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
50 #define MDS_DIR_DUMMY_START 0xffffffffffffffffULL
52 const char lfsck_bookmark_name[] = "lfsck_bookmark";
56 static inline struct mdd_device *mdd_lfsck2mdd(struct md_lfsck *lfsck)
58 return container_of0(lfsck, struct mdd_device, mdd_lfsck);
61 static inline char *mdd_lfsck2name(struct md_lfsck *lfsck)
63 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
65 return mdd2obd_dev(mdd)->obd_name;
68 static inline void mdd_lfsck_component_put(const struct lu_env *env,
69 struct lfsck_component *com)
71 if (atomic_dec_and_test(&com->lc_ref)) {
72 if (com->lc_obj != NULL)
73 lu_object_put(env, &com->lc_obj->do_lu);
74 if (com->lc_file_ram != NULL)
75 OBD_FREE(com->lc_file_ram, com->lc_file_size);
76 if (com->lc_file_disk != NULL)
77 OBD_FREE(com->lc_file_disk, com->lc_file_size);
82 static inline struct lfsck_component *
83 __mdd_lfsck_component_find(struct md_lfsck *lfsck, __u16 type, cfs_list_t *list)
85 struct lfsck_component *com;
87 cfs_list_for_each_entry(com, list, lc_link) {
88 if (com->lc_type == type)
94 static void mdd_lfsck_component_cleanup(const struct lu_env *env,
95 struct lfsck_component *com)
97 if (!cfs_list_empty(&com->lc_link))
98 cfs_list_del_init(&com->lc_link);
99 if (!cfs_list_empty(&com->lc_link_dir))
100 cfs_list_del_init(&com->lc_link_dir);
102 mdd_lfsck_component_put(env, com);
105 static void mdd_lfsck_pos_fill(const struct lu_env *env, struct md_lfsck *lfsck,
106 struct lfsck_position *pos, bool oit_processed,
109 const struct dt_it_ops *iops = &lfsck->ml_obj_oit->do_index_ops->dio_it;
111 spin_lock(&lfsck->ml_lock);
112 if (unlikely(lfsck->ml_di_oit == NULL)) {
113 spin_unlock(&lfsck->ml_lock);
114 memset(pos, 0, sizeof(*pos));
118 pos->lp_oit_cookie = iops->store(env, lfsck->ml_di_oit);
120 LASSERT(pos->lp_oit_cookie > 0);
123 pos->lp_oit_cookie--;
125 if (lfsck->ml_di_dir != NULL) {
126 struct dt_object *dto = lfsck->ml_obj_dir;
128 pos->lp_dir_parent = *lu_object_fid(&dto->do_lu);
129 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
132 LASSERT(pos->lp_dir_cookie != MDS_DIR_DUMMY_START);
134 if (pos->lp_dir_cookie == MDS_DIR_END_OFF)
135 LASSERT(dir_processed);
137 /* For the dir which just to be processed,
138 * lp_dir_cookie will become MDS_DIR_DUMMY_START,
139 * which can be correctly handled by mdd_lfsck_prep. */
141 pos->lp_dir_cookie--;
143 fid_zero(&pos->lp_dir_parent);
144 pos->lp_dir_cookie = 0;
146 spin_unlock(&lfsck->ml_lock);
149 static inline int mdd_lfsck_pos_is_zero(const struct lfsck_position *pos)
151 return pos->lp_oit_cookie == 0 && fid_is_zero(&pos->lp_dir_parent);
154 static inline int mdd_lfsck_pos_is_eq(const struct lfsck_position *pos1,
155 const struct lfsck_position *pos2)
157 if (pos1->lp_oit_cookie < pos2->lp_oit_cookie)
160 if (pos1->lp_oit_cookie > pos2->lp_oit_cookie)
163 if (fid_is_zero(&pos1->lp_dir_parent) &&
164 !fid_is_zero(&pos2->lp_dir_parent))
167 if (!fid_is_zero(&pos1->lp_dir_parent) &&
168 fid_is_zero(&pos2->lp_dir_parent))
171 if (fid_is_zero(&pos1->lp_dir_parent) &&
172 fid_is_zero(&pos2->lp_dir_parent))
175 LASSERT(lu_fid_eq(&pos1->lp_dir_parent, &pos2->lp_dir_parent));
177 if (pos1->lp_dir_cookie < pos2->lp_dir_cookie)
180 if (pos1->lp_dir_cookie > pos2->lp_dir_cookie)
186 static void mdd_lfsck_close_dir(const struct lu_env *env,
187 struct md_lfsck *lfsck)
189 struct dt_object *dir_obj = lfsck->ml_obj_dir;
190 const struct dt_it_ops *dir_iops = &dir_obj->do_index_ops->dio_it;
191 struct dt_it *dir_di = lfsck->ml_di_dir;
193 spin_lock(&lfsck->ml_lock);
194 lfsck->ml_di_dir = NULL;
195 spin_unlock(&lfsck->ml_lock);
197 dir_iops->put(env, dir_di);
198 dir_iops->fini(env, dir_di);
199 lfsck->ml_obj_dir = NULL;
200 lu_object_put(env, &dir_obj->do_lu);
203 static void __mdd_lfsck_set_speed(struct md_lfsck *lfsck, __u32 limit)
205 lfsck->ml_bookmark_ram.lb_speed_limit = limit;
206 if (limit != LFSCK_SPEED_NO_LIMIT) {
207 if (limit > CFS_HZ) {
208 lfsck->ml_sleep_rate = limit / CFS_HZ;
209 lfsck->ml_sleep_jif = 1;
211 lfsck->ml_sleep_rate = 1;
212 lfsck->ml_sleep_jif = CFS_HZ / limit;
215 lfsck->ml_sleep_jif = 0;
216 lfsck->ml_sleep_rate = 0;
220 static void mdd_lfsck_control_speed(struct md_lfsck *lfsck)
222 struct ptlrpc_thread *thread = &lfsck->ml_thread;
223 struct l_wait_info lwi;
225 if (lfsck->ml_sleep_jif > 0 &&
226 lfsck->ml_new_scanned >= lfsck->ml_sleep_rate) {
227 spin_lock(&lfsck->ml_lock);
228 if (likely(lfsck->ml_sleep_jif > 0 &&
229 lfsck->ml_new_scanned >= lfsck->ml_sleep_rate)) {
230 lwi = LWI_TIMEOUT_INTR(lfsck->ml_sleep_jif, NULL,
231 LWI_ON_SIGNAL_NOOP, NULL);
232 spin_unlock(&lfsck->ml_lock);
234 l_wait_event(thread->t_ctl_waitq,
235 !thread_is_running(thread),
237 lfsck->ml_new_scanned = 0;
239 spin_unlock(&lfsck->ml_lock);
244 /* lfsck_bookmark file ops */
246 static void inline mdd_lfsck_bookmark_to_cpu(struct lfsck_bookmark *des,
247 struct lfsck_bookmark *src)
249 des->lb_magic = le32_to_cpu(src->lb_magic);
250 des->lb_version = le16_to_cpu(src->lb_version);
251 des->lb_param = le16_to_cpu(src->lb_param);
252 des->lb_speed_limit = le32_to_cpu(src->lb_speed_limit);
255 static void inline mdd_lfsck_bookmark_to_le(struct lfsck_bookmark *des,
256 struct lfsck_bookmark *src)
258 des->lb_magic = cpu_to_le32(src->lb_magic);
259 des->lb_version = cpu_to_le16(src->lb_version);
260 des->lb_param = cpu_to_le16(src->lb_param);
261 des->lb_speed_limit = cpu_to_le32(src->lb_speed_limit);
264 static int mdd_lfsck_bookmark_load(const struct lu_env *env,
265 struct md_lfsck *lfsck)
268 int len = sizeof(struct lfsck_bookmark);
271 rc = dt_record_read(env, lfsck->ml_bookmark_obj,
272 mdd_buf_get(env, &lfsck->ml_bookmark_disk, len),
275 struct lfsck_bookmark *bm = &lfsck->ml_bookmark_ram;
277 mdd_lfsck_bookmark_to_cpu(bm, &lfsck->ml_bookmark_disk);
278 if (bm->lb_magic != LFSCK_BOOKMARK_MAGIC) {
279 CWARN("%.16s: invalid lfsck_bookmark magic "
280 "0x%x != 0x%x\n", mdd_lfsck2name(lfsck),
281 bm->lb_magic, LFSCK_BOOKMARK_MAGIC);
282 /* Process it as new lfsck_bookmark. */
286 if (rc == -EFAULT && pos == 0)
287 /* return -ENODATA for empty lfsck_bookmark. */
290 CERROR("%.16s: fail to load lfsck_bookmark, "
291 "expected = %d, rc = %d\n",
292 mdd_lfsck2name(lfsck), len, rc);
297 static int mdd_lfsck_bookmark_store(const struct lu_env *env,
298 struct md_lfsck *lfsck)
300 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
301 struct thandle *handle;
302 struct dt_object *obj = lfsck->ml_bookmark_obj;
304 int len = sizeof(struct lfsck_bookmark);
308 mdd_lfsck_bookmark_to_le(&lfsck->ml_bookmark_disk,
309 &lfsck->ml_bookmark_ram);
310 handle = dt_trans_create(env, mdd->mdd_bottom);
311 if (IS_ERR(handle)) {
312 rc = PTR_ERR(handle);
313 CERROR("%.16s: fail to create trans for storing "
314 "lfsck_bookmark: %d\n,", mdd_lfsck2name(lfsck), rc);
318 rc = dt_declare_record_write(env, obj, len, 0, handle);
320 CERROR("%.16s: fail to declare trans for storing "
321 "lfsck_bookmark: %d\n,", mdd_lfsck2name(lfsck), rc);
325 rc = dt_trans_start_local(env, mdd->mdd_bottom, handle);
327 CERROR("%.16s: fail to start trans for storing "
328 "lfsck_bookmark: %d\n,", mdd_lfsck2name(lfsck), rc);
332 rc = dt_record_write(env, obj,
333 mdd_buf_get(env, &lfsck->ml_bookmark_disk, len),
336 CERROR("%.16s: fail to store lfsck_bookmark, expected = %d, "
337 "rc = %d\n", mdd_lfsck2name(lfsck), len, rc);
342 dt_trans_stop(env, mdd->mdd_bottom, handle);
346 static int mdd_lfsck_bookmark_init(const struct lu_env *env,
347 struct md_lfsck *lfsck)
349 struct lfsck_bookmark *mb = &lfsck->ml_bookmark_ram;
352 memset(mb, 0, sizeof(mb));
353 mb->lb_magic = LFSCK_BOOKMARK_MAGIC;
354 mb->lb_version = LFSCK_VERSION_V1;
355 mutex_lock(&lfsck->ml_mutex);
356 rc = mdd_lfsck_bookmark_store(env, lfsck);
357 mutex_unlock(&lfsck->ml_mutex);
361 /* helper functions for framework */
363 static int object_is_client_visible(const struct lu_env *env,
364 struct mdd_device *mdd,
365 struct mdd_object *obj)
367 struct lu_fid *fid = &mdd_env_info(env)->mti_fid;
371 LASSERT(S_ISDIR(mdd_object_type(obj)));
374 if (mdd_is_root(mdd, mdo2fid(obj))) {
376 mdd_object_put(env, obj);
380 mdd_read_lock(env, obj, MOR_TGT_CHILD);
381 if (unlikely(mdd_is_dead_obj(obj))) {
382 mdd_read_unlock(env, obj);
384 mdd_object_put(env, obj);
388 rc = dt_xattr_get(env, mdd_object_child(obj),
389 mdd_buf_get(env, NULL, 0), XATTR_NAME_LINK,
391 mdd_read_unlock(env, obj);
394 mdd_object_put(env, obj);
398 if (rc < 0 && rc != -ENODATA) {
400 mdd_object_put(env, obj);
404 rc = mdd_parent_fid(env, obj, fid);
406 mdd_object_put(env, obj);
410 if (unlikely(lu_fid_eq(fid, &mdd->mdd_local_root_fid)))
413 obj = mdd_object_find(env, mdd, fid);
416 else if (IS_ERR(obj))
419 /* XXX: need more processing for remote object in the future. */
420 if (!mdd_object_exists(obj) || mdd_object_remote(obj)) {
421 mdd_object_put(env, obj);
430 static void mdd_lfsck_unpack_ent(struct lu_dirent *ent)
432 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
433 ent->lde_hash = le64_to_cpu(ent->lde_hash);
434 ent->lde_reclen = le16_to_cpu(ent->lde_reclen);
435 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
436 ent->lde_attrs = le32_to_cpu(ent->lde_attrs);
438 /* Make sure the name is terminated with '0'.
439 * The data (type) after ent::lde_name maybe
440 * broken, but we do not care. */
441 ent->lde_name[ent->lde_namelen] = 0;
444 /* LFSCK wrap functions */
446 static void mdd_lfsck_fail(const struct lu_env *env, struct md_lfsck *lfsck,
447 bool oit, bool new_checked)
449 struct lfsck_component *com;
451 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
452 com->lc_ops->lfsck_fail(env, com, oit, new_checked);
456 static int mdd_lfsck_checkpoint(const struct lu_env *env,
457 struct md_lfsck *lfsck, bool oit)
459 struct lfsck_component *com;
462 if (likely(cfs_time_beforeq(cfs_time_current(),
463 lfsck->ml_time_next_checkpoint)))
466 mdd_lfsck_pos_fill(env, lfsck, &lfsck->ml_pos_current, oit, !oit);
467 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
468 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
473 lfsck->ml_time_last_checkpoint = cfs_time_current();
474 lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
475 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
479 static int mdd_lfsck_prep(struct lu_env *env, struct md_lfsck *lfsck)
481 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
482 struct mdd_object *obj = NULL;
483 struct dt_object *dt_obj;
484 struct lfsck_component *com;
485 struct lfsck_component *next;
486 struct lfsck_position *pos = NULL;
487 const struct dt_it_ops *iops =
488 &lfsck->ml_obj_oit->do_index_ops->dio_it;
493 LASSERT(lfsck->ml_obj_dir == NULL);
494 LASSERT(lfsck->ml_di_dir == NULL);
496 cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_scan, lc_link) {
497 com->lc_new_checked = 0;
498 if (lfsck->ml_bookmark_ram.lb_param & LPF_DRYRUN)
501 rc = com->lc_ops->lfsck_prep(env, com);
506 (!mdd_lfsck_pos_is_zero(&com->lc_pos_start) &&
507 mdd_lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
508 pos = &com->lc_pos_start;
511 /* Init otable-based iterator. */
513 rc = iops->load(env, lfsck->ml_di_oit, 0);
514 GOTO(out, rc = (rc >= 0 ? 0 : rc));
517 rc = iops->load(env, lfsck->ml_di_oit, pos->lp_oit_cookie);
521 if (fid_is_zero(&pos->lp_dir_parent))
524 /* Find the directory for namespace-based traverse. */
525 obj = mdd_object_find(env, mdd, &pos->lp_dir_parent);
528 else if (IS_ERR(obj))
529 RETURN(PTR_ERR(obj));
531 /* XXX: need more processing for remote object in the future. */
532 if (!mdd_object_exists(obj) || mdd_object_remote(obj) ||
533 unlikely(!S_ISDIR(mdd_object_type(obj))))
536 if (unlikely(mdd_is_dead_obj(obj)))
539 dt_obj = mdd_object_child(obj);
540 if (unlikely(!dt_try_as_dir(env, dt_obj)))
541 GOTO(out, rc = -ENOTDIR);
543 /* Init the namespace-based directory traverse. */
544 iops = &dt_obj->do_index_ops->dio_it;
545 di = iops->init(env, dt_obj, lfsck->ml_args_dir, BYPASS_CAPA);
547 GOTO(out, rc = PTR_ERR(di));
549 rc = iops->load(env, di, pos->lp_dir_cookie);
551 rc = iops->next(env, di);
561 lfsck->ml_obj_dir = dt_obj;
562 spin_lock(&lfsck->ml_lock);
563 lfsck->ml_di_dir = di;
564 spin_unlock(&lfsck->ml_lock);
571 mdd_object_put(env, obj);
574 return (rc > 0 ? 0 : rc);
576 mdd_lfsck_pos_fill(env, lfsck, &lfsck->ml_pos_current, false, false);
577 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
578 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
583 lfsck->ml_time_last_checkpoint = cfs_time_current();
584 lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
585 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
589 static int mdd_lfsck_exec_oit(const struct lu_env *env, struct md_lfsck *lfsck,
590 struct mdd_object *obj)
592 struct lfsck_component *com;
593 struct dt_object *dt_obj;
594 const struct dt_it_ops *iops;
599 LASSERT(lfsck->ml_obj_dir == NULL);
601 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
602 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
607 if (!S_ISDIR(mdd_object_type(obj)) ||
608 cfs_list_empty(&lfsck->ml_list_dir))
611 rc = object_is_client_visible(env, mdd_lfsck2mdd(lfsck), obj);
615 if (unlikely(mdd_is_dead_obj(obj)))
618 dt_obj = mdd_object_child(obj);
619 if (unlikely(!dt_try_as_dir(env, dt_obj)))
620 GOTO(out, rc = -ENOTDIR);
622 iops = &dt_obj->do_index_ops->dio_it;
623 di = iops->init(env, dt_obj, lfsck->ml_args_dir, BYPASS_CAPA);
625 GOTO(out, rc = PTR_ERR(di));
627 rc = iops->load(env, di, 0);
629 rc = iops->next(env, di);
640 lfsck->ml_obj_dir = dt_obj;
641 spin_lock(&lfsck->ml_lock);
642 lfsck->ml_di_dir = di;
643 spin_unlock(&lfsck->ml_lock);
649 mdd_lfsck_fail(env, lfsck, false, false);
650 return (rc > 0 ? 0 : rc);
653 static int mdd_lfsck_exec_dir(const struct lu_env *env, struct md_lfsck *lfsck,
654 struct mdd_object *obj, struct lu_dirent *ent)
656 struct lfsck_component *com;
659 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
660 rc = com->lc_ops->lfsck_exec_dir(env, com, obj, ent);
667 static int mdd_lfsck_post(const struct lu_env *env, struct md_lfsck *lfsck,
670 struct lfsck_component *com;
671 struct lfsck_component *next;
674 mdd_lfsck_pos_fill(env, lfsck, &lfsck->ml_pos_current, true, true);
675 cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_scan, lc_link) {
676 rc = com->lc_ops->lfsck_post(env, com, result);
681 lfsck->ml_time_last_checkpoint = cfs_time_current();
682 lfsck->ml_time_next_checkpoint = lfsck->ml_time_last_checkpoint +
683 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
687 static int mdd_lfsck_double_scan(const struct lu_env *env,
688 struct md_lfsck *lfsck)
690 struct lfsck_component *com;
691 struct lfsck_component *next;
694 cfs_list_for_each_entry_safe(com, next, &lfsck->ml_list_double_scan,
696 if (lfsck->ml_bookmark_ram.lb_param & LPF_DRYRUN)
699 rc = com->lc_ops->lfsck_double_scan(env, com);
708 static int mdd_lfsck_dir_engine(const struct lu_env *env,
709 struct md_lfsck *lfsck)
711 struct mdd_thread_info *info = mdd_env_info(env);
712 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
713 const struct dt_it_ops *iops =
714 &lfsck->ml_obj_dir->do_index_ops->dio_it;
715 struct dt_it *di = lfsck->ml_di_dir;
716 struct lu_dirent *ent = &info->mti_ent;
717 struct lu_fid *fid = &info->mti_fid;
718 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
719 struct ptlrpc_thread *thread = &lfsck->ml_thread;
724 struct mdd_object *child;
726 lfsck->ml_new_scanned++;
727 rc = iops->rec(env, di, (struct dt_rec *)ent,
730 mdd_lfsck_fail(env, lfsck, false, true);
731 if (bk->lb_param & LPF_FAILOUT)
737 mdd_lfsck_unpack_ent(ent);
738 if (ent->lde_attrs & LUDA_IGNORE)
742 child = mdd_object_find(env, mdd, fid);
745 } else if (IS_ERR(child)) {
746 mdd_lfsck_fail(env, lfsck, false, true);
747 if (bk->lb_param & LPF_FAILOUT)
748 RETURN(PTR_ERR(child));
753 /* XXX: need more processing for remote object in the future. */
754 if (mdd_object_exists(child) && !mdd_object_remote(child))
755 rc = mdd_lfsck_exec_dir(env, lfsck, child, ent);
756 mdd_object_put(env, child);
757 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
761 rc = mdd_lfsck_checkpoint(env, lfsck, false);
762 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
766 mdd_lfsck_control_speed(lfsck);
767 if (unlikely(!thread_is_running(thread)))
770 rc = iops->next(env, di);
773 if (rc > 0 && !lfsck->ml_oit_over)
774 mdd_lfsck_close_dir(env, lfsck);
779 static int mdd_lfsck_oit_engine(const struct lu_env *env,
780 struct md_lfsck *lfsck)
782 struct mdd_thread_info *info = mdd_env_info(env);
783 struct mdd_device *mdd = mdd_lfsck2mdd(lfsck);
784 const struct dt_it_ops *iops =
785 &lfsck->ml_obj_oit->do_index_ops->dio_it;
786 struct dt_it *di = lfsck->ml_di_oit;
787 struct lu_fid *fid = &info->mti_fid;
788 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
789 struct ptlrpc_thread *thread = &lfsck->ml_thread;
794 struct mdd_object *target;
796 if (lfsck->ml_di_dir != NULL) {
797 rc = mdd_lfsck_dir_engine(env, lfsck);
802 if (unlikely(lfsck->ml_oit_over))
805 lfsck->ml_new_scanned++;
806 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
808 mdd_lfsck_fail(env, lfsck, true, true);
809 if (bk->lb_param & LPF_FAILOUT)
815 target = mdd_object_find(env, mdd, fid);
816 if (target == NULL) {
818 } else if (IS_ERR(target)) {
819 mdd_lfsck_fail(env, lfsck, true, true);
820 if (bk->lb_param & LPF_FAILOUT)
821 RETURN(PTR_ERR(target));
826 /* XXX: In fact, low layer otable-based iteration should not
827 * return agent object. But before LU-2646 resolved, we
828 * need more processing for agent object. */
829 if (mdd_object_exists(target) && !mdd_object_remote(target))
830 rc = mdd_lfsck_exec_oit(env, lfsck, target);
831 mdd_object_put(env, target);
832 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
836 rc = mdd_lfsck_checkpoint(env, lfsck, true);
837 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
841 mdd_lfsck_control_speed(lfsck);
843 rc = iops->next(env, di);
845 lfsck->ml_oit_over = 1;
847 if (unlikely(!thread_is_running(thread)))
849 } while (rc == 0 || lfsck->ml_di_dir != NULL);
854 static int mdd_lfsck_main(void *args)
857 struct md_lfsck *lfsck = (struct md_lfsck *)args;
858 struct ptlrpc_thread *thread = &lfsck->ml_thread;
859 struct dt_object *oit_obj = lfsck->ml_obj_oit;
860 const struct dt_it_ops *oit_iops = &oit_obj->do_index_ops->dio_it;
861 struct dt_it *oit_di;
865 cfs_daemonize("lfsck");
866 rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
868 CERROR("%s: LFSCK, fail to init env, rc = %d\n",
869 mdd_lfsck2name(lfsck), rc);
873 oit_di = oit_iops->init(&env, oit_obj, lfsck->ml_args_oit, BYPASS_CAPA);
874 if (IS_ERR(oit_di)) {
875 rc = PTR_ERR(oit_di);
876 CERROR("%s: LFSCK, fail to init iteration, rc = %d\n",
877 mdd_lfsck2name(lfsck), rc);
881 spin_lock(&lfsck->ml_lock);
882 lfsck->ml_di_oit = oit_di;
883 spin_unlock(&lfsck->ml_lock);
884 rc = mdd_lfsck_prep(&env, lfsck);
888 CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = 0x%x, dir_flags = 0x%x, "
889 "oit_cookie = "LPU64", dir_cookie = "LPU64", parent = "DFID
890 ", pid = %d\n", lfsck->ml_args_oit, lfsck->ml_args_dir,
891 lfsck->ml_pos_current.lp_oit_cookie,
892 lfsck->ml_pos_current.lp_dir_cookie,
893 PFID(&lfsck->ml_pos_current.lp_dir_parent),
896 spin_lock(&lfsck->ml_lock);
897 thread_set_flags(thread, SVC_RUNNING);
898 spin_unlock(&lfsck->ml_lock);
899 cfs_waitq_broadcast(&thread->t_ctl_waitq);
901 if (!cfs_list_empty(&lfsck->ml_list_scan) ||
902 cfs_list_empty(&lfsck->ml_list_double_scan))
903 rc = mdd_lfsck_oit_engine(&env, lfsck);
907 CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = 0x%x, dir_flags = 0x%x, "
908 "oit_cookie = "LPU64", dir_cookie = "LPU64", parent = "DFID
909 ", pid = %d, rc = %d\n", lfsck->ml_args_oit, lfsck->ml_args_dir,
910 lfsck->ml_pos_current.lp_oit_cookie,
911 lfsck->ml_pos_current.lp_dir_cookie,
912 PFID(&lfsck->ml_pos_current.lp_dir_parent),
913 cfs_curproc_pid(), rc);
915 if (lfsck->ml_paused && cfs_list_empty(&lfsck->ml_list_scan))
916 oit_iops->put(&env, oit_di);
918 rc = mdd_lfsck_post(&env, lfsck, rc);
919 if (lfsck->ml_di_dir != NULL)
920 mdd_lfsck_close_dir(&env, lfsck);
923 spin_lock(&lfsck->ml_lock);
924 lfsck->ml_di_oit = NULL;
925 spin_unlock(&lfsck->ml_lock);
927 oit_iops->fini(&env, oit_di);
929 if (!cfs_list_empty(&lfsck->ml_list_double_scan))
930 rc = mdd_lfsck_double_scan(&env, lfsck);
935 /* XXX: Purge the pinned objects in the future. */
941 spin_lock(&lfsck->ml_lock);
942 thread_set_flags(thread, SVC_STOPPED);
943 cfs_waitq_broadcast(&thread->t_ctl_waitq);
944 spin_unlock(&lfsck->ml_lock);
948 /* external interfaces */
950 int mdd_lfsck_set_speed(const struct lu_env *env, struct md_lfsck *lfsck,
955 mutex_lock(&lfsck->ml_mutex);
956 __mdd_lfsck_set_speed(lfsck, limit);
957 rc = mdd_lfsck_bookmark_store(env, lfsck);
958 mutex_unlock(&lfsck->ml_mutex);
962 int mdd_lfsck_start(const struct lu_env *env, struct md_lfsck *lfsck,
963 struct lfsck_start *start)
965 struct lfsck_bookmark *bk = &lfsck->ml_bookmark_ram;
966 struct ptlrpc_thread *thread = &lfsck->ml_thread;
967 struct lfsck_component *com;
968 struct l_wait_info lwi = { 0 };
975 if (lfsck->ml_obj_oit == NULL)
978 /* start == NULL means auto trigger paused LFSCK. */
979 if (start == NULL && cfs_list_empty(&lfsck->ml_list_scan))
982 mutex_lock(&lfsck->ml_mutex);
983 spin_lock(&lfsck->ml_lock);
984 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
985 spin_unlock(&lfsck->ml_lock);
986 mutex_unlock(&lfsck->ml_mutex);
990 spin_unlock(&lfsck->ml_lock);
992 lfsck->ml_paused = 0;
993 lfsck->ml_oit_over = 0;
994 lfsck->ml_drop_dryrun = 0;
995 lfsck->ml_new_scanned = 0;
997 /* For auto trigger. */
1001 start->ls_version = bk->lb_version;
1002 if (start->ls_valid & LSV_SPEED_LIMIT) {
1003 __mdd_lfsck_set_speed(lfsck, start->ls_speed_limit);
1007 if (start->ls_valid & LSV_ERROR_HANDLE) {
1008 valid |= DOIV_ERROR_HANDLE;
1009 if (start->ls_flags & LPF_FAILOUT)
1010 flags |= DOIF_FAILOUT;
1012 if ((start->ls_flags & LPF_FAILOUT) &&
1013 !(bk->lb_param & LPF_FAILOUT)) {
1014 bk->lb_param |= LPF_FAILOUT;
1016 } else if (!(start->ls_flags & LPF_FAILOUT) &&
1017 (bk->lb_param & LPF_FAILOUT)) {
1018 bk->lb_param &= ~LPF_FAILOUT;
1023 if (start->ls_valid & LSV_DRYRUN) {
1024 if ((start->ls_flags & LPF_DRYRUN) &&
1025 !(bk->lb_param & LPF_DRYRUN)) {
1026 bk->lb_param |= LPF_DRYRUN;
1028 } else if (!(start->ls_flags & LPF_DRYRUN) &&
1029 (bk->lb_param & LPF_DRYRUN)) {
1030 bk->lb_param &= ~LPF_DRYRUN;
1031 lfsck->ml_drop_dryrun = 1;
1037 rc = mdd_lfsck_bookmark_store(env, lfsck);
1042 if (start->ls_flags & LPF_RESET)
1043 flags |= DOIF_RESET;
1045 if (start->ls_active != 0) {
1046 struct lfsck_component *next;
1049 if (start->ls_active == LFSCK_TYPES_ALL)
1050 start->ls_active = LFSCK_TYPES_SUPPORTED;
1052 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
1053 start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
1054 GOTO(out, rc = -ENOTSUPP);
1057 cfs_list_for_each_entry_safe(com, next,
1058 &lfsck->ml_list_scan, lc_link) {
1059 if (!(com->lc_type & start->ls_active)) {
1060 rc = com->lc_ops->lfsck_post(env, com, 0);
1066 while (start->ls_active != 0) {
1067 if (type & start->ls_active) {
1068 com = __mdd_lfsck_component_find(lfsck, type,
1069 &lfsck->ml_list_idle);
1071 /* The component status will be updated
1072 * when its prep() is called later by
1073 * the LFSCK main engine. */
1074 cfs_list_del_init(&com->lc_link);
1075 cfs_list_add_tail(&com->lc_link,
1076 &lfsck->ml_list_scan);
1078 start->ls_active &= ~type;
1084 cfs_list_for_each_entry(com, &lfsck->ml_list_scan, lc_link) {
1085 start->ls_active |= com->lc_type;
1086 if (flags & DOIF_RESET) {
1087 rc = com->lc_ops->lfsck_reset(env, com, false);
1094 lfsck->ml_args_dir = LUDA_64BITHASH | LUDA_VERIFY;
1095 if (bk->lb_param & LPF_DRYRUN)
1096 lfsck->ml_args_dir |= LUDA_VERIFY_DRYRUN;
1098 if (bk->lb_param & LPF_FAILOUT) {
1099 valid |= DOIV_ERROR_HANDLE;
1100 flags |= DOIF_FAILOUT;
1103 if (!cfs_list_empty(&lfsck->ml_list_scan))
1104 flags |= DOIF_OUTUSED;
1106 lfsck->ml_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
1107 thread_set_flags(thread, 0);
1108 rc = cfs_create_thread(mdd_lfsck_main, lfsck, 0);
1110 CERROR("%s: cannot start LFSCK thread, rc = %d\n",
1111 mdd_lfsck2name(lfsck), rc);
1113 l_wait_event(thread->t_ctl_waitq,
1114 thread_is_running(thread) ||
1115 thread_is_stopped(thread),
1121 mutex_unlock(&lfsck->ml_mutex);
1122 return (rc < 0 ? rc : 0);
1125 int mdd_lfsck_stop(const struct lu_env *env, struct md_lfsck *lfsck,
1128 struct ptlrpc_thread *thread = &lfsck->ml_thread;
1129 struct l_wait_info lwi = { 0 };
1132 mutex_lock(&lfsck->ml_mutex);
1133 spin_lock(&lfsck->ml_lock);
1134 if (thread_is_init(thread) || thread_is_stopped(thread)) {
1135 spin_unlock(&lfsck->ml_lock);
1136 mutex_unlock(&lfsck->ml_mutex);
1141 lfsck->ml_paused = 1;
1142 thread_set_flags(thread, SVC_STOPPING);
1143 /* The LFSCK thread may be sleeping on low layer wait queue,
1145 if (likely(lfsck->ml_di_oit != NULL))
1146 lfsck->ml_obj_oit->do_index_ops->dio_it.put(env,
1148 spin_unlock(&lfsck->ml_lock);
1150 cfs_waitq_broadcast(&thread->t_ctl_waitq);
1151 l_wait_event(thread->t_ctl_waitq,
1152 thread_is_stopped(thread),
1154 mutex_unlock(&lfsck->ml_mutex);
1159 static const struct lu_fid lfsck_it_fid = { .f_seq = FID_SEQ_LOCAL_FILE,
1160 .f_oid = OTABLE_IT_OID,
1163 int mdd_lfsck_setup(const struct lu_env *env, struct mdd_device *mdd)
1165 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
1166 struct dt_object *obj;
1170 LASSERT(!lfsck->ml_initialized);
1172 lfsck->ml_initialized = 1;
1173 mutex_init(&lfsck->ml_mutex);
1174 spin_lock_init(&lfsck->ml_lock);
1175 CFS_INIT_LIST_HEAD(&lfsck->ml_list_scan);
1176 CFS_INIT_LIST_HEAD(&lfsck->ml_list_dir);
1177 CFS_INIT_LIST_HEAD(&lfsck->ml_list_double_scan);
1178 CFS_INIT_LIST_HEAD(&lfsck->ml_list_idle);
1179 cfs_waitq_init(&lfsck->ml_thread.t_ctl_waitq);
1181 obj = dt_locate(env, mdd->mdd_bottom, &lfsck_it_fid);
1183 RETURN(PTR_ERR(obj));
1185 lfsck->ml_obj_oit = obj;
1186 rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
1188 if (rc == -ENOTSUPP)
1194 obj = dt_store_open(env, mdd->mdd_bottom, "", lfsck_bookmark_name,
1195 &mdd_env_info(env)->mti_fid);
1197 RETURN(PTR_ERR(obj));
1199 lfsck->ml_bookmark_obj = obj;
1200 rc = mdd_lfsck_bookmark_load(env, lfsck);
1202 rc = mdd_lfsck_bookmark_init(env, lfsck);
1204 /* XXX: LFSCK components initialization to be added here. */
1209 void mdd_lfsck_cleanup(const struct lu_env *env, struct mdd_device *mdd)
1211 struct md_lfsck *lfsck = &mdd->mdd_lfsck;
1212 struct ptlrpc_thread *thread = &lfsck->ml_thread;
1213 struct lfsck_component *com;
1215 if (!lfsck->ml_initialized)
1218 LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
1220 if (lfsck->ml_obj_oit != NULL) {
1221 lu_object_put(env, &lfsck->ml_obj_oit->do_lu);
1222 lfsck->ml_obj_oit = NULL;
1225 LASSERT(lfsck->ml_obj_dir == NULL);
1227 if (lfsck->ml_bookmark_obj != NULL) {
1228 lu_object_put(env, &lfsck->ml_bookmark_obj->do_lu);
1229 lfsck->ml_bookmark_obj = NULL;
1232 while (!cfs_list_empty(&lfsck->ml_list_scan)) {
1233 com = cfs_list_entry(lfsck->ml_list_scan.next,
1234 struct lfsck_component,
1236 mdd_lfsck_component_cleanup(env, com);
1239 LASSERT(cfs_list_empty(&lfsck->ml_list_dir));
1241 while (!cfs_list_empty(&lfsck->ml_list_double_scan)) {
1242 com = cfs_list_entry(lfsck->ml_list_double_scan.next,
1243 struct lfsck_component,
1245 mdd_lfsck_component_cleanup(env, com);
1248 while (!cfs_list_empty(&lfsck->ml_list_idle)) {
1249 com = cfs_list_entry(lfsck->ml_list_idle.next,
1250 struct lfsck_component,
1252 mdd_lfsck_component_cleanup(env, com);