);
2552 spin_lock(&lfsck->ml_lock);
2553 lfsck->ml_di_oit = oit_di;
2554 spin_unlock(&lfsck->ml_lock);
2555 rc = mdd_lfsck_prep(&env, lfsck);
2559 CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = 0x%x, dir_flags = 0x%x, "
2560 "oit_cookie = "LPU64", dir_cookie = "LPU64", parent = "DFID
2561 ", pid = %d\n", lfsck->ml_args_oit, lfsck->ml_args_dir,
2562 lfsck->ml_pos_current.lp_oit_cookie,
2563 lfsck->ml_pos_current.lp_dir_cookie,
2564 PFID(&lfsck->ml_pos_current.lp_dir_parent),
2567 spin_lock(&lfsck->ml_lock);
2568 thread_set_flags(thread, SVC_RUNNING);
2569 spin_unlock(&lfsck->ml_lock);
2570 cfs_waitq_broadcast(&thread->t_ctl_waitq);
2572 if (!cfs_list_empty(&lfsck->ml_list_scan) ||
2573 cfs_list_empty(&lfsck->ml_list_double_scan))
2574 rc = mdd_lfsck_oit_engine(&env, lfsck);
2578 CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = 0x%x, dir_flags = 0x%x, "
2579 "oit_cookie = "LPU64", dir_cookie = "LPU64", parent = "DFID
2580 ", pid = %d, rc = %d\n", lfsck->ml_args_oit, lfsck->ml_args_dir,
2581 lfsck->ml_pos_current.lp_oit_cookie,
2582 lfsck->ml_pos_current.lp_dir_cookie,
2583 PFID(&lfsck->ml_pos_current.lp_dir_parent),
2584 cfs_curproc_pid(), rc);
2586 if (lfsck->ml_paused && cfs_list_empty(&lfsck->ml_list_scan))
2587 oit_iops->put(&env, oit_di);
2589 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
2590 rc = mdd_lfsck_post(&env, lfsck, rc);
2591 if (lfsck->ml_di_dir != NULL)
2592 mdd_lfsck_close_dir(&env, lfsck);
2595 spin_lock(&lfsck->ml_lock);
2596 lfsck->ml_di_oit = NULL;
2597 spin_unlock(&lfsck->ml_lock);
2599 oit_iops->fini(&env, oit_di);
2601 if (!cfs_list_empty(&lfsck->ml_list_double_scan))
2602 rc = mdd_lfsck_double_scan(&env, lfsck);
2607 /* XXX: Purge the pinned objects in the future. */
2613 spin_lock(&lfsck->ml_lock);
2614 thread_set_flags(thread, SVC_STOPPED);
2615 cfs_waitq_broadcast(&thread->t_ctl_waitq);
2616 spin_unlock(&lfsck->ml_lock);
2620 /* external interfaces */
2622 int mdd_lfsck_set_speed(const struct lu_env *env, struct md_lfsck *lfsck,
2627 mutex_lock(&lfsck->ml_mutex);
2628 __mdd_lfsck_set_speed(lfsck, limit);
2629 rc = mdd_lfsck_bookmark_store(env, lfsck);
2630 mutex_unlock(&lfsck->ml_mutex);
2634 int mdd_lfsck_dump(const struct lu_env *env, struct md_lfsck *lfsck,
2635 __u16 type, char *buf, int len)
2637 struct lfsck_component *com;
2640 if (!lfsck->ml_initialized)
2643 com = mdd_lfsck_component_find(lfsck, type);
2647 rc = com->lc_ops->lfsck_dump(env, com, buf, len);
2648 mdd_lfsck_component_put(env, com);
2652 int mdd_lfsck_start(const struct lu_env *env, struct md_lfsck *lfsck,
2653 struct lfsck_start *start)
2655 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
2656 struct ptlrpc_thread *thread = &lfsck->ml_thread;
2657 struct lfsck_component *com;
2658 struct l_wait_info lwi = { 0 };
2665 if (lfsck->ml_obj_oit == NULL)
2668 /* start == NULL means auto trigger paused LFSCK. */
2669 if ((start == NULL) &&
2670 (cfs_list_empty(&lfsck->ml_list_scan) ||
2671 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO)))
2674 mutex_lock(&lfsck->ml_mutex);
2675 spin_lock(&lfsck->ml_lock);
2676 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
2677 spin_unlock(&lfsck->ml_lock);
2678 mutex_unlock(&lfsck->ml_mutex);
2682 spin_unlock(&lfsck->ml_lock);
2684 lfsck->ml_paused = 0;
2685 lfsck->ml_oit_over = 0;
2686 lfsck->ml_drop_dryrun = 0;
2687 lfsck->ml_new_scanned = 0;
2689 /* For auto trigger. */
2693 start->ls_version = bk->lb_version;
2694 if (start->ls_valid & LSV_SPEED_LIMIT) {
2695 __mdd_lfsck_set_speed(lfsck, start->ls_speed_limit);
2699 if (start->ls_valid & LSV_ERROR_HANDLE) {
2700 valid |= DOIV_ERROR_HANDLE;
2701 if (start->ls_flags & LPF_FAILOUT)
2702 flags |= DOIF_FAILOUT;
2704 if ((start->ls_flags & LPF_FAILOUT) &&
2705 !(bk->lb_param & LPF_FAILOUT)) {
2706 bk->lb_param |= LPF_FAILOUT;
2708 } else if (!(start->ls_flags & LPF_FAILOUT) &&
2709 (bk->lb_param & LPF_FAILOUT)) {
2710 bk->lb_param &= ~LPF_FAILOUT;
2715 if (start->ls_valid & LSV_DRYRUN) {
2716 if ((start->ls_flags & LPF_DRYRUN) &&
2717 !(bk->lb_param & LPF_DRYRUN)) {
2718 bk->lb_param |= LPF_DRYRUN;
2720 } else if (!(start->ls_flags & LPF_DRYRUN) &&
2721 (bk->lb_param & LPF_DRYRUN)) {
2722 bk->lb_param &= ~LPF_DRYRUN;
2723 lfsck->ml_drop_dryrun = 1;
2729 rc = mdd_lfsck_bookmark_store(env, lfsck);
2734 if (start->ls_flags & LPF_RESET)
2735 flags |= DOIF_RESET;
2737 if (start->ls_active != 0) {
2738 struct lfsck_component *next;
2741 if (start->ls_active == LFSCK_TYPES_ALL)
2742 start->ls_active = LFSCK_TYPES_SUPPORTED;
2744 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
2745 start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
2746 GOTO(out, rc = -ENOTSUPP);
2749 cfs_list_for_each_entry_safe(com, next,
2750 &lfsck->ml_list_scan, lc_link) {
2751 if (!(com->lc_type & start->ls_active)) {
2752 rc = com->lc_ops->lfsck_post(env, com, 0,
2759 while (start->ls_active != 0) {
2760 if (type & start->ls_active) {
2761 com = __mdd_lfsck_component_find(lfsck, type,
2762 &lfsck->ml_list_idle);
2764 /* The component status will be updated
2765 * when its prep() is called later by
2766 * the LFSCK main engine. */
2767 cfs_list_del_init(&com->lc_link);
2768 cfs_list_add_tail(&com->lc_link,
2769 &lfsck->ml_list_scan);
2771 start->ls_active &= ~type;
2777 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
2778 start->ls_active |= com->lc_type;
2779 if (flags & DOIF_RESET) {
2780 rc = com->lc_ops->lfsck_reset(env, com, false);
2787 lfsck->ml_args_dir = LUDA_64BITHASH | LUDA_VERIFY;
2788 if (bk->lb_param & LPF_DRYRUN)
2789 lfsck->ml_args_dir |= LUDA_VERIFY_DRYRUN;
2791 if (bk->lb_param & LPF_FAILOUT) {
2792 valid |= DOIV_ERROR_HANDLE;
2793 flags |= DOIF_FAILOUT;
2796 if (!cfs_list_empty(&lfsck->ml_list_scan))
2797 flags |= DOIF_OUTUSED;
2799 lfsck->ml_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
2800 thread_set_flags(thread, 0);
2801 rc = cfs_create_thread(mdd_lfsck_main, lfsck, 0);
2803 CERROR("%s: cannot start LFSCK thread, rc = %d\n",
2804 mdd_lfsck2name(lfsck), rc);
2806 l_wait_event(thread->t_ctl_waitq,
2807 thread_is_running(thread) ||
2808 thread_is_stopped(thread),
2814 mutex_unlock(&lfsck->ml_mutex);
2815 return (rc < 0 ? rc : 0);
2818 int mdd_lfsck_stop(const struct lu_env *env, struct md_lfsck *lfsck,
2821 struct ptlrpc_thread *thread = &lfsck->ml_thread;
2822 struct l_wait_info lwi = { 0 };
2825 if (!lfsck->ml_initialized)
2828 mutex_lock(&lfsck->ml_mutex);
2829 spin_lock(&lfsck->ml_lock);
2830 if (thread_is_init(thread) || thread_is_stopped(thread)) {
2831 spin_unlock(&lfsck->ml_lock);
2832 mutex_unlock(&lfsck->ml_mutex);
2837 lfsck->ml_paused = 1;
2838 thread_set_flags(thread, SVC_STOPPING);
2839 /* The LFSCK thread may be sleeping on low layer wait queue,
2841 if (likely(lfsck->ml_di_oit != NULL))
2842 lfsck->ml_obj_oit->do_index_ops->dio_it.put(env,
2844 spin_unlock(&lfsck->ml_lock);
2846 cfs_waitq_broadcast(&thread->t_ctl_waitq);
2847 l_wait_event(thread->t_ctl_waitq,
2848 thread_is_stopped(thread),
2850 mutex_unlock(&lfsck->ml_mutex);
2855 static const struct lu_fid lfsck_it_fid = { .f_seq = FID_SEQ_LOCAL_FILE,
2856 .f_oid = OTABLE_IT_OID,
2859 int mdd_lfsck_setup(const struct lu_env *env, struct mdd_device *mdd)
2861 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
2862 struct dt_object *obj;
2868 LASSERT(!lfsck->ml_initialized);
2870 lfsck->ml_initialized = 1;
2871 mutex_init(&lfsck->ml_mutex);
2872 spin_lock_init(&lfsck->ml_lock);
2873 CFS_INIT_LIST_HEAD(&lfsck->ml_list_scan);
2874 CFS_INIT_LIST_HEAD(&lfsck->ml_list_dir);
2875 CFS_INIT_LIST_HEAD(&lfsck->ml_list_double_scan);
2876 CFS_INIT_LIST_HEAD(&lfsck->ml_list_idle);
2877 cfs_waitq_init(&lfsck->ml_thread.t_ctl_waitq);
2879 obj = dt_locate(env, mdd->mdd_bottom, &lfsck_it_fid);
2881 RETURN(PTR_ERR(obj));
2883 lfsck->ml_obj_oit = obj;
2884 rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
2886 if (rc == -ENOTSUPP)
2891 /* LFSCK bookmark */
2893 rc = mdd_local_file_create(env, mdd, &mdd->mdd_local_root_fid,
2894 lfsck_bookmark_name,
2895 S_IFREG | S_IRUGO | S_IWUSR, &fid);
2899 obj = dt_locate(env, mdd->mdd_bottom, &fid);
2901 GOTO(out, rc = PTR_ERR(obj));
2903 LASSERT(lu_object_exists(&obj->do_lu));
2904 lfsck->ml_bookmark_obj = obj;
2906 rc = mdd_lfsck_bookmark_load(env, lfsck);
2908 rc = mdd_lfsck_bookmark_init(env, lfsck);
2912 rc = mdd_lfsck_namespace_setup(env, lfsck);
2915 /* XXX: LFSCK components initialization to be added here. */
2918 lu_object_put(env, &lfsck->ml_obj_oit->do_lu);
2919 lfsck->ml_obj_oit = NULL;
2923 void mdd_lfsck_cleanup(const struct lu_env *env, struct mdd_device *mdd)
2925 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
2926 struct ptlrpc_thread *thread = &lfsck->ml_thread;
2927 struct lfsck_component *com;
2929 if (!lfsck->ml_initialized)
2932 LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
2934 if (lfsck->ml_obj_oit != NULL) {
2935 lu_object_put(env, &lfsck->ml_obj_oit->do_lu);
2936 lfsck->ml_obj_oit = NULL;
2939 LASSERT(lfsck->ml_obj_dir == NULL);
2941 if (lfsck->ml_bookmark_obj != NULL) {
2942 lu_object_put(env, &lfsck->ml_bookmark_obj->do_lu);
2943 lfsck->ml_bookmark_obj = NULL;
2946 while (!cfs_list_empty(&lfsck->ml_list_scan)) {
2947 com = cfs_list_entry(lfsck->ml_list_scan.next,
2948 struct lfsck_component,
2950 mdd_lfsck_component_cleanup(env, com);
2953 LASSERT(cfs_list_empty(&lfsck->ml_list_dir));
2955 while (!cfs_list_empty(&lfsck->ml_list_double_scan)) {
2956 com = cfs_list_entry(lfsck->ml_list_double_scan.next,
2957 struct lfsck_component,
2959 mdd_lfsck_component_cleanup(env, com);
2962 while (!cfs_list_empty(&lfsck->ml_list_idle)) {
2963 com = cfs_list_entry(lfsck->ml_list_idle.next,
2964 struct lfsck_component,
2966 mdd_lfsck_component_cleanup(env, com);