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) 2013, 2017, Intel Corporation.
26 * lustre/lfsck/lfsck_lib.c
28 * Author: Fan, Yong <fan.yong@intel.com>
31 #define DEBUG_SUBSYSTEM S_LFSCK
33 #include <linux/kthread.h>
34 #include <linux/sched.h>
35 #include <linux/list.h>
36 #include <lu_object.h>
37 #include <dt_object.h>
38 #include <md_object.h>
39 #include <lustre_fld.h>
40 #include <lustre_lib.h>
41 #include <lustre_net.h>
42 #include <lustre_lfsck.h>
44 #include "lfsck_internal.h"
46 #define LFSCK_CHECKPOINT_SKIP 1
48 /* define lfsck thread key */
49 LU_KEY_INIT(lfsck, struct lfsck_thread_info);
51 static void lfsck_key_fini(const struct lu_context *ctx,
52 struct lu_context_key *key, void *data)
54 struct lfsck_thread_info *info = data;
56 lu_buf_free(&info->lti_linkea_buf);
57 lu_buf_free(&info->lti_linkea_buf2);
58 lu_buf_free(&info->lti_big_buf);
62 LU_CONTEXT_KEY_DEFINE(lfsck, LCT_MD_THREAD | LCT_DT_THREAD);
63 LU_KEY_INIT_GENERIC(lfsck);
65 static struct list_head lfsck_instance_list;
66 static struct list_head lfsck_ost_orphan_list;
67 static struct list_head lfsck_mdt_orphan_list;
68 static DEFINE_SPINLOCK(lfsck_instance_lock);
70 const char *lfsck_flags_names[] = {
79 const char *lfsck_param_names[] = {
89 "delay_create_ostobj",
93 enum lfsck_verify_lpf_types {
95 LVLT_BY_NAMEENTRY = 1,
99 lfsck_reset_ltd_status(struct lfsck_tgt_desc *ltd, enum lfsck_type type)
101 if (type == LFSCK_TYPE_LAYOUT) {
102 ltd->ltd_layout_status = LS_MAX;
103 ltd->ltd_layout_repaired = 0;
105 ltd->ltd_namespace_status = LS_MAX;
106 ltd->ltd_namespace_repaired = 0;
110 static int lfsck_tgt_descs_init(struct lfsck_tgt_descs *ltds)
112 spin_lock_init(<ds->ltd_lock);
113 init_rwsem(<ds->ltd_rw_sem);
114 INIT_LIST_HEAD(<ds->ltd_orphan);
115 ltds->ltd_tgts_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
116 if (ltds->ltd_tgts_bitmap == NULL)
122 static void lfsck_tgt_descs_fini(struct lfsck_tgt_descs *ltds)
124 struct lfsck_tgt_desc *ltd;
125 struct lfsck_tgt_desc *next;
128 down_write(<ds->ltd_rw_sem);
130 list_for_each_entry_safe(ltd, next, <ds->ltd_orphan,
132 list_del_init(<d->ltd_orphan_list);
136 if (unlikely(ltds->ltd_tgts_bitmap == NULL)) {
137 up_write(<ds->ltd_rw_sem);
142 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
143 ltd = lfsck_ltd2tgt(ltds, idx);
144 if (likely(ltd != NULL)) {
145 LASSERT(list_empty(<d->ltd_layout_list));
146 LASSERT(list_empty(<d->ltd_layout_phase_list));
147 LASSERT(list_empty(<d->ltd_namespace_list));
148 LASSERT(list_empty(<d->ltd_namespace_phase_list));
151 cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx);
152 lfsck_assign_tgt(ltds, NULL, idx);
157 LASSERTF(ltds->ltd_tgtnr == 0, "tgt count unmatched: %d\n",
160 for (idx = 0; idx < TGT_PTRS; idx++) {
161 if (ltds->ltd_tgts_idx[idx] != NULL) {
162 OBD_FREE_PTR(ltds->ltd_tgts_idx[idx]);
163 ltds->ltd_tgts_idx[idx] = NULL;
167 CFS_FREE_BITMAP(ltds->ltd_tgts_bitmap);
168 ltds->ltd_tgts_bitmap = NULL;
169 up_write(<ds->ltd_rw_sem);
172 static int __lfsck_add_target(const struct lu_env *env,
173 struct lfsck_instance *lfsck,
174 struct lfsck_tgt_desc *ltd,
175 bool for_ost, bool locked)
177 struct lfsck_tgt_descs *ltds;
178 __u32 index = ltd->ltd_index;
183 ltds = &lfsck->li_ost_descs;
185 ltds = &lfsck->li_mdt_descs;
188 down_write(<ds->ltd_rw_sem);
190 LASSERT(ltds->ltd_tgts_bitmap != NULL);
192 if (index >= ltds->ltd_tgts_bitmap->size) {
193 __u32 newsize = max((__u32)ltds->ltd_tgts_bitmap->size,
194 (__u32)BITS_PER_LONG);
195 struct cfs_bitmap *old_bitmap = ltds->ltd_tgts_bitmap;
196 struct cfs_bitmap *new_bitmap;
198 while (newsize < index + 1)
201 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
202 if (new_bitmap == NULL)
203 GOTO(unlock, rc = -ENOMEM);
205 if (ltds->ltd_tgtnr > 0)
206 cfs_bitmap_copy(new_bitmap, old_bitmap);
207 ltds->ltd_tgts_bitmap = new_bitmap;
208 CFS_FREE_BITMAP(old_bitmap);
211 if (cfs_bitmap_check(ltds->ltd_tgts_bitmap, index)) {
212 CERROR("%s: the device %s (%u) is registered already\n",
213 lfsck_lfsck2name(lfsck),
214 ltd->ltd_tgt->dd_lu_dev.ld_obd->obd_name, index);
215 GOTO(unlock, rc = -EEXIST);
218 if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
219 OBD_ALLOC_PTR(ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK]);
220 if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL)
221 GOTO(unlock, rc = -ENOMEM);
224 lfsck_assign_tgt(ltds, ltd, index);
225 cfs_bitmap_set(ltds->ltd_tgts_bitmap, index);
228 GOTO(unlock, rc = 0);
232 up_write(<ds->ltd_rw_sem);
237 static int lfsck_add_target_from_orphan(const struct lu_env *env,
238 struct lfsck_instance *lfsck)
240 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
241 struct lfsck_tgt_desc *ltd;
242 struct lfsck_tgt_desc *next;
243 struct list_head *head = &lfsck_ost_orphan_list;
248 spin_lock(&lfsck_instance_lock);
249 list_for_each_entry_safe(ltd, next, head, ltd_orphan_list) {
250 if (ltd->ltd_key == lfsck->li_bottom)
251 list_move_tail(<d->ltd_orphan_list,
254 spin_unlock(&lfsck_instance_lock);
256 down_write(<ds->ltd_rw_sem);
257 while (!list_empty(<ds->ltd_orphan)) {
258 ltd = list_entry(ltds->ltd_orphan.next,
259 struct lfsck_tgt_desc,
261 list_del_init(<d->ltd_orphan_list);
262 rc = __lfsck_add_target(env, lfsck, ltd, for_ost, true);
263 /* Do not hold the semaphore for too long time. */
264 up_write(<ds->ltd_rw_sem);
268 down_write(<ds->ltd_rw_sem);
270 up_write(<ds->ltd_rw_sem);
273 ltds = &lfsck->li_mdt_descs;
274 head = &lfsck_mdt_orphan_list;
282 static inline struct lfsck_component *
283 __lfsck_component_find(struct lfsck_instance *lfsck, __u16 type,
284 struct list_head *list)
286 struct lfsck_component *com;
288 list_for_each_entry(com, list, lc_link) {
289 if (com->lc_type == type)
295 struct lfsck_component *
296 lfsck_component_find(struct lfsck_instance *lfsck, __u16 type)
298 struct lfsck_component *com;
300 spin_lock(&lfsck->li_lock);
301 com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
305 com = __lfsck_component_find(lfsck, type,
306 &lfsck->li_list_double_scan);
310 com = __lfsck_component_find(lfsck, type, &lfsck->li_list_idle);
314 lfsck_component_get(com);
315 spin_unlock(&lfsck->li_lock);
319 void lfsck_component_cleanup(const struct lu_env *env,
320 struct lfsck_component *com)
322 if (!list_empty(&com->lc_link))
323 list_del_init(&com->lc_link);
324 if (!list_empty(&com->lc_link_dir))
325 list_del_init(&com->lc_link_dir);
327 lfsck_component_put(env, com);
330 int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
331 struct lu_fid *fid, bool locked)
333 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
338 mutex_lock(&lfsck->li_mutex);
340 rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
342 bk->lb_last_fid = *fid;
343 /* We do not care about whether the subsequent sub-operations
344 * failed or not. The worst case is that one FID is lost that
345 * is not a big issue for the LFSCK since it is relative rare
346 * for LFSCK create. */
347 rc = lfsck_bookmark_store(env, lfsck);
351 mutex_unlock(&lfsck->li_mutex);
356 static int __lfsck_ibits_lock(const struct lu_env *env,
357 struct lfsck_instance *lfsck,
358 struct dt_object *obj, struct ldlm_res_id *resid,
359 struct lustre_handle *lh, __u64 bits,
362 struct lfsck_thread_info *info = lfsck_env_info(env);
363 union ldlm_policy_data *policy = &info->lti_policy;
364 __u64 flags = LDLM_FL_ATOMIC_CB;
367 LASSERT(lfsck->li_namespace != NULL);
369 memset(policy, 0, sizeof(*policy));
370 policy->l_inodebits.bits = bits;
371 if (dt_object_remote(obj)) {
372 struct ldlm_enqueue_info *einfo = &info->lti_einfo;
374 memset(einfo, 0, sizeof(*einfo));
375 einfo->ei_type = LDLM_IBITS;
376 einfo->ei_mode = mode;
377 einfo->ei_cb_bl = ldlm_blocking_ast;
378 einfo->ei_cb_cp = ldlm_completion_ast;
379 einfo->ei_res_id = resid;
381 rc = dt_object_lock(env, obj, lh, einfo, policy);
383 rc = ldlm_cli_enqueue_local(env, lfsck->li_namespace, resid,
384 LDLM_IBITS, policy, mode,
385 &flags, ldlm_blocking_ast,
386 ldlm_completion_ast, NULL, NULL,
387 0, LVB_T_NONE, NULL, lh);
390 if (rc == ELDLM_OK) {
393 memset(lh, 0, sizeof(*lh));
401 * Request the specified ibits lock for the given object.
403 * Before the LFSCK modifying on the namespace visible object,
404 * it needs to acquire related ibits ldlm lock.
406 * \param[in] env pointer to the thread context
407 * \param[in] lfsck pointer to the lfsck instance
408 * \param[in] obj pointer to the dt_object to be locked
409 * \param[out] lh pointer to the lock handle
410 * \param[in] bits the bits for the ldlm lock to be acquired
411 * \param[in] mode the mode for the ldlm lock to be acquired
413 * \retval 0 for success
414 * \retval negative error number on failure
416 int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
417 struct dt_object *obj, struct lustre_handle *lh,
418 __u64 bits, enum ldlm_mode mode)
420 struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
422 LASSERT(!lustre_handle_is_used(lh));
424 fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
425 return __lfsck_ibits_lock(env, lfsck, obj, resid, lh, bits, mode);
429 * Release the the specified ibits lock.
431 * If the lock has been acquired before, release it
432 * and cleanup the handle. Otherwise, do nothing.
434 * \param[in] lh pointer to the lock handle
435 * \param[in] mode the mode for the ldlm lock to be released
437 void lfsck_ibits_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
439 if (lustre_handle_is_used(lh)) {
440 ldlm_lock_decref(lh, mode);
441 memset(lh, 0, sizeof(*lh));
446 * Request compound ibits locks for the given <obj, name> pairs.
448 * Before the LFSCK modifying on the namespace visible object, it needs to
449 * acquire related ibits ldlm lock. Usually, we can use lfsck_ibits_lock for
450 * the lock purpose. But the simple lfsck_ibits_lock for directory-based
451 * modificationis (such as insert name entry to the directory) may be too
452 * coarse-grained and not efficient.
454 * The lfsck_lock() will request compound ibits locks on the specified
455 * <obj, name> pairs: the PDO (Parallel Directory Operations) ibits (UPDATE)
456 * lock on the directory object, and the regular ibits lock on the name hash.
458 * \param[in] env pointer to the thread context
459 * \param[in] lfsck pointer to the lfsck instance
460 * \param[in] obj pointer to the dt_object to be locked
461 * \param[in] name used for building the PDO lock resource
462 * \param[out] llh pointer to the lfsck_lock_handle
463 * \param[in] bits the bits for the ldlm lock to be acquired
464 * \param[in] mode the mode for the ldlm lock to be acquired
466 * \retval 0 for success
467 * \retval negative error number on failure
469 int lfsck_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
470 struct dt_object *obj, const char *name,
471 struct lfsck_lock_handle *llh, __u64 bits, enum ldlm_mode mode)
473 struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
476 LASSERT(S_ISDIR(lfsck_object_type(obj)));
477 LASSERT(name != NULL);
478 LASSERT(name[0] != 0);
479 LASSERT(!lustre_handle_is_used(&llh->llh_pdo_lh));
480 LASSERT(!lustre_handle_is_used(&llh->llh_reg_lh));
484 llh->llh_pdo_mode = LCK_EX;
487 llh->llh_pdo_mode = LCK_CW;
490 llh->llh_pdo_mode = LCK_CR;
493 CDEBUG(D_LFSCK, "%s: unexpected PDO lock mode %u on the obj "
494 DFID"\n", lfsck_lfsck2name(lfsck), mode,
495 PFID(lfsck_dto2fid(obj)));
499 fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
500 rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_pdo_lh,
501 MDS_INODELOCK_UPDATE, llh->llh_pdo_mode);
505 llh->llh_reg_mode = mode;
506 resid->name[LUSTRE_RES_ID_HSH_OFF] = ll_full_name_hash(NULL, name,
508 LASSERT(resid->name[LUSTRE_RES_ID_HSH_OFF] != 0);
509 rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_reg_lh,
510 bits, llh->llh_reg_mode);
512 lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
518 * Release the the compound ibits locks.
520 * \param[in] llh pointer to the lfsck_lock_handle to be released
522 void lfsck_unlock(struct lfsck_lock_handle *llh)
524 lfsck_ibits_unlock(&llh->llh_reg_lh, llh->llh_reg_mode);
525 lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
528 int lfsck_find_mdt_idx_by_fid(const struct lu_env *env,
529 struct lfsck_instance *lfsck,
530 const struct lu_fid *fid)
532 struct seq_server_site *ss = lfsck_dev_site(lfsck);
533 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
536 if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
537 /* "ROOT" is always on the MDT0. */
538 if (lu_fid_eq(fid, &lfsck->li_global_root_fid))
541 return lfsck_dev_idx(lfsck);
544 fld_range_set_mdt(range);
545 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
547 rc = range->lsr_index;
552 const char dot[] = ".";
553 const char dotdot[] = "..";
554 static const char dotlustre[] = ".lustre";
555 static const char lostfound[] = "lost+found";
558 * Remove the name entry from the .lustre/lost+found directory.
560 * No need to care about the object referenced by the name entry,
561 * either the name entry is invalid or redundant, or the referenced
562 * object has been processed or will be handled by others.
564 * \param[in] env pointer to the thread context
565 * \param[in] lfsck pointer to the lfsck instance
566 * \param[in] name the name for the name entry to be removed
568 * \retval 0 for success
569 * \retval negative error number on failure
571 static int lfsck_lpf_remove_name_entry(const struct lu_env *env,
572 struct lfsck_instance *lfsck,
575 struct dt_object *parent = lfsck->li_lpf_root_obj;
576 struct dt_device *dev = lfsck_obj2dev(parent);
578 struct lfsck_lock_handle *llh = &lfsck_env_info(env)->lti_llh;
582 rc = lfsck_lock(env, lfsck, parent, name, llh,
583 MDS_INODELOCK_UPDATE, LCK_PW);
587 th = dt_trans_create(env, dev);
589 GOTO(unlock, rc = PTR_ERR(th));
591 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
595 rc = dt_declare_ref_del(env, parent, th);
599 rc = dt_trans_start_local(env, dev, th);
603 rc = dt_delete(env, parent, (const struct dt_key *)name, th);
607 dt_write_lock(env, parent, 0);
608 rc = dt_ref_del(env, parent, th);
609 dt_write_unlock(env, parent);
614 dt_trans_stop(env, dev, th);
619 CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s: rc = %d\n",
620 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(parent)), name, rc);
625 static int lfsck_create_lpf_local(const struct lu_env *env,
626 struct lfsck_instance *lfsck,
627 struct dt_object *child,
629 struct dt_object_format *dof,
632 struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec;
633 struct dt_object *parent = lfsck->li_lpf_root_obj;
634 struct dt_device *dev = lfsck_obj2dev(child);
635 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
636 struct dt_object *bk_obj = lfsck->li_bookmark_obj;
637 const struct lu_fid *cfid = lfsck_dto2fid(child);
638 struct thandle *th = NULL;
639 struct linkea_data ldata = { NULL };
640 struct lu_buf linkea_buf;
641 const struct lu_name *cname;
643 int len = sizeof(struct lfsck_bookmark);
647 cname = lfsck_name_get_const(env, name, strlen(name));
648 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
649 cname, lfsck_dto2fid(parent));
653 th = dt_trans_create(env, dev);
657 /* 1a. create child */
658 rc = dt_declare_create(env, child, la, NULL, dof, th);
662 if (!dt_try_as_dir(env, child))
663 GOTO(stop, rc = -ENOTDIR);
665 /* 2a. increase child nlink */
666 rc = dt_declare_ref_add(env, child, th);
670 /* 3a. insert dot into child dir */
671 rec->rec_type = S_IFDIR;
673 rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
674 (const struct dt_key *)dot, th);
678 /* 4a. insert dotdot into child dir */
679 rec->rec_fid = &LU_LPF_FID;
680 rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
681 (const struct dt_key *)dotdot, th);
685 /* 5a. insert linkEA for child */
686 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
687 ldata.ld_leh->leh_len);
688 rc = dt_declare_xattr_set(env, child, &linkea_buf,
689 XATTR_NAME_LINK, 0, th);
693 /* 6a. insert name into parent dir */
694 rec->rec_type = S_IFDIR;
696 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
697 (const struct dt_key *)name, th);
701 /* 7a. increase parent nlink */
702 rc = dt_declare_ref_add(env, parent, th);
706 /* 8a. update bookmark */
707 rc = dt_declare_record_write(env, bk_obj,
708 lfsck_buf_get(env, bk, len), 0, th);
712 rc = dt_trans_start_local(env, dev, th);
716 dt_write_lock(env, child, 0);
717 /* 1b. create child */
718 rc = dt_create(env, child, la, NULL, dof, th);
722 /* 2b. increase child nlink */
723 rc = dt_ref_add(env, child, th);
727 /* 3b. insert dot into child dir */
729 rc = dt_insert(env, child, (const struct dt_rec *)rec,
730 (const struct dt_key *)dot, th);
734 /* 4b. insert dotdot into child dir */
735 rec->rec_fid = &LU_LPF_FID;
736 rc = dt_insert(env, child, (const struct dt_rec *)rec,
737 (const struct dt_key *)dotdot, th);
741 /* 5b. insert linkEA for child. */
742 rc = dt_xattr_set(env, child, &linkea_buf,
743 XATTR_NAME_LINK, 0, th);
744 dt_write_unlock(env, child);
748 /* 6b. insert name into parent dir */
750 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
751 (const struct dt_key *)name, th);
755 dt_write_lock(env, parent, 0);
756 /* 7b. increase parent nlink */
757 rc = dt_ref_add(env, parent, th);
758 dt_write_unlock(env, parent);
762 bk->lb_lpf_fid = *cfid;
763 lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
765 /* 8b. update bookmark */
766 rc = dt_record_write(env, bk_obj,
767 lfsck_buf_get(env, bk, len), &pos, th);
772 dt_write_unlock(env, child);
775 dt_trans_stop(env, dev, th);
780 static int lfsck_create_lpf_remote(const struct lu_env *env,
781 struct lfsck_instance *lfsck,
782 struct dt_object *child,
784 struct dt_object_format *dof,
787 struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec;
788 struct dt_object *parent = lfsck->li_lpf_root_obj;
789 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
790 struct dt_object *bk_obj = lfsck->li_bookmark_obj;
791 const struct lu_fid *cfid = lfsck_dto2fid(child);
792 struct thandle *th = NULL;
793 struct linkea_data ldata = { NULL };
794 struct lu_buf linkea_buf;
795 const struct lu_name *cname;
796 struct dt_device *dev;
798 int len = sizeof(struct lfsck_bookmark);
802 cname = lfsck_name_get_const(env, name, strlen(name));
803 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
804 cname, lfsck_dto2fid(parent));
808 /* Create .lustre/lost+found/MDTxxxx. */
810 /* XXX: Currently, cross-MDT create operation needs to create the child
811 * object firstly, then insert name into the parent directory. For
812 * this case, the child object resides on current MDT (local), but
813 * the parent ".lustre/lost+found" may be on remote MDT. It is not
814 * easy to contain all the sub-modifications orderly within single
817 * To avoid more inconsistency, we split the create operation into
820 * 1) create the child and update the lfsck_bookmark::lb_lpf_fid
822 * 2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
825 * If 1) done, but 2) failed, then go ahead, the LFSCK will try to
826 * repair such inconsistency when LFSCK run next time. */
828 /* Transaction I: locally */
830 dev = lfsck_obj2dev(child);
831 th = dt_trans_create(env, dev);
835 /* 1a. create child */
836 rc = dt_declare_create(env, child, la, NULL, dof, th);
840 if (!dt_try_as_dir(env, child))
841 GOTO(stop, rc = -ENOTDIR);
843 /* 2a. increase child nlink */
844 rc = dt_declare_ref_add(env, child, th);
848 /* 3a. insert dot into child dir */
849 rec->rec_type = S_IFDIR;
851 rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
852 (const struct dt_key *)dot, th);
856 /* 4a. insert dotdot into child dir */
857 rec->rec_fid = &LU_LPF_FID;
858 rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
859 (const struct dt_key *)dotdot, th);
863 /* 5a. insert linkEA for child */
864 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
865 ldata.ld_leh->leh_len);
866 rc = dt_declare_xattr_set(env, child, &linkea_buf,
867 XATTR_NAME_LINK, 0, th);
871 /* 6a. update bookmark */
872 rc = dt_declare_record_write(env, bk_obj,
873 lfsck_buf_get(env, bk, len), 0, th);
877 rc = dt_trans_start_local(env, dev, th);
881 dt_write_lock(env, child, 0);
882 /* 1b. create child */
883 rc = dt_create(env, child, la, NULL, dof, th);
887 /* 2b. increase child nlink */
888 rc = dt_ref_add(env, child, th);
892 /* 3b. insert dot into child dir */
893 rec->rec_type = S_IFDIR;
895 rc = dt_insert(env, child, (const struct dt_rec *)rec,
896 (const struct dt_key *)dot, th);
900 /* 4b. insert dotdot into child dir */
901 rec->rec_fid = &LU_LPF_FID;
902 rc = dt_insert(env, child, (const struct dt_rec *)rec,
903 (const struct dt_key *)dotdot, th);
907 /* 5b. insert linkEA for child */
908 rc = dt_xattr_set(env, child, &linkea_buf,
909 XATTR_NAME_LINK, 0, th);
913 bk->lb_lpf_fid = *cfid;
914 lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
916 /* 6b. update bookmark */
917 rc = dt_record_write(env, bk_obj,
918 lfsck_buf_get(env, bk, len), &pos, th);
920 dt_write_unlock(env, child);
921 dt_trans_stop(env, dev, th);
925 /* Transaction II: remotely */
927 dev = lfsck_obj2dev(parent);
928 th = dt_trans_create(env, dev);
933 /* 5a. insert name into parent dir */
935 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
936 (const struct dt_key *)name, th);
940 /* 6a. increase parent nlink */
941 rc = dt_declare_ref_add(env, parent, th);
945 rc = dt_trans_start_local(env, dev, th);
949 /* 5b. insert name into parent dir */
950 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
951 (const struct dt_key *)name, th);
955 dt_write_lock(env, parent, 0);
956 /* 6b. increase parent nlink */
957 rc = dt_ref_add(env, parent, th);
958 dt_write_unlock(env, parent);
963 dt_write_unlock(env, child);
965 dt_trans_stop(env, dev, th);
967 if (rc != 0 && dev == lfsck_obj2dev(parent))
968 CDEBUG(D_LFSCK, "%s: partially created the object "DFID
969 "for orphans, but failed to insert the name %s "
970 "to the .lustre/lost+found/. Such inconsistency "
971 "will be repaired when LFSCK run next time: rc = %d\n",
972 lfsck_lfsck2name(lfsck), PFID(cfid), name, rc);
978 * Create the MDTxxxx directory under /ROOT/.lustre/lost+found/
980 * The /ROOT/.lustre/lost+found/MDTxxxx/ directory is used for holding
981 * orphans and other uncertain inconsistent objects found during the
982 * LFSCK. Such directory will be created by the LFSCK engine on the
983 * local MDT before the LFSCK scanning.
985 * \param[in] env pointer to the thread context
986 * \param[in] lfsck pointer to the lfsck instance
988 * \retval 0 for success
989 * \retval negative error number on failure
991 static int lfsck_create_lpf(const struct lu_env *env,
992 struct lfsck_instance *lfsck)
994 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
995 struct lfsck_thread_info *info = lfsck_env_info(env);
996 struct lu_fid *cfid = &info->lti_fid2;
997 struct lu_attr *la = &info->lti_la;
998 struct dt_object_format *dof = &info->lti_dof;
999 struct dt_object *parent = lfsck->li_lpf_root_obj;
1000 struct dt_object *child = NULL;
1001 struct lfsck_lock_handle *llh = &info->lti_llh;
1003 int node = lfsck_dev_idx(lfsck);
1007 LASSERT(lfsck->li_master);
1008 LASSERT(parent != NULL);
1009 LASSERT(lfsck->li_lpf_obj == NULL);
1011 snprintf(name, 8, "MDT%04x", node);
1012 rc = lfsck_lock(env, lfsck, parent, name, llh,
1013 MDS_INODELOCK_UPDATE, LCK_PW);
1017 if (fid_is_zero(&bk->lb_lpf_fid)) {
1018 /* There is corner case that: in former LFSCK scanning we have
1019 * created the .lustre/lost+found/MDTxxxx but failed to update
1020 * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
1021 * it from MDT0 firstly. */
1022 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1023 (const struct dt_key *)name);
1024 if (rc != 0 && rc != -ENOENT)
1028 bk->lb_lpf_fid = *cfid;
1029 rc = lfsck_bookmark_store(env, lfsck);
1031 rc = lfsck_fid_alloc(env, lfsck, cfid, true);
1036 *cfid = bk->lb_lpf_fid;
1039 child = lfsck_object_find_bottom_new(env, lfsck, cfid);
1041 GOTO(unlock, rc = PTR_ERR(child));
1043 if (dt_object_exists(child) != 0) {
1044 if (unlikely(!dt_try_as_dir(env, child)))
1047 lfsck->li_lpf_obj = child;
1052 memset(la, 0, sizeof(*la));
1053 la->la_atime = la->la_mtime = la->la_ctime = ktime_get_real_seconds();
1054 la->la_mode = S_IFDIR | S_IRWXU;
1055 la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1057 memset(dof, 0, sizeof(*dof));
1058 dof->dof_type = dt_mode_to_dft(S_IFDIR);
1061 rc = lfsck_create_lpf_local(env, lfsck, child, la, dof, name);
1063 rc = lfsck_create_lpf_remote(env, lfsck, child, la, dof, name);
1065 lfsck->li_lpf_obj = child;
1071 if (rc != 0 && child != NULL && !IS_ERR(child))
1072 lfsck_object_put(env, child);
1078 * Scan .lustre/lost+found for bad name entries and remove them.
1080 * The valid name entry should be "MDTxxxx", the "xxxx" is the MDT device
1081 * index in the system. Any other formatted name is invalid and should be
1084 * \param[in] env pointer to the thread context
1085 * \param[in] lfsck pointer to the lfsck instance
1087 * \retval 0 for success
1088 * \retval negative error number on failure
1090 static int lfsck_scan_lpf_bad_entries(const struct lu_env *env,
1091 struct lfsck_instance *lfsck)
1093 struct dt_object *parent = lfsck->li_lpf_root_obj;
1094 struct lu_dirent *ent =
1095 (struct lu_dirent *)lfsck_env_info(env)->lti_key;
1096 const struct dt_it_ops *iops = &parent->do_index_ops->dio_it;
1101 it = iops->init(env, parent, LUDA_64BITHASH);
1103 RETURN(PTR_ERR(it));
1105 rc = iops->load(env, it, 0);
1107 rc = iops->next(env, it);
1114 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
1118 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
1119 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
1122 /* name length must be strlen("MDTxxxx") */
1123 if (ent->lde_namelen != 7)
1126 if (memcmp(ent->lde_name, "MDT", off) != 0)
1129 while (off < 7 && isxdigit(ent->lde_name[off]))
1135 rc = lfsck_lpf_remove_name_entry(env, lfsck,
1142 rc = iops->next(env, it);
1146 iops->fini(env, it);
1148 RETURN(rc > 0 ? 0 : rc);
1151 static int lfsck_update_lpf_entry(const struct lu_env *env,
1152 struct lfsck_instance *lfsck,
1153 struct dt_object *parent,
1154 struct dt_object *child,
1156 enum lfsck_verify_lpf_types type)
1160 if (type == LVLT_BY_BOOKMARK) {
1161 rc = lfsck_update_name_entry(env, lfsck, parent, name,
1162 lfsck_dto2fid(child), S_IFDIR);
1163 } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1164 lfsck->li_bookmark_ram.lb_lpf_fid = *lfsck_dto2fid(child);
1165 rc = lfsck_bookmark_store(env, lfsck);
1167 CDEBUG(D_LFSCK, "%s: update LPF fid "DFID
1168 " in the bookmark file: rc = %d\n",
1169 lfsck_lfsck2name(lfsck),
1170 PFID(lfsck_dto2fid(child)), rc);
1177 * Check whether the @child back references the @parent.
1180 * 1) The child's FID is stored in the bookmark file. If the child back
1181 * references the parent (LU_LPF_FID object) via its ".." entry, then
1182 * insert the name (MDTxxxx) to the .lustre/lost+found; otherwise, if
1183 * the child back references another parent2, then:
1184 * 1.1) If the parent2 recognizes the child, then update the bookmark file;
1185 * 1.2) Otherwise, the LFSCK cannot know whether there will be parent3 that
1186 * references the child. So keep them there. As the LFSCK processing,
1187 * the parent3 may be found, then when the LFSCK run next time, the
1188 * inconsistency can be repaired.
1190 * 2) The child's FID is stored in the .lustre/lost+found/ sub-directory name
1191 * entry (MDTxxxx). If the child back references the parent (LU_LPF_FID obj)
1192 * via its ".." entry, then update the bookmark file, otherwise, if the child
1193 * back references another parent2, then:
1194 * 2.1) If the parent2 recognizes the child, then remove the sub-directory
1195 * from .lustre/lost+found/;
1196 * 2.2) Otherwise, if the parent2 does not recognizes the child, trust the
1197 * sub-directory name entry and update the child;
1198 * 2.3) Otherwise, if we do not know whether the parent2 recognizes the child
1199 * or not, then keep them there.
1201 * \param[in] env pointer to the thread context
1202 * \param[in] lfsck pointer to the lfsck instance
1203 * \param[in] child pointer to the lost+found sub-directory object
1204 * \param[in] name the name for lost+found sub-directory object
1205 * \param[out] fid pointer to the buffer to hold the FID of the object
1206 * (called it as parent2) that is referenced via the
1207 * child's dotdot entry; it also can be the FID that
1208 * is referenced by the name entry under the parent2.
1209 * \param[in] type to indicate where the child's FID is stored in
1211 * \retval positive number for uncertain inconsistency
1212 * \retval 0 for success
1213 * \retval negative error number on failure
1215 static int lfsck_verify_lpf_pairs(const struct lu_env *env,
1216 struct lfsck_instance *lfsck,
1217 struct dt_object *child, const char *name,
1219 enum lfsck_verify_lpf_types type)
1221 struct dt_object *parent = lfsck->li_lpf_root_obj;
1222 struct lfsck_thread_info *info = lfsck_env_info(env);
1223 char *name2 = info->lti_key;
1224 struct lu_fid *fid2 = &info->lti_fid3;
1225 struct dt_object *parent2 = NULL;
1226 struct lustre_handle lh = { 0 };
1231 rc = dt_lookup(env, child, (struct dt_rec *)fid,
1232 (const struct dt_key *)dotdot);
1236 if (!fid_is_sane(fid))
1237 GOTO(linkea, rc = -EINVAL);
1239 if (lu_fid_eq(fid, &LU_LPF_FID)) {
1240 const struct lu_name *cname;
1242 if (lfsck->li_lpf_obj == NULL) {
1243 lu_object_get(&child->do_lu);
1244 lfsck->li_lpf_obj = child;
1247 cname = lfsck_name_get_const(env, name, strlen(name));
1248 rc = lfsck_verify_linkea(env, child, cname, &LU_LPF_FID);
1250 rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1256 parent2 = lfsck_object_find_bottom(env, lfsck, fid);
1257 if (IS_ERR(parent2))
1258 GOTO(linkea, parent2);
1260 if (!dt_object_exists(parent2)) {
1261 lfsck_object_put(env, parent2);
1263 GOTO(linkea, parent2 = ERR_PTR(-ENOENT));
1266 if (!dt_try_as_dir(env, parent2)) {
1267 lfsck_object_put(env, parent2);
1269 GOTO(linkea, parent2 = ERR_PTR(-ENOTDIR));
1273 /* To prevent rename/unlink race */
1274 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
1275 MDS_INODELOCK_UPDATE, LCK_PR);
1279 dt_read_lock(env, child, 0);
1280 rc = lfsck_links_get_first(env, child, name2, fid2);
1282 dt_read_unlock(env, child);
1283 lfsck_ibits_unlock(&lh, LCK_PR);
1285 GOTO(out_put, rc = 1);
1288 /* It is almost impossible that the bookmark file (or the name entry)
1289 * and the linkEA hit the same data corruption. Trust the linkEA. */
1290 if (lu_fid_eq(fid2, &LU_LPF_FID) && strcmp(name, name2) == 0) {
1291 dt_read_unlock(env, child);
1292 lfsck_ibits_unlock(&lh, LCK_PR);
1295 if (lfsck->li_lpf_obj == NULL) {
1296 lu_object_get(&child->do_lu);
1297 lfsck->li_lpf_obj = child;
1300 /* Update the child's dotdot entry */
1301 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1302 &LU_LPF_FID, S_IFDIR);
1304 rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1310 if (parent2 == NULL || IS_ERR(parent2)) {
1311 dt_read_unlock(env, child);
1312 lfsck_ibits_unlock(&lh, LCK_PR);
1314 GOTO(out_done, rc = 1);
1317 rc = dt_lookup(env, parent2, (struct dt_rec *)fid,
1318 (const struct dt_key *)name2);
1319 dt_read_unlock(env, child);
1320 lfsck_ibits_unlock(&lh, LCK_PR);
1321 if (rc != 0 && rc != -ENOENT)
1324 if (rc == -ENOENT || !lu_fid_eq(fid, lfsck_dto2fid(child))) {
1325 if (type == LVLT_BY_BOOKMARK)
1326 GOTO(out_put, rc = 1);
1328 /* Trust the name entry, update the child's dotdot entry. */
1329 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1330 &LU_LPF_FID, S_IFDIR);
1335 if (type == LVLT_BY_BOOKMARK) {
1336 /* Invalid FID record in the bookmark file, reset it. */
1337 fid_zero(&lfsck->li_bookmark_ram.lb_lpf_fid);
1338 rc = lfsck_bookmark_store(env, lfsck);
1340 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1341 " in the bookmark file: rc = %d\n",
1342 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), rc);
1343 } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1344 /* The name entry is wrong, remove it. */
1345 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1351 if (parent2 != NULL && !IS_ERR(parent2))
1352 lfsck_object_put(env, parent2);
1359 * Verify the /ROOT/.lustre/lost+found/ directory.
1361 * /ROOT/.lustre/lost+found/ is a special directory to hold the objects that
1362 * the LFSCK does not exactly know how to handle, such as orphans. So before
1363 * the LFSCK scanning the system, the consistency of such directory needs to
1364 * be verified firstly to allow the users to use it during the LFSCK.
1366 * \param[in] env pointer to the thread context
1367 * \param[in] lfsck pointer to the lfsck instance
1369 * \retval positive number for uncertain inconsistency
1370 * \retval 0 for success
1371 * \retval negative error number on failure
1373 int lfsck_verify_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
1375 struct lfsck_thread_info *info = lfsck_env_info(env);
1376 struct lu_fid *pfid = &info->lti_fid;
1377 struct lu_fid *cfid = &info->lti_fid2;
1378 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1379 struct dt_object *parent;
1380 /* child1's FID is in the bookmark file. */
1381 struct dt_object *child1 = NULL;
1382 /* child2's FID is in the name entry MDTxxxx. */
1383 struct dt_object *child2 = NULL;
1384 const struct lu_name *cname;
1386 int node = lfsck_dev_idx(lfsck);
1390 LASSERT(lfsck->li_master);
1392 if (lfsck->li_lpf_root_obj != NULL)
1396 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
1399 struct lfsck_tgt_desc *ltd;
1401 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
1402 if (unlikely(ltd == NULL))
1405 parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
1411 RETURN(PTR_ERR(parent));
1413 LASSERT(dt_object_exists(parent));
1415 if (unlikely(!dt_try_as_dir(env, parent))) {
1416 lfsck_object_put(env, parent);
1418 GOTO(put, rc = -ENOTDIR);
1421 lfsck->li_lpf_root_obj = parent;
1423 rc = lfsck_scan_lpf_bad_entries(env, lfsck);
1425 CDEBUG(D_LFSCK, "%s: scan .lustre/lost+found/ "
1426 "for bad sub-directories: rc = %d\n",
1427 lfsck_lfsck2name(lfsck), rc);
1431 snprintf(name, 8, "MDT%04x", node);
1432 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1433 (const struct dt_key *)name);
1434 if (rc == -ENOENT) {
1442 /* Invalid FID in the name entry, remove the name entry. */
1443 if (!fid_is_norm(cfid)) {
1444 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1451 child2 = lfsck_object_find_bottom(env, lfsck, cfid);
1453 GOTO(put, rc = PTR_ERR(child2));
1455 if (unlikely(!dt_object_exists(child2) ||
1456 dt_object_remote(child2)) ||
1457 !S_ISDIR(lfsck_object_type(child2))) {
1458 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1465 if (unlikely(!dt_try_as_dir(env, child2))) {
1466 lfsck_object_put(env, child2);
1472 if (fid_is_zero(&bk->lb_lpf_fid))
1475 if (likely(lu_fid_eq(cfid, &bk->lb_lpf_fid))) {
1476 if (lfsck->li_lpf_obj == NULL) {
1477 lu_object_get(&child2->do_lu);
1478 lfsck->li_lpf_obj = child2;
1481 cname = lfsck_name_get_const(env, name, strlen(name));
1482 rc = lfsck_verify_linkea(env, child2, cname, &LU_LPF_FID);
1487 if (unlikely(!fid_is_norm(&bk->lb_lpf_fid))) {
1488 struct lu_fid tfid = bk->lb_lpf_fid;
1490 /* Invalid FID record in the bookmark file, reset it. */
1491 fid_zero(&bk->lb_lpf_fid);
1492 rc = lfsck_bookmark_store(env, lfsck);
1494 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1495 " in the bookmark file: rc = %d\n",
1496 lfsck_lfsck2name(lfsck), PFID(&tfid), rc);
1504 child1 = lfsck_object_find_bottom(env, lfsck, &bk->lb_lpf_fid);
1505 if (IS_ERR(child1)) {
1510 if (unlikely(!dt_object_exists(child1) ||
1511 dt_object_remote(child1)) ||
1512 !S_ISDIR(lfsck_object_type(child1))) {
1513 /* Invalid FID record in the bookmark file, reset it. */
1514 fid_zero(&bk->lb_lpf_fid);
1515 rc = lfsck_bookmark_store(env, lfsck);
1517 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1518 " in the bookmark file: rc = %d\n",
1519 lfsck_lfsck2name(lfsck),
1520 PFID(lfsck_dto2fid(child1)), rc);
1525 lfsck_object_put(env, child1);
1530 if (unlikely(!dt_try_as_dir(env, child1))) {
1531 lfsck_object_put(env, child1);
1537 rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name, pfid,
1539 if (lu_fid_eq(pfid, &LU_LPF_FID))
1544 rc = lfsck_verify_lpf_pairs(env, lfsck, child2, name,
1545 pfid, LVLT_BY_NAMEENTRY);
1550 if (lfsck->li_lpf_obj != NULL) {
1551 if (unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj))) {
1552 lfsck_object_put(env, lfsck->li_lpf_obj);
1553 lfsck->li_lpf_obj = NULL;
1556 } else if (rc == 0) {
1557 rc = lfsck_create_lpf(env, lfsck);
1560 if (child2 != NULL && !IS_ERR(child2))
1561 lfsck_object_put(env, child2);
1562 if (child1 != NULL && !IS_ERR(child1))
1563 lfsck_object_put(env, child1);
1568 static int lfsck_fid_init(struct lfsck_instance *lfsck)
1570 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1571 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1576 if (unlikely(ss == NULL))
1579 OBD_ALLOC_PTR(lfsck->li_seq);
1580 if (lfsck->li_seq == NULL)
1583 OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
1585 GOTO(out, rc = -ENOMEM);
1587 snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
1588 rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
1590 OBD_FREE(prefix, MAX_OBD_NAME + 7);
1594 if (fid_is_sane(&bk->lb_last_fid))
1595 lfsck->li_seq->lcs_fid = bk->lb_last_fid;
1600 OBD_FREE_PTR(lfsck->li_seq);
1601 lfsck->li_seq = NULL;
1606 static void lfsck_fid_fini(struct lfsck_instance *lfsck)
1608 if (lfsck->li_seq != NULL) {
1609 seq_client_fini(lfsck->li_seq);
1610 OBD_FREE_PTR(lfsck->li_seq);
1611 lfsck->li_seq = NULL;
1615 void lfsck_instance_cleanup(const struct lu_env *env,
1616 struct lfsck_instance *lfsck)
1618 struct ptlrpc_thread *thread = &lfsck->li_thread;
1619 struct lfsck_component *com;
1620 struct lfsck_component *next;
1621 struct lfsck_lmv_unit *llu;
1622 struct lfsck_lmv_unit *llu_next;
1623 struct lfsck_lmv *llmv;
1626 LASSERT(list_empty(&lfsck->li_link));
1627 LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
1629 if (lfsck->li_obj_oit != NULL) {
1630 lfsck_object_put(env, lfsck->li_obj_oit);
1631 lfsck->li_obj_oit = NULL;
1634 LASSERT(lfsck->li_obj_dir == NULL);
1635 LASSERT(lfsck->li_lmv == NULL);
1637 list_for_each_entry_safe(llu, llu_next, &lfsck->li_list_lmv, llu_link) {
1638 llmv = &llu->llu_lmv;
1640 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
1641 "still in using: %u\n",
1642 atomic_read(&llmv->ll_ref));
1644 lfsck_lmv_put(env, llmv);
1647 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
1648 lfsck_component_cleanup(env, com);
1651 LASSERT(list_empty(&lfsck->li_list_dir));
1653 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
1655 lfsck_component_cleanup(env, com);
1658 list_for_each_entry_safe(com, next, &lfsck->li_list_idle, lc_link) {
1659 lfsck_component_cleanup(env, com);
1662 lfsck_tgt_descs_fini(&lfsck->li_ost_descs);
1663 lfsck_tgt_descs_fini(&lfsck->li_mdt_descs);
1665 if (lfsck->li_lfsck_dir != NULL) {
1666 lfsck_object_put(env, lfsck->li_lfsck_dir);
1667 lfsck->li_lfsck_dir = NULL;
1670 if (lfsck->li_bookmark_obj != NULL) {
1671 lfsck_object_put(env, lfsck->li_bookmark_obj);
1672 lfsck->li_bookmark_obj = NULL;
1675 if (lfsck->li_lpf_obj != NULL) {
1676 lfsck_object_put(env, lfsck->li_lpf_obj);
1677 lfsck->li_lpf_obj = NULL;
1680 if (lfsck->li_lpf_root_obj != NULL) {
1681 lfsck_object_put(env, lfsck->li_lpf_root_obj);
1682 lfsck->li_lpf_root_obj = NULL;
1685 if (lfsck->li_los != NULL) {
1686 local_oid_storage_fini(env, lfsck->li_los);
1687 lfsck->li_los = NULL;
1690 lfsck_fid_fini(lfsck);
1692 OBD_FREE_PTR(lfsck);
1695 static inline struct lfsck_instance *
1696 __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink)
1698 struct lfsck_instance *lfsck;
1700 list_for_each_entry(lfsck, &lfsck_instance_list, li_link) {
1701 if (lfsck->li_bottom == key) {
1703 lfsck_instance_get(lfsck);
1705 list_del_init(&lfsck->li_link);
1714 struct lfsck_instance *lfsck_instance_find(struct dt_device *key, bool ref,
1717 struct lfsck_instance *lfsck;
1719 spin_lock(&lfsck_instance_lock);
1720 lfsck = __lfsck_instance_find(key, ref, unlink);
1721 spin_unlock(&lfsck_instance_lock);
1726 static inline int lfsck_instance_add(struct lfsck_instance *lfsck)
1728 struct lfsck_instance *tmp;
1730 spin_lock(&lfsck_instance_lock);
1731 list_for_each_entry(tmp, &lfsck_instance_list, li_link) {
1732 if (lfsck->li_bottom == tmp->li_bottom) {
1733 spin_unlock(&lfsck_instance_lock);
1738 list_add_tail(&lfsck->li_link, &lfsck_instance_list);
1739 spin_unlock(&lfsck_instance_lock);
1743 void lfsck_bits_dump(struct seq_file *m, int bits, const char *names[],
1748 bool newline = (bits != 0 ? false : true);
1750 seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1752 for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
1755 if (names[i] != NULL) {
1759 seq_printf(m, "%s%c", names[i],
1760 newline ? '\n' : ',');
1769 void lfsck_time_dump(struct seq_file *m, time64_t time, const char *name)
1772 seq_printf(m, "%s_time: N/A\n", name);
1773 seq_printf(m, "time_since_%s: N/A\n", name);
1775 seq_printf(m, "%s_time: %lld\n", name, time);
1776 seq_printf(m, "time_since_%s: %lld seconds\n",
1777 name, ktime_get_real_seconds() - time);
1781 void lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos,
1784 if (fid_is_zero(&pos->lp_dir_parent)) {
1785 if (pos->lp_oit_cookie == 0) {
1786 seq_printf(m, "%s: N/A, N/A, N/A\n", prefix);
1789 seq_printf(m, "%s: %llu, N/A, N/A\n",
1790 prefix, pos->lp_oit_cookie);
1792 seq_printf(m, "%s: %llu, "DFID", %#llx\n",
1793 prefix, pos->lp_oit_cookie,
1794 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
1798 void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
1799 struct lfsck_position *pos, bool init)
1801 const struct dt_it_ops *iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
1803 if (unlikely(lfsck->li_di_oit == NULL)) {
1804 memset(pos, 0, sizeof(*pos));
1808 pos->lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
1809 if (!lfsck->li_current_oit_processed && !init)
1810 pos->lp_oit_cookie--;
1812 if (unlikely(pos->lp_oit_cookie == 0))
1813 pos->lp_oit_cookie = 1;
1815 if (lfsck->li_di_dir != NULL) {
1816 struct dt_object *dto = lfsck->li_obj_dir;
1818 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
1821 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
1822 fid_zero(&pos->lp_dir_parent);
1823 pos->lp_dir_cookie = 0;
1825 pos->lp_dir_parent = *lfsck_dto2fid(dto);
1828 fid_zero(&pos->lp_dir_parent);
1829 pos->lp_dir_cookie = 0;
1833 bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
1837 if (limit != LFSCK_SPEED_NO_LIMIT) {
1838 if (limit > msecs_to_jiffies(MSEC_PER_SEC)) {
1839 lfsck->li_sleep_rate = limit /
1840 msecs_to_jiffies(MSEC_PER_SEC);
1841 lfsck->li_sleep_jif = 1;
1843 lfsck->li_sleep_rate = 1;
1844 lfsck->li_sleep_jif = msecs_to_jiffies(MSEC_PER_SEC) /
1848 lfsck->li_sleep_jif = 0;
1849 lfsck->li_sleep_rate = 0;
1852 if (lfsck->li_bookmark_ram.lb_speed_limit != limit) {
1853 lfsck->li_bookmark_ram.lb_speed_limit = limit;
1860 void lfsck_control_speed(struct lfsck_instance *lfsck)
1862 struct ptlrpc_thread *thread = &lfsck->li_thread;
1863 struct l_wait_info lwi;
1865 if (lfsck->li_sleep_jif > 0 &&
1866 lfsck->li_new_scanned >= lfsck->li_sleep_rate) {
1867 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1868 LWI_ON_SIGNAL_NOOP, NULL);
1870 l_wait_event(thread->t_ctl_waitq,
1871 !thread_is_running(thread),
1873 lfsck->li_new_scanned = 0;
1877 void lfsck_control_speed_by_self(struct lfsck_component *com)
1879 struct lfsck_instance *lfsck = com->lc_lfsck;
1880 struct ptlrpc_thread *thread = &lfsck->li_thread;
1881 struct l_wait_info lwi;
1883 if (lfsck->li_sleep_jif > 0 &&
1884 com->lc_new_scanned >= lfsck->li_sleep_rate) {
1885 lwi = LWI_TIMEOUT_INTR(lfsck->li_sleep_jif, NULL,
1886 LWI_ON_SIGNAL_NOOP, NULL);
1888 l_wait_event(thread->t_ctl_waitq,
1889 !thread_is_running(thread),
1891 com->lc_new_scanned = 0;
1895 static struct lfsck_thread_args *
1896 lfsck_thread_args_init(struct lfsck_instance *lfsck,
1897 struct lfsck_component *com,
1898 struct lfsck_start_param *lsp)
1900 struct lfsck_thread_args *lta;
1905 return ERR_PTR(-ENOMEM);
1907 rc = lu_env_init(<a->lta_env, LCT_MD_THREAD | LCT_DT_THREAD);
1913 lta->lta_lfsck = lfsck_instance_get(lfsck);
1915 lta->lta_com = lfsck_component_get(com);
1922 void lfsck_thread_args_fini(struct lfsck_thread_args *lta)
1924 if (lta->lta_com != NULL)
1925 lfsck_component_put(<a->lta_env, lta->lta_com);
1926 lfsck_instance_put(<a->lta_env, lta->lta_lfsck);
1927 lu_env_fini(<a->lta_env);
1931 struct lfsck_assistant_data *
1932 lfsck_assistant_data_init(struct lfsck_assistant_operations *lao,
1935 struct lfsck_assistant_data *lad;
1939 lad->lad_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
1940 if (lad->lad_bitmap == NULL) {
1945 INIT_LIST_HEAD(&lad->lad_req_list);
1946 spin_lock_init(&lad->lad_lock);
1947 INIT_LIST_HEAD(&lad->lad_ost_list);
1948 INIT_LIST_HEAD(&lad->lad_ost_phase1_list);
1949 INIT_LIST_HEAD(&lad->lad_ost_phase2_list);
1950 INIT_LIST_HEAD(&lad->lad_mdt_list);
1951 INIT_LIST_HEAD(&lad->lad_mdt_phase1_list);
1952 INIT_LIST_HEAD(&lad->lad_mdt_phase2_list);
1953 init_waitqueue_head(&lad->lad_thread.t_ctl_waitq);
1955 lad->lad_name = name;
1961 struct lfsck_assistant_object *
1962 lfsck_assistant_object_init(const struct lu_env *env, const struct lu_fid *fid,
1963 const struct lu_attr *attr, __u64 cookie,
1966 struct lfsck_assistant_object *lso;
1970 return ERR_PTR(-ENOMEM);
1972 lso->lso_fid = *fid;
1974 lso->lso_attr = *attr;
1976 atomic_set(&lso->lso_ref, 1);
1977 lso->lso_oit_cookie = cookie;
1979 lso->lso_is_dir = 1;
1985 lfsck_assistant_object_load(const struct lu_env *env,
1986 struct lfsck_instance *lfsck,
1987 struct lfsck_assistant_object *lso)
1989 struct dt_object *obj;
1991 obj = lfsck_object_find_bottom(env, lfsck, &lso->lso_fid);
1995 if (unlikely(!dt_object_exists(obj) || lfsck_is_dead_obj(obj))) {
1997 lfsck_object_put(env, obj);
1999 return ERR_PTR(-ENOENT);
2002 if (lso->lso_is_dir && unlikely(!dt_try_as_dir(env, obj))) {
2003 lfsck_object_put(env, obj);
2005 return ERR_PTR(-ENOTDIR);
2012 * Generic LFSCK asynchronous communication interpretor function.
2013 * The LFSCK RPC reply for both the event notification and status
2014 * querying will be handled here.
2016 * \param[in] env pointer to the thread context
2017 * \param[in] req pointer to the LFSCK request
2018 * \param[in] args pointer to the lfsck_async_interpret_args
2019 * \param[in] rc the result for handling the LFSCK request
2021 * \retval 0 for success
2022 * \retval negative error number on failure
2024 int lfsck_async_interpret_common(const struct lu_env *env,
2025 struct ptlrpc_request *req,
2028 struct lfsck_async_interpret_args *laia = args;
2029 struct lfsck_component *com = laia->laia_com;
2030 struct lfsck_assistant_data *lad = com->lc_data;
2031 struct lfsck_tgt_descs *ltds = laia->laia_ltds;
2032 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
2033 struct lfsck_request *lr = laia->laia_lr;
2035 LASSERT(com->lc_lfsck->li_master);
2037 switch (lr->lr_event) {
2039 if (unlikely(rc == -EINPROGRESS)) {
2040 ltd->ltd_retry_start = 1;
2045 CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s "
2047 lfsck_lfsck2name(com->lc_lfsck),
2048 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2049 ltd->ltd_index, lad->lad_name, rc);
2051 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2052 struct lfsck_layout *lo = com->lc_file_ram;
2054 if (lr->lr_flags & LEF_TO_OST)
2055 lfsck_lad_set_bitmap(env, com,
2058 lo->ll_flags |= LF_INCOMPLETE;
2060 struct lfsck_namespace *ns = com->lc_file_ram;
2062 /* If some MDT does not join the namespace
2063 * LFSCK, then we cannot know whether there
2064 * is some name entry on such MDT that with
2065 * the referenced MDT-object on this MDT or
2066 * not. So the namespace LFSCK on this MDT
2067 * cannot handle orphan MDT-objects properly.
2068 * So we mark the LFSCK as LF_INCOMPLETE and
2069 * skip orphan MDT-objects handling. */
2070 ns->ln_flags |= LF_INCOMPLETE;
2075 spin_lock(<ds->ltd_lock);
2076 if (ltd->ltd_dead) {
2077 spin_unlock(<ds->ltd_lock);
2081 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2082 struct list_head *list;
2083 struct list_head *phase_list;
2085 if (ltd->ltd_layout_done) {
2086 spin_unlock(<ds->ltd_lock);
2090 if (lr->lr_flags & LEF_TO_OST) {
2091 list = &lad->lad_ost_list;
2092 phase_list = &lad->lad_ost_phase1_list;
2094 list = &lad->lad_mdt_list;
2095 phase_list = &lad->lad_mdt_phase1_list;
2098 if (list_empty(<d->ltd_layout_list))
2099 list_add_tail(<d->ltd_layout_list, list);
2100 if (list_empty(<d->ltd_layout_phase_list))
2101 list_add_tail(<d->ltd_layout_phase_list,
2104 if (ltd->ltd_namespace_done) {
2105 spin_unlock(<ds->ltd_lock);
2109 if (list_empty(<d->ltd_namespace_list))
2110 list_add_tail(<d->ltd_namespace_list,
2111 &lad->lad_mdt_list);
2112 if (list_empty(<d->ltd_namespace_phase_list))
2113 list_add_tail(<d->ltd_namespace_phase_list,
2114 &lad->lad_mdt_phase1_list);
2116 spin_unlock(<ds->ltd_lock);
2119 case LE_PHASE1_DONE:
2120 case LE_PHASE2_DONE:
2122 if (rc != 0 && rc != -EALREADY)
2123 CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s: "
2124 "event = %d, rc = %d\n",
2125 lfsck_lfsck2name(com->lc_lfsck),
2126 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2127 ltd->ltd_index, lad->lad_name, lr->lr_event, rc);
2130 struct lfsck_reply *reply;
2131 struct list_head *list;
2132 struct list_head *phase_list;
2134 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2135 list = <d->ltd_layout_list;
2136 phase_list = <d->ltd_layout_phase_list;
2138 list = <d->ltd_namespace_list;
2139 phase_list = <d->ltd_namespace_phase_list;
2143 if (lr->lr_flags & LEF_QUERY_ALL) {
2144 lfsck_reset_ltd_status(ltd, com->lc_type);
2148 spin_lock(<ds->ltd_lock);
2149 list_del_init(phase_list);
2150 list_del_init(list);
2151 spin_unlock(<ds->ltd_lock);
2155 reply = req_capsule_server_get(&req->rq_pill,
2157 if (reply == NULL) {
2159 CDEBUG(D_LFSCK, "%s: invalid query reply for %s: "
2160 "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2163 if (lr->lr_flags & LEF_QUERY_ALL) {
2164 lfsck_reset_ltd_status(ltd, com->lc_type);
2168 spin_lock(<ds->ltd_lock);
2169 list_del_init(phase_list);
2170 list_del_init(list);
2171 spin_unlock(<ds->ltd_lock);
2175 if (lr->lr_flags & LEF_QUERY_ALL) {
2176 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2177 ltd->ltd_layout_status = reply->lr_status;
2178 ltd->ltd_layout_repaired = reply->lr_repaired;
2180 ltd->ltd_namespace_status = reply->lr_status;
2181 ltd->ltd_namespace_repaired =
2187 switch (reply->lr_status) {
2188 case LS_SCANNING_PHASE1:
2190 case LS_SCANNING_PHASE2:
2191 spin_lock(<ds->ltd_lock);
2192 list_del_init(phase_list);
2193 if (ltd->ltd_dead) {
2194 spin_unlock(<ds->ltd_lock);
2198 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2199 if (ltd->ltd_layout_done) {
2200 spin_unlock(<ds->ltd_lock);
2204 if (lr->lr_flags & LEF_TO_OST)
2205 list_add_tail(phase_list,
2206 &lad->lad_ost_phase2_list);
2208 list_add_tail(phase_list,
2209 &lad->lad_mdt_phase2_list);
2211 if (ltd->ltd_namespace_done) {
2212 spin_unlock(<ds->ltd_lock);
2216 list_add_tail(phase_list,
2217 &lad->lad_mdt_phase2_list);
2219 spin_unlock(<ds->ltd_lock);
2222 spin_lock(<ds->ltd_lock);
2223 list_del_init(phase_list);
2224 list_del_init(list);
2225 spin_unlock(<ds->ltd_lock);
2231 CDEBUG(D_LFSCK, "%s: unexpected event: rc = %d\n",
2232 lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
2236 if (!laia->laia_shared) {
2238 lfsck_component_put(env, com);
2244 static void lfsck_interpret(const struct lu_env *env,
2245 struct lfsck_instance *lfsck,
2246 struct ptlrpc_request *req, void *args, int result)
2248 struct lfsck_async_interpret_args *laia = args;
2249 struct lfsck_component *com;
2251 LASSERT(laia->laia_com == NULL);
2252 LASSERT(laia->laia_shared);
2254 spin_lock(&lfsck->li_lock);
2255 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
2256 laia->laia_com = com;
2257 lfsck_async_interpret_common(env, req, laia, result);
2260 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
2261 laia->laia_com = com;
2262 lfsck_async_interpret_common(env, req, laia, result);
2264 spin_unlock(&lfsck->li_lock);
2267 static int lfsck_stop_notify(const struct lu_env *env,
2268 struct lfsck_instance *lfsck,
2269 struct lfsck_tgt_descs *ltds,
2270 struct lfsck_tgt_desc *ltd, __u16 type)
2272 struct lfsck_component *com;
2276 LASSERT(lfsck->li_master);
2278 spin_lock(&lfsck->li_lock);
2279 com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
2281 com = __lfsck_component_find(lfsck, type,
2282 &lfsck->li_list_double_scan);
2284 lfsck_component_get(com);
2285 spin_unlock(&lfsck->li_lock);
2288 struct lfsck_thread_info *info = lfsck_env_info(env);
2289 struct lfsck_async_interpret_args *laia = &info->lti_laia;
2290 struct lfsck_request *lr = &info->lti_lr;
2291 struct lfsck_assistant_data *lad = com->lc_data;
2292 struct list_head *list;
2293 struct list_head *phase_list;
2294 struct ptlrpc_request_set *set;
2296 set = ptlrpc_prep_set();
2298 lfsck_component_put(env, com);
2303 if (type == LFSCK_TYPE_LAYOUT) {
2304 list = <d->ltd_layout_list;
2305 phase_list = <d->ltd_layout_phase_list;
2307 list = <d->ltd_namespace_list;
2308 phase_list = <d->ltd_namespace_phase_list;
2311 spin_lock(<ds->ltd_lock);
2312 if (list_empty(list)) {
2313 LASSERT(list_empty(phase_list));
2314 spin_unlock(<ds->ltd_lock);
2315 ptlrpc_set_destroy(set);
2320 list_del_init(phase_list);
2321 list_del_init(list);
2322 spin_unlock(<ds->ltd_lock);
2324 memset(lr, 0, sizeof(*lr));
2325 lr->lr_index = lfsck_dev_idx(lfsck);
2326 lr->lr_event = LE_PEER_EXIT;
2327 lr->lr_active = type;
2328 lr->lr_status = LS_CO_PAUSED;
2329 if (ltds == &lfsck->li_ost_descs)
2330 lr->lr_flags = LEF_TO_OST;
2332 memset(laia, 0, sizeof(*laia));
2333 laia->laia_com = com;
2334 laia->laia_ltds = ltds;
2335 atomic_inc(<d->ltd_ref);
2336 laia->laia_ltd = ltd;
2339 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2340 lfsck_async_interpret_common,
2341 laia, LFSCK_NOTIFY);
2343 CDEBUG(D_LFSCK, "%s: fail to notify %s %x for "
2344 "co-stop for %s: rc = %d\n",
2345 lfsck_lfsck2name(lfsck),
2346 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2347 ltd->ltd_index, lad->lad_name, rc);
2350 rc = ptlrpc_set_wait(env, set);
2353 ptlrpc_set_destroy(set);
2354 lfsck_component_put(env, com);
2360 static int lfsck_async_interpret(const struct lu_env *env,
2361 struct ptlrpc_request *req,
2364 struct lfsck_async_interpret_args *laia = args;
2365 struct lfsck_instance *lfsck;
2367 lfsck = container_of0(laia->laia_ltds, struct lfsck_instance,
2369 lfsck_interpret(env, lfsck, req, laia, rc);
2370 lfsck_tgt_put(laia->laia_ltd);
2371 if (rc != 0 && laia->laia_result != -EALREADY)
2372 laia->laia_result = rc;
2377 int lfsck_async_request(const struct lu_env *env, struct obd_export *exp,
2378 struct lfsck_request *lr,
2379 struct ptlrpc_request_set *set,
2380 ptlrpc_interpterer_t interpreter,
2381 void *args, int request)
2383 struct lfsck_async_interpret_args *laia;
2384 struct ptlrpc_request *req;
2385 struct lfsck_request *tmp;
2386 struct req_format *format;
2391 format = &RQF_LFSCK_NOTIFY;
2394 format = &RQF_LFSCK_QUERY;
2397 CDEBUG(D_LFSCK, "%s: unknown async request %d: rc = %d\n",
2398 exp->exp_obd->obd_name, request, -EINVAL);
2402 req = ptlrpc_request_alloc(class_exp2cliimp(exp), format);
2406 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, request);
2408 ptlrpc_request_free(req);
2413 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2415 ptlrpc_request_set_replen(req);
2417 laia = ptlrpc_req_async_args(req);
2418 *laia = *(struct lfsck_async_interpret_args *)args;
2419 if (laia->laia_com != NULL)
2420 lfsck_component_get(laia->laia_com);
2421 req->rq_interpret_reply = interpreter;
2422 req->rq_allow_intr = 1;
2423 req->rq_no_delay = 1;
2424 ptlrpc_set_add_req(set, req);
2429 int lfsck_query_all(const struct lu_env *env, struct lfsck_component *com)
2431 struct lfsck_thread_info *info = lfsck_env_info(env);
2432 struct lfsck_request *lr = &info->lti_lr;
2433 struct lfsck_async_interpret_args *laia = &info->lti_laia;
2434 struct lfsck_instance *lfsck = com->lc_lfsck;
2435 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
2436 struct lfsck_tgt_desc *ltd;
2437 struct ptlrpc_request_set *set;
2442 memset(lr, 0, sizeof(*lr));
2443 lr->lr_event = LE_QUERY;
2444 lr->lr_active = com->lc_type;
2445 lr->lr_flags = LEF_QUERY_ALL;
2447 memset(laia, 0, sizeof(*laia));
2448 laia->laia_com = com;
2451 set = ptlrpc_prep_set();
2456 laia->laia_ltds = ltds;
2457 down_read(<ds->ltd_rw_sem);
2458 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2459 ltd = lfsck_tgt_get(ltds, idx);
2460 LASSERT(ltd != NULL);
2462 laia->laia_ltd = ltd;
2463 up_read(<ds->ltd_rw_sem);
2464 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2465 lfsck_async_interpret_common,
2468 struct lfsck_assistant_data *lad = com->lc_data;
2470 CDEBUG(D_LFSCK, "%s: Fail to query %s %x for stat %s: "
2471 "rc = %d\n", lfsck_lfsck2name(lfsck),
2472 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2473 ltd->ltd_index, lad->lad_name, rc);
2474 lfsck_reset_ltd_status(ltd, com->lc_type);
2477 down_read(<ds->ltd_rw_sem);
2479 up_read(<ds->ltd_rw_sem);
2481 if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST)) {
2482 ltds = &lfsck->li_ost_descs;
2483 lr->lr_flags |= LEF_TO_OST;
2487 rc = ptlrpc_set_wait(env, set);
2488 ptlrpc_set_destroy(set);
2493 int lfsck_start_assistant(const struct lu_env *env, struct lfsck_component *com,
2494 struct lfsck_start_param *lsp)
2496 struct lfsck_instance *lfsck = com->lc_lfsck;
2497 struct lfsck_assistant_data *lad = com->lc_data;
2498 struct ptlrpc_thread *mthread = &lfsck->li_thread;
2499 struct ptlrpc_thread *athread = &lad->lad_thread;
2500 struct lfsck_thread_args *lta;
2501 struct task_struct *task;
2505 lad->lad_assistant_status = 0;
2506 lad->lad_post_result = 0;
2507 lad->lad_to_post = 0;
2508 lad->lad_to_double_scan = 0;
2509 lad->lad_in_double_scan = 0;
2511 lad->lad_advance_lock = false;
2512 thread_set_flags(athread, 0);
2514 lta = lfsck_thread_args_init(lfsck, com, lsp);
2516 RETURN(PTR_ERR(lta));
2518 task = kthread_run(lfsck_assistant_engine, lta, lad->lad_name);
2521 CERROR("%s: cannot start LFSCK assistant thread for %s: "
2522 "rc = %d\n", lfsck_lfsck2name(lfsck), lad->lad_name, rc);
2523 lfsck_thread_args_fini(lta);
2525 struct l_wait_info lwi = { 0 };
2527 l_wait_event(mthread->t_ctl_waitq,
2528 thread_is_running(athread) ||
2529 thread_is_stopped(athread) ||
2530 !thread_is_starting(mthread),
2532 if (unlikely(!thread_is_starting(mthread)))
2533 /* stopped by race */
2535 else if (unlikely(!thread_is_running(athread)))
2536 rc = lad->lad_assistant_status;
2544 int lfsck_checkpoint_generic(const struct lu_env *env,
2545 struct lfsck_component *com)
2547 struct lfsck_assistant_data *lad = com->lc_data;
2548 struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread;
2549 struct ptlrpc_thread *athread = &lad->lad_thread;
2550 struct l_wait_info lwi = { 0 };
2552 l_wait_event(mthread->t_ctl_waitq,
2553 list_empty(&lad->lad_req_list) ||
2554 !thread_is_running(mthread) ||
2555 thread_is_stopped(athread),
2558 if (!thread_is_running(mthread) || thread_is_stopped(athread))
2559 return LFSCK_CHECKPOINT_SKIP;
2564 void lfsck_post_generic(const struct lu_env *env,
2565 struct lfsck_component *com, int *result)
2567 struct lfsck_assistant_data *lad = com->lc_data;
2568 struct ptlrpc_thread *athread = &lad->lad_thread;
2569 struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread;
2570 struct l_wait_info lwi = { 0 };
2572 lad->lad_post_result = *result;
2575 lad->lad_to_post = 1;
2577 CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s post, rc = %d\n",
2578 lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2580 wake_up_all(&athread->t_ctl_waitq);
2581 l_wait_event(mthread->t_ctl_waitq,
2582 (*result > 0 && list_empty(&lad->lad_req_list)) ||
2583 thread_is_stopped(athread),
2586 if (lad->lad_assistant_status < 0)
2587 *result = lad->lad_assistant_status;
2589 CDEBUG(D_LFSCK, "%s: the assistant has done %s post, rc = %d\n",
2590 lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2593 int lfsck_double_scan_generic(const struct lu_env *env,
2594 struct lfsck_component *com, int status)
2596 struct lfsck_assistant_data *lad = com->lc_data;
2597 struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread;
2598 struct ptlrpc_thread *athread = &lad->lad_thread;
2599 struct l_wait_info lwi = { 0 };
2601 if (status != LS_SCANNING_PHASE2)
2604 lad->lad_to_double_scan = 1;
2606 CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s double_scan, "
2608 lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, status);
2610 wake_up_all(&athread->t_ctl_waitq);
2611 l_wait_event(mthread->t_ctl_waitq,
2612 lad->lad_in_double_scan ||
2613 thread_is_stopped(athread),
2616 CDEBUG(D_LFSCK, "%s: the assistant has done %s double_scan, "
2617 "status %d\n", lfsck_lfsck2name(com->lc_lfsck), lad->lad_name,
2618 lad->lad_assistant_status);
2620 if (lad->lad_assistant_status < 0)
2621 return lad->lad_assistant_status;
2626 void lfsck_quit_generic(const struct lu_env *env,
2627 struct lfsck_component *com)
2629 struct lfsck_assistant_data *lad = com->lc_data;
2630 struct ptlrpc_thread *mthread = &com->lc_lfsck->li_thread;
2631 struct ptlrpc_thread *athread = &lad->lad_thread;
2632 struct l_wait_info lwi = { 0 };
2635 wake_up_all(&athread->t_ctl_waitq);
2636 l_wait_event(mthread->t_ctl_waitq,
2637 thread_is_init(athread) ||
2638 thread_is_stopped(athread),
2642 int lfsck_load_one_trace_file(const struct lu_env *env,
2643 struct lfsck_component *com,
2644 struct dt_object *parent,
2645 struct dt_object **child,
2646 const struct dt_index_features *ft,
2647 const char *name, bool reset)
2649 struct lfsck_instance *lfsck = com->lc_lfsck;
2650 struct dt_object *obj;
2654 if (*child != NULL) {
2656 const struct dt_it_ops *iops;
2657 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid3;
2663 rc = obj->do_ops->do_index_try(env, obj, ft);
2665 /* unlink by force */
2668 iops = &obj->do_index_ops->dio_it;
2669 it = iops->init(env, obj, 0);
2671 /* unlink by force */
2675 rc = iops->get(env, it, (const struct dt_key *)fid);
2677 rc = iops->next(env, it);
2680 iops->fini(env, it);
2682 /* "rc > 0" means the index file is empty. */
2686 /* The old index is not empty, remove it firstly. */
2687 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
2688 CDEBUG_LIMIT(rc ? D_ERROR : D_LFSCK,
2689 "%s: unlink lfsck sub trace file %s: rc = %d\n",
2690 lfsck_lfsck2name(com->lc_lfsck), name, rc);
2695 lfsck_object_put(env, *child);
2702 obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
2703 S_IFREG | S_IRUGO | S_IWUSR, ft);
2705 RETURN(PTR_ERR(obj));
2707 rc = obj->do_ops->do_index_try(env, obj, ft);
2709 lfsck_object_put(env, obj);
2710 CDEBUG(D_LFSCK, "%s: LFSCK fail to load "
2711 "sub trace file %s: rc = %d\n",
2712 lfsck_lfsck2name(com->lc_lfsck), name, rc);
2720 int lfsck_load_sub_trace_files(const struct lu_env *env,
2721 struct lfsck_component *com,
2722 const struct dt_index_features *ft,
2723 const char *prefix, bool reset)
2725 char *name = lfsck_env_info(env)->lti_key;
2726 struct lfsck_sub_trace_obj *lsto;
2730 for (i = 0, rc = 0, lsto = &com->lc_sub_trace_objs[0];
2731 i < LFSCK_STF_COUNT && rc == 0; i++, lsto++) {
2732 snprintf(name, NAME_MAX, "%s_%02d", prefix, i);
2733 rc = lfsck_load_one_trace_file(env, com,
2734 com->lc_lfsck->li_lfsck_dir,
2735 &lsto->lsto_obj, ft, name, reset);
2741 /* external interfaces */
2742 int lfsck_get_speed(struct seq_file *m, char *buf, struct dt_device *key)
2745 struct lfsck_instance *lfsck;
2749 rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2753 lfsck = lfsck_instance_find(key, true, false);
2754 if (likely(lfsck != NULL)) {
2756 seq_printf(m, "%u\n",
2757 lfsck->li_bookmark_ram.lb_speed_limit);
2759 rc = sprintf(buf, "%u\n",
2760 lfsck->li_bookmark_ram.lb_speed_limit);
2762 lfsck_instance_put(&env, lfsck);
2771 EXPORT_SYMBOL(lfsck_get_speed);
2773 int lfsck_set_speed(struct dt_device *key, __u32 val)
2776 struct lfsck_instance *lfsck;
2780 rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2784 lfsck = lfsck_instance_find(key, true, false);
2785 if (likely(lfsck != NULL)) {
2786 mutex_lock(&lfsck->li_mutex);
2787 if (__lfsck_set_speed(lfsck, val))
2788 rc = lfsck_bookmark_store(&env, lfsck);
2789 mutex_unlock(&lfsck->li_mutex);
2790 lfsck_instance_put(&env, lfsck);
2799 EXPORT_SYMBOL(lfsck_set_speed);
2801 int lfsck_get_windows(struct seq_file *m, struct dt_device *key)
2804 struct lfsck_instance *lfsck;
2808 rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2812 lfsck = lfsck_instance_find(key, true, false);
2813 if (likely(lfsck != NULL)) {
2814 seq_printf(m, "%u\n", lfsck->li_bookmark_ram.lb_async_windows);
2815 lfsck_instance_put(&env, lfsck);
2824 EXPORT_SYMBOL(lfsck_get_windows);
2826 int lfsck_set_windows(struct dt_device *key, unsigned int val)
2829 struct lfsck_instance *lfsck;
2833 rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2837 lfsck = lfsck_instance_find(key, true, false);
2838 if (likely(lfsck != NULL)) {
2839 if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
2840 CWARN("%s: invalid async windows size that may "
2841 "cause memory issues. The valid range is "
2843 lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
2845 } else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
2846 mutex_lock(&lfsck->li_mutex);
2847 lfsck->li_bookmark_ram.lb_async_windows = val;
2848 rc = lfsck_bookmark_store(&env, lfsck);
2849 mutex_unlock(&lfsck->li_mutex);
2851 lfsck_instance_put(&env, lfsck);
2860 EXPORT_SYMBOL(lfsck_set_windows);
2862 int lfsck_dump(struct seq_file *m, struct dt_device *key, enum lfsck_type type)
2865 struct lfsck_instance *lfsck;
2866 struct lfsck_component *com;
2870 rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2874 lfsck = lfsck_instance_find(key, true, false);
2875 if (likely(lfsck != NULL)) {
2876 com = lfsck_component_find(lfsck, type);
2877 if (likely(com != NULL)) {
2878 com->lc_ops->lfsck_dump(&env, com, m);
2879 lfsck_component_put(&env, com);
2884 lfsck_instance_put(&env, lfsck);
2893 EXPORT_SYMBOL(lfsck_dump);
2895 static int lfsck_stop_all(const struct lu_env *env,
2896 struct lfsck_instance *lfsck,
2897 struct lfsck_stop *stop)
2899 struct lfsck_thread_info *info = lfsck_env_info(env);
2900 struct lfsck_request *lr = &info->lti_lr;
2901 struct lfsck_async_interpret_args *laia = &info->lti_laia;
2902 struct ptlrpc_request_set *set;
2903 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
2904 struct lfsck_tgt_desc *ltd;
2905 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2911 LASSERT(stop->ls_flags & LPF_BROADCAST);
2913 set = ptlrpc_prep_set();
2914 if (unlikely(set == NULL))
2917 memset(lr, 0, sizeof(*lr));
2918 lr->lr_event = LE_STOP;
2919 lr->lr_index = lfsck_dev_idx(lfsck);
2920 lr->lr_status = stop->ls_status;
2921 lr->lr_version = bk->lb_version;
2922 lr->lr_active = LFSCK_TYPES_ALL;
2923 lr->lr_param = stop->ls_flags;
2925 memset(laia, 0, sizeof(*laia));
2926 laia->laia_ltds = ltds;
2928 laia->laia_shared = 1;
2930 down_read(<ds->ltd_rw_sem);
2931 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2932 ltd = lfsck_tgt_get(ltds, idx);
2933 LASSERT(ltd != NULL);
2935 laia->laia_ltd = ltd;
2936 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2937 lfsck_async_interpret, laia,
2940 lfsck_interpret(env, lfsck, NULL, laia, rc);
2942 CERROR("%s: cannot notify MDT %x for LFSCK stop: "
2943 "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc);
2947 up_read(<ds->ltd_rw_sem);
2949 rc = ptlrpc_set_wait(env, set);
2950 ptlrpc_set_destroy(set);
2953 rc = laia->laia_result;
2955 if (rc == -EALREADY)
2959 CERROR("%s: fail to stop LFSCK on some MDTs: rc = %d\n",
2960 lfsck_lfsck2name(lfsck), rc);
2962 RETURN(rc != 0 ? rc : rc1);
2965 static int lfsck_start_all(const struct lu_env *env,
2966 struct lfsck_instance *lfsck,
2967 struct lfsck_start *start)
2969 struct lfsck_thread_info *info = lfsck_env_info(env);
2970 struct lfsck_request *lr = &info->lti_lr;
2971 struct lfsck_async_interpret_args *laia = &info->lti_laia;
2972 struct ptlrpc_request_set *set;
2973 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
2974 struct lfsck_tgt_desc *ltd;
2975 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2981 LASSERT(start->ls_flags & LPF_BROADCAST);
2983 memset(lr, 0, sizeof(*lr));
2984 lr->lr_event = LE_START;
2985 lr->lr_index = lfsck_dev_idx(lfsck);
2986 lr->lr_speed = bk->lb_speed_limit;
2987 lr->lr_version = bk->lb_version;
2988 lr->lr_active = start->ls_active;
2989 lr->lr_param = start->ls_flags;
2990 lr->lr_async_windows = bk->lb_async_windows;
2991 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
2992 LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ |
2995 memset(laia, 0, sizeof(*laia));
2996 laia->laia_ltds = ltds;
2998 laia->laia_shared = 1;
3001 set = ptlrpc_prep_set();
3005 down_read(<ds->ltd_rw_sem);
3006 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3007 ltd = lfsck_tgt_get(ltds, idx);
3008 LASSERT(ltd != NULL);
3010 if (retry && !ltd->ltd_retry_start) {
3015 laia->laia_ltd = ltd;
3016 ltd->ltd_retry_start = 0;
3017 ltd->ltd_layout_done = 0;
3018 ltd->ltd_namespace_done = 0;
3019 ltd->ltd_synced_failures = 0;
3020 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
3021 lfsck_async_interpret, laia,
3024 lfsck_interpret(env, lfsck, NULL, laia, rc);
3026 CERROR("%s: cannot notify MDT %x for LFSCK "
3027 "start, failout: rc = %d\n",
3028 lfsck_lfsck2name(lfsck), idx, rc);
3032 up_read(<ds->ltd_rw_sem);
3035 ptlrpc_set_destroy(set);
3040 rc = ptlrpc_set_wait(env, set);
3041 ptlrpc_set_destroy(set);
3044 rc = laia->laia_result;
3046 if (unlikely(rc == -EINPROGRESS)) {
3048 set_current_state(TASK_INTERRUPTIBLE);
3049 schedule_timeout(msecs_to_jiffies(MSEC_PER_SEC));
3050 set_current_state(TASK_RUNNING);
3051 if (!signal_pending(current) &&
3052 thread_is_running(&lfsck->li_thread))
3059 struct lfsck_stop *stop = &info->lti_stop;
3061 CERROR("%s: cannot start LFSCK on some MDTs, "
3062 "stop all: rc = %d\n",
3063 lfsck_lfsck2name(lfsck), rc);
3064 if (rc != -EALREADY) {
3065 stop->ls_status = LS_FAILED;
3066 stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST;
3067 lfsck_stop_all(env, lfsck, stop);
3074 int lfsck_start(const struct lu_env *env, struct dt_device *key,
3075 struct lfsck_start_param *lsp)
3077 struct lfsck_start *start = lsp->lsp_start;
3078 struct lfsck_instance *lfsck;
3079 struct lfsck_bookmark *bk;
3080 struct ptlrpc_thread *thread;
3081 struct lfsck_component *com;
3082 struct l_wait_info lwi = { 0 };
3083 struct lfsck_thread_args *lta;
3084 struct task_struct *task;
3085 struct lfsck_tgt_descs *ltds;
3086 struct lfsck_tgt_desc *ltd;
3097 lfsck = lfsck_instance_find(key, true, false);
3098 if (unlikely(lfsck == NULL))
3101 if (unlikely(lfsck->li_stopping))
3102 GOTO(put, rc = -ENXIO);
3104 /* System is not ready, try again later. */
3105 if (unlikely(lfsck->li_namespace == NULL ||
3106 lfsck_dev_site(lfsck)->ss_server_fld == NULL))
3107 GOTO(put, rc = -EINPROGRESS);
3109 /* start == NULL means auto trigger paused LFSCK. */
3111 if (list_empty(&lfsck->li_list_scan) ||
3112 OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO))
3114 } else if (start->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3115 CERROR("%s: only allow to specify '-A | -o' via MDS\n",
3116 lfsck_lfsck2name(lfsck));
3118 GOTO(put, rc = -EPERM);
3121 bk = &lfsck->li_bookmark_ram;
3122 thread = &lfsck->li_thread;
3123 mutex_lock(&lfsck->li_mutex);
3124 spin_lock(&lfsck->li_lock);
3125 if (unlikely(thread_is_stopping(thread))) {
3126 /* Someone is stopping the LFSCK. */
3127 spin_unlock(&lfsck->li_lock);
3128 GOTO(out, rc = -EBUSY);
3131 if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
3133 if (unlikely(start == NULL)) {
3134 spin_unlock(&lfsck->li_lock);
3138 while (start->ls_active != 0) {
3139 if (!(type & start->ls_active)) {
3144 com = __lfsck_component_find(lfsck, type,
3145 &lfsck->li_list_scan);
3147 com = __lfsck_component_find(lfsck, type,
3148 &lfsck->li_list_double_scan);
3154 if (com->lc_ops->lfsck_join != NULL) {
3155 rc = com->lc_ops->lfsck_join( env, com, lsp);
3156 if (rc != 0 && rc != -EALREADY)
3159 start->ls_active &= ~type;
3162 spin_unlock(&lfsck->li_lock);
3165 spin_unlock(&lfsck->li_lock);
3167 lfsck->li_status = 0;
3168 lfsck->li_oit_over = 0;
3169 lfsck->li_start_unplug = 0;
3170 lfsck->li_drop_dryrun = 0;
3171 lfsck->li_new_scanned = 0;
3173 /* For auto trigger. */
3177 start->ls_version = bk->lb_version;
3179 if (start->ls_active != 0) {
3180 struct lfsck_component *next;
3182 if (start->ls_active == LFSCK_TYPES_ALL)
3183 start->ls_active = LFSCK_TYPES_SUPPORTED;
3185 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
3186 start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
3187 GOTO(out, rc = -ENOTSUPP);
3190 list_for_each_entry_safe(com, next,
3191 &lfsck->li_list_scan, lc_link) {
3192 if (!(com->lc_type & start->ls_active)) {
3193 rc = com->lc_ops->lfsck_post(env, com, 0,
3200 while (start->ls_active != 0) {
3201 if (type & start->ls_active) {
3202 com = __lfsck_component_find(lfsck, type,
3203 &lfsck->li_list_idle);
3205 /* The component status will be updated
3206 * when its prep() is called later by
3207 * the LFSCK main engine. */
3208 list_move_tail(&com->lc_link,
3209 &lfsck->li_list_scan);
3210 start->ls_active &= ~type;
3216 if (list_empty(&lfsck->li_list_scan)) {
3217 /* The speed limit will be used to control both the LFSCK and
3218 * low layer scrub (if applied), need to be handled firstly. */
3219 if (start->ls_valid & LSV_SPEED_LIMIT) {
3220 if (__lfsck_set_speed(lfsck, start->ls_speed_limit)) {
3221 rc = lfsck_bookmark_store(env, lfsck);
3230 if (start->ls_flags & LPF_RESET)
3231 flags |= DOIF_RESET;
3233 rc = lfsck_set_param(env, lfsck, start, !!(flags & DOIF_RESET));
3237 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3238 start->ls_active |= com->lc_type;
3239 if (flags & DOIF_RESET) {
3240 rc = com->lc_ops->lfsck_reset(env, com, false);
3246 ltds = &lfsck->li_mdt_descs;
3247 down_read(<ds->ltd_rw_sem);
3248 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3249 ltd = lfsck_ltd2tgt(ltds, idx);
3250 LASSERT(ltd != NULL);
3252 ltd->ltd_layout_done = 0;
3253 ltd->ltd_namespace_done = 0;
3254 ltd->ltd_synced_failures = 0;
3255 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_NAMESPACE);
3256 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3257 list_del_init(<d->ltd_layout_phase_list);
3258 list_del_init(<d->ltd_layout_list);
3259 list_del_init(<d->ltd_namespace_phase_list);
3260 list_del_init(<d->ltd_namespace_list);
3262 up_read(<ds->ltd_rw_sem);
3264 ltds = &lfsck->li_ost_descs;
3265 down_read(<ds->ltd_rw_sem);
3266 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3267 ltd = lfsck_ltd2tgt(ltds, idx);
3268 LASSERT(ltd != NULL);
3270 ltd->ltd_layout_done = 0;
3271 ltd->ltd_synced_failures = 0;
3272 lfsck_reset_ltd_status(ltd, LFSCK_TYPE_LAYOUT);
3273 list_del_init(<d->ltd_layout_phase_list);
3274 list_del_init(<d->ltd_layout_list);
3276 up_read(<ds->ltd_rw_sem);
3279 lfsck->li_args_dir = LUDA_64BITHASH | LUDA_VERIFY | LUDA_TYPE;
3280 if (bk->lb_param & LPF_DRYRUN)
3281 lfsck->li_args_dir |= LUDA_VERIFY_DRYRUN;
3283 if (start != NULL && start->ls_valid & LSV_ERROR_HANDLE) {
3284 valid |= DOIV_ERROR_HANDLE;
3285 if (start->ls_flags & LPF_FAILOUT)
3286 flags |= DOIF_FAILOUT;
3289 if (start != NULL && start->ls_valid & LSV_DRYRUN) {
3290 valid |= DOIV_DRYRUN;
3291 if (start->ls_flags & LPF_DRYRUN)
3292 flags |= DOIF_DRYRUN;
3295 if (!list_empty(&lfsck->li_list_scan))
3296 flags |= DOIF_OUTUSED;
3298 lfsck->li_args_oit = (flags << DT_OTABLE_IT_FLAGS_SHIFT) | valid;
3299 lta = lfsck_thread_args_init(lfsck, NULL, lsp);
3301 GOTO(out, rc = PTR_ERR(lta));
3303 __lfsck_set_speed(lfsck, bk->lb_speed_limit);
3304 spin_lock(&lfsck->li_lock);
3305 thread_set_flags(thread, SVC_STARTING);
3306 spin_unlock(&lfsck->li_lock);
3307 task = kthread_run(lfsck_master_engine, lta, "lfsck");
3310 CERROR("%s: cannot start LFSCK thread: rc = %d\n",
3311 lfsck_lfsck2name(lfsck), rc);
3312 lfsck_thread_args_fini(lta);
3317 l_wait_event(thread->t_ctl_waitq,
3318 thread_is_running(thread) ||
3319 thread_is_stopped(thread),
3321 if (start == NULL || !(start->ls_flags & LPF_BROADCAST)) {
3322 lfsck->li_start_unplug = 1;
3323 wake_up_all(&thread->t_ctl_waitq);
3328 /* release lfsck::li_mutex to avoid deadlock. */
3329 mutex_unlock(&lfsck->li_mutex);
3330 rc = lfsck_start_all(env, lfsck, start);
3332 spin_lock(&lfsck->li_lock);
3333 if (thread_is_stopped(thread)) {
3334 spin_unlock(&lfsck->li_lock);
3336 lfsck->li_status = LS_FAILED;
3337 lfsck->li_flags = 0;
3338 thread_set_flags(thread, SVC_STOPPING);
3339 spin_unlock(&lfsck->li_lock);
3341 lfsck->li_start_unplug = 1;
3342 wake_up_all(&thread->t_ctl_waitq);
3343 l_wait_event(thread->t_ctl_waitq,
3344 thread_is_stopped(thread),
3348 lfsck->li_start_unplug = 1;
3349 wake_up_all(&thread->t_ctl_waitq);
3355 mutex_unlock(&lfsck->li_mutex);
3358 lfsck_instance_put(env, lfsck);
3360 return rc < 0 ? rc : 0;
3362 EXPORT_SYMBOL(lfsck_start);
3364 int lfsck_stop(const struct lu_env *env, struct dt_device *key,
3365 struct lfsck_stop *stop)
3367 struct lfsck_instance *lfsck;
3368 struct ptlrpc_thread *thread;
3369 struct l_wait_info lwi = { 0 };
3374 lfsck = lfsck_instance_find(key, true, false);
3375 if (unlikely(lfsck == NULL))
3378 thread = &lfsck->li_thread;
3379 if (stop && stop->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3380 CERROR("%s: only allow to specify '-A' via MDS\n",
3381 lfsck_lfsck2name(lfsck));
3382 GOTO(put, rc = -EPERM);
3385 spin_lock(&lfsck->li_lock);
3386 /* The target is umounted */
3387 if (stop && stop->ls_status == LS_PAUSED)
3388 lfsck->li_stopping = 1;
3390 if (thread_is_init(thread) || thread_is_stopped(thread))
3391 /* no error if LFSCK stopped already, or not started */
3392 GOTO(unlock, rc = 0);
3394 if (thread_is_stopping(thread))
3395 /* Someone is stopping LFSCK. */
3396 GOTO(unlock, rc = -EINPROGRESS);
3399 lfsck->li_status = stop->ls_status;
3400 lfsck->li_flags = stop->ls_flags;
3402 lfsck->li_status = LS_STOPPED;
3403 lfsck->li_flags = 0;
3406 thread_set_flags(thread, SVC_STOPPING);
3408 LASSERT(lfsck->li_task != NULL);
3409 force_sig(SIGINT, lfsck->li_task);
3411 if (lfsck->li_master) {
3412 struct lfsck_component *com;
3413 struct lfsck_assistant_data *lad;
3415 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
3417 spin_lock(&lad->lad_lock);
3418 if (lad->lad_task != NULL)
3419 force_sig(SIGINT, lad->lad_task);
3420 spin_unlock(&lad->lad_lock);
3423 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
3425 spin_lock(&lad->lad_lock);
3426 if (lad->lad_task != NULL)
3427 force_sig(SIGINT, lad->lad_task);
3428 spin_unlock(&lad->lad_lock);
3432 wake_up_all(&thread->t_ctl_waitq);
3433 spin_unlock(&lfsck->li_lock);
3434 if (stop && stop->ls_flags & LPF_BROADCAST)
3435 rc1 = lfsck_stop_all(env, lfsck, stop);
3437 /* It was me set the status as 'stopping' just now, if it is not
3438 * 'stopping' now, then either stopped, or re-started by race. */
3439 l_wait_event(thread->t_ctl_waitq,
3440 !thread_is_stopping(thread),
3446 spin_unlock(&lfsck->li_lock);
3448 lfsck_instance_put(env, lfsck);
3450 return rc != 0 ? rc : rc1;
3452 EXPORT_SYMBOL(lfsck_stop);
3454 int lfsck_in_notify_local(const struct lu_env *env, struct dt_device *key,
3455 struct lfsck_req_local *lrl, struct thandle *th)
3457 struct lfsck_instance *lfsck;
3458 struct lfsck_component *com;
3459 int rc = -EOPNOTSUPP;
3462 lfsck = lfsck_instance_find(key, true, false);
3463 if (unlikely(!lfsck))
3466 com = lfsck_component_find(lfsck, lrl->lrl_active);
3467 if (likely(com && com->lc_ops->lfsck_in_notify_local)) {
3468 rc = com->lc_ops->lfsck_in_notify_local(env, com, lrl, th);
3469 lfsck_component_put(env, com);
3472 lfsck_instance_put(env, lfsck);
3476 EXPORT_SYMBOL(lfsck_in_notify_local);
3478 int lfsck_in_notify(const struct lu_env *env, struct dt_device *key,
3479 struct lfsck_request *lr)
3481 int rc = -EOPNOTSUPP;
3484 switch (lr->lr_event) {
3486 struct lfsck_start *start = &lfsck_env_info(env)->lti_start;
3487 struct lfsck_start_param lsp;
3489 memset(start, 0, sizeof(*start));
3490 start->ls_valid = lr->lr_valid;
3491 start->ls_speed_limit = lr->lr_speed;
3492 start->ls_version = lr->lr_version;
3493 start->ls_active = lr->lr_active;
3494 start->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3495 start->ls_async_windows = lr->lr_async_windows;
3497 lsp.lsp_start = start;
3498 lsp.lsp_index = lr->lr_index;
3499 lsp.lsp_index_valid = 1;
3500 rc = lfsck_start(env, key, &lsp);
3504 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
3506 memset(stop, 0, sizeof(*stop));
3507 stop->ls_status = lr->lr_status;
3508 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
3509 rc = lfsck_stop(env, key, stop);
3512 case LE_PHASE1_DONE:
3513 case LE_PHASE2_DONE:
3515 case LE_CONDITIONAL_DESTROY:
3516 case LE_SET_LMV_MASTER:
3517 case LE_SET_LMV_SLAVE:
3518 case LE_PAIRS_VERIFY: {
3519 struct lfsck_instance *lfsck;
3520 struct lfsck_component *com;
3522 lfsck = lfsck_instance_find(key, true, false);
3523 if (unlikely(lfsck == NULL))
3526 com = lfsck_component_find(lfsck, lr->lr_active);
3528 rc = com->lc_ops->lfsck_in_notify(env, com, lr);
3529 lfsck_component_put(env, com);
3532 lfsck_instance_put(env, lfsck);
3541 EXPORT_SYMBOL(lfsck_in_notify);
3543 int lfsck_query(const struct lu_env *env, struct dt_device *key,
3544 struct lfsck_request *req, struct lfsck_reply *rep,
3545 struct lfsck_query *que)
3547 struct lfsck_instance *lfsck;
3548 struct lfsck_component *com;
3554 lfsck = lfsck_instance_find(key, true, false);
3555 if (unlikely(lfsck == NULL))
3559 if (que->lu_types == LFSCK_TYPES_ALL)
3561 LFSCK_TYPES_SUPPORTED & ~LFSCK_TYPE_SCRUB;
3563 if (que->lu_types & ~LFSCK_TYPES_SUPPORTED) {
3564 que->lu_types &= ~LFSCK_TYPES_SUPPORTED;
3566 GOTO(out, rc = -ENOTSUPP);
3569 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3570 i++, type = 1 << i) {
3571 if (!(que->lu_types & type))
3575 com = lfsck_component_find(lfsck, type);
3576 if (unlikely(com == NULL))
3577 GOTO(out, rc = -ENOTSUPP);
3579 memset(que->lu_mdts_count[i], 0,
3580 sizeof(__u32) * (LS_MAX + 1));
3581 memset(que->lu_osts_count[i], 0,
3582 sizeof(__u32) * (LS_MAX + 1));
3583 que->lu_repaired[i] = 0;
3584 rc = com->lc_ops->lfsck_query(env, com, req, rep,
3586 lfsck_component_put(env, com);
3591 if (!(que->lu_flags & LPF_WAIT))
3594 for (i = 0, type = 1 << i; i < LFSCK_TYPE_BITS;
3595 i++, type = 1 << i) {
3596 if (!(que->lu_types & type))
3599 if (que->lu_mdts_count[i][LS_SCANNING_PHASE1] != 0 ||
3600 que->lu_mdts_count[i][LS_SCANNING_PHASE2] != 0 ||
3601 que->lu_osts_count[i][LS_SCANNING_PHASE1] != 0 ||
3602 que->lu_osts_count[i][LS_SCANNING_PHASE2] != 0) {
3603 struct l_wait_info lwi;
3605 /* If it is required to wait, then sleep
3606 * 3 seconds and try to query again. */
3607 lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(3),
3611 rc = l_wait_event(lfsck->li_thread.t_ctl_waitq,
3613 if (rc == -ETIMEDOUT)
3618 com = lfsck_component_find(lfsck, req->lr_active);
3619 if (likely(com != NULL)) {
3620 rc = com->lc_ops->lfsck_query(env, com, req, rep,
3622 lfsck_component_put(env, com);
3631 lfsck_instance_put(env, lfsck);
3634 EXPORT_SYMBOL(lfsck_query);
3636 int lfsck_register_namespace(const struct lu_env *env, struct dt_device *key,
3637 struct ldlm_namespace *ns)
3639 struct lfsck_instance *lfsck;
3642 lfsck = lfsck_instance_find(key, true, false);
3643 if (likely(lfsck != NULL)) {
3644 lfsck->li_namespace = ns;
3645 lfsck_instance_put(env, lfsck);
3651 EXPORT_SYMBOL(lfsck_register_namespace);
3653 int lfsck_register(const struct lu_env *env, struct dt_device *key,
3654 struct dt_device *next, struct obd_device *obd,
3655 lfsck_out_notify notify, void *notify_data, bool master)
3657 struct lfsck_instance *lfsck;
3658 struct dt_object *root = NULL;
3659 struct dt_object *obj = NULL;
3660 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
3664 lfsck = lfsck_instance_find(key, false, false);
3665 if (unlikely(lfsck != NULL))
3668 OBD_ALLOC_PTR(lfsck);
3672 mutex_init(&lfsck->li_mutex);
3673 spin_lock_init(&lfsck->li_lock);
3674 INIT_LIST_HEAD(&lfsck->li_link);
3675 INIT_LIST_HEAD(&lfsck->li_list_scan);
3676 INIT_LIST_HEAD(&lfsck->li_list_dir);
3677 INIT_LIST_HEAD(&lfsck->li_list_double_scan);
3678 INIT_LIST_HEAD(&lfsck->li_list_idle);
3679 INIT_LIST_HEAD(&lfsck->li_list_lmv);
3680 atomic_set(&lfsck->li_ref, 1);
3681 atomic_set(&lfsck->li_double_scan_count, 0);
3682 init_waitqueue_head(&lfsck->li_thread.t_ctl_waitq);
3683 lfsck->li_out_notify = notify;
3684 lfsck->li_out_notify_data = notify_data;
3685 lfsck->li_next = next;
3686 lfsck->li_bottom = key;
3687 lfsck->li_obd = obd;
3689 rc = lfsck_tgt_descs_init(&lfsck->li_ost_descs);
3693 rc = lfsck_tgt_descs_init(&lfsck->li_mdt_descs);
3697 fid->f_seq = FID_SEQ_LOCAL_NAME;
3700 rc = local_oid_storage_init(env, key, fid, &lfsck->li_los);
3704 rc = dt_root_get(env, key, fid);
3708 root = dt_locate(env, key, fid);
3710 GOTO(out, rc = PTR_ERR(root));
3712 if (unlikely(!dt_try_as_dir(env, root)))
3713 GOTO(out, rc = -ENOTDIR);
3715 lfsck->li_local_root_fid = *fid;
3717 lfsck->li_master = 1;
3718 if (lfsck_dev_idx(lfsck) == 0) {
3719 struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid2;
3720 const struct lu_name *cname;
3722 rc = dt_lookup(env, root,
3723 (struct dt_rec *)(&lfsck->li_global_root_fid),
3724 (const struct dt_key *)"ROOT");
3728 obj = dt_locate(env, key, &lfsck->li_global_root_fid);
3730 GOTO(out, rc = PTR_ERR(obj));
3732 if (unlikely(!dt_try_as_dir(env, obj)))
3733 GOTO(out, rc = -ENOTDIR);
3735 rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3736 (const struct dt_key *)dotlustre);
3740 lfsck_object_put(env, obj);
3741 obj = dt_locate(env, key, fid);
3743 GOTO(out, rc = PTR_ERR(obj));
3745 cname = lfsck_name_get_const(env, dotlustre,
3747 rc = lfsck_verify_linkea(env, obj, cname,
3748 &lfsck->li_global_root_fid);
3752 if (unlikely(!dt_try_as_dir(env, obj)))
3753 GOTO(out, rc = -ENOTDIR);
3756 rc = dt_lookup(env, obj, (struct dt_rec *)fid,
3757 (const struct dt_key *)lostfound);
3761 lfsck_object_put(env, obj);
3762 obj = dt_locate(env, key, fid);
3764 GOTO(out, rc = PTR_ERR(obj));
3766 cname = lfsck_name_get_const(env, lostfound,
3768 rc = lfsck_verify_linkea(env, obj, cname, pfid);
3772 lfsck_object_put(env, obj);
3777 fid->f_seq = FID_SEQ_LOCAL_FILE;
3778 fid->f_oid = OTABLE_IT_OID;
3780 obj = dt_locate(env, key, fid);
3782 GOTO(out, rc = PTR_ERR(obj));
3784 rc = obj->do_ops->do_index_try(env, obj, &dt_otable_features);
3788 lfsck->li_obj_oit = obj;
3789 obj = local_file_find_or_create(env, lfsck->li_los, root, LFSCK_DIR,
3790 S_IFDIR | S_IRUGO | S_IWUSR);
3792 GOTO(out, rc = PTR_ERR(obj));
3794 lu_object_get(&obj->do_lu);
3795 lfsck->li_lfsck_dir = obj;
3796 rc = lfsck_bookmark_setup(env, lfsck);
3801 rc = lfsck_fid_init(lfsck);
3805 rc = lfsck_namespace_setup(env, lfsck);
3810 rc = lfsck_layout_setup(env, lfsck);
3814 /* XXX: more LFSCK components initialization to be added here. */
3816 rc = lfsck_instance_add(lfsck);
3818 rc = lfsck_add_target_from_orphan(env, lfsck);
3820 if (obj != NULL && !IS_ERR(obj))
3821 lfsck_object_put(env, obj);
3822 if (root != NULL && !IS_ERR(root))
3823 lfsck_object_put(env, root);
3825 lfsck_instance_cleanup(env, lfsck);
3828 EXPORT_SYMBOL(lfsck_register);
3830 void lfsck_degister(const struct lu_env *env, struct dt_device *key)
3832 struct lfsck_instance *lfsck;
3834 lfsck = lfsck_instance_find(key, false, true);
3836 lfsck_instance_put(env, lfsck);
3838 EXPORT_SYMBOL(lfsck_degister);
3840 int lfsck_add_target(const struct lu_env *env, struct dt_device *key,
3841 struct dt_device *tgt, struct obd_export *exp,
3842 __u32 index, bool for_ost)
3844 struct lfsck_instance *lfsck;
3845 struct lfsck_tgt_desc *ltd;
3856 INIT_LIST_HEAD(<d->ltd_orphan_list);
3857 INIT_LIST_HEAD(<d->ltd_layout_list);
3858 INIT_LIST_HEAD(<d->ltd_layout_phase_list);
3859 INIT_LIST_HEAD(<d->ltd_namespace_list);
3860 INIT_LIST_HEAD(<d->ltd_namespace_phase_list);
3861 atomic_set(<d->ltd_ref, 1);
3862 ltd->ltd_index = index;
3864 spin_lock(&lfsck_instance_lock);
3865 lfsck = __lfsck_instance_find(key, true, false);
3866 if (lfsck == NULL) {
3868 list_add_tail(<d->ltd_orphan_list,
3869 &lfsck_ost_orphan_list);
3871 list_add_tail(<d->ltd_orphan_list,
3872 &lfsck_mdt_orphan_list);
3873 spin_unlock(&lfsck_instance_lock);
3877 spin_unlock(&lfsck_instance_lock);
3879 rc = __lfsck_add_target(env, lfsck, ltd, for_ost, false);
3883 lfsck_instance_put(env, lfsck);
3887 EXPORT_SYMBOL(lfsck_add_target);
3889 void lfsck_del_target(const struct lu_env *env, struct dt_device *key,
3890 struct dt_device *tgt, __u32 index, bool for_ost)
3892 struct lfsck_instance *lfsck;
3893 struct lfsck_tgt_descs *ltds;
3894 struct lfsck_tgt_desc *ltd;
3895 struct list_head *head;
3898 head = &lfsck_ost_orphan_list;
3900 head = &lfsck_mdt_orphan_list;
3902 spin_lock(&lfsck_instance_lock);
3903 list_for_each_entry(ltd, head, ltd_orphan_list) {
3904 if (ltd->ltd_tgt == tgt) {
3905 list_del_init(<d->ltd_orphan_list);
3906 spin_unlock(&lfsck_instance_lock);
3914 lfsck = __lfsck_instance_find(key, true, false);
3915 spin_unlock(&lfsck_instance_lock);
3916 if (unlikely(lfsck == NULL))
3920 ltds = &lfsck->li_ost_descs;
3922 ltds = &lfsck->li_mdt_descs;
3924 down_write(<ds->ltd_rw_sem);
3925 LASSERT(ltds->ltd_tgts_bitmap != NULL);
3927 if (unlikely(index >= ltds->ltd_tgts_bitmap->size))
3930 ltd = lfsck_ltd2tgt(ltds, index);
3931 if (unlikely(ltd == NULL))
3934 LASSERT(ltds->ltd_tgtnr > 0);
3937 cfs_bitmap_clear(ltds->ltd_tgts_bitmap, index);
3938 lfsck_assign_tgt(ltds, NULL, index);
3943 head = &lfsck->li_ost_descs.ltd_orphan;
3945 head = &lfsck->li_mdt_descs.ltd_orphan;
3947 list_for_each_entry(ltd, head, ltd_orphan_list) {
3948 if (ltd->ltd_tgt == tgt) {
3949 list_del_init(<d->ltd_orphan_list);
3955 up_write(<ds->ltd_rw_sem);
3957 spin_lock(<ds->ltd_lock);
3959 spin_unlock(<ds->ltd_lock);
3960 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_NAMESPACE);
3961 lfsck_stop_notify(env, lfsck, ltds, ltd, LFSCK_TYPE_LAYOUT);
3965 lfsck_instance_put(env, lfsck);
3967 EXPORT_SYMBOL(lfsck_del_target);
3969 static int __init lfsck_init(void)
3973 INIT_LIST_HEAD(&lfsck_instance_list);
3974 INIT_LIST_HEAD(&lfsck_ost_orphan_list);
3975 INIT_LIST_HEAD(&lfsck_mdt_orphan_list);
3976 lfsck_key_init_generic(&lfsck_thread_key, NULL);
3977 rc = lu_context_key_register(&lfsck_thread_key);
3979 tgt_register_lfsck_in_notify_local(lfsck_in_notify_local);
3980 tgt_register_lfsck_in_notify(lfsck_in_notify);
3981 tgt_register_lfsck_query(lfsck_query);
3987 static void __exit lfsck_exit(void)
3989 struct lfsck_tgt_desc *ltd;
3990 struct lfsck_tgt_desc *next;
3992 LASSERT(list_empty(&lfsck_instance_list));
3994 list_for_each_entry_safe(ltd, next, &lfsck_ost_orphan_list,
3996 list_del_init(<d->ltd_orphan_list);
4000 list_for_each_entry_safe(ltd, next, &lfsck_mdt_orphan_list,
4002 list_del_init(<d->ltd_orphan_list);
4006 lu_context_key_degister(&lfsck_thread_key);
4009 MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
4010 MODULE_DESCRIPTION("Lustre File System Checker");
4011 MODULE_VERSION(LUSTRE_VERSION_STRING);
4012 MODULE_LICENSE("GPL");
4014 module_init(lfsck_init);
4015 module_exit(lfsck_exit);