Whamcloud - gitweb
LU-9679 lustre: use LIST_HEAD() for local lists.
[fs/lustre-release.git] / lustre / lfsck / lfsck_lib.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9
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.
15
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
19  *
20  * GPL HEADER END
21  */
22 /*
23  * Copyright (c) 2013, 2017, Intel Corporation.
24  */
25 /*
26  * lustre/lfsck/lfsck_lib.c
27  *
28  * Author: Fan, Yong <fan.yong@intel.com>
29  */
30
31 #define DEBUG_SUBSYSTEM S_LFSCK
32
33 #include <linux/kthread.h>
34 #include <linux/sched.h>
35 #include <linux/list.h>
36 #include <linux/delay.h>
37 #include <lu_object.h>
38 #include <dt_object.h>
39 #include <md_object.h>
40 #include <lustre_fld.h>
41 #include <lustre_lib.h>
42 #include <lustre_net.h>
43 #include <lustre_lfsck.h>
44
45 #include "lfsck_internal.h"
46
47 #define LFSCK_CHECKPOINT_SKIP   1
48
49 /* define lfsck thread key */
50 LU_KEY_INIT(lfsck, struct lfsck_thread_info);
51
52 static void lfsck_key_fini(const struct lu_context *ctx,
53                            struct lu_context_key *key, void *data)
54 {
55         struct lfsck_thread_info *info = data;
56
57         lu_buf_free(&info->lti_linkea_buf);
58         lu_buf_free(&info->lti_linkea_buf2);
59         lu_buf_free(&info->lti_big_buf);
60         OBD_FREE_PTR(info);
61 }
62
63 LU_CONTEXT_KEY_DEFINE(lfsck, LCT_MD_THREAD | LCT_DT_THREAD);
64 LU_KEY_INIT_GENERIC(lfsck);
65
66 static LIST_HEAD(lfsck_instance_list);
67 static LIST_HEAD(lfsck_ost_orphan_list);
68 static LIST_HEAD(lfsck_mdt_orphan_list);
69 static DEFINE_SPINLOCK(lfsck_instance_lock);
70
71 const char *lfsck_flags_names[] = {
72         "scanned-once",
73         "inconsistent",
74         "upgrade",
75         "incomplete",
76         "crashed_lastid",
77         NULL
78 };
79
80 const char *lfsck_param_names[] = {
81         NULL,
82         "failout",
83         "dryrun",
84         "all_targets",
85         "broadcast",
86         "orphan",
87         "create_ostobj",
88         "create_mdtobj",
89         NULL,
90         "delay_create_ostobj",
91         NULL
92 };
93
94 enum lfsck_verify_lpf_types {
95         LVLT_BY_BOOKMARK        = 0,
96         LVLT_BY_NAMEENTRY       = 1,
97 };
98
99 static inline void
100 lfsck_reset_ltd_status(struct lfsck_tgt_desc *ltd, enum lfsck_type type)
101 {
102         if (type == LFSCK_TYPE_LAYOUT) {
103                 ltd->ltd_layout_status = LS_MAX;
104                 ltd->ltd_layout_repaired = 0;
105         } else {
106                 ltd->ltd_namespace_status = LS_MAX;
107                 ltd->ltd_namespace_repaired = 0;
108         }
109 }
110
111 static int lfsck_tgt_descs_init(struct lfsck_tgt_descs *ltds)
112 {
113         spin_lock_init(&ltds->ltd_lock);
114         init_rwsem(&ltds->ltd_rw_sem);
115         INIT_LIST_HEAD(&ltds->ltd_orphan);
116         ltds->ltd_tgts_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
117         if (ltds->ltd_tgts_bitmap == NULL)
118                 return -ENOMEM;
119
120         return 0;
121 }
122
123 static void lfsck_tgt_descs_fini(struct lfsck_tgt_descs *ltds)
124 {
125         struct lfsck_tgt_desc   *ltd;
126         struct lfsck_tgt_desc   *next;
127         int                      idx;
128
129         down_write(&ltds->ltd_rw_sem);
130
131         list_for_each_entry_safe(ltd, next, &ltds->ltd_orphan,
132                                  ltd_orphan_list) {
133                 list_del_init(&ltd->ltd_orphan_list);
134                 lfsck_tgt_put(ltd);
135         }
136
137         if (unlikely(ltds->ltd_tgts_bitmap == NULL)) {
138                 up_write(&ltds->ltd_rw_sem);
139
140                 return;
141         }
142
143         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
144                 ltd = lfsck_ltd2tgt(ltds, idx);
145                 if (likely(ltd != NULL)) {
146                         LASSERT(list_empty(&ltd->ltd_layout_list));
147                         LASSERT(list_empty(&ltd->ltd_layout_phase_list));
148                         LASSERT(list_empty(&ltd->ltd_namespace_list));
149                         LASSERT(list_empty(&ltd->ltd_namespace_phase_list));
150
151                         ltds->ltd_tgtnr--;
152                         cfs_bitmap_clear(ltds->ltd_tgts_bitmap, idx);
153                         lfsck_assign_tgt(ltds, NULL, idx);
154                         lfsck_tgt_put(ltd);
155                 }
156         }
157
158         LASSERTF(ltds->ltd_tgtnr == 0, "tgt count unmatched: %d\n",
159                  ltds->ltd_tgtnr);
160
161         for (idx = 0; idx < TGT_PTRS; idx++) {
162                 if (ltds->ltd_tgts_idx[idx] != NULL) {
163                         OBD_FREE_PTR(ltds->ltd_tgts_idx[idx]);
164                         ltds->ltd_tgts_idx[idx] = NULL;
165                 }
166         }
167
168         CFS_FREE_BITMAP(ltds->ltd_tgts_bitmap);
169         ltds->ltd_tgts_bitmap = NULL;
170         up_write(&ltds->ltd_rw_sem);
171 }
172
173 static int __lfsck_add_target(const struct lu_env *env,
174                               struct lfsck_instance *lfsck,
175                               struct lfsck_tgt_desc *ltd,
176                               bool for_ost, bool locked)
177 {
178         struct lfsck_tgt_descs *ltds;
179         __u32                   index = ltd->ltd_index;
180         int                     rc    = 0;
181         ENTRY;
182
183         if (for_ost)
184                 ltds = &lfsck->li_ost_descs;
185         else
186                 ltds = &lfsck->li_mdt_descs;
187
188         if (!locked)
189                 down_write(&ltds->ltd_rw_sem);
190
191         LASSERT(ltds->ltd_tgts_bitmap != NULL);
192
193         if (index >= ltds->ltd_tgts_bitmap->size) {
194                 __u32 newsize = max((__u32)ltds->ltd_tgts_bitmap->size,
195                                     (__u32)BITS_PER_LONG);
196                 struct cfs_bitmap *old_bitmap = ltds->ltd_tgts_bitmap;
197                 struct cfs_bitmap *new_bitmap;
198
199                 while (newsize < index + 1)
200                         newsize <<= 1;
201
202                 new_bitmap = CFS_ALLOCATE_BITMAP(newsize);
203                 if (new_bitmap == NULL)
204                         GOTO(unlock, rc = -ENOMEM);
205
206                 if (ltds->ltd_tgtnr > 0)
207                         cfs_bitmap_copy(new_bitmap, old_bitmap);
208                 ltds->ltd_tgts_bitmap = new_bitmap;
209                 CFS_FREE_BITMAP(old_bitmap);
210         }
211
212         if (cfs_bitmap_check(ltds->ltd_tgts_bitmap, index)) {
213                 CERROR("%s: the device %s (%u) is registered already\n",
214                        lfsck_lfsck2name(lfsck),
215                        ltd->ltd_tgt->dd_lu_dev.ld_obd->obd_name, index);
216                 GOTO(unlock, rc = -EEXIST);
217         }
218
219         if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL) {
220                 OBD_ALLOC_PTR(ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK]);
221                 if (ltds->ltd_tgts_idx[index / TGT_PTRS_PER_BLOCK] == NULL)
222                         GOTO(unlock, rc = -ENOMEM);
223         }
224
225         lfsck_assign_tgt(ltds, ltd, index);
226         cfs_bitmap_set(ltds->ltd_tgts_bitmap, index);
227         ltds->ltd_tgtnr++;
228
229         GOTO(unlock, rc = 0);
230
231 unlock:
232         if (!locked)
233                 up_write(&ltds->ltd_rw_sem);
234
235         return rc;
236 }
237
238 static int lfsck_add_target_from_orphan(const struct lu_env *env,
239                                         struct lfsck_instance *lfsck)
240 {
241         struct lfsck_tgt_descs  *ltds    = &lfsck->li_ost_descs;
242         struct lfsck_tgt_desc   *ltd;
243         struct lfsck_tgt_desc   *next;
244         struct list_head        *head    = &lfsck_ost_orphan_list;
245         int                      rc;
246         bool                     for_ost = true;
247
248 again:
249         spin_lock(&lfsck_instance_lock);
250         list_for_each_entry_safe(ltd, next, head, ltd_orphan_list) {
251                 if (ltd->ltd_key == lfsck->li_bottom)
252                         list_move_tail(&ltd->ltd_orphan_list,
253                                        &ltds->ltd_orphan);
254         }
255         spin_unlock(&lfsck_instance_lock);
256
257         down_write(&ltds->ltd_rw_sem);
258         while (!list_empty(&ltds->ltd_orphan)) {
259                 ltd = list_entry(ltds->ltd_orphan.next,
260                                  struct lfsck_tgt_desc,
261                                  ltd_orphan_list);
262                 list_del_init(&ltd->ltd_orphan_list);
263                 rc = __lfsck_add_target(env, lfsck, ltd, for_ost, true);
264                 /* Do not hold the semaphore for too long time. */
265                 up_write(&ltds->ltd_rw_sem);
266                 if (rc != 0)
267                         return rc;
268
269                 down_write(&ltds->ltd_rw_sem);
270         }
271         up_write(&ltds->ltd_rw_sem);
272
273         if (for_ost) {
274                 ltds = &lfsck->li_mdt_descs;
275                 head = &lfsck_mdt_orphan_list;
276                 for_ost = false;
277                 goto again;
278         }
279
280         return 0;
281 }
282
283 static inline struct lfsck_component *
284 __lfsck_component_find(struct lfsck_instance *lfsck, __u16 type,
285                        struct list_head *list)
286 {
287         struct lfsck_component *com;
288
289         list_for_each_entry(com, list, lc_link) {
290                 if (com->lc_type == type)
291                         return com;
292         }
293         return NULL;
294 }
295
296 struct lfsck_component *
297 lfsck_component_find(struct lfsck_instance *lfsck, __u16 type)
298 {
299         struct lfsck_component *com;
300
301         spin_lock(&lfsck->li_lock);
302         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
303         if (com != NULL)
304                 goto unlock;
305
306         com = __lfsck_component_find(lfsck, type,
307                                      &lfsck->li_list_double_scan);
308         if (com != NULL)
309                 goto unlock;
310
311         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_idle);
312
313 unlock:
314         if (com != NULL)
315                 lfsck_component_get(com);
316         spin_unlock(&lfsck->li_lock);
317         return com;
318 }
319
320 void lfsck_component_cleanup(const struct lu_env *env,
321                              struct lfsck_component *com)
322 {
323         if (!list_empty(&com->lc_link))
324                 list_del_init(&com->lc_link);
325         if (!list_empty(&com->lc_link_dir))
326                 list_del_init(&com->lc_link_dir);
327
328         lfsck_component_put(env, com);
329 }
330
331 int lfsck_fid_alloc(const struct lu_env *env, struct lfsck_instance *lfsck,
332                     struct lu_fid *fid, bool locked)
333 {
334         struct lfsck_bookmark   *bk = &lfsck->li_bookmark_ram;
335         int                      rc = 0;
336         ENTRY;
337
338         if (!locked)
339                 mutex_lock(&lfsck->li_mutex);
340
341         rc = seq_client_alloc_fid(env, lfsck->li_seq, fid);
342         if (rc >= 0) {
343                 bk->lb_last_fid = *fid;
344                 /* We do not care about whether the subsequent sub-operations
345                  * failed or not. The worst case is that one FID is lost that
346                  * is not a big issue for the LFSCK since it is relative rare
347                  * for LFSCK create. */
348                 rc = lfsck_bookmark_store(env, lfsck);
349         }
350
351         if (!locked)
352                 mutex_unlock(&lfsck->li_mutex);
353
354         RETURN(rc);
355 }
356
357 static int __lfsck_ibits_lock(const struct lu_env *env,
358                               struct lfsck_instance *lfsck,
359                               struct dt_object *obj, struct ldlm_res_id *resid,
360                               struct lustre_handle *lh, __u64 bits,
361                               enum ldlm_mode mode)
362 {
363         struct lfsck_thread_info        *info   = lfsck_env_info(env);
364         union ldlm_policy_data          *policy = &info->lti_policy;
365         __u64                            flags  = LDLM_FL_ATOMIC_CB;
366         int                              rc;
367
368         LASSERT(lfsck->li_namespace != NULL);
369
370         memset(policy, 0, sizeof(*policy));
371         policy->l_inodebits.bits = bits;
372         if (dt_object_remote(obj)) {
373                 struct ldlm_enqueue_info *einfo = &info->lti_einfo;
374
375                 memset(einfo, 0, sizeof(*einfo));
376                 einfo->ei_type = LDLM_IBITS;
377                 einfo->ei_mode = mode;
378                 einfo->ei_cb_bl = ldlm_blocking_ast;
379                 einfo->ei_cb_cp = ldlm_completion_ast;
380                 einfo->ei_res_id = resid;
381
382                 rc = dt_object_lock(env, obj, lh, einfo, policy);
383                 /* for regular checks LFSCK doesn't use LDLM locking,
384                  * so the state isn't coherent. here we just took LDLM
385                  * lock for coherency and it's time to invalidate
386                  * previous state */
387                 if (rc == ELDLM_OK)
388                         dt_invalidate(env, obj);
389         } else {
390                 rc = ldlm_cli_enqueue_local(env, lfsck->li_namespace, resid,
391                                             LDLM_IBITS, policy, mode,
392                                             &flags, ldlm_blocking_ast,
393                                             ldlm_completion_ast, NULL, NULL,
394                                             0, LVB_T_NONE, NULL, lh);
395         }
396
397         if (rc == ELDLM_OK) {
398                 rc = 0;
399         } else {
400                 memset(lh, 0, sizeof(*lh));
401                 rc = -EIO;
402         }
403
404         return rc;
405 }
406
407 /**
408  * Request the specified ibits lock for the given object.
409  *
410  * Before the LFSCK modifying on the namespace visible object,
411  * it needs to acquire related ibits ldlm lock.
412  *
413  * \param[in] env       pointer to the thread context
414  * \param[in] lfsck     pointer to the lfsck instance
415  * \param[in] obj       pointer to the dt_object to be locked
416  * \param[out] lh       pointer to the lock handle
417  * \param[in] bits      the bits for the ldlm lock to be acquired
418  * \param[in] mode      the mode for the ldlm lock to be acquired
419  *
420  * \retval              0 for success
421  * \retval              negative error number on failure
422  */
423 int lfsck_ibits_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
424                      struct dt_object *obj, struct lustre_handle *lh,
425                      __u64 bits, enum ldlm_mode mode)
426 {
427         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
428
429         LASSERT(!lustre_handle_is_used(lh));
430
431         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
432         return __lfsck_ibits_lock(env, lfsck, obj, resid, lh, bits, mode);
433 }
434
435 /**
436  * Request the remote LOOKUP lock for the given object.
437  *
438  * If \a pobj is remote, the LOOKUP lock of \a obj is on the MDT where
439  * \a pobj is, acquire LOOKUP lock there.
440  *
441  * \param[in] env       pointer to the thread context
442  * \param[in] lfsck     pointer to the lfsck instance
443  * \param[in] pobj      pointer to parent dt_object
444  * \param[in] obj       pointer to the dt_object to be locked
445  * \param[out] lh       pointer to the lock handle
446  * \param[in] mode      the mode for the ldlm lock to be acquired
447  *
448  * \retval              0 for success
449  * \retval              negative error number on failure
450  */
451 int lfsck_remote_lookup_lock(const struct lu_env *env,
452                              struct lfsck_instance *lfsck,
453                              struct dt_object *pobj, struct dt_object *obj,
454                              struct lustre_handle *lh, enum ldlm_mode mode)
455 {
456         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
457
458         LASSERT(!lustre_handle_is_used(lh));
459
460         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
461         return __lfsck_ibits_lock(env, lfsck, pobj, resid, lh,
462                                   MDS_INODELOCK_LOOKUP, mode);
463 }
464
465 /**
466  * Release the the specified ibits lock.
467  *
468  * If the lock has been acquired before, release it
469  * and cleanup the handle. Otherwise, do nothing.
470  *
471  * \param[in] lh        pointer to the lock handle
472  * \param[in] mode      the mode for the ldlm lock to be released
473  */
474 void lfsck_ibits_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
475 {
476         if (lustre_handle_is_used(lh)) {
477                 ldlm_lock_decref(lh, mode);
478                 memset(lh, 0, sizeof(*lh));
479         }
480 }
481
482 /**
483  * Request compound ibits locks for the given <obj, name> pairs.
484  *
485  * Before the LFSCK modifying on the namespace visible object, it needs to
486  * acquire related ibits ldlm lock. Usually, we can use lfsck_ibits_lock for
487  * the lock purpose. But the simple lfsck_ibits_lock for directory-based
488  * modificationis (such as insert name entry to the directory) may be too
489  * coarse-grained and not efficient.
490  *
491  * The lfsck_lock() will request compound ibits locks on the specified
492  * <obj, name> pairs: the PDO (Parallel Directory Operations) ibits (UPDATE)
493  * lock on the directory object, and the regular ibits lock on the name hash.
494  *
495  * \param[in] env       pointer to the thread context
496  * \param[in] lfsck     pointer to the lfsck instance
497  * \param[in] obj       pointer to the dt_object to be locked
498  * \param[in] name      used for building the PDO lock resource
499  * \param[out] llh      pointer to the lfsck_lock_handle
500  * \param[in] bits      the bits for the ldlm lock to be acquired
501  * \param[in] mode      the mode for the ldlm lock to be acquired
502  *
503  * \retval              0 for success
504  * \retval              negative error number on failure
505  */
506 int lfsck_lock(const struct lu_env *env, struct lfsck_instance *lfsck,
507                struct dt_object *obj, const char *name,
508                struct lfsck_lock_handle *llh, __u64 bits, enum ldlm_mode mode)
509 {
510         struct ldlm_res_id *resid = &lfsck_env_info(env)->lti_resid;
511         int                 rc;
512
513         LASSERT(S_ISDIR(lfsck_object_type(obj)));
514         LASSERT(name != NULL);
515         LASSERT(name[0] != 0);
516         LASSERT(!lustre_handle_is_used(&llh->llh_pdo_lh));
517         LASSERT(!lustre_handle_is_used(&llh->llh_reg_lh));
518
519         switch (mode) {
520         case LCK_EX:
521                 llh->llh_pdo_mode = LCK_EX;
522                 break;
523         case LCK_PW:
524                 llh->llh_pdo_mode = LCK_CW;
525                 break;
526         case LCK_PR:
527                 llh->llh_pdo_mode = LCK_CR;
528                 break;
529         default:
530                 CDEBUG(D_LFSCK, "%s: unexpected PDO lock mode %u on the obj "
531                        DFID"\n", lfsck_lfsck2name(lfsck), mode,
532                        PFID(lfsck_dto2fid(obj)));
533                 LBUG();
534         }
535
536         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
537         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_pdo_lh,
538                                 MDS_INODELOCK_UPDATE, llh->llh_pdo_mode);
539         if (rc != 0)
540                 return rc;
541
542         llh->llh_reg_mode = mode;
543         resid->name[LUSTRE_RES_ID_HSH_OFF] = ll_full_name_hash(NULL, name,
544                                                                strlen(name));
545         LASSERT(resid->name[LUSTRE_RES_ID_HSH_OFF] != 0);
546         rc = __lfsck_ibits_lock(env, lfsck, obj, resid, &llh->llh_reg_lh,
547                                 bits, llh->llh_reg_mode);
548         if (rc != 0)
549                 lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
550
551         return rc;
552 }
553
554 /**
555  * Release the the compound ibits locks.
556  *
557  * \param[in] llh       pointer to the lfsck_lock_handle to be released
558  */
559 void lfsck_unlock(struct lfsck_lock_handle *llh)
560 {
561         lfsck_ibits_unlock(&llh->llh_reg_lh, llh->llh_reg_mode);
562         lfsck_ibits_unlock(&llh->llh_pdo_lh, llh->llh_pdo_mode);
563 }
564
565 int lfsck_find_mdt_idx_by_fid(const struct lu_env *env,
566                               struct lfsck_instance *lfsck,
567                               const struct lu_fid *fid)
568 {
569         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
570         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
571         int                      rc;
572
573         if (unlikely(fid_seq(fid) == FID_SEQ_LOCAL_FILE)) {
574                 /* "ROOT" is always on the MDT0. */
575                 if (lu_fid_eq(fid, &lfsck->li_global_root_fid))
576                         return 0;
577
578                 return lfsck_dev_idx(lfsck);
579         }
580
581         fld_range_set_mdt(range);
582         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
583         if (rc == 0)
584                 rc = range->lsr_index;
585
586         return rc;
587 }
588
589 const char dot[] = ".";
590 const char dotdot[] = "..";
591 static const char dotlustre[] = ".lustre";
592 static const char lostfound[] = "lost+found";
593
594 /**
595  * Remove the name entry from the .lustre/lost+found directory.
596  *
597  * No need to care about the object referenced by the name entry,
598  * either the name entry is invalid or redundant, or the referenced
599  * object has been processed or will be handled by others.
600  *
601  * \param[in] env       pointer to the thread context
602  * \param[in] lfsck     pointer to the lfsck instance
603  * \param[in] name      the name for the name entry to be removed
604  *
605  * \retval              0 for success
606  * \retval              negative error number on failure
607  */
608 static int lfsck_lpf_remove_name_entry(const struct lu_env *env,
609                                        struct lfsck_instance *lfsck,
610                                        const char *name)
611 {
612         struct dt_object        *parent = lfsck->li_lpf_root_obj;
613         struct dt_device        *dev    = lfsck_obj2dev(parent);
614         struct thandle          *th;
615         struct lfsck_lock_handle *llh   = &lfsck_env_info(env)->lti_llh;
616         int                      rc;
617         ENTRY;
618
619         rc = lfsck_lock(env, lfsck, parent, name, llh,
620                         MDS_INODELOCK_UPDATE, LCK_PW);
621         if (rc != 0)
622                 RETURN(rc);
623
624         th = dt_trans_create(env, dev);
625         if (IS_ERR(th))
626                 GOTO(unlock, rc = PTR_ERR(th));
627
628         rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
629         if (rc != 0)
630                 GOTO(stop, rc);
631
632         rc = dt_declare_ref_del(env, parent, th);
633         if (rc != 0)
634                 GOTO(stop, rc);
635
636         rc = dt_trans_start_local(env, dev, th);
637         if (rc != 0)
638                 GOTO(stop, rc);
639
640         rc = dt_delete(env, parent, (const struct dt_key *)name, th);
641         if (rc != 0)
642                 GOTO(stop, rc);
643
644         dt_write_lock(env, parent, 0);
645         rc = dt_ref_del(env, parent, th);
646         dt_write_unlock(env, parent);
647
648         GOTO(stop, rc);
649
650 stop:
651         dt_trans_stop(env, dev, th);
652
653 unlock:
654         lfsck_unlock(llh);
655
656         CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s: rc = %d\n",
657                lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(parent)), name, rc);
658
659         return rc;
660 }
661
662 static int lfsck_create_lpf_local(const struct lu_env *env,
663                                   struct lfsck_instance *lfsck,
664                                   struct dt_object *child,
665                                   struct lu_attr *la,
666                                   struct dt_object_format *dof,
667                                   const char *name)
668 {
669         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
670         struct dt_object        *parent = lfsck->li_lpf_root_obj;
671         struct dt_device        *dev    = lfsck_obj2dev(child);
672         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
673         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
674         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
675         struct thandle          *th     = NULL;
676         struct linkea_data       ldata  = { NULL };
677         struct lu_buf            linkea_buf;
678         const struct lu_name    *cname;
679         loff_t                   pos    = 0;
680         int                      len    = sizeof(struct lfsck_bookmark);
681         int                      rc;
682         ENTRY;
683
684         cname = lfsck_name_get_const(env, name, strlen(name));
685         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
686                               cname, lfsck_dto2fid(parent));
687         if (rc != 0)
688                 RETURN(rc);
689
690         th = dt_trans_create(env, dev);
691         if (IS_ERR(th))
692                 RETURN(PTR_ERR(th));
693
694         /* 1a. create child */
695         rc = dt_declare_create(env, child, la, NULL, dof, th);
696         if (rc != 0)
697                 GOTO(stop, rc);
698
699         if (!dt_try_as_dir(env, child))
700                 GOTO(stop, rc = -ENOTDIR);
701
702         /* 2a. increase child nlink */
703         rc = dt_declare_ref_add(env, child, th);
704         if (rc != 0)
705                 GOTO(stop, rc);
706
707         /* 3a. insert dot into child dir */
708         rec->rec_type = S_IFDIR;
709         rec->rec_fid = cfid;
710         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
711                                (const struct dt_key *)dot, th);
712         if (rc != 0)
713                 GOTO(stop, rc);
714
715         /* 4a. insert dotdot into child dir */
716         rec->rec_fid = &LU_LPF_FID;
717         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
718                                (const struct dt_key *)dotdot, th);
719         if (rc != 0)
720                 GOTO(stop, rc);
721
722         /* 5a. insert linkEA for child */
723         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
724                        ldata.ld_leh->leh_len);
725         rc = dt_declare_xattr_set(env, child, &linkea_buf,
726                                   XATTR_NAME_LINK, 0, th);
727         if (rc != 0)
728                 GOTO(stop, rc);
729
730         /* 6a. insert name into parent dir */
731         rec->rec_type = S_IFDIR;
732         rec->rec_fid = cfid;
733         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
734                                (const struct dt_key *)name, th);
735         if (rc != 0)
736                 GOTO(stop, rc);
737
738         /* 7a. increase parent nlink */
739         rc = dt_declare_ref_add(env, parent, th);
740         if (rc != 0)
741                 GOTO(stop, rc);
742
743         /* 8a. update bookmark */
744         rc = dt_declare_record_write(env, bk_obj,
745                                      lfsck_buf_get(env, bk, len), 0, th);
746         if (rc != 0)
747                 GOTO(stop, rc);
748
749         rc = dt_trans_start_local(env, dev, th);
750         if (rc != 0)
751                 GOTO(stop, rc);
752
753         dt_write_lock(env, child, 0);
754         /* 1b. create child */
755         rc = dt_create(env, child, la, NULL, dof, th);
756         if (rc != 0)
757                 GOTO(unlock, rc);
758
759         /* 2b. increase child nlink */
760         rc = dt_ref_add(env, child, th);
761         if (rc != 0)
762                 GOTO(unlock, rc);
763
764         /* 3b. insert dot into child dir */
765         rec->rec_fid = cfid;
766         rc = dt_insert(env, child, (const struct dt_rec *)rec,
767                        (const struct dt_key *)dot, th);
768         if (rc != 0)
769                 GOTO(unlock, rc);
770
771         /* 4b. insert dotdot into child dir */
772         rec->rec_fid = &LU_LPF_FID;
773         rc = dt_insert(env, child, (const struct dt_rec *)rec,
774                        (const struct dt_key *)dotdot, th);
775         if (rc != 0)
776                 GOTO(unlock, rc);
777
778         /* 5b. insert linkEA for child. */
779         rc = dt_xattr_set(env, child, &linkea_buf,
780                           XATTR_NAME_LINK, 0, th);
781         dt_write_unlock(env, child);
782         if (rc != 0)
783                 GOTO(stop, rc);
784
785         /* 6b. insert name into parent dir */
786         rec->rec_fid = cfid;
787         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
788                        (const struct dt_key *)name, th);
789         if (rc != 0)
790                 GOTO(stop, rc);
791
792         dt_write_lock(env, parent, 0);
793         /* 7b. increase parent nlink */
794         rc = dt_ref_add(env, parent, th);
795         dt_write_unlock(env, parent);
796         if (rc != 0)
797                 GOTO(stop, rc);
798
799         bk->lb_lpf_fid = *cfid;
800         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
801
802         /* 8b. update bookmark */
803         rc = dt_record_write(env, bk_obj,
804                              lfsck_buf_get(env, bk, len), &pos, th);
805
806         GOTO(stop, rc);
807
808 unlock:
809         dt_write_unlock(env, child);
810
811 stop:
812         dt_trans_stop(env, dev, th);
813
814         return rc;
815 }
816
817 static int lfsck_create_lpf_remote(const struct lu_env *env,
818                                    struct lfsck_instance *lfsck,
819                                    struct dt_object *child,
820                                    struct lu_attr *la,
821                                    struct dt_object_format *dof,
822                                    const char *name)
823 {
824         struct dt_insert_rec    *rec    = &lfsck_env_info(env)->lti_dt_rec;
825         struct dt_object        *parent = lfsck->li_lpf_root_obj;
826         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
827         struct dt_object        *bk_obj = lfsck->li_bookmark_obj;
828         const struct lu_fid     *cfid   = lfsck_dto2fid(child);
829         struct thandle          *th     = NULL;
830         struct linkea_data       ldata  = { NULL };
831         struct lu_buf            linkea_buf;
832         const struct lu_name    *cname;
833         struct dt_device        *dev;
834         loff_t                   pos    = 0;
835         int                      len    = sizeof(struct lfsck_bookmark);
836         int                      rc;
837         ENTRY;
838
839         cname = lfsck_name_get_const(env, name, strlen(name));
840         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf2,
841                               cname, lfsck_dto2fid(parent));
842         if (rc != 0)
843                 RETURN(rc);
844
845         /* Create .lustre/lost+found/MDTxxxx. */
846
847         /* XXX: Currently, cross-MDT create operation needs to create the child
848          *      object firstly, then insert name into the parent directory. For
849          *      this case, the child object resides on current MDT (local), but
850          *      the parent ".lustre/lost+found" may be on remote MDT. It is not
851          *      easy to contain all the sub-modifications orderly within single
852          *      transaction.
853          *
854          *      To avoid more inconsistency, we split the create operation into
855          *      two transactions:
856          *
857          *      1) create the child and update the lfsck_bookmark::lb_lpf_fid
858          *         locally.
859          *      2) insert the name "MDTXXXX" in the parent ".lustre/lost+found"
860          *         remotely.
861          *
862          *      If 1) done, but 2) failed, then go ahead, the LFSCK will try to
863          *      repair such inconsistency when LFSCK run next time. */
864
865         /* Transaction I: locally */
866
867         dev = lfsck_obj2dev(child);
868         th = dt_trans_create(env, dev);
869         if (IS_ERR(th))
870                 RETURN(PTR_ERR(th));
871
872         /* 1a. create child */
873         rc = dt_declare_create(env, child, la, NULL, dof, th);
874         if (rc != 0)
875                 GOTO(stop, rc);
876
877         if (!dt_try_as_dir(env, child))
878                 GOTO(stop, rc = -ENOTDIR);
879
880         /* 2a. increase child nlink */
881         rc = dt_declare_ref_add(env, child, th);
882         if (rc != 0)
883                 GOTO(stop, rc);
884
885         /* 3a. insert dot into child dir */
886         rec->rec_type = S_IFDIR;
887         rec->rec_fid = cfid;
888         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
889                                (const struct dt_key *)dot, th);
890         if (rc != 0)
891                 GOTO(stop, rc);
892
893         /* 4a. insert dotdot into child dir */
894         rec->rec_fid = &LU_LPF_FID;
895         rc = dt_declare_insert(env, child, (const struct dt_rec *)rec,
896                                (const struct dt_key *)dotdot, th);
897         if (rc != 0)
898                 GOTO(stop, rc);
899
900         /* 5a. insert linkEA for child */
901         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
902                        ldata.ld_leh->leh_len);
903         rc = dt_declare_xattr_set(env, child, &linkea_buf,
904                                   XATTR_NAME_LINK, 0, th);
905         if (rc != 0)
906                 GOTO(stop, rc);
907
908         /* 6a. update bookmark */
909         rc = dt_declare_record_write(env, bk_obj,
910                                      lfsck_buf_get(env, bk, len), 0, th);
911         if (rc != 0)
912                 GOTO(stop, rc);
913
914         rc = dt_trans_start_local(env, dev, th);
915         if (rc != 0)
916                 GOTO(stop, rc);
917
918         dt_write_lock(env, child, 0);
919         /* 1b. create child */
920         rc = dt_create(env, child, la, NULL, dof, th);
921         if (rc != 0)
922                 GOTO(unlock, rc);
923
924         /* 2b. increase child nlink */
925         rc = dt_ref_add(env, child, th);
926         if (rc != 0)
927                 GOTO(unlock, rc);
928
929         /* 3b. insert dot into child dir */
930         rec->rec_type = S_IFDIR;
931         rec->rec_fid = cfid;
932         rc = dt_insert(env, child, (const struct dt_rec *)rec,
933                        (const struct dt_key *)dot, th);
934         if (rc != 0)
935                 GOTO(unlock, rc);
936
937         /* 4b. insert dotdot into child dir */
938         rec->rec_fid = &LU_LPF_FID;
939         rc = dt_insert(env, child, (const struct dt_rec *)rec,
940                        (const struct dt_key *)dotdot, th);
941         if (rc != 0)
942                 GOTO(unlock, rc);
943
944         /* 5b. insert linkEA for child */
945         rc = dt_xattr_set(env, child, &linkea_buf,
946                           XATTR_NAME_LINK, 0, th);
947         if (rc != 0)
948                 GOTO(unlock, rc);
949
950         bk->lb_lpf_fid = *cfid;
951         lfsck_bookmark_cpu_to_le(&lfsck->li_bookmark_disk, bk);
952
953         /* 6b. update bookmark */
954         rc = dt_record_write(env, bk_obj,
955                              lfsck_buf_get(env, bk, len), &pos, th);
956
957         dt_write_unlock(env, child);
958         dt_trans_stop(env, dev, th);
959         if (rc != 0)
960                 RETURN(rc);
961
962         /* Transaction II: remotely */
963
964         dev = lfsck_obj2dev(parent);
965         th = dt_trans_create(env, dev);
966         if (IS_ERR(th))
967                 RETURN(PTR_ERR(th));
968
969         th->th_sync = 1;
970         /* 5a. insert name into parent dir */
971         rec->rec_fid = cfid;
972         rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
973                                (const struct dt_key *)name, th);
974         if (rc != 0)
975                 GOTO(stop, rc);
976
977         /* 6a. increase parent nlink */
978         rc = dt_declare_ref_add(env, parent, th);
979         if (rc != 0)
980                 GOTO(stop, rc);
981
982         rc = dt_trans_start_local(env, dev, th);
983         if (rc != 0)
984                 GOTO(stop, rc);
985
986         /* 5b. insert name into parent dir */
987         rc = dt_insert(env, parent, (const struct dt_rec *)rec,
988                        (const struct dt_key *)name, th);
989         if (rc != 0)
990                 GOTO(stop, rc);
991
992         dt_write_lock(env, parent, 0);
993         /* 6b. increase parent nlink */
994         rc = dt_ref_add(env, parent, th);
995         dt_write_unlock(env, parent);
996
997         GOTO(stop, rc);
998
999 unlock:
1000         dt_write_unlock(env, child);
1001 stop:
1002         dt_trans_stop(env, dev, th);
1003
1004         if (rc != 0 && dev == lfsck_obj2dev(parent))
1005                 CDEBUG(D_LFSCK, "%s: partially created the object "DFID
1006                        "for orphans, but failed to insert the name %s "
1007                        "to the .lustre/lost+found/. Such inconsistency "
1008                        "will be repaired when LFSCK run next time: rc = %d\n",
1009                        lfsck_lfsck2name(lfsck), PFID(cfid), name, rc);
1010
1011         return rc;
1012 }
1013
1014 /**
1015  * Create the MDTxxxx directory under /ROOT/.lustre/lost+found/
1016  *
1017  * The /ROOT/.lustre/lost+found/MDTxxxx/ directory is used for holding
1018  * orphans and other uncertain inconsistent objects found during the
1019  * LFSCK. Such directory will be created by the LFSCK engine on the
1020  * local MDT before the LFSCK scanning.
1021  *
1022  * \param[in] env       pointer to the thread context
1023  * \param[in] lfsck     pointer to the lfsck instance
1024  *
1025  * \retval              0 for success
1026  * \retval              negative error number on failure
1027  */
1028 static int lfsck_create_lpf(const struct lu_env *env,
1029                             struct lfsck_instance *lfsck)
1030 {
1031         struct lfsck_bookmark    *bk    = &lfsck->li_bookmark_ram;
1032         struct lfsck_thread_info *info  = lfsck_env_info(env);
1033         struct lu_fid            *cfid  = &info->lti_fid2;
1034         struct lu_attr           *la    = &info->lti_la;
1035         struct dt_object_format  *dof   = &info->lti_dof;
1036         struct dt_object         *parent = lfsck->li_lpf_root_obj;
1037         struct dt_object         *child = NULL;
1038         struct lfsck_lock_handle *llh   = &info->lti_llh;
1039         char                      name[8];
1040         int                       node  = lfsck_dev_idx(lfsck);
1041         int                       rc    = 0;
1042         ENTRY;
1043
1044         LASSERT(lfsck->li_master);
1045         LASSERT(parent != NULL);
1046         LASSERT(lfsck->li_lpf_obj == NULL);
1047
1048         snprintf(name, 8, "MDT%04x", node);
1049         rc = lfsck_lock(env, lfsck, parent, name, llh,
1050                         MDS_INODELOCK_UPDATE, LCK_PW);
1051         if (rc != 0)
1052                 RETURN(rc);
1053
1054         if (fid_is_zero(&bk->lb_lpf_fid)) {
1055                 /* There is corner case that: in former LFSCK scanning we have
1056                  * created the .lustre/lost+found/MDTxxxx but failed to update
1057                  * the lfsck_bookmark::lb_lpf_fid successfully. So need lookup
1058                  * it from MDT0 firstly. */
1059                 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1060                                (const struct dt_key *)name);
1061                 if (rc != 0 && rc != -ENOENT)
1062                         GOTO(unlock, rc);
1063
1064                 if (rc == 0) {
1065                         bk->lb_lpf_fid = *cfid;
1066                         rc = lfsck_bookmark_store(env, lfsck);
1067                 } else {
1068                         rc = lfsck_fid_alloc(env, lfsck, cfid, true);
1069                 }
1070                 if (rc != 0)
1071                         GOTO(unlock, rc);
1072         } else {
1073                 *cfid = bk->lb_lpf_fid;
1074         }
1075
1076         child = lfsck_object_find_bottom_new(env, lfsck, cfid);
1077         if (IS_ERR(child))
1078                 GOTO(unlock, rc = PTR_ERR(child));
1079
1080         if (dt_object_exists(child) != 0) {
1081                 if (unlikely(!dt_try_as_dir(env, child)))
1082                         rc = -ENOTDIR;
1083                 else
1084                         lfsck->li_lpf_obj = child;
1085
1086                 GOTO(unlock, rc);
1087         }
1088
1089         memset(la, 0, sizeof(*la));
1090         la->la_atime = la->la_mtime = la->la_ctime = ktime_get_real_seconds();
1091         la->la_mode = S_IFDIR | S_IRWXU;
1092         la->la_valid = LA_ATIME | LA_MTIME | LA_CTIME | LA_MODE |
1093                        LA_UID | LA_GID;
1094         memset(dof, 0, sizeof(*dof));
1095         dof->dof_type = dt_mode_to_dft(S_IFDIR);
1096
1097         if (node == 0)
1098                 rc = lfsck_create_lpf_local(env, lfsck, child, la, dof, name);
1099         else
1100                 rc = lfsck_create_lpf_remote(env, lfsck, child, la, dof, name);
1101         if (rc == 0)
1102                 lfsck->li_lpf_obj = child;
1103
1104         GOTO(unlock, rc);
1105
1106 unlock:
1107         lfsck_unlock(llh);
1108         if (rc != 0 && child != NULL && !IS_ERR(child))
1109                 lfsck_object_put(env, child);
1110
1111         return rc;
1112 }
1113
1114 /**
1115  * Scan .lustre/lost+found for bad name entries and remove them.
1116  *
1117  * The valid name entry should be "MDTxxxx", the "xxxx" is the MDT device
1118  * index in the system. Any other formatted name is invalid and should be
1119  * removed.
1120  *
1121  * \param[in] env       pointer to the thread context
1122  * \param[in] lfsck     pointer to the lfsck instance
1123  *
1124  * \retval              0 for success
1125  * \retval              negative error number on failure
1126  */
1127 static int lfsck_scan_lpf_bad_entries(const struct lu_env *env,
1128                                       struct lfsck_instance *lfsck)
1129 {
1130         struct dt_object        *parent = lfsck->li_lpf_root_obj;
1131         struct lu_dirent        *ent    =
1132                         (struct lu_dirent *)lfsck_env_info(env)->lti_key;
1133         const struct dt_it_ops  *iops   = &parent->do_index_ops->dio_it;
1134         struct dt_it            *it;
1135         int                      rc;
1136         ENTRY;
1137
1138         it = iops->init(env, parent, LUDA_64BITHASH);
1139         if (IS_ERR(it))
1140                 RETURN(PTR_ERR(it));
1141
1142         rc = iops->load(env, it, 0);
1143         if (rc == 0)
1144                 rc = iops->next(env, it);
1145         else if (rc > 0)
1146                 rc = 0;
1147
1148         while (rc == 0) {
1149                 int off = 3;
1150
1151                 rc = iops->rec(env, it, (struct dt_rec *)ent, LUDA_64BITHASH);
1152                 if (rc != 0)
1153                         break;
1154
1155                 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
1156                 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
1157                         goto next;
1158
1159                 /* name length must be strlen("MDTxxxx") */
1160                 if (ent->lde_namelen != 7)
1161                         goto remove;
1162
1163                 if (memcmp(ent->lde_name, "MDT", off) != 0)
1164                         goto remove;
1165
1166                 while (off < 7 && isxdigit(ent->lde_name[off]))
1167                         off++;
1168
1169                 if (off != 7) {
1170
1171 remove:
1172                         rc = lfsck_lpf_remove_name_entry(env, lfsck,
1173                                                          ent->lde_name);
1174                         if (rc != 0)
1175                                 break;
1176                 }
1177
1178 next:
1179                 rc = iops->next(env, it);
1180         }
1181
1182         iops->put(env, it);
1183         iops->fini(env, it);
1184
1185         RETURN(rc > 0 ? 0 : rc);
1186 }
1187
1188 static int lfsck_update_lpf_entry(const struct lu_env *env,
1189                                   struct lfsck_instance *lfsck,
1190                                   struct dt_object *parent,
1191                                   struct dt_object *child,
1192                                   const char *name,
1193                                   enum lfsck_verify_lpf_types type)
1194 {
1195         int rc;
1196
1197         if (type == LVLT_BY_BOOKMARK) {
1198                 rc = lfsck_update_name_entry(env, lfsck, parent, name,
1199                                              lfsck_dto2fid(child), S_IFDIR);
1200         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1201                 lfsck->li_bookmark_ram.lb_lpf_fid = *lfsck_dto2fid(child);
1202                 rc = lfsck_bookmark_store(env, lfsck);
1203
1204                 CDEBUG(D_LFSCK, "%s: update LPF fid "DFID
1205                        " in the bookmark file: rc = %d\n",
1206                        lfsck_lfsck2name(lfsck),
1207                        PFID(lfsck_dto2fid(child)), rc);
1208         }
1209
1210         return rc;
1211 }
1212
1213 /**
1214  * Check whether the @child back references the @parent.
1215  *
1216  * Two cases:
1217  * 1) The child's FID is stored in the bookmark file. If the child back
1218  *    references the parent (LU_LPF_FID object) via its ".." entry, then
1219  *    insert the name (MDTxxxx) to the .lustre/lost+found; otherwise, if
1220  *    the child back references another parent2, then:
1221  * 1.1) If the parent2 recognizes the child, then update the bookmark file;
1222  * 1.2) Otherwise, the LFSCK cannot know whether there will be parent3 that
1223  *      references the child. So keep them there. As the LFSCK processing,
1224  *      the parent3 may be found, then when the LFSCK run next time, the
1225  *      inconsistency can be repaired.
1226  *
1227  * 2) The child's FID is stored in the .lustre/lost+found/ sub-directory name
1228  *    entry (MDTxxxx). If the child back references the parent (LU_LPF_FID obj)
1229  *    via its ".." entry, then update the bookmark file, otherwise, if the child
1230  *    back references another parent2, then:
1231  * 2.1) If the parent2 recognizes the child, then remove the sub-directory
1232  *      from .lustre/lost+found/;
1233  * 2.2) Otherwise, if the parent2 does not recognizes the child, trust the
1234  *      sub-directory name entry and update the child;
1235  * 2.3) Otherwise, if we do not know whether the parent2 recognizes the child
1236  *      or not, then keep them there.
1237  *
1238  * \param[in] env       pointer to the thread context
1239  * \param[in] lfsck     pointer to the lfsck instance
1240  * \param[in] child     pointer to the lost+found sub-directory object
1241  * \param[in] name      the name for lost+found sub-directory object
1242  * \param[out] fid      pointer to the buffer to hold the FID of the object
1243  *                      (called it as parent2) that is referenced via the
1244  *                      child's dotdot entry; it also can be the FID that
1245  *                      is referenced by the name entry under the parent2.
1246  * \param[in] type      to indicate where the child's FID is stored in
1247  *
1248  * \retval              positive number for uncertain inconsistency
1249  * \retval              0 for success
1250  * \retval              negative error number on failure
1251  */
1252 static int lfsck_verify_lpf_pairs(const struct lu_env *env,
1253                                   struct lfsck_instance *lfsck,
1254                                   struct dt_object *child, const char *name,
1255                                   struct lu_fid *fid,
1256                                   enum lfsck_verify_lpf_types type)
1257 {
1258         struct dt_object         *parent  = lfsck->li_lpf_root_obj;
1259         struct lfsck_thread_info *info    = lfsck_env_info(env);
1260         char                     *name2   = info->lti_key;
1261         struct lu_fid            *fid2    = &info->lti_fid3;
1262         struct dt_object         *parent2 = NULL;
1263         struct lustre_handle      lh      = { 0 };
1264         int                       rc;
1265         ENTRY;
1266
1267         fid_zero(fid);
1268         rc = dt_lookup(env, child, (struct dt_rec *)fid,
1269                        (const struct dt_key *)dotdot);
1270         if (rc != 0)
1271                 GOTO(linkea, rc);
1272
1273         if (!fid_is_sane(fid))
1274                 GOTO(linkea, rc = -EINVAL);
1275
1276         if (lu_fid_eq(fid, &LU_LPF_FID)) {
1277                 const struct lu_name *cname;
1278
1279                 if (lfsck->li_lpf_obj == NULL) {
1280                         lu_object_get(&child->do_lu);
1281                         lfsck->li_lpf_obj = child;
1282                 }
1283
1284                 cname = lfsck_name_get_const(env, name, strlen(name));
1285                 rc = lfsck_verify_linkea(env, child, cname, &LU_LPF_FID);
1286                 if (rc == 0)
1287                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1288                                                     name, type);
1289
1290                 GOTO(out_done, rc);
1291         }
1292
1293         parent2 = lfsck_object_find_bottom(env, lfsck, fid);
1294         if (IS_ERR(parent2))
1295                 GOTO(linkea, parent2);
1296
1297         if (!dt_object_exists(parent2)) {
1298                 lfsck_object_put(env, parent2);
1299
1300                 GOTO(linkea, parent2 = ERR_PTR(-ENOENT));
1301         }
1302
1303         if (!dt_try_as_dir(env, parent2)) {
1304                 lfsck_object_put(env, parent2);
1305
1306                 GOTO(linkea, parent2 = ERR_PTR(-ENOTDIR));
1307         }
1308
1309 linkea:
1310         /* To prevent rename/unlink race */
1311         rc = lfsck_ibits_lock(env, lfsck, child, &lh,
1312                               MDS_INODELOCK_UPDATE, LCK_PR);
1313         if (rc != 0)
1314                 GOTO(out_put, rc);
1315
1316         dt_read_lock(env, child, 0);
1317         rc = lfsck_links_get_first(env, child, name2, fid2);
1318         if (rc != 0) {
1319                 dt_read_unlock(env, child);
1320                 lfsck_ibits_unlock(&lh, LCK_PR);
1321
1322                 GOTO(out_put, rc = 1);
1323         }
1324
1325         /* It is almost impossible that the bookmark file (or the name entry)
1326          * and the linkEA hit the same data corruption. Trust the linkEA. */
1327         if (lu_fid_eq(fid2, &LU_LPF_FID) && strcmp(name, name2) == 0) {
1328                 dt_read_unlock(env, child);
1329                 lfsck_ibits_unlock(&lh, LCK_PR);
1330
1331                 *fid = *fid2;
1332                 if (lfsck->li_lpf_obj == NULL) {
1333                         lu_object_get(&child->do_lu);
1334                         lfsck->li_lpf_obj = child;
1335                 }
1336
1337                 /* Update the child's dotdot entry */
1338                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1339                                              &LU_LPF_FID, S_IFDIR);
1340                 if (rc == 0)
1341                         rc = lfsck_update_lpf_entry(env, lfsck, parent, child,
1342                                                     name, type);
1343
1344                 GOTO(out_put, rc);
1345         }
1346
1347         if (parent2 == NULL || IS_ERR(parent2)) {
1348                 dt_read_unlock(env, child);
1349                 lfsck_ibits_unlock(&lh, LCK_PR);
1350
1351                 GOTO(out_done, rc = 1);
1352         }
1353
1354         rc = dt_lookup(env, parent2, (struct dt_rec *)fid,
1355                        (const struct dt_key *)name2);
1356         dt_read_unlock(env, child);
1357         lfsck_ibits_unlock(&lh, LCK_PR);
1358         if (rc != 0 && rc != -ENOENT)
1359                 GOTO(out_put, rc);
1360
1361         if (rc == -ENOENT || !lu_fid_eq(fid, lfsck_dto2fid(child))) {
1362                 if (type == LVLT_BY_BOOKMARK)
1363                         GOTO(out_put, rc = 1);
1364
1365                 /* Trust the name entry, update the child's dotdot entry. */
1366                 rc = lfsck_update_name_entry(env, lfsck, child, dotdot,
1367                                              &LU_LPF_FID, S_IFDIR);
1368
1369                 GOTO(out_put, rc);
1370         }
1371
1372         if (type == LVLT_BY_BOOKMARK) {
1373                 /* Invalid FID record in the bookmark file, reset it. */
1374                 fid_zero(&lfsck->li_bookmark_ram.lb_lpf_fid);
1375                 rc = lfsck_bookmark_store(env, lfsck);
1376
1377                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1378                        " in the bookmark file: rc = %d\n",
1379                        lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), rc);
1380         } else /* if (type == LVLT_BY_NAMEENTRY) */ {
1381                 /* The name entry is wrong, remove it. */
1382                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1383         }
1384
1385         GOTO(out_put, rc);
1386
1387 out_put:
1388         if (parent2 != NULL && !IS_ERR(parent2))
1389                 lfsck_object_put(env, parent2);
1390
1391 out_done:
1392         return rc;
1393 }
1394
1395 /**
1396  * Verify the /ROOT/.lustre/lost+found/ directory.
1397  *
1398  * /ROOT/.lustre/lost+found/ is a special directory to hold the objects that
1399  * the LFSCK does not exactly know how to handle, such as orphans. So before
1400  * the LFSCK scanning the system, the consistency of such directory needs to
1401  * be verified firstly to allow the users to use it during the LFSCK.
1402  *
1403  * \param[in] env       pointer to the thread context
1404  * \param[in] lfsck     pointer to the lfsck instance
1405  *
1406  * \retval              positive number for uncertain inconsistency
1407  * \retval              0 for success
1408  * \retval              negative error number on failure
1409  */
1410 int lfsck_verify_lpf(const struct lu_env *env, struct lfsck_instance *lfsck)
1411 {
1412         struct lfsck_thread_info *info   = lfsck_env_info(env);
1413         struct lu_fid            *pfid   = &info->lti_fid;
1414         struct lu_fid            *cfid   = &info->lti_fid2;
1415         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1416         struct dt_object         *parent;
1417         /* child1's FID is in the bookmark file. */
1418         struct dt_object         *child1 = NULL;
1419         /* child2's FID is in the name entry MDTxxxx. */
1420         struct dt_object         *child2 = NULL;
1421         const struct lu_name     *cname;
1422         char                      name[8];
1423         int                       node   = lfsck_dev_idx(lfsck);
1424         int                       rc     = 0;
1425         ENTRY;
1426
1427         LASSERT(lfsck->li_master);
1428
1429         if (lfsck->li_lpf_root_obj != NULL)
1430                 RETURN(0);
1431
1432         if (node == 0) {
1433                 parent = lfsck_object_find_by_dev(env, lfsck->li_bottom,
1434                                                   &LU_LPF_FID);
1435         } else {
1436                 struct lfsck_tgt_desc *ltd;
1437
1438                 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, 0);
1439                 if (unlikely(ltd == NULL))
1440                         RETURN(-ENXIO);
1441
1442                 parent = lfsck_object_find_by_dev(env, ltd->ltd_tgt,
1443                                                   &LU_LPF_FID);
1444                 lfsck_tgt_put(ltd);
1445         }
1446
1447         if (IS_ERR(parent))
1448                 RETURN(PTR_ERR(parent));
1449
1450         LASSERT(dt_object_exists(parent));
1451
1452         if (unlikely(!dt_try_as_dir(env, parent))) {
1453                 lfsck_object_put(env, parent);
1454
1455                 GOTO(put, rc = -ENOTDIR);
1456         }
1457
1458         lfsck->li_lpf_root_obj = parent;
1459         if (node == 0) {
1460                 rc = lfsck_scan_lpf_bad_entries(env, lfsck);
1461                 if (rc != 0)
1462                         CDEBUG(D_LFSCK, "%s: scan .lustre/lost+found/ "
1463                                "for bad sub-directories: rc = %d\n",
1464                                lfsck_lfsck2name(lfsck), rc);
1465         }
1466
1467         /* child2 */
1468         snprintf(name, 8, "MDT%04x", node);
1469         rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1470                        (const struct dt_key *)name);
1471         if (rc == -ENOENT) {
1472                 rc = 0;
1473                 goto find_child1;
1474         }
1475
1476         if (rc != 0)
1477                 GOTO(put, rc);
1478
1479         /* Invalid FID in the name entry, remove the name entry. */
1480         if (!fid_is_norm(cfid)) {
1481                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1482                 if (rc != 0)
1483                         GOTO(put, rc);
1484
1485                 goto find_child1;
1486         }
1487
1488         child2 = lfsck_object_find_bottom(env, lfsck, cfid);
1489         if (IS_ERR(child2))
1490                 GOTO(put, rc = PTR_ERR(child2));
1491
1492         if (unlikely(!dt_object_exists(child2) ||
1493                      dt_object_remote(child2)) ||
1494                      !S_ISDIR(lfsck_object_type(child2))) {
1495                 rc = lfsck_lpf_remove_name_entry(env, lfsck, name);
1496                 if (rc != 0)
1497                         GOTO(put, rc);
1498
1499                 goto find_child1;
1500         }
1501
1502         if (unlikely(!dt_try_as_dir(env, child2))) {
1503                 lfsck_object_put(env, child2);
1504                 child2 = NULL;
1505                 rc = -ENOTDIR;
1506         }
1507
1508 find_child1:
1509         if (fid_is_zero(&bk->lb_lpf_fid))
1510                 goto check_child2;
1511
1512         if (likely(lu_fid_eq(cfid, &bk->lb_lpf_fid))) {
1513                 if (lfsck->li_lpf_obj == NULL) {
1514                         lu_object_get(&child2->do_lu);
1515                         lfsck->li_lpf_obj = child2;
1516                 }
1517
1518                 cname = lfsck_name_get_const(env, name, strlen(name));
1519                 rc = lfsck_verify_linkea(env, child2, cname, &LU_LPF_FID);
1520
1521                 GOTO(put, rc);
1522         }
1523
1524         if (unlikely(!fid_is_norm(&bk->lb_lpf_fid))) {
1525                 struct lu_fid tfid = bk->lb_lpf_fid;
1526
1527                 /* Invalid FID record in the bookmark file, reset it. */
1528                 fid_zero(&bk->lb_lpf_fid);
1529                 rc = lfsck_bookmark_store(env, lfsck);
1530
1531                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1532                        " in the bookmark file: rc = %d\n",
1533                        lfsck_lfsck2name(lfsck), PFID(&tfid), rc);
1534
1535                 if (rc != 0)
1536                         GOTO(put, rc);
1537
1538                 goto check_child2;
1539         }
1540
1541         child1 = lfsck_object_find_bottom(env, lfsck, &bk->lb_lpf_fid);
1542         if (IS_ERR(child1)) {
1543                 child1 = NULL;
1544                 goto check_child2;
1545         }
1546
1547         if (unlikely(!dt_object_exists(child1) ||
1548                      dt_object_remote(child1)) ||
1549                      !S_ISDIR(lfsck_object_type(child1))) {
1550                 /* Invalid FID record in the bookmark file, reset it. */
1551                 fid_zero(&bk->lb_lpf_fid);
1552                 rc = lfsck_bookmark_store(env, lfsck);
1553
1554                 CDEBUG(D_LFSCK, "%s: reset invalid LPF fid "DFID
1555                        " in the bookmark file: rc = %d\n",
1556                        lfsck_lfsck2name(lfsck),
1557                        PFID(lfsck_dto2fid(child1)), rc);
1558
1559                 if (rc != 0)
1560                         GOTO(put, rc);
1561
1562                 lfsck_object_put(env, child1);
1563                 child1 = NULL;
1564                 goto check_child2;
1565         }
1566
1567         if (unlikely(!dt_try_as_dir(env, child1))) {
1568                 lfsck_object_put(env, child1);
1569                 child1 = NULL;
1570                 rc = -ENOTDIR;
1571                 goto check_child2;
1572         }
1573
1574         rc = lfsck_verify_lpf_pairs(env, lfsck, child1, name, pfid,
1575                                     LVLT_BY_BOOKMARK);
1576         if (lu_fid_eq(pfid, &LU_LPF_FID))
1577                 GOTO(put, rc);
1578
1579 check_child2:
1580         if (child2 != NULL)
1581                 rc = lfsck_verify_lpf_pairs(env, lfsck, child2, name,
1582                                             pfid, LVLT_BY_NAMEENTRY);
1583
1584         GOTO(put, rc);
1585
1586 put:
1587         if (lfsck->li_lpf_obj != NULL) {
1588                 if (unlikely(!dt_try_as_dir(env, lfsck->li_lpf_obj))) {
1589                         lfsck_object_put(env, lfsck->li_lpf_obj);
1590                         lfsck->li_lpf_obj = NULL;
1591                         rc = -ENOTDIR;
1592                 }
1593         } else if (rc == 0) {
1594                 rc = lfsck_create_lpf(env, lfsck);
1595         }
1596
1597         if (child2 != NULL && !IS_ERR(child2))
1598                 lfsck_object_put(env, child2);
1599         if (child1 != NULL && !IS_ERR(child1))
1600                 lfsck_object_put(env, child1);
1601
1602         return rc;
1603 }
1604
1605 static int lfsck_fid_init(struct lfsck_instance *lfsck)
1606 {
1607         struct lfsck_bookmark   *bk     = &lfsck->li_bookmark_ram;
1608         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
1609         char                    *prefix;
1610         int                      rc     = 0;
1611         ENTRY;
1612
1613         if (unlikely(ss == NULL))
1614                 RETURN(-ENXIO);
1615
1616         OBD_ALLOC_PTR(lfsck->li_seq);
1617         if (lfsck->li_seq == NULL)
1618                 RETURN(-ENOMEM);
1619
1620         OBD_ALLOC(prefix, MAX_OBD_NAME + 7);
1621         if (prefix == NULL)
1622                 GOTO(out, rc = -ENOMEM);
1623
1624         snprintf(prefix, MAX_OBD_NAME + 7, "lfsck-%s", lfsck_lfsck2name(lfsck));
1625         rc = seq_client_init(lfsck->li_seq, NULL, LUSTRE_SEQ_METADATA, prefix,
1626                              ss->ss_server_seq);
1627         OBD_FREE(prefix, MAX_OBD_NAME + 7);
1628         if (rc != 0)
1629                 GOTO(out, rc);
1630
1631         if (fid_is_sane(&bk->lb_last_fid))
1632                 lfsck->li_seq->lcs_fid = bk->lb_last_fid;
1633
1634         RETURN(0);
1635
1636 out:
1637         OBD_FREE_PTR(lfsck->li_seq);
1638         lfsck->li_seq = NULL;
1639
1640         return rc;
1641 }
1642
1643 static void lfsck_fid_fini(struct lfsck_instance *lfsck)
1644 {
1645         if (lfsck->li_seq != NULL) {
1646                 seq_client_fini(lfsck->li_seq);
1647                 OBD_FREE_PTR(lfsck->li_seq);
1648                 lfsck->li_seq = NULL;
1649         }
1650 }
1651
1652 void lfsck_instance_cleanup(const struct lu_env *env,
1653                             struct lfsck_instance *lfsck)
1654 {
1655         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1656         struct lfsck_component  *com;
1657         struct lfsck_component  *next;
1658         struct lfsck_lmv_unit   *llu;
1659         struct lfsck_lmv_unit   *llu_next;
1660         struct lfsck_lmv        *llmv;
1661         ENTRY;
1662
1663         LASSERT(list_empty(&lfsck->li_link));
1664         LASSERT(thread_is_init(thread) || thread_is_stopped(thread));
1665
1666         if (lfsck->li_obj_oit != NULL) {
1667                 lfsck_object_put(env, lfsck->li_obj_oit);
1668                 lfsck->li_obj_oit = NULL;
1669         }
1670
1671         LASSERT(lfsck->li_obj_dir == NULL);
1672         LASSERT(lfsck->li_lmv == NULL);
1673
1674         list_for_each_entry_safe(llu, llu_next, &lfsck->li_list_lmv, llu_link) {
1675                 llmv = &llu->llu_lmv;
1676
1677                 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
1678                          "still in using: %u\n",
1679                          atomic_read(&llmv->ll_ref));
1680
1681                 lfsck_lmv_put(env, llmv);
1682         }
1683
1684         list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
1685                 lfsck_component_cleanup(env, com);
1686         }
1687
1688         LASSERT(list_empty(&lfsck->li_list_dir));
1689
1690         list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
1691                                  lc_link) {
1692                 lfsck_component_cleanup(env, com);
1693         }
1694
1695         list_for_each_entry_safe(com, next, &lfsck->li_list_idle, lc_link) {
1696                 lfsck_component_cleanup(env, com);
1697         }
1698
1699         lfsck_tgt_descs_fini(&lfsck->li_ost_descs);
1700         lfsck_tgt_descs_fini(&lfsck->li_mdt_descs);
1701
1702         if (lfsck->li_lfsck_dir != NULL) {
1703                 lfsck_object_put(env, lfsck->li_lfsck_dir);
1704                 lfsck->li_lfsck_dir = NULL;
1705         }
1706
1707         if (lfsck->li_bookmark_obj != NULL) {
1708                 lfsck_object_put(env, lfsck->li_bookmark_obj);
1709                 lfsck->li_bookmark_obj = NULL;
1710         }
1711
1712         if (lfsck->li_lpf_obj != NULL) {
1713                 lfsck_object_put(env, lfsck->li_lpf_obj);
1714                 lfsck->li_lpf_obj = NULL;
1715         }
1716
1717         if (lfsck->li_lpf_root_obj != NULL) {
1718                 lfsck_object_put(env, lfsck->li_lpf_root_obj);
1719                 lfsck->li_lpf_root_obj = NULL;
1720         }
1721
1722         if (lfsck->li_los != NULL) {
1723                 local_oid_storage_fini(env, lfsck->li_los);
1724                 lfsck->li_los = NULL;
1725         }
1726
1727         lfsck_fid_fini(lfsck);
1728
1729         OBD_FREE_PTR(lfsck);
1730 }
1731
1732 static inline struct lfsck_instance *
1733 __lfsck_instance_find(struct dt_device *key, bool ref, bool unlink)
1734 {
1735         struct lfsck_instance *lfsck;
1736
1737         list_for_each_entry(lfsck, &lfsck_instance_list, li_link) {
1738                 if (lfsck->li_bottom == key) {
1739                         if (ref)
1740                                 lfsck_instance_get(lfsck);
1741                         if (unlink)
1742                                 list_del_init(&lfsck->li_link);
1743
1744                         return lfsck;
1745                 }
1746         }
1747
1748         return NULL;
1749 }
1750
1751 struct lfsck_instance *lfsck_instance_find(struct dt_device *key, bool ref,
1752                                            bool unlink)
1753 {
1754         struct lfsck_instance *lfsck;
1755
1756         spin_lock(&lfsck_instance_lock);
1757         lfsck = __lfsck_instance_find(key, ref, unlink);
1758         spin_unlock(&lfsck_instance_lock);
1759
1760         return lfsck;
1761 }
1762
1763 static inline int lfsck_instance_add(struct lfsck_instance *lfsck)
1764 {
1765         struct lfsck_instance *tmp;
1766
1767         spin_lock(&lfsck_instance_lock);
1768         list_for_each_entry(tmp, &lfsck_instance_list, li_link) {
1769                 if (lfsck->li_bottom == tmp->li_bottom) {
1770                         spin_unlock(&lfsck_instance_lock);
1771                         return -EEXIST;
1772                 }
1773         }
1774
1775         list_add_tail(&lfsck->li_link, &lfsck_instance_list);
1776         spin_unlock(&lfsck_instance_lock);
1777         return 0;
1778 }
1779
1780 void lfsck_bits_dump(struct seq_file *m, int bits, const char *names[],
1781                      const char *prefix)
1782 {
1783         int flag;
1784         int i;
1785         bool newline = (bits != 0 ? false : true);
1786
1787         seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
1788
1789         for (i = 0, flag = 1; bits != 0; i++, flag = 1 << i) {
1790                 if (flag & bits) {
1791                         bits &= ~flag;
1792                         if (names[i] != NULL) {
1793                                 if (bits == 0)
1794                                         newline = true;
1795
1796                                 seq_printf(m, "%s%c", names[i],
1797                                            newline ? '\n' : ',');
1798                         }
1799                 }
1800         }
1801
1802         if (!newline)
1803                 seq_putc(m, '\n');
1804 }
1805
1806 void lfsck_time_dump(struct seq_file *m, time64_t time, const char *name)
1807 {
1808         if (time == 0) {
1809                 seq_printf(m, "%s_time: N/A\n", name);
1810                 seq_printf(m, "time_since_%s: N/A\n", name);
1811         } else {
1812                 seq_printf(m, "%s_time: %lld\n", name, time);
1813                 seq_printf(m, "time_since_%s: %lld seconds\n",
1814                            name, ktime_get_real_seconds() - time);
1815         }
1816 }
1817
1818 void lfsck_pos_dump(struct seq_file *m, struct lfsck_position *pos,
1819                     const char *prefix)
1820 {
1821         if (fid_is_zero(&pos->lp_dir_parent)) {
1822                 if (pos->lp_oit_cookie == 0) {
1823                         seq_printf(m, "%s: N/A, N/A, N/A\n", prefix);
1824                         return;
1825                 }
1826                 seq_printf(m, "%s: %llu, N/A, N/A\n",
1827                            prefix, pos->lp_oit_cookie);
1828         } else {
1829                 seq_printf(m, "%s: %llu, "DFID", %#llx\n",
1830                            prefix, pos->lp_oit_cookie,
1831                            PFID(&pos->lp_dir_parent), pos->lp_dir_cookie);
1832         }
1833 }
1834
1835 void lfsck_pos_fill(const struct lu_env *env, struct lfsck_instance *lfsck,
1836                     struct lfsck_position *pos, bool init)
1837 {
1838         const struct dt_it_ops *iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
1839
1840         if (unlikely(lfsck->li_di_oit == NULL)) {
1841                 memset(pos, 0, sizeof(*pos));
1842                 return;
1843         }
1844
1845         pos->lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
1846         if (!lfsck->li_current_oit_processed && !init)
1847                 pos->lp_oit_cookie--;
1848
1849         if (unlikely(pos->lp_oit_cookie == 0))
1850                 pos->lp_oit_cookie = 1;
1851
1852         if (lfsck->li_di_dir != NULL) {
1853                 struct dt_object *dto = lfsck->li_obj_dir;
1854
1855                 pos->lp_dir_cookie = dto->do_index_ops->dio_it.store(env,
1856                                                         lfsck->li_di_dir);
1857
1858                 if (pos->lp_dir_cookie >= MDS_DIR_END_OFF) {
1859                         fid_zero(&pos->lp_dir_parent);
1860                         pos->lp_dir_cookie = 0;
1861                 } else {
1862                         pos->lp_dir_parent = *lfsck_dto2fid(dto);
1863                 }
1864         } else {
1865                 fid_zero(&pos->lp_dir_parent);
1866                 pos->lp_dir_cookie = 0;
1867         }
1868 }
1869
1870 bool __lfsck_set_speed(struct lfsck_instance *lfsck, __u32 limit)
1871 {
1872         bool dirty = false;
1873
1874         if (limit != LFSCK_SPEED_NO_LIMIT) {
1875                 if (limit > cfs_time_seconds(1)) {
1876                         lfsck->li_sleep_rate = limit / cfs_time_seconds(1);
1877                         lfsck->li_sleep_jif = 1;
1878                 } else {
1879                         lfsck->li_sleep_rate = 1;
1880                         lfsck->li_sleep_jif = cfs_time_seconds(1) / limit;
1881                 }
1882         } else {
1883                 lfsck->li_sleep_jif = 0;
1884                 lfsck->li_sleep_rate = 0;
1885         }
1886
1887         if (lfsck->li_bookmark_ram.lb_speed_limit != limit) {
1888                 lfsck->li_bookmark_ram.lb_speed_limit = limit;
1889                 dirty = true;
1890         }
1891
1892         return dirty;
1893 }
1894
1895 void lfsck_control_speed(struct lfsck_instance *lfsck)
1896 {
1897         struct ptlrpc_thread *thread = &lfsck->li_thread;
1898
1899         if (lfsck->li_sleep_jif > 0 &&
1900             lfsck->li_new_scanned >= lfsck->li_sleep_rate) {
1901                 wait_event_idle_timeout(thread->t_ctl_waitq,
1902                                         !thread_is_running(thread),
1903                                         lfsck->li_sleep_jif);
1904                 lfsck->li_new_scanned = 0;
1905         }
1906 }
1907
1908 void lfsck_control_speed_by_self(struct lfsck_component *com)
1909 {
1910         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1911         struct ptlrpc_thread    *thread = &lfsck->li_thread;
1912
1913         if (lfsck->li_sleep_jif > 0 &&
1914             com->lc_new_scanned >= lfsck->li_sleep_rate) {
1915                 wait_event_idle_timeout(thread->t_ctl_waitq,
1916                                         !thread_is_running(thread),
1917                                         lfsck->li_sleep_jif);
1918                 com->lc_new_scanned = 0;
1919         }
1920 }
1921
1922 static struct lfsck_thread_args *
1923 lfsck_thread_args_init(struct lfsck_instance *lfsck,
1924                        struct lfsck_component *com,
1925                        struct lfsck_start_param *lsp)
1926 {
1927         struct lfsck_thread_args *lta;
1928         int                       rc;
1929
1930         OBD_ALLOC_PTR(lta);
1931         if (lta == NULL)
1932                 return ERR_PTR(-ENOMEM);
1933
1934         rc = lu_env_init(&lta->lta_env, LCT_MD_THREAD | LCT_DT_THREAD);
1935         if (rc != 0) {
1936                 OBD_FREE_PTR(lta);
1937                 return ERR_PTR(rc);
1938         }
1939
1940         lta->lta_lfsck = lfsck_instance_get(lfsck);
1941         if (com != NULL)
1942                 lta->lta_com = lfsck_component_get(com);
1943
1944         lta->lta_lsp = lsp;
1945
1946         return lta;
1947 }
1948
1949 void lfsck_thread_args_fini(struct lfsck_thread_args *lta)
1950 {
1951         if (lta->lta_com != NULL)
1952                 lfsck_component_put(&lta->lta_env, lta->lta_com);
1953         lfsck_instance_put(&lta->lta_env, lta->lta_lfsck);
1954         lu_env_fini(&lta->lta_env);
1955         OBD_FREE_PTR(lta);
1956 }
1957
1958 struct lfsck_assistant_data *
1959 lfsck_assistant_data_init(struct lfsck_assistant_operations *lao,
1960                           const char *name)
1961 {
1962         struct lfsck_assistant_data *lad;
1963
1964         OBD_ALLOC_PTR(lad);
1965         if (lad != NULL) {
1966                 lad->lad_bitmap = CFS_ALLOCATE_BITMAP(BITS_PER_LONG);
1967                 if (lad->lad_bitmap == NULL) {
1968                         OBD_FREE_PTR(lad);
1969                         return NULL;
1970                 }
1971
1972                 INIT_LIST_HEAD(&lad->lad_req_list);
1973                 spin_lock_init(&lad->lad_lock);
1974                 INIT_LIST_HEAD(&lad->lad_ost_list);
1975                 INIT_LIST_HEAD(&lad->lad_ost_phase1_list);
1976                 INIT_LIST_HEAD(&lad->lad_ost_phase2_list);
1977                 INIT_LIST_HEAD(&lad->lad_mdt_list);
1978                 INIT_LIST_HEAD(&lad->lad_mdt_phase1_list);
1979                 INIT_LIST_HEAD(&lad->lad_mdt_phase2_list);
1980                 init_waitqueue_head(&lad->lad_thread.t_ctl_waitq);
1981                 lad->lad_ops = lao;
1982                 lad->lad_name = name;
1983         }
1984
1985         return lad;
1986 }
1987
1988 struct lfsck_assistant_object *
1989 lfsck_assistant_object_init(const struct lu_env *env, const struct lu_fid *fid,
1990                             const struct lu_attr *attr, __u64 cookie,
1991                             bool is_dir)
1992 {
1993         struct lfsck_assistant_object   *lso;
1994
1995         OBD_ALLOC_PTR(lso);
1996         if (lso == NULL)
1997                 return ERR_PTR(-ENOMEM);
1998
1999         lso->lso_fid = *fid;
2000         if (attr != NULL)
2001                 lso->lso_attr = *attr;
2002
2003         atomic_set(&lso->lso_ref, 1);
2004         lso->lso_oit_cookie = cookie;
2005         if (is_dir)
2006                 lso->lso_is_dir = 1;
2007
2008         return lso;
2009 }
2010
2011 struct dt_object *
2012 lfsck_assistant_object_load(const struct lu_env *env,
2013                             struct lfsck_instance *lfsck,
2014                             struct lfsck_assistant_object *lso)
2015 {
2016         struct dt_object *obj;
2017
2018         obj = lfsck_object_find_bottom(env, lfsck, &lso->lso_fid);
2019         if (IS_ERR(obj))
2020                 return obj;
2021
2022         if (unlikely(!dt_object_exists(obj) || lfsck_is_dead_obj(obj))) {
2023                 lso->lso_dead = 1;
2024                 lfsck_object_put(env, obj);
2025
2026                 return ERR_PTR(-ENOENT);
2027         }
2028
2029         if (lso->lso_is_dir && unlikely(!dt_try_as_dir(env, obj))) {
2030                 lfsck_object_put(env, obj);
2031
2032                 return ERR_PTR(-ENOTDIR);
2033         }
2034
2035         return obj;
2036 }
2037
2038 /**
2039  * Generic LFSCK asynchronous communication interpretor function.
2040  * The LFSCK RPC reply for both the event notification and status
2041  * querying will be handled here.
2042  *
2043  * \param[in] env       pointer to the thread context
2044  * \param[in] req       pointer to the LFSCK request
2045  * \param[in] args      pointer to the lfsck_async_interpret_args
2046  * \param[in] rc        the result for handling the LFSCK request
2047  *
2048  * \retval              0 for success
2049  * \retval              negative error number on failure
2050  */
2051 int lfsck_async_interpret_common(const struct lu_env *env,
2052                                  struct ptlrpc_request *req,
2053                                  void *args, int rc)
2054 {
2055         struct lfsck_async_interpret_args *laia = args;
2056         struct lfsck_component            *com  = laia->laia_com;
2057         struct lfsck_assistant_data       *lad  = com->lc_data;
2058         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
2059         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
2060         struct lfsck_request              *lr   = laia->laia_lr;
2061
2062         LASSERT(com->lc_lfsck->li_master);
2063
2064         switch (lr->lr_event) {
2065         case LE_START:
2066                 if (unlikely(rc == -EINPROGRESS)) {
2067                         ltd->ltd_retry_start = 1;
2068                         break;
2069                 }
2070
2071                 if (rc != 0) {
2072                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s "
2073                                "start: rc = %d\n",
2074                                lfsck_lfsck2name(com->lc_lfsck),
2075                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2076                                ltd->ltd_index, lad->lad_name, rc);
2077
2078                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2079                                 struct lfsck_layout *lo = com->lc_file_ram;
2080
2081                                 if (lr->lr_flags & LEF_TO_OST)
2082                                         lfsck_lad_set_bitmap(env, com,
2083                                                              ltd->ltd_index);
2084                                 else
2085                                         lo->ll_flags |= LF_INCOMPLETE;
2086                         } else {
2087                                 struct lfsck_namespace *ns = com->lc_file_ram;
2088
2089                                 /* If some MDT does not join the namespace
2090                                  * LFSCK, then we cannot know whether there
2091                                  * is some name entry on such MDT that with
2092                                  * the referenced MDT-object on this MDT or
2093                                  * not. So the namespace LFSCK on this MDT
2094                                  * cannot handle orphan MDT-objects properly.
2095                                  * So we mark the LFSCK as LF_INCOMPLETE and
2096                                  * skip orphan MDT-objects handling. */
2097                                 ns->ln_flags |= LF_INCOMPLETE;
2098                         }
2099                         break;
2100                 }
2101
2102                 spin_lock(&ltds->ltd_lock);
2103                 if (ltd->ltd_dead) {
2104                         spin_unlock(&ltds->ltd_lock);
2105                         break;
2106                 }
2107
2108                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2109                         struct list_head *list;
2110                         struct list_head *phase_list;
2111
2112                         if (ltd->ltd_layout_done) {
2113                                 spin_unlock(&ltds->ltd_lock);
2114                                 break;
2115                         }
2116
2117                         if (lr->lr_flags & LEF_TO_OST) {
2118                                 list = &lad->lad_ost_list;
2119                                 phase_list = &lad->lad_ost_phase1_list;
2120                         } else {
2121                                 list = &lad->lad_mdt_list;
2122                                 phase_list = &lad->lad_mdt_phase1_list;
2123                         }
2124
2125                         if (list_empty(&ltd->ltd_layout_list))
2126                                 list_add_tail(&ltd->ltd_layout_list, list);
2127                         if (list_empty(&ltd->ltd_layout_phase_list))
2128                                 list_add_tail(&ltd->ltd_layout_phase_list,
2129                                               phase_list);
2130                 } else {
2131                         if (ltd->ltd_namespace_done) {
2132                                 spin_unlock(&ltds->ltd_lock);
2133                                 break;
2134                         }
2135
2136                         if (list_empty(&ltd->ltd_namespace_list))
2137                                 list_add_tail(&ltd->ltd_namespace_list,
2138                                               &lad->lad_mdt_list);
2139                         if (list_empty(&ltd->ltd_namespace_phase_list))
2140                                 list_add_tail(&ltd->ltd_namespace_phase_list,
2141                                               &lad->lad_mdt_phase1_list);
2142                 }
2143                 spin_unlock(&ltds->ltd_lock);
2144                 break;
2145         case LE_STOP:
2146         case LE_PHASE1_DONE:
2147         case LE_PHASE2_DONE:
2148         case LE_PEER_EXIT:
2149                 if (rc != 0 && rc != -EALREADY)
2150                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for %s: "
2151                               "event = %d, rc = %d\n",
2152                               lfsck_lfsck2name(com->lc_lfsck),
2153                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2154                               ltd->ltd_index, lad->lad_name, lr->lr_event, rc);
2155                 break;
2156         case LE_QUERY: {
2157                 struct lfsck_reply *reply;
2158                 struct list_head *list;
2159                 struct list_head *phase_list;
2160
2161                 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2162                         list = &ltd->ltd_layout_list;
2163                         phase_list = &ltd->ltd_layout_phase_list;
2164                 } else {
2165                         list = &ltd->ltd_namespace_list;
2166                         phase_list = &ltd->ltd_namespace_phase_list;
2167                 }
2168
2169                 if (rc != 0) {
2170                         if (lr->lr_flags & LEF_QUERY_ALL) {
2171                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2172                                 break;
2173                         }
2174
2175                         spin_lock(&ltds->ltd_lock);
2176                         list_del_init(phase_list);
2177                         list_del_init(list);
2178                         spin_unlock(&ltds->ltd_lock);
2179                         break;
2180                 }
2181
2182                 reply = req_capsule_server_get(&req->rq_pill,
2183                                                &RMF_LFSCK_REPLY);
2184                 if (reply == NULL) {
2185                         rc = -EPROTO;
2186                         CDEBUG(D_LFSCK, "%s: invalid query reply for %s: "
2187                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2188                                lad->lad_name, rc);
2189
2190                         if (lr->lr_flags & LEF_QUERY_ALL) {
2191                                 lfsck_reset_ltd_status(ltd, com->lc_type);
2192                                 break;
2193                         }
2194
2195                         spin_lock(&ltds->ltd_lock);
2196                         list_del_init(phase_list);
2197                         list_del_init(list);
2198                         spin_unlock(&ltds->ltd_lock);
2199                         break;
2200                 }
2201
2202                 if (lr->lr_flags & LEF_QUERY_ALL) {
2203                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2204                                 ltd->ltd_layout_status = reply->lr_status;
2205                                 ltd->ltd_layout_repaired = reply->lr_repaired;
2206                         } else {
2207                                 ltd->ltd_namespace_status = reply->lr_status;
2208                                 ltd->ltd_namespace_repaired =
2209                                                         reply->lr_repaired;
2210                         }
2211                         break;
2212                 }
2213
2214                 switch (reply->lr_status) {
2215                 case LS_SCANNING_PHASE1:
2216                         break;
2217                 case LS_SCANNING_PHASE2:
2218                         spin_lock(&ltds->ltd_lock);
2219                         list_del_init(phase_list);
2220                         if (ltd->ltd_dead) {
2221                                 spin_unlock(&ltds->ltd_lock);
2222                                 break;
2223                         }
2224
2225                         if (com->lc_type == LFSCK_TYPE_LAYOUT) {
2226                                 if (ltd->ltd_layout_done) {
2227                                         spin_unlock(&ltds->ltd_lock);
2228                                         break;
2229                                 }
2230
2231                                 if (lr->lr_flags & LEF_TO_OST)
2232                                         list_add_tail(phase_list,
2233                                                 &lad->lad_ost_phase2_list);
2234                                 else
2235                                         list_add_tail(phase_list,
2236                                                 &lad->lad_mdt_phase2_list);
2237                         } else {
2238                                 if (ltd->ltd_namespace_done) {
2239                                         spin_unlock(&ltds->ltd_lock);
2240                                         break;
2241                                 }
2242
2243                                 list_add_tail(phase_list,
2244                                               &lad->lad_mdt_phase2_list);
2245                         }
2246                         spin_unlock(&ltds->ltd_lock);
2247                         break;
2248                 default:
2249                         spin_lock(&ltds->ltd_lock);
2250                         list_del_init(phase_list);
2251                         list_del_init(list);
2252                         spin_unlock(&ltds->ltd_lock);
2253                         break;
2254                 }
2255                 break;
2256         }
2257         default:
2258                 CDEBUG(D_LFSCK, "%s: unexpected event: rc = %d\n",
2259                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
2260                 break;
2261         }
2262
2263         if (!laia->laia_shared) {
2264                 lfsck_tgt_put(ltd);
2265                 lfsck_component_put(env, com);
2266         }
2267
2268         return 0;
2269 }
2270
2271 static void lfsck_interpret(const struct lu_env *env,
2272                             struct lfsck_instance *lfsck,
2273                             struct ptlrpc_request *req, void *args, int result)
2274 {
2275         struct lfsck_async_interpret_args *laia = args;
2276         struct lfsck_component            *com;
2277
2278         LASSERT(laia->laia_com == NULL);
2279         LASSERT(laia->laia_shared);
2280
2281         spin_lock(&lfsck->li_lock);
2282         list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
2283                 laia->laia_com = com;
2284                 lfsck_async_interpret_common(env, req, laia, result);
2285         }
2286
2287         list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
2288                 laia->laia_com = com;
2289                 lfsck_async_interpret_common(env, req, laia, result);
2290         }
2291         spin_unlock(&lfsck->li_lock);
2292 }
2293
2294 static int lfsck_stop_notify(const struct lu_env *env,
2295                              struct lfsck_instance *lfsck,
2296                              struct lfsck_tgt_descs *ltds,
2297                              struct lfsck_tgt_desc *ltd, __u16 type)
2298 {
2299         struct lfsck_component *com;
2300         int                     rc = 0;
2301         ENTRY;
2302
2303         LASSERT(lfsck->li_master);
2304
2305         spin_lock(&lfsck->li_lock);
2306         com = __lfsck_component_find(lfsck, type, &lfsck->li_list_scan);
2307         if (com == NULL)
2308                 com = __lfsck_component_find(lfsck, type,
2309                                              &lfsck->li_list_double_scan);
2310         if (com != NULL)
2311                 lfsck_component_get(com);
2312         spin_unlock(&lfsck->li_lock);
2313
2314         if (com != NULL) {
2315                 struct lfsck_thread_info          *info  = lfsck_env_info(env);
2316                 struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2317                 struct lfsck_request              *lr    = &info->lti_lr;
2318                 struct lfsck_assistant_data       *lad   = com->lc_data;
2319                 struct list_head                  *list;
2320                 struct list_head                  *phase_list;
2321                 struct ptlrpc_request_set         *set;
2322
2323                 set = ptlrpc_prep_set();
2324                 if (set == NULL) {
2325                         lfsck_component_put(env, com);
2326
2327                         RETURN(-ENOMEM);
2328                 }
2329
2330                 if (type == LFSCK_TYPE_LAYOUT) {
2331                         list = &ltd->ltd_layout_list;
2332                         phase_list = &ltd->ltd_layout_phase_list;
2333                 } else {
2334                         list = &ltd->ltd_namespace_list;
2335                         phase_list = &ltd->ltd_namespace_phase_list;
2336                 }
2337
2338                 spin_lock(&ltds->ltd_lock);
2339                 if (list_empty(list)) {
2340                         LASSERT(list_empty(phase_list));
2341                         spin_unlock(&ltds->ltd_lock);
2342                         ptlrpc_set_destroy(set);
2343
2344                         RETURN(0);
2345                 }
2346
2347                 list_del_init(phase_list);
2348                 list_del_init(list);
2349                 spin_unlock(&ltds->ltd_lock);
2350
2351                 memset(lr, 0, sizeof(*lr));
2352                 lr->lr_index = lfsck_dev_idx(lfsck);
2353                 lr->lr_event = LE_PEER_EXIT;
2354                 lr->lr_active = type;
2355                 lr->lr_status = LS_CO_PAUSED;
2356                 if (ltds == &lfsck->li_ost_descs)
2357                         lr->lr_flags = LEF_TO_OST;
2358
2359                 memset(laia, 0, sizeof(*laia));
2360                 laia->laia_com = com;
2361                 laia->laia_ltds = ltds;
2362                 atomic_inc(&ltd->ltd_ref);
2363                 laia->laia_ltd = ltd;
2364                 laia->laia_lr = lr;
2365
2366                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2367                                          lfsck_async_interpret_common,
2368                                          laia, LFSCK_NOTIFY);
2369                 if (rc != 0) {
2370                         CDEBUG(D_LFSCK, "%s: fail to notify %s %x for "
2371                                "co-stop for %s: rc = %d\n",
2372                                lfsck_lfsck2name(lfsck),
2373                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2374                                ltd->ltd_index, lad->lad_name, rc);
2375                         lfsck_tgt_put(ltd);
2376                 } else {
2377                         rc = ptlrpc_set_wait(env, set);
2378                 }
2379
2380                 ptlrpc_set_destroy(set);
2381                 lfsck_component_put(env, com);
2382         }
2383
2384         RETURN(rc);
2385 }
2386
2387 static int lfsck_async_interpret(const struct lu_env *env,
2388                                  struct ptlrpc_request *req,
2389                                  void *args, int rc)
2390 {
2391         struct lfsck_async_interpret_args *laia = args;
2392         struct lfsck_instance             *lfsck;
2393
2394         lfsck = container_of0(laia->laia_ltds, struct lfsck_instance,
2395                               li_mdt_descs);
2396         lfsck_interpret(env, lfsck, req, laia, rc);
2397         lfsck_tgt_put(laia->laia_ltd);
2398         if (rc != 0 && laia->laia_result != -EALREADY)
2399                 laia->laia_result = rc;
2400
2401         return 0;
2402 }
2403
2404 int lfsck_async_request(const struct lu_env *env, struct obd_export *exp,
2405                         struct lfsck_request *lr,
2406                         struct ptlrpc_request_set *set,
2407                         ptlrpc_interpterer_t interpreter,
2408                         void *args, int request)
2409 {
2410         struct lfsck_async_interpret_args *laia;
2411         struct ptlrpc_request             *req;
2412         struct lfsck_request              *tmp;
2413         struct req_format                 *format;
2414         int                                rc;
2415
2416         switch (request) {
2417         case LFSCK_NOTIFY:
2418                 format = &RQF_LFSCK_NOTIFY;
2419                 break;
2420         case LFSCK_QUERY:
2421                 format = &RQF_LFSCK_QUERY;
2422                 break;
2423         default:
2424                 CDEBUG(D_LFSCK, "%s: unknown async request %d: rc = %d\n",
2425                        exp->exp_obd->obd_name, request, -EINVAL);
2426                 return -EINVAL;
2427         }
2428
2429         req = ptlrpc_request_alloc(class_exp2cliimp(exp), format);
2430         if (req == NULL)
2431                 return -ENOMEM;
2432
2433         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, request);
2434         if (rc != 0) {
2435                 ptlrpc_request_free(req);
2436
2437                 return rc;
2438         }
2439
2440         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2441         *tmp = *lr;
2442         ptlrpc_request_set_replen(req);
2443
2444         laia = ptlrpc_req_async_args(laia, req);
2445         *laia = *(struct lfsck_async_interpret_args *)args;
2446         if (laia->laia_com != NULL)
2447                 lfsck_component_get(laia->laia_com);
2448         req->rq_interpret_reply = interpreter;
2449         req->rq_allow_intr = 1;
2450         req->rq_no_delay = 1;
2451         ptlrpc_set_add_req(set, req);
2452
2453         return 0;
2454 }
2455
2456 int lfsck_query_all(const struct lu_env *env, struct lfsck_component *com)
2457 {
2458         struct lfsck_thread_info          *info  = lfsck_env_info(env);
2459         struct lfsck_request              *lr    = &info->lti_lr;
2460         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
2461         struct lfsck_instance             *lfsck = com->lc_lfsck;
2462         struct lfsck_tgt_descs            *ltds  = &lfsck->li_mdt_descs;
2463         struct lfsck_tgt_desc             *ltd;
2464         struct ptlrpc_request_set         *set;
2465         int                                idx;
2466         int                                rc;
2467         ENTRY;
2468
2469         memset(lr, 0, sizeof(*lr));
2470         lr->lr_event = LE_QUERY;
2471         lr->lr_active = com->lc_type;
2472         lr->lr_flags = LEF_QUERY_ALL;
2473
2474         memset(laia, 0, sizeof(*laia));
2475         laia->laia_com = com;
2476         laia->laia_lr = lr;
2477
2478         set = ptlrpc_prep_set();
2479         if (set == NULL)
2480                 RETURN(-ENOMEM);
2481
2482 again:
2483         laia->laia_ltds = ltds;
2484         down_read(&ltds->ltd_rw_sem);
2485         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2486                 ltd = lfsck_tgt_get(ltds, idx);
2487                 LASSERT(ltd != NULL);
2488
2489                 laia->laia_ltd = ltd;
2490                 up_read(&ltds->ltd_rw_sem);
2491                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2492                                          lfsck_async_interpret_common,
2493                                          laia, LFSCK_QUERY);
2494                 if (rc != 0) {
2495                         struct lfsck_assistant_data *lad = com->lc_data;
2496
2497                         CDEBUG(D_LFSCK, "%s: Fail to query %s %x for stat %s: "
2498                                "rc = %d\n", lfsck_lfsck2name(lfsck),
2499                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
2500                                ltd->ltd_index, lad->lad_name, rc);
2501                         lfsck_reset_ltd_status(ltd, com->lc_type);
2502                         lfsck_tgt_put(ltd);
2503                 }
2504                 down_read(&ltds->ltd_rw_sem);
2505         }
2506         up_read(&ltds->ltd_rw_sem);
2507
2508         if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST)) {
2509                 ltds = &lfsck->li_ost_descs;
2510                 lr->lr_flags |= LEF_TO_OST;
2511                 goto again;
2512         }
2513
2514         rc = ptlrpc_set_wait(env, set);
2515         ptlrpc_set_destroy(set);
2516
2517         RETURN(rc);
2518 }
2519
2520 int lfsck_start_assistant(const struct lu_env *env, struct lfsck_component *com,
2521                           struct lfsck_start_param *lsp)
2522 {
2523         struct lfsck_instance           *lfsck   = com->lc_lfsck;
2524         struct lfsck_assistant_data     *lad     = com->lc_data;
2525         struct ptlrpc_thread            *mthread = &lfsck->li_thread;
2526         struct ptlrpc_thread            *athread = &lad->lad_thread;
2527         struct lfsck_thread_args        *lta;
2528         struct task_struct              *task;
2529         int                              rc;
2530         ENTRY;
2531
2532         lad->lad_assistant_status = 0;
2533         lad->lad_post_result = 0;
2534         lad->lad_flags = 0;
2535         lad->lad_advance_lock = false;
2536         thread_set_flags(athread, 0);
2537
2538         lta = lfsck_thread_args_init(lfsck, com, lsp);
2539         if (IS_ERR(lta))
2540                 RETURN(PTR_ERR(lta));
2541
2542         task = kthread_run(lfsck_assistant_engine, lta, lad->lad_name);
2543         if (IS_ERR(task)) {
2544                 rc = PTR_ERR(task);
2545                 CERROR("%s: cannot start LFSCK assistant thread for %s: "
2546                        "rc = %d\n", lfsck_lfsck2name(lfsck), lad->lad_name, rc);
2547                 lfsck_thread_args_fini(lta);
2548         } else {
2549                 wait_event_idle(mthread->t_ctl_waitq,
2550                                 thread_is_running(athread) ||
2551                                 thread_is_stopped(athread) ||
2552                                 !thread_is_starting(mthread));
2553                 if (unlikely(!thread_is_starting(mthread)))
2554                         /* stopped by race */
2555                         rc = -ESRCH;
2556                 else if (unlikely(!thread_is_running(athread)))
2557                         rc = lad->lad_assistant_status;
2558                 else
2559                         rc = 0;
2560         }
2561
2562         RETURN(rc);
2563 }
2564
2565 int lfsck_checkpoint_generic(const struct lu_env *env,
2566                              struct lfsck_component *com)
2567 {
2568         struct lfsck_assistant_data     *lad     = com->lc_data;
2569         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2570         struct ptlrpc_thread            *athread = &lad->lad_thread;
2571
2572         wait_event_idle(mthread->t_ctl_waitq,
2573                         list_empty(&lad->lad_req_list) ||
2574                         !thread_is_running(mthread) ||
2575                         thread_is_stopped(athread));
2576
2577         if (!thread_is_running(mthread) || thread_is_stopped(athread))
2578                 return LFSCK_CHECKPOINT_SKIP;
2579
2580         return 0;
2581 }
2582
2583 void lfsck_post_generic(const struct lu_env *env,
2584                         struct lfsck_component *com, int *result)
2585 {
2586         struct lfsck_assistant_data     *lad     = com->lc_data;
2587         struct ptlrpc_thread            *athread = &lad->lad_thread;
2588         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2589
2590         lad->lad_post_result = *result;
2591         if (*result <= 0)
2592                 set_bit(LAD_EXIT, &lad->lad_flags);
2593         set_bit(LAD_TO_POST, &lad->lad_flags);
2594
2595         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s post, rc = %d\n",
2596                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2597
2598         wake_up_all(&athread->t_ctl_waitq);
2599         wait_event_idle(mthread->t_ctl_waitq,
2600                         (*result > 0 && list_empty(&lad->lad_req_list)) ||
2601                         thread_is_stopped(athread));
2602
2603         if (lad->lad_assistant_status < 0)
2604                 *result = lad->lad_assistant_status;
2605
2606         CDEBUG(D_LFSCK, "%s: the assistant has done %s post, rc = %d\n",
2607                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, *result);
2608 }
2609
2610 int lfsck_double_scan_generic(const struct lu_env *env,
2611                               struct lfsck_component *com, int status)
2612 {
2613         struct lfsck_assistant_data     *lad     = com->lc_data;
2614         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2615         struct ptlrpc_thread            *athread = &lad->lad_thread;
2616
2617         if (status != LS_SCANNING_PHASE2)
2618                 set_bit(LAD_EXIT, &lad->lad_flags);
2619         else
2620                 set_bit(LAD_TO_DOUBLE_SCAN, &lad->lad_flags);
2621
2622         CDEBUG(D_LFSCK, "%s: waiting for assistant to do %s double_scan, "
2623                "status %d\n",
2624                lfsck_lfsck2name(com->lc_lfsck), lad->lad_name, status);
2625
2626         wake_up_all(&athread->t_ctl_waitq);
2627         wait_event_idle(mthread->t_ctl_waitq,
2628                         test_bit(LAD_IN_DOUBLE_SCAN, &lad->lad_flags) ||
2629                         thread_is_stopped(athread));
2630
2631         CDEBUG(D_LFSCK, "%s: the assistant has done %s double_scan, "
2632                "status %d\n", lfsck_lfsck2name(com->lc_lfsck), lad->lad_name,
2633                lad->lad_assistant_status);
2634
2635         if (lad->lad_assistant_status < 0)
2636                 return lad->lad_assistant_status;
2637
2638         return 0;
2639 }
2640
2641 void lfsck_quit_generic(const struct lu_env *env,
2642                         struct lfsck_component *com)
2643 {
2644         struct lfsck_assistant_data     *lad     = com->lc_data;
2645         struct ptlrpc_thread            *mthread = &com->lc_lfsck->li_thread;
2646         struct ptlrpc_thread            *athread = &lad->lad_thread;
2647
2648         set_bit(LAD_EXIT, &lad->lad_flags);
2649         wake_up_all(&athread->t_ctl_waitq);
2650         wait_event_idle(mthread->t_ctl_waitq,
2651                         thread_is_init(athread) ||
2652                         thread_is_stopped(athread));
2653 }
2654
2655 int lfsck_load_one_trace_file(const struct lu_env *env,
2656                               struct lfsck_component *com,
2657                               struct dt_object *parent,
2658                               struct dt_object **child,
2659                               const struct dt_index_features *ft,
2660                               const char *name, bool reset)
2661 {
2662         struct lfsck_instance *lfsck = com->lc_lfsck;
2663         struct dt_object *obj;
2664         int rc;
2665         ENTRY;
2666
2667         if (*child != NULL) {
2668                 struct dt_it *it;
2669                 const struct dt_it_ops *iops;
2670                 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid3;
2671
2672                 if (!reset)
2673                         RETURN(0);
2674
2675                 obj = *child;
2676                 rc = obj->do_ops->do_index_try(env, obj, ft);
2677                 if (rc)
2678                         /* unlink by force */
2679                         goto unlink;
2680
2681                 iops = &obj->do_index_ops->dio_it;
2682                 it = iops->init(env, obj, 0);
2683                 if (IS_ERR(it))
2684                         /* unlink by force */
2685                         goto unlink;
2686
2687                 fid_zero(fid);
2688                 rc = iops->get(env, it, (const struct dt_key *)fid);
2689                 if (rc >= 0) {
2690                         rc = iops->next(env, it);
2691                         iops->put(env, it);
2692                 }
2693                 iops->fini(env, it);
2694                 if (rc > 0)
2695                         /* "rc > 0" means the index file is empty. */
2696                         RETURN(0);
2697
2698 unlink:
2699                 /* The old index is not empty, remove it firstly. */
2700                 rc = local_object_unlink(env, lfsck->li_bottom, parent, name);
2701                 CDEBUG_LIMIT(rc ? D_ERROR : D_LFSCK,
2702                              "%s: unlink lfsck sub trace file %s: rc = %d\n",
2703                              lfsck_lfsck2name(com->lc_lfsck), name, rc);
2704                 if (rc)
2705                         RETURN(rc);
2706
2707                 if (*child) {
2708                         lfsck_object_put(env, *child);
2709                         *child = NULL;
2710                 }
2711         } else if (reset) {
2712                 goto unlink;
2713         }
2714
2715         obj = local_index_find_or_create(env, lfsck->li_los, parent, name,
2716                                          S_IFREG | S_IRUGO | S_IWUSR, ft);
2717         if (IS_ERR(obj))
2718                 RETURN(PTR_ERR(obj));
2719
2720         rc = obj->do_ops->do_index_try(env, obj, ft);
2721         if (rc) {
2722                 lfsck_object_put(env, obj);
2723                 CDEBUG(D_LFSCK, "%s: LFSCK fail to load "
2724                        "sub trace file %s: rc = %d\n",
2725                        lfsck_lfsck2name(com->lc_lfsck), name, rc);
2726         } else {
2727                 *child = obj;
2728         }
2729
2730         RETURN(rc);
2731 }
2732
2733 int lfsck_load_sub_trace_files(const struct lu_env *env,
2734                                struct lfsck_component *com,
2735                                const struct dt_index_features *ft,
2736                                const char *prefix, bool reset)
2737 {
2738         char *name = lfsck_env_info(env)->lti_key;
2739         struct lfsck_sub_trace_obj *lsto;
2740         int rc;
2741         int i;
2742
2743         for (i = 0, rc = 0, lsto = &com->lc_sub_trace_objs[0];
2744              i < LFSCK_STF_COUNT && rc == 0; i++, lsto++) {
2745                 snprintf(name, NAME_MAX, "%s_%02d", prefix, i);
2746                 rc = lfsck_load_one_trace_file(env, com,
2747                                 com->lc_lfsck->li_lfsck_dir,
2748                                 &lsto->lsto_obj, ft, name, reset);
2749         }
2750
2751         return rc;
2752 }
2753
2754 /* external interfaces */
2755 int lfsck_get_speed(char *buf, struct dt_device *key)
2756 {
2757         struct lu_env           env;
2758         struct lfsck_instance  *lfsck;
2759         int                     rc;
2760         ENTRY;
2761
2762         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2763         if (rc != 0)
2764                 RETURN(rc);
2765
2766         lfsck = lfsck_instance_find(key, true, false);
2767         if (lfsck && buf) {
2768                 rc = sprintf(buf, "%u\n",
2769                              lfsck->li_bookmark_ram.lb_speed_limit);
2770                 lfsck_instance_put(&env, lfsck);
2771         } else {
2772                 rc = -ENXIO;
2773         }
2774
2775         lu_env_fini(&env);
2776
2777         RETURN(rc);
2778 }
2779 EXPORT_SYMBOL(lfsck_get_speed);
2780
2781 int lfsck_set_speed(struct dt_device *key, __u32 val)
2782 {
2783         struct lu_env           env;
2784         struct lfsck_instance  *lfsck;
2785         int                     rc;
2786         ENTRY;
2787
2788         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2789         if (rc != 0)
2790                 RETURN(rc);
2791
2792         lfsck = lfsck_instance_find(key, true, false);
2793         if (likely(lfsck != NULL)) {
2794                 mutex_lock(&lfsck->li_mutex);
2795                 if (__lfsck_set_speed(lfsck, val))
2796                         rc = lfsck_bookmark_store(&env, lfsck);
2797                 mutex_unlock(&lfsck->li_mutex);
2798                 lfsck_instance_put(&env, lfsck);
2799         } else {
2800                 rc = -ENXIO;
2801         }
2802
2803         lu_env_fini(&env);
2804
2805         RETURN(rc);
2806 }
2807 EXPORT_SYMBOL(lfsck_set_speed);
2808
2809 int lfsck_get_windows(char *buf, struct dt_device *key)
2810 {
2811         struct lu_env           env;
2812         struct lfsck_instance  *lfsck;
2813         int                     rc;
2814         ENTRY;
2815
2816         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2817         if (rc != 0)
2818                 RETURN(rc);
2819
2820         lfsck = lfsck_instance_find(key, true, false);
2821         if (likely(lfsck != NULL)) {
2822                 rc = sprintf(buf, "%u\n",
2823                              lfsck->li_bookmark_ram.lb_async_windows);
2824                 lfsck_instance_put(&env, lfsck);
2825         } else {
2826                 rc = -ENXIO;
2827         }
2828
2829         lu_env_fini(&env);
2830
2831         RETURN(rc);
2832 }
2833 EXPORT_SYMBOL(lfsck_get_windows);
2834
2835 int lfsck_set_windows(struct dt_device *key, unsigned int val)
2836 {
2837         struct lu_env           env;
2838         struct lfsck_instance  *lfsck;
2839         int                     rc;
2840         ENTRY;
2841
2842         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2843         if (rc != 0)
2844                 RETURN(rc);
2845
2846         lfsck = lfsck_instance_find(key, true, false);
2847         if (likely(lfsck != NULL)) {
2848                 if (val < 1 || val > LFSCK_ASYNC_WIN_MAX) {
2849                         CWARN("%s: invalid async windows size that may "
2850                               "cause memory issues. The valid range is "
2851                               "[1 - %u].\n",
2852                               lfsck_lfsck2name(lfsck), LFSCK_ASYNC_WIN_MAX);
2853                         rc = -EINVAL;
2854                 } else if (lfsck->li_bookmark_ram.lb_async_windows != val) {
2855                         mutex_lock(&lfsck->li_mutex);
2856                         lfsck->li_bookmark_ram.lb_async_windows = val;
2857                         rc = lfsck_bookmark_store(&env, lfsck);
2858                         mutex_unlock(&lfsck->li_mutex);
2859                 }
2860                 lfsck_instance_put(&env, lfsck);
2861         } else {
2862                 rc = -ENXIO;
2863         }
2864
2865         lu_env_fini(&env);
2866
2867         RETURN(rc);
2868 }
2869 EXPORT_SYMBOL(lfsck_set_windows);
2870
2871 int lfsck_dump(struct seq_file *m, struct dt_device *key, enum lfsck_type type)
2872 {
2873         struct lu_env           env;
2874         struct lfsck_instance  *lfsck;
2875         struct lfsck_component *com;
2876         int                     rc;
2877         ENTRY;
2878
2879         rc = lu_env_init(&env, LCT_MD_THREAD | LCT_DT_THREAD);
2880         if (rc != 0)
2881                 RETURN(rc);
2882
2883         lfsck = lfsck_instance_find(key, true, false);
2884         if (likely(lfsck != NULL)) {
2885                 com = lfsck_component_find(lfsck, type);
2886                 if (likely(com != NULL)) {
2887                         com->lc_ops->lfsck_dump(&env, com, m);
2888                         lfsck_component_put(&env, com);
2889                 } else {
2890                         rc = -ENOTSUPP;
2891                 }
2892
2893                 lfsck_instance_put(&env, lfsck);
2894         } else {
2895                 rc = -ENXIO;
2896         }
2897
2898         lu_env_fini(&env);
2899
2900         RETURN(rc);
2901 }
2902 EXPORT_SYMBOL(lfsck_dump);
2903
2904 static int lfsck_stop_all(const struct lu_env *env,
2905                           struct lfsck_instance *lfsck,
2906                           struct lfsck_stop *stop)
2907 {
2908         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2909         struct lfsck_request              *lr     = &info->lti_lr;
2910         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2911         struct ptlrpc_request_set         *set;
2912         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2913         struct lfsck_tgt_desc             *ltd;
2914         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2915         __u32                              idx;
2916         int                                rc     = 0;
2917         int                                rc1    = 0;
2918         ENTRY;
2919
2920         LASSERT(stop->ls_flags & LPF_BROADCAST);
2921
2922         set = ptlrpc_prep_set();
2923         if (unlikely(set == NULL))
2924                 RETURN(-ENOMEM);
2925
2926         memset(lr, 0, sizeof(*lr));
2927         lr->lr_event = LE_STOP;
2928         lr->lr_index = lfsck_dev_idx(lfsck);
2929         lr->lr_status = stop->ls_status;
2930         lr->lr_version = bk->lb_version;
2931         lr->lr_active = LFSCK_TYPES_ALL;
2932         lr->lr_param = stop->ls_flags;
2933
2934         memset(laia, 0, sizeof(*laia));
2935         laia->laia_ltds = ltds;
2936         laia->laia_lr = lr;
2937         laia->laia_shared = 1;
2938
2939         down_read(&ltds->ltd_rw_sem);
2940         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
2941                 ltd = lfsck_tgt_get(ltds, idx);
2942                 LASSERT(ltd != NULL);
2943
2944                 laia->laia_ltd = ltd;
2945                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
2946                                          lfsck_async_interpret, laia,
2947                                          LFSCK_NOTIFY);
2948                 if (rc != 0) {
2949                         lfsck_interpret(env, lfsck, NULL, laia, rc);
2950                         lfsck_tgt_put(ltd);
2951                         CERROR("%s: cannot notify MDT %x for LFSCK stop: "
2952                                "rc = %d\n", lfsck_lfsck2name(lfsck), idx, rc);
2953                         rc1 = rc;
2954                 }
2955         }
2956         up_read(&ltds->ltd_rw_sem);
2957
2958         rc = ptlrpc_set_wait(env, set);
2959         ptlrpc_set_destroy(set);
2960
2961         if (rc == 0)
2962                 rc = laia->laia_result;
2963
2964         if (rc == -EALREADY)
2965                 rc = 0;
2966
2967         if (rc != 0)
2968                 CERROR("%s: fail to stop LFSCK on some MDTs: rc = %d\n",
2969                        lfsck_lfsck2name(lfsck), rc);
2970
2971         RETURN(rc != 0 ? rc : rc1);
2972 }
2973
2974 static int lfsck_start_all(const struct lu_env *env,
2975                            struct lfsck_instance *lfsck,
2976                            struct lfsck_start *start)
2977 {
2978         struct lfsck_thread_info          *info   = lfsck_env_info(env);
2979         struct lfsck_request              *lr     = &info->lti_lr;
2980         struct lfsck_async_interpret_args *laia   = &info->lti_laia;
2981         struct ptlrpc_request_set         *set;
2982         struct lfsck_tgt_descs            *ltds   = &lfsck->li_mdt_descs;
2983         struct lfsck_tgt_desc             *ltd;
2984         struct lfsck_bookmark             *bk     = &lfsck->li_bookmark_ram;
2985         __u32                              idx;
2986         int                                rc     = 0;
2987         bool retry = false;
2988         ENTRY;
2989
2990         LASSERT(start->ls_flags & LPF_BROADCAST);
2991
2992         memset(lr, 0, sizeof(*lr));
2993         lr->lr_event = LE_START;
2994         lr->lr_index = lfsck_dev_idx(lfsck);
2995         lr->lr_speed = bk->lb_speed_limit;
2996         lr->lr_version = bk->lb_version;
2997         lr->lr_active = start->ls_active;
2998         lr->lr_param = start->ls_flags;
2999         lr->lr_async_windows = bk->lb_async_windows;
3000         lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
3001                        LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ |
3002                        LSV_CREATE_MDTOBJ;
3003
3004         memset(laia, 0, sizeof(*laia));
3005         laia->laia_ltds = ltds;
3006         laia->laia_lr = lr;
3007         laia->laia_shared = 1;
3008
3009 again:
3010         set = ptlrpc_prep_set();
3011         if (unlikely(!set))
3012                 RETURN(-ENOMEM);
3013
3014         down_read(&ltds->ltd_rw_sem);
3015         cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
3016                 ltd = lfsck_tgt_get(ltds, idx);
3017                 LASSERT(ltd != NULL);
3018
3019                 if (retry && !ltd->ltd_retry_start) {
3020                         lfsck_tgt_put(ltd);
3021                         continue;
3022                 }
3023
3024                 laia->laia_ltd = ltd;
3025                 ltd->ltd_retry_start = 0;
3026                 ltd->ltd_layout_done = 0;
3027                 ltd->ltd_namespace_done = 0;
3028                 ltd->ltd_synced_failures = 0;
3029                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
3030                                          lfsck_async_interpret, laia,
3031                                          LFSCK_NOTIFY);
3032                 if (rc != 0) {
3033                         lfsck_interpret(env, lfsck, NULL, laia, rc);
3034                         lfsck_tgt_put(ltd);
3035                         CERROR("%s: cannot notify MDT %x for LFSCK "
3036                                "start, failout: rc = %d\n",
3037                                lfsck_lfsck2name(lfsck), idx, rc);
3038                         break;
3039                 }
3040         }
3041         up_read(&ltds->ltd_rw_sem);
3042
3043         if (rc != 0) {
3044                 ptlrpc_set_destroy(set);
3045
3046                 RETURN(rc);
3047         }
3048
3049         rc = ptlrpc_set_wait(env, set);
3050         ptlrpc_set_destroy(set);
3051
3052         if (rc == 0)
3053                 rc = laia->laia_result;
3054
3055         if (unlikely(rc == -EINPROGRESS)) {
3056                 retry = true;
3057                 set_current_state(TASK_INTERRUPTIBLE);
3058                 schedule_timeout(cfs_time_seconds(1));
3059                 set_current_state(TASK_RUNNING);
3060                 if (!signal_pending(current) &&
3061                     thread_is_running(&lfsck->li_thread))
3062                         goto again;
3063
3064                 rc = -EINTR;
3065         }
3066
3067         if (rc != 0) {
3068                 struct lfsck_stop *stop = &info->lti_stop;
3069
3070                 CERROR("%s: cannot start LFSCK on some MDTs, "
3071                        "stop all: rc = %d\n",
3072                        lfsck_lfsck2name(lfsck), rc);
3073                 if (rc != -EALREADY) {
3074                         stop->ls_status = LS_FAILED;
3075                         stop->ls_flags = LPF_ALL_TGT | LPF_BROADCAST;
3076                         lfsck_stop_all(env, lfsck, stop);
3077                 }
3078         }
3079
3080         RETURN(rc);
3081 }
3082
3083 int lfsck_start(const struct lu_env *env, struct dt_device *key,
3084                 struct lfsck_start_param *lsp)
3085 {
3086         struct lfsck_start              *start  = lsp->lsp_start;
3087         struct lfsck_instance           *lfsck;
3088         struct lfsck_bookmark           *bk;
3089         struct ptlrpc_thread            *thread;
3090         struct lfsck_component          *com;
3091         struct lfsck_thread_args        *lta;
3092         struct task_struct              *task;
3093         struct lfsck_tgt_descs          *ltds;
3094         struct lfsck_tgt_desc           *ltd;
3095         __u32                            idx;
3096         int                              rc     = 0;
3097         __u16                            valid  = 0;
3098         __u16                            flags  = 0;
3099         __u16                            type   = 1;
3100         ENTRY;
3101
3102         if (key->dd_rdonly)
3103                 RETURN(-EROFS);
3104
3105         lfsck = lfsck_instance_find(key, true, false);
3106         if (unlikely(lfsck == NULL))
3107                 RETURN(-ENXIO);
3108
3109         if (unlikely(lfsck->li_stopping))
3110                 GOTO(put, rc = -ENXIO);
3111
3112         /* System is not ready, try again later. */
3113         if (unlikely(lfsck->li_namespace == NULL ||
3114                      lfsck_dev_site(lfsck)->ss_server_fld == NULL))
3115                 GOTO(put, rc = -EINPROGRESS);
3116
3117         /* start == NULL means auto trigger paused LFSCK. */
3118         if (!start) {
3119                 if (list_empty(&lfsck->li_list_scan) ||
3120                     OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_AUTO))
3121                         GOTO(put, rc = 0);
3122         } else if (start->ls_flags & LPF_BROADCAST && !lfsck->li_master) {
3123                 CERROR("%s: only allow to specify '-A | -o' via MDS\n",
3124                        lfsck_lfsck2name(lfsck));
3125
3126                 GOTO(put, rc = -EPERM);
3127         }
3128
3129         bk = &lfsck->li_bookmark_ram;
3130         thread = &lfsck->li_thread;
3131         mutex_lock(&lfsck->li_mutex);
3132         spin_lock(&lfsck->li_lock);
3133         if (unlikely(thread_is_stopping(thread))) {
3134                 /* Someone is stopping the LFSCK. */
3135                 spin_unlock(&lfsck->li_lock);
3136                 GOTO(out, rc = -EBUSY);
3137         }
3138
3139         if (!thread_is_init(thread) && !thread_is_stopped(thread)) {
3140                 rc = -EALREADY;
3141                 if (unlikely(start == NULL)) {
3142                         spin_unlock(&lfsck->li_lock);
3143                         GOTO(out, rc);
3144                 }
3145
3146                 while (start->ls_active != 0) {
3147                         if (!(type & start->ls_active)) {
3148                                 type <<= 1;
3149                                 continue;
3150                         }
3151
3152                         com = __lfsck_component_find(lfsck, type,
3153                                                      &lfsck->li_list_scan);
3154                         if (com == NULL)
3155                                 com = __lfsck_component_find(lfsck, type,
3156                                                 &lfsck->li_list_double_scan);
3157                         if (com == NULL) {
3158                                 rc = -EOPNOTSUPP;
3159                                 break;
3160                         }
3161
3162                         if (com->lc_ops->lfsck_join != NULL) {
3163                                 rc = com->lc_ops->lfsck_join( env, com, lsp);
3164                                 if (rc != 0 && rc != -EALREADY)
3165                                         break;
3166                         }
3167                         start->ls_active &= ~type;
3168                         type <<= 1;
3169                 }
3170                 spin_unlock(&lfsck->li_lock);
3171                 GOTO(out, rc);
3172         }
3173         spin_unlock(&lfsck->li_lock);
3174
3175         lfsck->li_status = 0;
3176         lfsck->li_oit_over = 0;
3177         lfsck->li_start_unplug = 0;
3178         lfsck->li_drop_dryrun = 0;
3179         lfsck->li_new_scanned = 0;
3180
3181         /* For auto trigger. */
3182         if (start == NULL)
3183                 goto trigger;
3184
3185         start->ls_version = bk->lb_version;
3186
3187         if (start->ls_active != 0) {
3188                 struct lfsck_component *next;
3189
3190                 if (start->ls_active == LFSCK_TYPES_ALL)
3191                         start->ls_active = LFSCK_TYPES_SUPPORTED;
3192
3193                 if (start->ls_active & ~LFSCK_TYPES_SUPPORTED) {
3194                         start->ls_active &= ~LFSCK_TYPES_SUPPORTED;
3195                         GOTO(out, rc = -ENOTSUPP);
3196                 }
3197
3198                 list_for_each_entry_safe(com, next,
3199                                          &lfsck->li_list_scan, lc_link) {
3200                         if (!(com->lc_type & start->ls_active)) {
3201                                 rc = com->lc_ops->lfsck_post(env, com, 0,
3202                                                              false);
3203                                 if (rc != 0)
3204                                         GOTO(out, rc);
3205                         }
3206                 }
3207