Whamcloud - gitweb
LU-8929 lfsck: dumper gets current position properly
[fs/lustre-release.git] / lustre / lfsck / lfsck_layout.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) 2014, 2016, Intel Corporation.
24  */
25 /*
26  * lustre/lfsck/lfsck_layout.c
27  *
28  * Author: Fan, Yong <fan.yong@intel.com>
29  */
30
31 #ifndef EXPORT_SYMTAB
32 # define EXPORT_SYMTAB
33 #endif
34 #define DEBUG_SUBSYSTEM S_LFSCK
35
36 #include <linux/bitops.h>
37 #include <linux/rbtree.h>
38
39 #include <lustre/lustre_idl.h>
40 #include <lu_object.h>
41 #include <dt_object.h>
42 #include <lustre_fid.h>
43 #include <lustre_lib.h>
44 #include <lustre_net.h>
45 #include <lustre/lustre_user.h>
46 #include <md_object.h>
47 #include <obd_class.h>
48
49 #include "lfsck_internal.h"
50
51 #define LFSCK_LAYOUT_MAGIC_V1           0xB173AE14
52 #define LFSCK_LAYOUT_MAGIC_V2           0xB1734D76
53
54 #define LFSCK_LAYOUT_MAGIC              LFSCK_LAYOUT_MAGIC_V2
55
56 struct lfsck_layout_seq {
57         struct list_head         lls_list;
58         __u64                    lls_seq;
59         __u64                    lls_lastid;
60         __u64                    lls_lastid_known;
61         struct dt_object        *lls_lastid_obj;
62         unsigned int             lls_dirty:1;
63 };
64
65 struct lfsck_layout_slave_target {
66         /* link into lfsck_layout_slave_data::llsd_master_list. */
67         struct list_head        llst_list;
68         /* The position for next record in the rbtree for iteration. */
69         struct lu_fid           llst_fid;
70         /* Dummy hash for iteration against the rbtree. */
71         __u64                   llst_hash;
72         __u64                   llst_gen;
73         atomic_t                llst_ref;
74         __u32                   llst_index;
75         /* How many times we have failed to get the master status. */
76         int                     llst_failures;
77 };
78
79 struct lfsck_layout_slave_data {
80         /* list for lfsck_layout_seq */
81         struct list_head         llsd_seq_list;
82
83         /* list for the masters involve layout verification. */
84         struct list_head         llsd_master_list;
85         spinlock_t               llsd_lock;
86         __u64                    llsd_touch_gen;
87         struct dt_object        *llsd_rb_obj;
88         struct rb_root           llsd_rb_root;
89         rwlock_t                 llsd_rb_lock;
90         unsigned int             llsd_rbtree_valid:1;
91 };
92
93 struct lfsck_layout_slave_async_args {
94         struct obd_export                *llsaa_exp;
95         struct lfsck_component           *llsaa_com;
96         struct lfsck_layout_slave_target *llsaa_llst;
97 };
98
99 static inline void
100 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
101 {
102         if (atomic_dec_and_test(&llst->llst_ref)) {
103                 LASSERT(list_empty(&llst->llst_list));
104
105                 OBD_FREE_PTR(llst);
106         }
107 }
108
109 static inline int
110 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
111 {
112         struct lfsck_layout_slave_target *llst;
113         struct lfsck_layout_slave_target *tmp;
114         int                               rc   = 0;
115
116         OBD_ALLOC_PTR(llst);
117         if (llst == NULL)
118                 return -ENOMEM;
119
120         INIT_LIST_HEAD(&llst->llst_list);
121         llst->llst_gen = 0;
122         llst->llst_index = index;
123         atomic_set(&llst->llst_ref, 1);
124
125         spin_lock(&llsd->llsd_lock);
126         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
127                 if (tmp->llst_index == index) {
128                         rc = -EALREADY;
129                         break;
130                 }
131         }
132         if (rc == 0)
133                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
134         spin_unlock(&llsd->llsd_lock);
135
136         if (rc != 0)
137                 OBD_FREE_PTR(llst);
138
139         return rc;
140 }
141
142 static inline void
143 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
144                       struct lfsck_layout_slave_target *llst)
145 {
146         bool del = false;
147
148         spin_lock(&llsd->llsd_lock);
149         if (!list_empty(&llst->llst_list)) {
150                 list_del_init(&llst->llst_list);
151                 del = true;
152         }
153         spin_unlock(&llsd->llsd_lock);
154
155         if (del)
156                 lfsck_layout_llst_put(llst);
157 }
158
159 static inline struct lfsck_layout_slave_target *
160 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
161                                __u32 index, bool unlink)
162 {
163         struct lfsck_layout_slave_target *llst;
164
165         spin_lock(&llsd->llsd_lock);
166         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
167                 if (llst->llst_index == index) {
168                         if (unlink)
169                                 list_del_init(&llst->llst_list);
170                         else
171                                 atomic_inc(&llst->llst_ref);
172                         spin_unlock(&llsd->llsd_lock);
173
174                         return llst;
175                 }
176         }
177         spin_unlock(&llsd->llsd_lock);
178
179         return NULL;
180 }
181
182 static struct lfsck_layout_req *
183 lfsck_layout_assistant_req_init(struct lfsck_assistant_object *lso,
184                                 struct dt_object *child, __u32 ost_idx,
185                                 __u32 lov_idx)
186 {
187         struct lfsck_layout_req *llr;
188
189         OBD_ALLOC_PTR(llr);
190         if (llr == NULL)
191                 return ERR_PTR(-ENOMEM);
192
193         INIT_LIST_HEAD(&llr->llr_lar.lar_list);
194         llr->llr_lar.lar_parent = lfsck_assistant_object_get(lso);
195         llr->llr_child = child;
196         llr->llr_ost_idx = ost_idx;
197         llr->llr_lov_idx = lov_idx;
198
199         return llr;
200 }
201
202 static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
203                                             struct lfsck_assistant_req *lar)
204 {
205         struct lfsck_layout_req *llr =
206                         container_of0(lar, struct lfsck_layout_req, llr_lar);
207
208         lfsck_object_put(env, llr->llr_child);
209         lfsck_assistant_object_put(env, lar->lar_parent);
210         OBD_FREE_PTR(llr);
211 }
212
213 static int
214 lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
215                                                struct ptlrpc_request *req,
216                                                void *args, int rc)
217 {
218         if (rc == 0) {
219                 struct lfsck_async_interpret_args *laia = args;
220                 struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
221
222                 ltd->ltd_synced_failures = 1;
223                 atomic_dec(laia->laia_count);
224         }
225
226         return 0;
227 }
228
229 /**
230  * Notify remote LFSCK instances about former failures.
231  *
232  * The local LFSCK instance has recorded which OSTs have ever failed to respond
233  * some LFSCK verification requests (maybe because of network issues or the OST
234  * itself trouble). During the respond gap, the OST may missed some OST-objects
235  * verification, then the OST cannot know whether related OST-objects have been
236  * referenced by related MDT-objects or not, then in the second-stage scanning,
237  * these OST-objects will be regarded as orphan, if the OST-object contains bad
238  * parent FID for back reference, then it will misguide the LFSCK to make wrong
239  * fixing for the fake orphan.
240  *
241  * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
242  * it will scan the bitmap for the ever failed OSTs, and notify them that they
243  * have ever missed some OST-object verification and should skip the handling
244  * for orphan OST-objects on all MDTs that are in the layout LFSCK.
245  *
246  * \param[in] env       pointer to the thread context
247  * \param[in] com       pointer to the lfsck component
248  * \param[in] lr        pointer to the lfsck request
249  */
250 static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
251                                                  struct lfsck_component *com,
252                                                  struct lfsck_request *lr)
253 {
254         struct lfsck_async_interpret_args *laia  =
255                                 &lfsck_env_info(env)->lti_laia2;
256         struct lfsck_assistant_data       *lad   = com->lc_data;
257         struct lfsck_layout               *lo    = com->lc_file_ram;
258         struct lfsck_instance             *lfsck = com->lc_lfsck;
259         struct lfsck_tgt_descs            *ltds  = &lfsck->li_ost_descs;
260         struct lfsck_tgt_desc             *ltd;
261         struct ptlrpc_request_set         *set;
262         atomic_t                           count;
263         __u32                              idx;
264         int                                rc    = 0;
265         ENTRY;
266
267         if (!lad->lad_incomplete || lo->ll_flags & LF_INCOMPLETE)
268                 RETURN_EXIT;
269
270         /* If the MDT has ever failed to verfiy some OST-objects,
271          * then sync failures with them firstly. */
272         lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
273
274         atomic_set(&count, 0);
275         memset(laia, 0, sizeof(*laia));
276         laia->laia_count = &count;
277         set = ptlrpc_prep_set();
278         if (set == NULL)
279                 GOTO(out, rc = -ENOMEM);
280
281         down_read(&ltds->ltd_rw_sem);
282         cfs_foreach_bit(lad->lad_bitmap, idx) {
283                 ltd = lfsck_ltd2tgt(ltds, idx);
284                 LASSERT(ltd != NULL);
285
286                 laia->laia_ltd = ltd;
287                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
288                                 lfsck_layout_assistant_sync_failures_interpret,
289                                 laia, LFSCK_NOTIFY);
290                 if (rc != 0) {
291                         CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
292                                "notify target %x for %s phase1 done: "
293                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
294                                ltd->ltd_index, lad->lad_name, rc);
295
296                         break;
297                 }
298
299                 atomic_inc(&count);
300         }
301         up_read(&ltds->ltd_rw_sem);
302
303         if (rc == 0 && atomic_read(&count) > 0)
304                 rc = ptlrpc_set_wait(set);
305
306         ptlrpc_set_destroy(set);
307
308         if (rc == 0 && atomic_read(&count) > 0)
309                 rc = -EINVAL;
310
311         GOTO(out, rc);
312
313 out:
314         if (rc != 0)
315                 /* If failed to sync failures with the OSTs, then have to
316                  * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
317                  * subsequent orphan OST-object handling. */
318                 lo->ll_flags |= LF_INCOMPLETE;
319
320         lr->lr_flags2 = lo->ll_flags;
321 }
322
323 static int lfsck_layout_get_lovea(const struct lu_env *env,
324                                   struct dt_object *obj, struct lu_buf *buf)
325 {
326         int rc;
327
328 again:
329         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
330         if (rc == -ERANGE) {
331                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
332                 if (rc <= 0)
333                         return rc;
334
335                 lu_buf_realloc(buf, rc);
336                 if (buf->lb_buf == NULL)
337                         return -ENOMEM;
338
339                 goto again;
340         }
341
342         if (rc == -ENODATA)
343                 rc = 0;
344
345         if (rc <= 0)
346                 return rc;
347
348         if (unlikely(buf->lb_buf == NULL)) {
349                 lu_buf_alloc(buf, rc);
350                 if (buf->lb_buf == NULL)
351                         return -ENOMEM;
352
353                 goto again;
354         }
355
356         return rc;
357 }
358
359 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
360 {
361         __u32 magic;
362         __u32 pattern;
363
364         magic = le32_to_cpu(lmm->lmm_magic);
365         /* If magic crashed, keep it there. Sometime later, during OST-object
366          * orphan handling, if some OST-object(s) back-point to it, it can be
367          * verified and repaired. */
368         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
369                 struct ost_id   oi;
370                 int             rc;
371
372                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
373                 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
374                         rc = -EOPNOTSUPP;
375                 else
376                         rc = -EINVAL;
377
378                 CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
379                        rc == -EINVAL ? "Unknown" : "Unsupported",
380                        magic, POSTID(&oi));
381
382                 return rc;
383         }
384
385         pattern = le32_to_cpu(lmm->lmm_pattern);
386         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
387         if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
388                 struct ost_id oi;
389
390                 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
391                 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
392                        pattern, POSTID(&oi));
393
394                 return -EOPNOTSUPP;
395         }
396
397         return 0;
398 }
399
400 #define LFSCK_RBTREE_BITMAP_SIZE        PAGE_SIZE
401 #define LFSCK_RBTREE_BITMAP_WIDTH       (LFSCK_RBTREE_BITMAP_SIZE << 3)
402 #define LFSCK_RBTREE_BITMAP_MASK        (LFSCK_RBTREE_BITMAP_WIDTH - 1)
403
404 struct lfsck_rbtree_node {
405         struct rb_node   lrn_node;
406         __u64            lrn_seq;
407         __u32            lrn_first_oid;
408         atomic_t         lrn_known_count;
409         atomic_t         lrn_accessed_count;
410         void            *lrn_known_bitmap;
411         void            *lrn_accessed_bitmap;
412 };
413
414 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
415                                    __u64 seq, __u32 oid)
416 {
417         if (seq < lrn->lrn_seq)
418                 return -1;
419
420         if (seq > lrn->lrn_seq)
421                 return 1;
422
423         if (oid < lrn->lrn_first_oid)
424                 return -1;
425
426         if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
427                 return 1;
428
429         return 0;
430 }
431
432 /* The caller should hold llsd->llsd_rb_lock. */
433 static struct lfsck_rbtree_node *
434 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
435                     const struct lu_fid *fid, bool *exact)
436 {
437         struct rb_node           *node  = llsd->llsd_rb_root.rb_node;
438         struct rb_node           *prev  = NULL;
439         struct lfsck_rbtree_node *lrn   = NULL;
440         int                       rc    = 0;
441
442         if (exact != NULL)
443                 *exact = true;
444
445         while (node != NULL) {
446                 prev = node;
447                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
448                 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
449                 if (rc < 0)
450                         node = node->rb_left;
451                 else if (rc > 0)
452                         node = node->rb_right;
453                 else
454                         return lrn;
455         }
456
457         if (exact == NULL)
458                 return NULL;
459
460         /* If there is no exactly matched one, then to the next valid one. */
461         *exact = false;
462
463         /* The rbtree is empty. */
464         if (rc == 0)
465                 return NULL;
466
467         if (rc < 0)
468                 return lrn;
469
470         node = rb_next(prev);
471
472         /* The end of the rbtree. */
473         if (node == NULL)
474                 return NULL;
475
476         lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
477
478         return lrn;
479 }
480
481 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
482                                                   const struct lu_fid *fid)
483 {
484         struct lfsck_rbtree_node *lrn;
485
486         OBD_ALLOC_PTR(lrn);
487         if (lrn == NULL)
488                 return ERR_PTR(-ENOMEM);
489
490         OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
491         if (lrn->lrn_known_bitmap == NULL) {
492                 OBD_FREE_PTR(lrn);
493
494                 return ERR_PTR(-ENOMEM);
495         }
496
497         OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
498         if (lrn->lrn_accessed_bitmap == NULL) {
499                 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
500                 OBD_FREE_PTR(lrn);
501
502                 return ERR_PTR(-ENOMEM);
503         }
504
505         RB_CLEAR_NODE(&lrn->lrn_node);
506         lrn->lrn_seq = fid_seq(fid);
507         lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
508         atomic_set(&lrn->lrn_known_count, 0);
509         atomic_set(&lrn->lrn_accessed_count, 0);
510
511         return lrn;
512 }
513
514 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
515 {
516         OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
517         OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
518         OBD_FREE_PTR(lrn);
519 }
520
521 /* The caller should hold lock. */
522 static struct lfsck_rbtree_node *
523 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
524                     struct lfsck_rbtree_node *lrn)
525 {
526         struct rb_node           **pos    = &llsd->llsd_rb_root.rb_node;
527         struct rb_node            *parent = NULL;
528         struct lfsck_rbtree_node  *tmp;
529         int                        rc;
530
531         while (*pos != NULL) {
532                 parent = *pos;
533                 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
534                 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
535                 if (rc < 0)
536                         pos = &(*pos)->rb_left;
537                 else if (rc > 0)
538                         pos = &(*pos)->rb_right;
539                 else
540                         return tmp;
541         }
542
543         rb_link_node(&lrn->lrn_node, parent, pos);
544         rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
545
546         return lrn;
547 }
548
549 extern const struct dt_index_operations lfsck_orphan_index_ops;
550
551 static int lfsck_rbtree_setup(const struct lu_env *env,
552                               struct lfsck_component *com)
553 {
554         struct lu_fid                   *fid    = &lfsck_env_info(env)->lti_fid;
555         struct lfsck_instance           *lfsck  = com->lc_lfsck;
556         struct dt_device                *dev    = lfsck->li_bottom;
557         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
558         struct dt_object                *obj;
559
560         fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
561         fid->f_oid = lfsck_dev_idx(lfsck);
562         fid->f_ver = 0;
563         obj = dt_locate(env, dev, fid);
564         if (IS_ERR(obj))
565                 RETURN(PTR_ERR(obj));
566
567         /* Generate an in-RAM object to stand for the layout rbtree.
568          * Scanning the layout rbtree will be via the iteration over
569          * the object. In the future, the rbtree may be written onto
570          * disk with the object.
571          *
572          * Mark the object to be as exist. */
573         obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
574         obj->do_index_ops = &lfsck_orphan_index_ops;
575         llsd->llsd_rb_obj = obj;
576         llsd->llsd_rbtree_valid = 1;
577         dev->dd_record_fid_accessed = 1;
578
579         CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
580                lfsck_lfsck2name(lfsck));
581
582         return 0;
583 }
584
585 static void lfsck_rbtree_cleanup(const struct lu_env *env,
586                                  struct lfsck_component *com)
587 {
588         struct lfsck_instance           *lfsck = com->lc_lfsck;
589         struct lfsck_layout_slave_data  *llsd  = com->lc_data;
590         struct rb_node                  *node  = rb_first(&llsd->llsd_rb_root);
591         struct rb_node                  *next;
592         struct lfsck_rbtree_node        *lrn;
593
594         lfsck->li_bottom->dd_record_fid_accessed = 0;
595         /* Invalid the rbtree, then no others will use it. */
596         write_lock(&llsd->llsd_rb_lock);
597         llsd->llsd_rbtree_valid = 0;
598         write_unlock(&llsd->llsd_rb_lock);
599
600         while (node != NULL) {
601                 next = rb_next(node);
602                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
603                 rb_erase(node, &llsd->llsd_rb_root);
604                 lfsck_rbtree_free(lrn);
605                 node = next;
606         }
607
608         if (llsd->llsd_rb_obj != NULL) {
609                 lfsck_object_put(env, llsd->llsd_rb_obj);
610                 llsd->llsd_rb_obj = NULL;
611         }
612
613         CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
614                lfsck_lfsck2name(lfsck));
615 }
616
617 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
618                                        struct lfsck_component *com,
619                                        const struct lu_fid *fid,
620                                        bool accessed)
621 {
622         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
623         struct lfsck_rbtree_node        *lrn;
624         bool                             insert = false;
625         int                              idx;
626         int                              rc     = 0;
627         ENTRY;
628
629         if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
630                 RETURN_EXIT;
631
632         if (!fid_is_idif(fid) && !fid_is_norm(fid))
633                 RETURN_EXIT;
634
635         read_lock(&llsd->llsd_rb_lock);
636         if (!llsd->llsd_rbtree_valid)
637                 GOTO(unlock, rc = 0);
638
639         lrn = lfsck_rbtree_search(llsd, fid, NULL);
640         if (lrn == NULL) {
641                 struct lfsck_rbtree_node *tmp;
642
643                 LASSERT(!insert);
644
645                 read_unlock(&llsd->llsd_rb_lock);
646                 tmp = lfsck_rbtree_new(env, fid);
647                 if (IS_ERR(tmp))
648                         GOTO(out, rc = PTR_ERR(tmp));
649
650                 insert = true;
651                 write_lock(&llsd->llsd_rb_lock);
652                 if (!llsd->llsd_rbtree_valid) {
653                         lfsck_rbtree_free(tmp);
654                         GOTO(unlock, rc = 0);
655                 }
656
657                 lrn = lfsck_rbtree_insert(llsd, tmp);
658                 if (lrn != tmp)
659                         lfsck_rbtree_free(tmp);
660         }
661
662         idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
663         /* Any accessed object must be a known object. */
664         if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
665                 atomic_inc(&lrn->lrn_known_count);
666         if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
667                 atomic_inc(&lrn->lrn_accessed_count);
668
669         GOTO(unlock, rc = 0);
670
671 unlock:
672         if (insert)
673                 write_unlock(&llsd->llsd_rb_lock);
674         else
675                 read_unlock(&llsd->llsd_rb_lock);
676 out:
677         if (rc != 0 && accessed) {
678                 struct lfsck_layout *lo = com->lc_file_ram;
679
680                 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
681                        "bitmap, and will cause incorrect LFSCK OST-object "
682                        "handling, so disable it to cancel orphan handling "
683                        "for related device. rc = %d\n",
684                        lfsck_lfsck2name(com->lc_lfsck), rc);
685
686                 lo->ll_flags |= LF_INCOMPLETE;
687                 lfsck_rbtree_cleanup(env, com);
688         }
689 }
690
691 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
692                                    const struct lfsck_layout *src)
693 {
694         int i;
695
696         des->ll_magic = le32_to_cpu(src->ll_magic);
697         des->ll_status = le32_to_cpu(src->ll_status);
698         des->ll_flags = le32_to_cpu(src->ll_flags);
699         des->ll_success_count = le32_to_cpu(src->ll_success_count);
700         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
701         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
702         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
703         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
704         des->ll_time_last_checkpoint =
705                                 le64_to_cpu(src->ll_time_last_checkpoint);
706         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
707         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
708         des->ll_pos_first_inconsistent =
709                         le64_to_cpu(src->ll_pos_first_inconsistent);
710         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
711         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
712         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
713         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
714         for (i = 0; i < LLIT_MAX; i++)
715                 des->ll_objs_repaired[i] =
716                                 le64_to_cpu(src->ll_objs_repaired[i]);
717         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
718         des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
719         fid_le_to_cpu(&des->ll_fid_latest_scanned_phase2,
720                       &src->ll_fid_latest_scanned_phase2);
721 }
722
723 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
724                                    const struct lfsck_layout *src)
725 {
726         int i;
727
728         des->ll_magic = cpu_to_le32(src->ll_magic);
729         des->ll_status = cpu_to_le32(src->ll_status);
730         des->ll_flags = cpu_to_le32(src->ll_flags);
731         des->ll_success_count = cpu_to_le32(src->ll_success_count);
732         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
733         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
734         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
735         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
736         des->ll_time_last_checkpoint =
737                                 cpu_to_le64(src->ll_time_last_checkpoint);
738         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
739         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
740         des->ll_pos_first_inconsistent =
741                         cpu_to_le64(src->ll_pos_first_inconsistent);
742         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
743         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
744         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
745         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
746         for (i = 0; i < LLIT_MAX; i++)
747                 des->ll_objs_repaired[i] =
748                                 cpu_to_le64(src->ll_objs_repaired[i]);
749         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
750         des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
751         fid_cpu_to_le(&des->ll_fid_latest_scanned_phase2,
752                       &src->ll_fid_latest_scanned_phase2);
753 }
754
755 /**
756  * Load the OST bitmap from the lfsck_layout trace file.
757  *
758  * \param[in] env       pointer to the thread context
759  * \param[in] com       pointer to the lfsck component
760  *
761  * \retval              0 for success
762  * \retval              negative error number on failure or data corruption
763  */
764 static int lfsck_layout_load_bitmap(const struct lu_env *env,
765                                     struct lfsck_component *com)
766 {
767         struct dt_object                *obj    = com->lc_obj;
768         struct lfsck_assistant_data     *lad    = com->lc_data;
769         struct lfsck_layout             *lo     = com->lc_file_ram;
770         struct cfs_bitmap                       *bitmap = lad->lad_bitmap;
771         loff_t                           pos    = com->lc_file_size;
772         ssize_t                          size;
773         __u32                            nbits;
774         int                              rc;
775         ENTRY;
776
777         if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
778             lo->ll_bitmap_size)
779                 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
780         else
781                 nbits = lo->ll_bitmap_size;
782
783         if (unlikely(nbits < BITS_PER_LONG))
784                 nbits = BITS_PER_LONG;
785
786         if (nbits > bitmap->size) {
787                 __u32 new_bits = bitmap->size;
788                 struct cfs_bitmap *new_bitmap;
789
790                 while (new_bits < nbits)
791                         new_bits <<= 1;
792
793                 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
794                 if (new_bitmap == NULL)
795                         RETURN(-ENOMEM);
796
797                 lad->lad_bitmap = new_bitmap;
798                 CFS_FREE_BITMAP(bitmap);
799                 bitmap = new_bitmap;
800         }
801
802         if (lo->ll_bitmap_size == 0) {
803                 lad->lad_incomplete = 0;
804                 CFS_RESET_BITMAP(bitmap);
805
806                 RETURN(0);
807         }
808
809         size = (lo->ll_bitmap_size + 7) >> 3;
810         rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
811         if (rc != size)
812                 RETURN(rc >= 0 ? -EINVAL : rc);
813
814         if (cfs_bitmap_check_empty(bitmap))
815                 lad->lad_incomplete = 0;
816         else
817                 lad->lad_incomplete = 1;
818
819         RETURN(0);
820 }
821
822 /**
823  * Load the layout LFSCK trace file from disk.
824  *
825  * The layout LFSCK trace file records the layout LFSCK status information
826  * and other statistics, such as how many objects have been scanned, and how
827  * many objects have been repaired, and etc. It also contains the bitmap for
828  * failed OSTs during the layout LFSCK. All these information will be loaded
829  * from disk to RAM when the layout LFSCK component setup.
830  *
831  * \param[in] env       pointer to the thread context
832  * \param[in] com       pointer to the lfsck component
833  *
834  * \retval              positive number for file data corruption, the caller
835  *                      should reset the layout LFSCK trace file
836  * \retval              0 for success
837  * \retval              negative error number on failure
838  */
839 static int lfsck_layout_load(const struct lu_env *env,
840                              struct lfsck_component *com)
841 {
842         struct lfsck_layout             *lo     = com->lc_file_ram;
843         ssize_t                          size   = com->lc_file_size;
844         loff_t                           pos    = 0;
845         int                              rc;
846
847         rc = dt_read(env, com->lc_obj,
848                      lfsck_buf_get(env, com->lc_file_disk, size), &pos);
849         if (rc == 0) {
850                 return -ENOENT;
851         } else if (rc < 0) {
852                 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
853                        lfsck_lfsck2name(com->lc_lfsck), rc);
854                 return rc;
855         } else if (rc != size) {
856                 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
857                        lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
858                 return 1;
859         }
860
861         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
862         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
863                 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
864                        "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
865                        lo->ll_magic, LFSCK_LAYOUT_MAGIC);
866                 return 1;
867         }
868
869         return 0;
870 }
871
872 /**
873  * Store the layout LFSCK trace file on disk.
874  *
875  * The layout LFSCK trace file records the layout LFSCK status information
876  * and other statistics, such as how many objects have been scanned, and how
877  * many objects have been repaired, and etc. It also contains the bitmap for
878  * failed OSTs during the layout LFSCK. All these information will be synced
879  * from RAM to disk periodically.
880  *
881  * \param[in] env       pointer to the thread context
882  * \param[in] com       pointer to the lfsck component
883  *
884  * \retval              0 for success
885  * \retval              negative error number on failure
886  */
887 static int lfsck_layout_store(const struct lu_env *env,
888                               struct lfsck_component *com)
889 {
890         struct dt_object        *obj    = com->lc_obj;
891         struct lfsck_instance   *lfsck  = com->lc_lfsck;
892         struct lfsck_layout     *lo_ram = com->lc_file_ram;
893         struct lfsck_layout     *lo     = com->lc_file_disk;
894         struct thandle          *th;
895         struct dt_device        *dev    = lfsck_obj2dev(obj);
896         struct cfs_bitmap       *bitmap = NULL;
897         loff_t                   pos;
898         ssize_t                  size   = com->lc_file_size;
899         __u32                    nbits  = 0;
900         int                      rc;
901         ENTRY;
902
903         if (lfsck->li_master) {
904                 struct lfsck_assistant_data *lad = com->lc_data;
905
906                 bitmap = lad->lad_bitmap;
907                 nbits = bitmap->size;
908
909                 LASSERT(nbits > 0);
910                 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
911         }
912
913         lo_ram->ll_bitmap_size = nbits;
914         lfsck_layout_cpu_to_le(lo, lo_ram);
915         th = dt_trans_create(env, dev);
916         if (IS_ERR(th))
917                 GOTO(log, rc = PTR_ERR(th));
918
919         rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
920                                      (loff_t)0, th);
921         if (rc != 0)
922                 GOTO(out, rc);
923
924         if (bitmap != NULL) {
925                 rc = dt_declare_record_write(env, obj,
926                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
927                                 (loff_t)size, th);
928                 if (rc != 0)
929                         GOTO(out, rc);
930         }
931
932         rc = dt_trans_start_local(env, dev, th);
933         if (rc != 0)
934                 GOTO(out, rc);
935
936         pos = 0;
937         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
938         if (rc != 0)
939                 GOTO(out, rc);
940
941         if (bitmap != NULL) {
942                 pos = size;
943                 rc = dt_record_write(env, obj,
944                                 lfsck_buf_get(env, bitmap->data, nbits >> 3),
945                                 &pos, th);
946         }
947
948         GOTO(out, rc);
949
950 out:
951         dt_trans_stop(env, dev, th);
952
953 log:
954         if (rc != 0)
955                 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
956                        lfsck_lfsck2name(lfsck), rc);
957
958         return rc;
959 }
960
961 static int lfsck_layout_init(const struct lu_env *env,
962                              struct lfsck_component *com)
963 {
964         struct lfsck_layout *lo = com->lc_file_ram;
965         int rc;
966
967         memset(lo, 0, com->lc_file_size);
968         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
969         lo->ll_status = LS_INIT;
970         down_write(&com->lc_sem);
971         rc = lfsck_layout_store(env, com);
972         if (rc == 0 && com->lc_lfsck->li_master)
973                 rc = lfsck_load_sub_trace_files(env, com,
974                         &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
975         up_write(&com->lc_sem);
976
977         return rc;
978 }
979
980 static int fid_is_for_ostobj(const struct lu_env *env,
981                              struct lfsck_instance *lfsck,
982                              struct dt_object *obj, const struct lu_fid *fid)
983 {
984         struct seq_server_site  *ss     = lfsck_dev_site(lfsck);
985         struct lu_seq_range     *range  = &lfsck_env_info(env)->lti_range;
986         struct lustre_mdt_attrs *lma;
987         int                      rc;
988
989         fld_range_set_any(range);
990         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
991         if (rc == 0) {
992                 if (fld_range_is_ost(range))
993                         return 1;
994
995                 return 0;
996         }
997
998         lma = &lfsck_env_info(env)->lti_lma;
999         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
1000                           XATTR_NAME_LMA);
1001         if (rc == sizeof(*lma)) {
1002                 lustre_lma_swab(lma);
1003
1004                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1005         }
1006
1007         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1008
1009         return rc > 0;
1010 }
1011
1012 static struct lfsck_layout_seq *
1013 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1014 {
1015         struct lfsck_layout_seq *lls;
1016
1017         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1018                 if (lls->lls_seq == seq)
1019                         return lls;
1020
1021                 if (lls->lls_seq > seq)
1022                         return NULL;
1023         }
1024
1025         return NULL;
1026 }
1027
1028 static void
1029 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1030                         struct lfsck_layout_seq *lls)
1031 {
1032         struct lfsck_layout_seq *tmp;
1033         struct list_head        *pos = &llsd->llsd_seq_list;
1034
1035         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1036                 if (lls->lls_seq < tmp->lls_seq) {
1037                         pos = &tmp->lls_list;
1038                         break;
1039                 }
1040         }
1041         list_add_tail(&lls->lls_list, pos);
1042 }
1043
1044 static int
1045 lfsck_layout_lastid_create(const struct lu_env *env,
1046                            struct lfsck_instance *lfsck,
1047                            struct dt_object *obj)
1048 {
1049         struct lfsck_thread_info *info   = lfsck_env_info(env);
1050         struct lu_attr           *la     = &info->lti_la;
1051         struct dt_object_format  *dof    = &info->lti_dof;
1052         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
1053         struct dt_device         *dt     = lfsck_obj2dev(obj);
1054         struct thandle           *th;
1055         __u64                     lastid = 0;
1056         loff_t                    pos    = 0;
1057         int                       rc;
1058         ENTRY;
1059
1060         if (bk->lb_param & LPF_DRYRUN)
1061                 return 0;
1062
1063         memset(la, 0, sizeof(*la));
1064         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
1065         la->la_valid = LA_MODE | LA_UID | LA_GID;
1066         memset(dof, 0, sizeof(*dof));
1067         dof->dof_type = dt_mode_to_dft(S_IFREG);
1068
1069         th = dt_trans_create(env, dt);
1070         if (IS_ERR(th))
1071                 GOTO(log, rc = PTR_ERR(th));
1072
1073         rc = dt_declare_create(env, obj, la, NULL, dof, th);
1074         if (rc != 0)
1075                 GOTO(stop, rc);
1076
1077         rc = dt_declare_record_write(env, obj,
1078                                      lfsck_buf_get(env, &lastid,
1079                                                    sizeof(lastid)),
1080                                      pos, th);
1081         if (rc != 0)
1082                 GOTO(stop, rc);
1083
1084         rc = dt_trans_start_local(env, dt, th);
1085         if (rc != 0)
1086                 GOTO(stop, rc);
1087
1088         dt_write_lock(env, obj, 0);
1089         if (likely(dt_object_exists(obj) == 0)) {
1090                 rc = dt_create(env, obj, la, NULL, dof, th);
1091                 if (rc == 0)
1092                         rc = dt_record_write(env, obj,
1093                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1094                                 &pos, th);
1095         }
1096         dt_write_unlock(env, obj);
1097
1098         GOTO(stop, rc);
1099
1100 stop:
1101         dt_trans_stop(env, dt, th);
1102
1103 log:
1104         CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1105                "%#llx: rc = %d\n",
1106                lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1107
1108         return rc;
1109 }
1110
1111 static int
1112 lfsck_layout_lastid_reload(const struct lu_env *env,
1113                            struct lfsck_component *com,
1114                            struct lfsck_layout_seq *lls)
1115 {
1116         __u64   lastid;
1117         loff_t  pos     = 0;
1118         int     rc;
1119
1120         dt_read_lock(env, lls->lls_lastid_obj, 0);
1121         rc = dt_record_read(env, lls->lls_lastid_obj,
1122                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1123         dt_read_unlock(env, lls->lls_lastid_obj);
1124         if (unlikely(rc != 0))
1125                 return rc;
1126
1127         lastid = le64_to_cpu(lastid);
1128         if (lastid < lls->lls_lastid_known) {
1129                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
1130                 struct lfsck_layout     *lo     = com->lc_file_ram;
1131
1132                 lls->lls_lastid = lls->lls_lastid_known;
1133                 lls->lls_dirty = 1;
1134                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1135                         LASSERT(lfsck->li_out_notify != NULL);
1136
1137                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1138                                              LE_LASTID_REBUILDING);
1139                         lo->ll_flags |= LF_CRASHED_LASTID;
1140
1141                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1142                                "LAST_ID file (1) for the sequence %#llx"
1143                                ", old value %llu, known value %llu\n",
1144                                lfsck_lfsck2name(lfsck), lls->lls_seq,
1145                                lastid, lls->lls_lastid);
1146                 }
1147         } else if (lastid >= lls->lls_lastid) {
1148                 lls->lls_lastid = lastid;
1149                 lls->lls_dirty = 0;
1150         }
1151
1152         return 0;
1153 }
1154
1155 static int
1156 lfsck_layout_lastid_store(const struct lu_env *env,
1157                           struct lfsck_component *com)
1158 {
1159         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1160         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1161         struct dt_device                *dt     = lfsck->li_bottom;
1162         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
1163         struct lfsck_layout_seq         *lls;
1164         struct thandle                  *th;
1165         __u64                            lastid;
1166         int                              rc     = 0;
1167         int                              rc1    = 0;
1168
1169         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1170                 loff_t pos = 0;
1171
1172                 if (!lls->lls_dirty)
1173                         continue;
1174
1175                 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1176                        "<seq> %#llx as <oid> %llu\n",
1177                        lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1178
1179                 if (bk->lb_param & LPF_DRYRUN) {
1180                         lls->lls_dirty = 0;
1181                         continue;
1182                 }
1183
1184                 th = dt_trans_create(env, dt);
1185                 if (IS_ERR(th)) {
1186                         rc1 = PTR_ERR(th);
1187                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1188                                "the LAST_ID for <seq> %#llx(1): rc = %d\n",
1189                                lfsck_lfsck2name(com->lc_lfsck),
1190                                lls->lls_seq, rc1);
1191                         continue;
1192                 }
1193
1194                 lastid = cpu_to_le64(lls->lls_lastid);
1195                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1196                                              lfsck_buf_get(env, &lastid,
1197                                                            sizeof(lastid)),
1198                                              pos, th);
1199                 if (rc != 0)
1200                         goto stop;
1201
1202                 rc = dt_trans_start_local(env, dt, th);
1203                 if (rc != 0)
1204                         goto stop;
1205
1206                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1207                 rc = dt_record_write(env, lls->lls_lastid_obj,
1208                                      lfsck_buf_get(env, &lastid,
1209                                      sizeof(lastid)), &pos, th);
1210                 dt_write_unlock(env, lls->lls_lastid_obj);
1211                 if (rc == 0)
1212                         lls->lls_dirty = 0;
1213
1214 stop:
1215                 dt_trans_stop(env, dt, th);
1216                 if (rc != 0) {
1217                         rc1 = rc;
1218                         CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1219                                "the LAST_ID for <seq> %#llx(2): rc = %d\n",
1220                                lfsck_lfsck2name(com->lc_lfsck),
1221                                lls->lls_seq, rc1);
1222                 }
1223         }
1224
1225         return rc1;
1226 }
1227
1228 static int
1229 lfsck_layout_lastid_load(const struct lu_env *env,
1230                          struct lfsck_component *com,
1231                          struct lfsck_layout_seq *lls)
1232 {
1233         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1234         struct lfsck_layout     *lo     = com->lc_file_ram;
1235         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1236         struct dt_object        *obj;
1237         loff_t                   pos    = 0;
1238         int                      rc;
1239         ENTRY;
1240
1241         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1242         obj = dt_locate(env, lfsck->li_bottom, fid);
1243         if (IS_ERR(obj))
1244                 RETURN(PTR_ERR(obj));
1245
1246         /* LAST_ID crashed, to be rebuilt */
1247         if (dt_object_exists(obj) == 0) {
1248                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1249                         LASSERT(lfsck->li_out_notify != NULL);
1250
1251                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1252                                              LE_LASTID_REBUILDING);
1253                         lo->ll_flags |= LF_CRASHED_LASTID;
1254
1255                         CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1256                                "LAST_ID file for sequence %#llx\n",
1257                                lfsck_lfsck2name(lfsck), lls->lls_seq);
1258
1259                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1260                             cfs_fail_val > 0) {
1261                                 struct l_wait_info lwi = LWI_TIMEOUT(
1262                                                 cfs_time_seconds(cfs_fail_val),
1263                                                 NULL, NULL);
1264
1265                                 /* Some others may changed the cfs_fail_val
1266                                  * as zero after above check, re-check it for
1267                                  * sure to avoid falling into wait for ever. */
1268                                 if (likely(lwi.lwi_timeout > 0)) {
1269                                         struct ptlrpc_thread *thread =
1270                                                 &lfsck->li_thread;
1271
1272                                         up_write(&com->lc_sem);
1273                                         l_wait_event(thread->t_ctl_waitq,
1274                                                      !thread_is_running(thread),
1275                                                      &lwi);
1276                                         down_write(&com->lc_sem);
1277                                 }
1278                         }
1279                 }
1280
1281                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1282         } else {
1283                 dt_read_lock(env, obj, 0);
1284                 rc = dt_read(env, obj,
1285                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1286                         &pos);
1287                 dt_read_unlock(env, obj);
1288                 if (rc != 0 && rc != sizeof(__u64))
1289                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1290
1291                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1292                         LASSERT(lfsck->li_out_notify != NULL);
1293
1294                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1295                                              LE_LASTID_REBUILDING);
1296                         lo->ll_flags |= LF_CRASHED_LASTID;
1297
1298                         CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1299                                "LAST_ID file for the sequence %#llx"
1300                                ": rc = %d\n",
1301                                lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1302                 }
1303
1304                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1305                 rc = 0;
1306         }
1307
1308         GOTO(out, rc);
1309
1310 out:
1311         if (rc != 0)
1312                 lfsck_object_put(env, obj);
1313         else
1314                 lls->lls_lastid_obj = obj;
1315
1316         return rc;
1317 }
1318
1319 static void lfsck_layout_record_failure(const struct lu_env *env,
1320                                         struct lfsck_instance *lfsck,
1321                                         struct lfsck_layout *lo)
1322 {
1323         __u64 cookie;
1324
1325         lo->ll_objs_failed_phase1++;
1326         cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1327                                                         lfsck->li_di_oit);
1328         if (lo->ll_pos_first_inconsistent == 0 ||
1329             lo->ll_pos_first_inconsistent < cookie) {
1330                 lo->ll_pos_first_inconsistent = cookie;
1331
1332                 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1333                        "inconsistency at the pos [%llu]\n",
1334                        lfsck_lfsck2name(lfsck),
1335                        lo->ll_pos_first_inconsistent);
1336         }
1337 }
1338
1339 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1340                                            struct lfsck_component *com,
1341                                            int rc)
1342 {
1343         struct lfsck_instance   *lfsck = com->lc_lfsck;
1344         struct lfsck_layout     *lo    = com->lc_file_ram;
1345
1346         down_write(&com->lc_sem);
1347         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1348                                   HALF_SEC - com->lc_time_last_checkpoint);
1349         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1350         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1351
1352         if (rc > 0) {
1353                 if (lo->ll_flags & LF_INCOMPLETE) {
1354                         lo->ll_status = LS_PARTIAL;
1355                 } else {
1356                         if (lfsck->li_master) {
1357                                 struct lfsck_assistant_data *lad = com->lc_data;
1358
1359                                 if (lad->lad_incomplete)
1360                                         lo->ll_status = LS_PARTIAL;
1361                                 else
1362                                         lo->ll_status = LS_COMPLETED;
1363                         } else {
1364                                 lo->ll_status = LS_COMPLETED;
1365                         }
1366                 }
1367                 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1368                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1369                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1370                 lo->ll_success_count++;
1371         } else if (rc == 0) {
1372                 if (lfsck->li_status != 0)
1373                         lo->ll_status = lfsck->li_status;
1374                 else
1375                         lo->ll_status = LS_STOPPED;
1376         } else {
1377                 lo->ll_status = LS_FAILED;
1378         }
1379
1380         rc = lfsck_layout_store(env, com);
1381         up_write(&com->lc_sem);
1382
1383         return rc;
1384 }
1385
1386 static int lfsck_layout_trans_stop(const struct lu_env *env,
1387                                    struct dt_device *dev,
1388                                    struct thandle *handle, int result)
1389 {
1390         int rc;
1391
1392         /* XXX: If there is something worng or it needs to repair nothing,
1393          *      then notify the lower to stop the modification. Currently,
1394          *      we use th_result for such purpose, that may be replaced by
1395          *      some rollback mechanism in the future. */
1396         handle->th_result = result;
1397         rc = dt_trans_stop(env, dev, handle);
1398         if (result != 0)
1399                 return result > 0 ? 0 : result;
1400
1401         return rc == 0 ? 1 : rc;
1402 }
1403
1404 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1405                                          struct lfsck_component *com,
1406                                          const struct lu_fid *pfid,
1407                                          const struct lu_fid *cfid,
1408                                          __u32 ea_off, __u32 ost_idx)
1409 {
1410         struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
1411         struct lu_fid *rec = &lfsck_env_info(env)->lti_fid4;
1412         struct dt_device *dev;
1413         struct dt_object *obj;
1414         struct thandle *th = NULL;
1415         int idx;
1416         int rc = 0;
1417         ENTRY;
1418
1419         idx = lfsck_sub_trace_file_fid2idx(pfid);
1420         obj = com->lc_sub_trace_objs[idx].lsto_obj;
1421         dev = lfsck_obj2dev(obj);
1422         fid_cpu_to_be(key, pfid);
1423         key->f_ver = cpu_to_be32(ea_off);
1424         fid_cpu_to_be(rec, cfid);
1425         rec->f_ver = cpu_to_be32(ost_idx);
1426
1427         mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1428
1429         th = dt_trans_create(env, dev);
1430         if (IS_ERR(th))
1431                 GOTO(unlock, rc = PTR_ERR(th));
1432
1433         rc = dt_declare_insert(env, obj,
1434                                (const struct dt_rec *)rec,
1435                                (const struct dt_key *)key, th);
1436         if (rc)
1437                 GOTO(unlock, rc);
1438
1439         rc = dt_trans_start_local(env, dev, th);
1440         if (rc)
1441                 GOTO(unlock, rc);
1442
1443         rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1444                        (const struct dt_key *)key, th, 1);
1445
1446         GOTO(unlock, rc);
1447
1448 unlock:
1449         if (th != NULL && !IS_ERR(th))
1450                 dt_trans_stop(env, dev, th);
1451
1452         mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1453
1454         CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", ea_off = %u, "
1455                "ost_idx = %u, into the trace file for further dangling check: "
1456                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1457                PFID(pfid), PFID(cfid), ea_off, ost_idx, rc);
1458
1459         return rc;
1460 }
1461
1462 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1463                                          struct lfsck_component *com,
1464                                          const struct lu_fid *fid,
1465                                          __u32 ea_off)
1466 {
1467         struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
1468         struct dt_device *dev;
1469         struct dt_object *obj;
1470         struct thandle *th = NULL;
1471         int idx;
1472         int rc = 0;
1473         ENTRY;
1474
1475         idx = lfsck_sub_trace_file_fid2idx(fid);
1476         obj = com->lc_sub_trace_objs[idx].lsto_obj;
1477         dev = lfsck_obj2dev(obj);
1478         fid_cpu_to_be(key, fid);
1479         key->f_ver = cpu_to_be32(ea_off);
1480
1481         mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1482
1483         th = dt_trans_create(env, dev);
1484         if (IS_ERR(th))
1485                 GOTO(unlock, rc = PTR_ERR(th));
1486
1487         rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1488         if (rc)
1489                 GOTO(unlock, rc);
1490
1491         rc = dt_trans_start_local(env, dev, th);
1492         if (rc)
1493                 GOTO(unlock, rc);
1494
1495         rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1496
1497         GOTO(unlock, rc);
1498
1499 unlock:
1500         if (th != NULL && !IS_ERR(th))
1501                 dt_trans_stop(env, dev, th);
1502
1503         mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1504
1505         CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1506                ", ea_off = %u from the trace file: rc = %d\n",
1507                lfsck_lfsck2name(com->lc_lfsck), PFID(fid), ea_off, rc);
1508
1509         return rc;
1510 }
1511
1512 /**
1513  * Get the system default stripe size.
1514  *
1515  * \param[in] env       pointer to the thread context
1516  * \param[in] lfsck     pointer to the lfsck instance
1517  * \param[out] size     pointer to the default stripe size
1518  *
1519  * \retval              0 for success
1520  * \retval              negative error number on failure
1521  */
1522 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1523                                            struct lfsck_instance *lfsck,
1524                                            __u32 *size)
1525 {
1526         struct lov_user_md      *lum = &lfsck_env_info(env)->lti_lum;
1527         struct dt_object        *root;
1528         int                      rc;
1529
1530         root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1531         if (IS_ERR(root))
1532                 return PTR_ERR(root);
1533
1534         /* Get the default stripe size via xattr_get on the backend root. */
1535         rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1536                           XATTR_NAME_LOV);
1537         if (rc > 0) {
1538                 /* The lum->lmm_stripe_size is LE mode. The *size also
1539                  * should be LE mode. So it is unnecessary to convert. */
1540                 *size = lum->lmm_stripe_size;
1541                 rc = 0;
1542         } else if (unlikely(rc == 0)) {
1543                 rc = -EINVAL;
1544         }
1545
1546         lfsck_object_put(env, root);
1547
1548         return rc;
1549 }
1550
1551 /**
1552  * \retval       +1: repaired
1553  * \retval        0: did nothing
1554  * \retval      -ve: on error
1555  */
1556 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1557                                      struct thandle *handle,
1558                                      struct dt_object *parent,
1559                                      struct lu_fid *cfid,
1560                                      struct lu_buf *buf,
1561                                      struct lov_ost_data_v1 *slot,
1562                                      int fl, __u32 ost_idx)
1563 {
1564         struct ost_id           *oi     = &lfsck_env_info(env)->lti_oi;
1565         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1566         struct lu_buf            ea_buf;
1567         int                      rc;
1568         __u32                    magic;
1569         __u16                    count;
1570         ENTRY;
1571
1572         magic = le32_to_cpu(lmm->lmm_magic);
1573         count = le16_to_cpu(lmm->lmm_stripe_count);
1574
1575         fid_to_ostid(cfid, oi);
1576         ostid_cpu_to_le(oi, &slot->l_ost_oi);
1577         slot->l_ost_gen = cpu_to_le32(0);
1578         slot->l_ost_idx = cpu_to_le32(ost_idx);
1579
1580         if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
1581                 struct lov_ost_data_v1 *objs;
1582                 int                     i;
1583
1584                 if (magic == LOV_MAGIC_V1)
1585                         objs = &lmm->lmm_objects[0];
1586                 else
1587                         objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1588                 for (i = 0; i < count; i++, objs++) {
1589                         if (objs != slot && lovea_slot_is_dummy(objs))
1590                                 break;
1591                 }
1592
1593                 /* If the @slot is the last dummy slot to be refilled,
1594                  * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1595                 if (i == count)
1596                         lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
1597         }
1598
1599         lfsck_buf_init(&ea_buf, lmm, lov_mds_md_size(count, magic));
1600         rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1601         if (rc == 0)
1602                 rc = 1;
1603
1604         RETURN(rc);
1605 }
1606
1607 /**
1608  * \retval       +1: repaired
1609  * \retval        0: did nothing
1610  * \retval      -ve: on error
1611  */
1612 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1613                                      struct lfsck_instance *lfsck,
1614                                      struct thandle *handle,
1615                                      struct dt_object *parent,
1616                                      struct lu_fid *cfid,
1617                                      struct lu_buf *buf, int fl,
1618                                      __u32 ost_idx, __u32 ea_off, bool reset)
1619 {
1620         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1621         struct lov_ost_data_v1  *objs;
1622         int                      rc;
1623         __u16                    count;
1624         bool                     hole   = false;
1625         ENTRY;
1626
1627         if (fl == LU_XATTR_CREATE || reset) {
1628                 __u32 pattern = LOV_PATTERN_RAID0;
1629
1630                 count = ea_off + 1;
1631                 LASSERT(buf->lb_len >= lov_mds_md_size(count, LOV_MAGIC_V1));
1632
1633                 if (ea_off != 0 || reset) {
1634                         pattern |= LOV_PATTERN_F_HOLE;
1635                         hole = true;
1636                 }
1637
1638                 memset(lmm, 0, buf->lb_len);
1639                 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1640                 lmm->lmm_pattern = cpu_to_le32(pattern);
1641                 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1642                 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1643
1644                 rc = lfsck_layout_get_def_stripesize(env, lfsck,
1645                                                      &lmm->lmm_stripe_size);
1646                 if (rc != 0)
1647                         RETURN(rc);
1648
1649                 objs = &lmm->lmm_objects[ea_off];
1650         } else {
1651                 __u32   magic = le32_to_cpu(lmm->lmm_magic);
1652                 int     gap;
1653
1654                 count = le16_to_cpu(lmm->lmm_stripe_count);
1655                 if (magic == LOV_MAGIC_V1)
1656                         objs = &lmm->lmm_objects[count];
1657                 else
1658                         objs = &((struct lov_mds_md_v3 *)lmm)->
1659                                                         lmm_objects[count];
1660
1661                 gap = ea_off - count;
1662                 if (gap >= 0)
1663                         count = ea_off + 1;
1664                 LASSERT(buf->lb_len >= lov_mds_md_size(count, magic));
1665
1666                 if (gap > 0) {
1667                         memset(objs, 0, gap * sizeof(*objs));
1668                         lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1669                         hole = true;
1670                 }
1671
1672                 lmm->lmm_layout_gen =
1673                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1674                 objs += gap;
1675         }
1676
1677         lmm->lmm_stripe_count = cpu_to_le16(count);
1678         rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1679                                        fl, ost_idx);
1680
1681         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1682                DFID": parent "DFID", OST-index %u, stripe-index %u, fl %d, "
1683                "reset %s, %s LOV EA hole: rc = %d\n",
1684                lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1685                ost_idx, ea_off, fl, reset ? "yes" : "no",
1686                hole ? "with" : "without", rc);
1687
1688         RETURN(rc);
1689 }
1690
1691 static int __lfsck_layout_update_pfid(const struct lu_env *env,
1692                                       struct dt_object *child,
1693                                       const struct lu_fid *pfid, __u32 offset)
1694 {
1695         struct dt_device        *dev    = lfsck_obj2dev(child);
1696         struct filter_fid       *ff     = &lfsck_env_info(env)->lti_new_pfid;
1697         struct thandle          *handle;
1698         struct lu_buf            buf    = { NULL };
1699         int                      rc;
1700
1701         ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1702         ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1703         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1704          * MDT-object's FID::f_ver, instead it is the OST-object index in its
1705          * parent MDT-object's layout EA. */
1706         ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
1707         lfsck_buf_init(&buf, ff, sizeof(struct filter_fid));
1708
1709         handle = dt_trans_create(env, dev);
1710         if (IS_ERR(handle))
1711                 RETURN(PTR_ERR(handle));
1712
1713         rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1714         if (rc != 0)
1715                 GOTO(stop, rc);
1716
1717         rc = dt_trans_start_local(env, dev, handle);
1718         if (rc != 0)
1719                 GOTO(stop, rc);
1720
1721         rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1722
1723         GOTO(stop, rc);
1724
1725 stop:
1726         dt_trans_stop(env, dev, handle);
1727
1728         return rc;
1729 }
1730
1731 /**
1732  * \retval       +1: repaired
1733  * \retval        0: did nothing
1734  * \retval      -ve: on error
1735  */
1736 static int lfsck_layout_update_pfid(const struct lu_env *env,
1737                                     struct lfsck_component *com,
1738                                     struct dt_object *parent,
1739                                     struct lu_fid *cfid,
1740                                     struct dt_device *cdev, __u32 ea_off)
1741 {
1742         struct dt_object        *child;
1743         int                      rc     = 0;
1744         ENTRY;
1745
1746         child = lfsck_object_find_by_dev(env, cdev, cfid);
1747         if (IS_ERR(child))
1748                 RETURN(PTR_ERR(child));
1749
1750         rc = __lfsck_layout_update_pfid(env, child,
1751                                         lu_object_fid(&parent->do_lu), ea_off);
1752         lfsck_object_put(env, child);
1753
1754         RETURN(rc == 0 ? 1 : rc);
1755 }
1756
1757 /**
1758  * This function will create the MDT-object with the given (partial) LOV EA.
1759  *
1760  * Under some data corruption cases, the MDT-object of the file may be lost,
1761  * but its OST-objects, or some of them are there. The layout LFSCK needs to
1762  * re-create the MDT-object with the orphan OST-object(s) information.
1763  *
1764  * On the other hand, the LFSCK may has created some OST-object for repairing
1765  * dangling LOV EA reference, but as the LFSCK processing, it may find that
1766  * the old OST-object is there and should replace the former new created OST
1767  * object. Unfortunately, some others have modified such newly created object.
1768  * To keep the data (both new and old), the LFSCK will create MDT-object with
1769  * new FID to reference the original OST-object.
1770  *
1771  * \param[in] env       pointer to the thread context
1772  * \param[in] com       pointer to the lfsck component
1773  * \param[in] ltd       pointer to target device descriptor
1774  * \param[in] rec       pointer to the record for the orphan OST-object
1775  * \param[in] cfid      pointer to FID for the orphan OST-object
1776  * \param[in] infix     additional information, such as the FID for original
1777  *                      MDT-object and the stripe offset in the LOV EA
1778  * \param[in] type      the type for describing why the orphan MDT-object is
1779  *                      created. The rules are as following:
1780  *
1781  *  type "C":           Multiple OST-objects claim the same MDT-object and the
1782  *                      same slot in the layout EA. Then the LFSCK will create
1783  *                      new MDT-object(s) to hold the conflict OST-object(s).
1784  *
1785  *  type "N":           The orphan OST-object does not know which one was the
1786  *                      real parent MDT-object, so the LFSCK uses new FID for
1787  *                      its parent MDT-object.
1788  *
1789  *  type "R":           The orphan OST-object knows its parent MDT-object FID,
1790  *                      but does not know the position (the file name) in the
1791  *                      layout.
1792  *
1793  *  type "D":           The MDT-object is a directory, it may knows its parent
1794  *                      but because there is no valid linkEA, the LFSCK cannot
1795  *                      know where to put it back to the namespace.
1796  *  type "O":           The MDT-object has no linkEA, and there is no name
1797  *                      entry that references the MDT-object.
1798  *
1799  *  type "P":           The orphan object to be created was a parent directory
1800  *                      of some MDT-object which linkEA shows that the @orphan
1801  *                      object is missing.
1802  *
1803  * The orphan name will be like:
1804  * ${FID}-${infix}-${type}-${conflict_version}
1805  *
1806  * \param[in] ea_off    the stripe offset in the LOV EA
1807  *
1808  * \retval              positive on repaired something
1809  * \retval              0 if needs to repair nothing
1810  * \retval              negative error number on failure
1811  */
1812 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1813                                         struct lfsck_component *com,
1814                                         struct lfsck_tgt_desc *ltd,
1815                                         struct lu_orphan_rec *rec,
1816                                         struct lu_fid *cfid,
1817                                         const char *infix,
1818                                         const char *type,
1819                                         __u32 ea_off)
1820 {
1821         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1822         struct dt_insert_rec            *dtrec  = &info->lti_dt_rec;
1823         char                            *name   = info->lti_key;
1824         struct lu_attr                  *la     = &info->lti_la2;
1825         struct dt_object_format         *dof    = &info->lti_dof;
1826         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1827         struct lu_fid                   *pfid   = &rec->lor_fid;
1828         struct lu_fid                   *tfid   = &info->lti_fid3;
1829         struct dt_device                *dev    = lfsck->li_bottom;
1830         struct dt_object                *lpf    = lfsck->li_lpf_obj;
1831         struct dt_object                *pobj   = NULL;
1832         struct dt_object                *cobj   = NULL;
1833         struct thandle                  *th     = NULL;
1834         struct lu_buf                   *ea_buf = &info->lti_big_buf;
1835         struct lu_buf                    lov_buf;
1836         struct lfsck_lock_handle        *llh    = &info->lti_llh;
1837         struct linkea_data               ldata  = { NULL };
1838         struct lu_buf                    linkea_buf;
1839         const struct lu_name            *pname;
1840         int                              size   = 0;
1841         int                              idx    = 0;
1842         int                              rc     = 0;
1843         ENTRY;
1844
1845         if (unlikely(lpf == NULL))
1846                 GOTO(log, rc = -ENXIO);
1847
1848         /* We use two separated transactions to repair the inconsistency.
1849          *
1850          * 1) create the MDT-object locally.
1851          * 2) update the OST-object's PFID EA if necessary.
1852          *
1853          * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
1854          * updated when the layout LFSCK run next time.
1855          *
1856          * If 1) failed, but 2) succeed, then such MDT-object will be re-created
1857          * when the layout LFSCK run next time. */
1858
1859         if (fid_is_zero(pfid)) {
1860                 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1861                 if (rc != 0)
1862                         GOTO(log, rc);
1863
1864                 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1865                 if (IS_ERR(cobj))
1866                         GOTO(log, rc = PTR_ERR(cobj));
1867         }
1868
1869         pobj = lfsck_object_find_by_dev(env, dev, pfid);
1870         if (IS_ERR(pobj))
1871                 GOTO(log, rc = PTR_ERR(pobj));
1872
1873         LASSERT(infix != NULL);
1874         LASSERT(type != NULL);
1875
1876         memset(la, 0, sizeof(*la));
1877         la->la_uid = rec->lor_uid;
1878         la->la_gid = rec->lor_gid;
1879         la->la_mode = S_IFREG | S_IRUSR;
1880         la->la_valid = LA_MODE | LA_UID | LA_GID;
1881
1882         memset(dof, 0, sizeof(*dof));
1883         dof->dof_type = dt_mode_to_dft(S_IFREG);
1884         /* Because the dof->dof_reg.striped = 0, the LOD will not create
1885          * the stripe(s). The LFSCK will specify the LOV EA via
1886          * lfsck_layout_extend_lovea(). */
1887
1888         size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1889         if (ea_buf->lb_len < size) {
1890                 lu_buf_realloc(ea_buf, size);
1891                 if (ea_buf->lb_buf == NULL)
1892                         GOTO(log, rc = -ENOMEM);
1893         }
1894
1895 again:
1896         do {
1897                 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
1898                          type, idx++);
1899                 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1900                                (const struct dt_key *)name);
1901                 if (rc != 0 && rc != -ENOENT)
1902                         GOTO(log, rc);
1903         } while (rc == 0);
1904
1905         rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
1906                         MDS_INODELOCK_UPDATE, LCK_PW);
1907         if (rc != 0)
1908                 GOTO(log, rc);
1909
1910         /* Re-check whether the name conflict with othrs after taken
1911          * the ldlm lock. */
1912         rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1913                        (const struct dt_key *)name);
1914         if (unlikely(rc == 0)) {
1915                 lfsck_unlock(llh);
1916                 goto again;
1917         }
1918
1919         if (rc != -ENOENT)
1920                 GOTO(unlock, rc);
1921
1922         pname = lfsck_name_get_const(env, name, strlen(name));
1923         rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
1924                               pname, lfsck_dto2fid(lfsck->li_lpf_obj));
1925         if (rc != 0)
1926                 GOTO(unlock, rc);
1927
1928         /* The 1st transaction. */
1929         th = dt_trans_create(env, dev);
1930         if (IS_ERR(th))
1931                 GOTO(unlock, rc = PTR_ERR(th));
1932
1933         rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1934         if (rc != 0)
1935                 GOTO(stop, rc);
1936
1937         lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
1938         rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
1939                                   LU_XATTR_REPLACE, th);
1940         if (rc != 0)
1941                 GOTO(stop, rc);
1942
1943         dtrec->rec_fid = pfid;
1944         dtrec->rec_type = S_IFREG;
1945         rc = dt_declare_insert(env, lpf,
1946                                (const struct dt_rec *)dtrec,
1947                                (const struct dt_key *)name, th);
1948         if (rc != 0)
1949                 GOTO(stop, rc);
1950
1951         lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1952                        ldata.ld_leh->leh_len);
1953         rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
1954                                   XATTR_NAME_LINK, 0, th);
1955         if (rc != 0)
1956                 GOTO(stop, rc);
1957
1958         rc = dt_trans_start_local(env, dev, th);
1959         if (rc != 0)
1960                 GOTO(stop, rc);
1961
1962         dt_write_lock(env, pobj, 0);
1963         rc = dt_create(env, pobj, la, NULL, dof, th);
1964         if (rc == 0)
1965                 rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
1966                                 &lov_buf, 0, ltd->ltd_index, ea_off, true);
1967         dt_write_unlock(env, pobj);
1968         if (rc < 0)
1969                 GOTO(stop, rc);
1970
1971         rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
1972                        (const struct dt_key *)name, th, 1);
1973         if (rc != 0)
1974                 GOTO(stop, rc);
1975
1976         rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
1977         if (rc == 0 && cobj != NULL) {
1978                 dt_trans_stop(env, dev, th);
1979                 th = NULL;
1980
1981                 /* The 2nd transaction. */
1982                 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ea_off);
1983         }
1984
1985         GOTO(stop, rc);
1986
1987 stop:
1988         if (th != NULL)
1989                 dt_trans_stop(env, dev, th);
1990
1991 unlock:
1992         lfsck_unlock(llh);
1993
1994 log:
1995         if (cobj != NULL && !IS_ERR(cobj))
1996                 lfsck_object_put(env, cobj);
1997         if (pobj != NULL && !IS_ERR(pobj))
1998                 lfsck_object_put(env, pobj);
1999
2000         if (rc < 0)
2001                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2002                        "recreate the lost MDT-object: parent "DFID
2003                        ", child "DFID", OST-index %u, stripe-index %u, "
2004                        "infix %s, type %s: rc = %d\n",
2005                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2006                        ltd->ltd_index, ea_off, infix, type, rc);
2007
2008         return rc >= 0 ? 1 : rc;
2009 }
2010
2011 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2012                                                    struct lfsck_component *com,
2013                                                    const struct lu_fid *fid,
2014                                                    __u32 index)
2015 {
2016         struct lfsck_thread_info *info  = lfsck_env_info(env);
2017         struct lfsck_request     *lr    = &info->lti_lr;
2018         struct lfsck_instance    *lfsck = com->lc_lfsck;
2019         struct lfsck_tgt_desc    *ltd;
2020         struct ptlrpc_request    *req;
2021         struct lfsck_request     *tmp;
2022         struct obd_export        *exp;
2023         int                       rc    = 0;
2024         ENTRY;
2025
2026         ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2027         if (unlikely(ltd == NULL))
2028                 RETURN(-ENXIO);
2029
2030         exp = ltd->ltd_exp;
2031         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2032                 GOTO(put, rc = -EOPNOTSUPP);
2033
2034         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2035         if (req == NULL)
2036                 GOTO(put, rc = -ENOMEM);
2037
2038         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2039         if (rc != 0) {
2040                 ptlrpc_request_free(req);
2041
2042                 GOTO(put, rc);
2043         }
2044
2045         memset(lr, 0, sizeof(*lr));
2046         lr->lr_event = LE_CONDITIONAL_DESTROY;
2047         lr->lr_active = LFSCK_TYPE_LAYOUT;
2048         lr->lr_fid = *fid;
2049
2050         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2051         *tmp = *lr;
2052         ptlrpc_request_set_replen(req);
2053
2054         rc = ptlrpc_queue_wait(req);
2055         ptlrpc_req_finished(req);
2056
2057         GOTO(put, rc);
2058
2059 put:
2060         lfsck_tgt_put(ltd);
2061
2062         return rc;
2063 }
2064
2065 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2066                                                   struct lfsck_component *com,
2067                                                   struct lfsck_request *lr)
2068 {
2069         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2070         struct lu_attr                  *la     = &info->lti_la;
2071         union ldlm_policy_data          *policy = &info->lti_policy;
2072         struct ldlm_res_id              *resid  = &info->lti_resid;
2073         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2074         struct dt_device                *dev    = lfsck->li_bottom;
2075         struct lu_fid                   *fid    = &lr->lr_fid;
2076         struct dt_object                *obj;
2077         struct thandle                  *th     = NULL;
2078         struct lustre_handle             lh     = { 0 };
2079         __u64                            flags  = 0;
2080         int                              rc     = 0;
2081         ENTRY;
2082
2083         obj = lfsck_object_find_by_dev(env, dev, fid);
2084         if (IS_ERR(obj))
2085                 RETURN(PTR_ERR(obj));
2086
2087         dt_read_lock(env, obj, 0);
2088         if (dt_object_exists(obj) == 0 ||
2089             lfsck_is_dead_obj(obj)) {
2090                 dt_read_unlock(env, obj);
2091
2092                 GOTO(put, rc = -ENOENT);
2093         }
2094
2095         /* Get obj's attr without lock firstly. */
2096         rc = dt_attr_get(env, obj, la);
2097         dt_read_unlock(env, obj);
2098         if (rc != 0)
2099                 GOTO(put, rc);
2100
2101         if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2102                 GOTO(put, rc = -ETXTBSY);
2103
2104         /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2105         LASSERT(lfsck->li_namespace != NULL);
2106
2107         memset(policy, 0, sizeof(*policy));
2108         policy->l_extent.end = OBD_OBJECT_EOF;
2109         ost_fid_build_resid(fid, resid);
2110         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2111                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
2112                                     ldlm_completion_ast, NULL, NULL, 0,
2113                                     LVB_T_NONE, NULL, &lh);
2114         if (rc != ELDLM_OK)
2115                 GOTO(put, rc = -EIO);
2116
2117         dt_write_lock(env, obj, 0);
2118         /* Get obj's attr within lock again. */
2119         rc = dt_attr_get(env, obj, la);
2120         if (rc != 0)
2121                 GOTO(unlock, rc);
2122
2123         if (la->la_ctime != 0)
2124                 GOTO(unlock, rc = -ETXTBSY);
2125
2126         th = dt_trans_create(env, dev);
2127         if (IS_ERR(th))
2128                 GOTO(unlock, rc = PTR_ERR(th));
2129
2130         rc = dt_declare_ref_del(env, obj, th);
2131         if (rc != 0)
2132                 GOTO(stop, rc);
2133
2134         rc = dt_declare_destroy(env, obj, th);
2135         if (rc != 0)
2136                 GOTO(stop, rc);
2137
2138         rc = dt_trans_start_local(env, dev, th);
2139         if (rc != 0)
2140                 GOTO(stop, rc);
2141
2142         rc = dt_ref_del(env, obj, th);
2143         if (rc != 0)
2144                 GOTO(stop, rc);
2145
2146         rc = dt_destroy(env, obj, th);
2147         if (rc == 0)
2148                 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2149                        "OST-object "DFID" that was created for reparing "
2150                        "dangling referenced case. But the original missing "
2151                        "OST-object is found now.\n",
2152                        lfsck_lfsck2name(lfsck), PFID(fid));
2153
2154         GOTO(stop, rc);
2155
2156 stop:
2157         dt_trans_stop(env, dev, th);
2158
2159 unlock:
2160         dt_write_unlock(env, obj);
2161         ldlm_lock_decref(&lh, LCK_EX);
2162
2163 put:
2164         lfsck_object_put(env, obj);
2165
2166         return rc;
2167 }
2168
2169 /**
2170  * Some OST-object has occupied the specified layout EA slot.
2171  * Such OST-object may be generated by the LFSCK when repair
2172  * dangling referenced MDT-object, which can be indicated by
2173  * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2174  * is true and such OST-object has not been modified yet, we
2175  * will replace it with the orphan OST-object; otherwise the
2176  * LFSCK will create new MDT-object to reference the orphan.
2177  *
2178  * \retval       +1: repaired
2179  * \retval        0: did nothing
2180  * \retval      -ve: on error
2181  */
2182 static int lfsck_layout_conflict_create(const struct lu_env *env,
2183                                         struct lfsck_component *com,
2184                                         struct lfsck_tgt_desc *ltd,
2185                                         struct lu_orphan_rec *rec,
2186                                         struct dt_object *parent,
2187                                         struct lu_fid *cfid,
2188                                         struct lu_buf *ea_buf,
2189                                         struct lov_ost_data_v1 *slot,
2190                                         __u32 ea_off)
2191 {
2192         struct lfsck_thread_info *info          = lfsck_env_info(env);
2193         struct lu_fid            *cfid2         = &info->lti_fid2;
2194         struct ost_id            *oi            = &info->lti_oi;
2195         struct lov_mds_md_v1     *lmm           = ea_buf->lb_buf;
2196         struct dt_device         *dev           = lfsck_obj2dev(parent);
2197         struct thandle           *th            = NULL;
2198         struct lustre_handle      lh            = { 0 };
2199         __u32                     ost_idx2      = le32_to_cpu(slot->l_ost_idx);
2200         int                       rc            = 0;
2201         ENTRY;
2202
2203         while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2204                 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2205                         RETURN(0);
2206         }
2207
2208         ostid_le_to_cpu(&slot->l_ost_oi, oi);
2209         rc = ostid_to_fid(cfid2, oi, ost_idx2);
2210         if (rc != 0)
2211                 GOTO(out, rc);
2212
2213         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2214                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2215                               LCK_EX);
2216         if (rc != 0)
2217                 GOTO(out, rc);
2218
2219         rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2220
2221         /* If the conflict OST-obejct is not created for fixing dangling
2222          * referenced MDT-object in former LFSCK check/repair, or it has
2223          * been modified by others, then we cannot destroy it. Re-create
2224          * a new MDT-object for the orphan OST-object. */
2225         if (rc == -ETXTBSY) {
2226                 /* No need the layout lock on the original parent. */
2227                 lfsck_ibits_unlock(&lh, LCK_EX);
2228
2229                 fid_zero(&rec->lor_fid);
2230                 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2231                          "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2232                          ea_off);
2233                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2234                                                 info->lti_tmpbuf, "C", ea_off);
2235
2236                 RETURN(rc);
2237         }
2238
2239         if (rc != 0 && rc != -ENOENT)
2240                 GOTO(unlock, rc);
2241
2242         th = dt_trans_create(env, dev);
2243         if (IS_ERR(th))
2244                 GOTO(unlock, rc = PTR_ERR(th));
2245
2246         rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2247                                   LU_XATTR_REPLACE, th);
2248         if (rc != 0)
2249                 GOTO(stop, rc);
2250
2251         rc = dt_trans_start_local(env, dev, th);
2252         if (rc != 0)
2253                 GOTO(stop, rc);
2254
2255         dt_write_lock(env, parent, 0);
2256         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2257         rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2258                                        LU_XATTR_REPLACE, ltd->ltd_index);
2259         dt_write_unlock(env, parent);
2260
2261         GOTO(stop, rc);
2262
2263 stop:
2264         dt_trans_stop(env, dev, th);
2265
2266 unlock:
2267         lfsck_ibits_unlock(&lh, LCK_EX);
2268
2269 out:
2270         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2271                "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2272                "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2273                lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2274                PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2275                ea_off, rc);
2276
2277         return rc >= 0 ? 1 : rc;
2278 }
2279
2280 /**
2281  * \retval       +1: repaired
2282  * \retval        0: did nothing
2283  * \retval      -ve: on error
2284  */
2285 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2286                                        struct lfsck_component *com,
2287                                        struct lfsck_tgt_desc *ltd,
2288                                        struct lu_orphan_rec *rec,
2289                                        struct dt_object *parent,
2290                                        struct lu_fid *cfid,
2291                                        __u32 ost_idx, __u32 ea_off)
2292 {
2293         struct lfsck_thread_info *info          = lfsck_env_info(env);
2294         struct lu_buf            *buf           = &info->lti_big_buf;
2295         struct lu_fid            *fid           = &info->lti_fid2;
2296         struct ost_id            *oi            = &info->lti_oi;
2297         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2298         struct dt_device         *dt            = lfsck_obj2dev(parent);
2299         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2300         struct thandle            *handle       = NULL;
2301         size_t                    lovea_size;
2302         struct lov_mds_md_v1     *lmm;
2303         struct lov_ost_data_v1   *objs;
2304         struct lustre_handle      lh            = { 0 };
2305         __u32                     magic;
2306         int                       fl            = 0;
2307         int                       rc            = 0;
2308         int                       rc1;
2309         int                       i;
2310         __u16                     count;
2311         bool                      locked        = false;
2312         ENTRY;
2313
2314         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2315                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2316                               LCK_EX);
2317         if (rc != 0) {
2318                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2319                        "LOV EA for "DFID": parent "DFID", OST-index %u, "
2320                        "stripe-index %u: rc = %d\n",
2321                        lfsck_lfsck2name(lfsck), PFID(cfid),
2322                        PFID(lfsck_dto2fid(parent)), ost_idx, ea_off, rc);
2323
2324                 RETURN(rc);
2325         }
2326
2327 again:
2328         if (locked) {
2329                 dt_write_unlock(env, parent);
2330                 locked = false;
2331         }
2332
2333         if (handle != NULL) {
2334                 dt_trans_stop(env, dt, handle);
2335                 handle = NULL;
2336         }
2337
2338         if (rc < 0)
2339                 GOTO(unlock_layout, rc);
2340
2341         lovea_size = rc;
2342         if (buf->lb_len < lovea_size) {
2343                 lu_buf_realloc(buf, lovea_size);
2344                 if (buf->lb_buf == NULL)
2345                         GOTO(unlock_layout, rc = -ENOMEM);
2346         }
2347
2348         if (!(bk->lb_param & LPF_DRYRUN)) {
2349                 handle = dt_trans_create(env, dt);
2350                 if (IS_ERR(handle))
2351                         GOTO(unlock_layout, rc = PTR_ERR(handle));
2352
2353                 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2354                                           fl, handle);
2355                 if (rc != 0)
2356                         GOTO(stop, rc);
2357
2358                 rc = dt_trans_start_local(env, dt, handle);
2359                 if (rc != 0)
2360                         GOTO(stop, rc);
2361         }
2362
2363         dt_write_lock(env, parent, 0);
2364         locked = true;
2365         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2366         if (rc == -ERANGE) {
2367                 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2368                 LASSERT(rc != 0);
2369                 goto again;
2370         } else if (rc == -ENODATA || rc == 0) {
2371                 lovea_size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2372                 /* If the declared is not big enough, re-try. */
2373                 if (buf->lb_len < lovea_size) {
2374                         rc = lovea_size;
2375                         goto again;
2376                 }
2377                 fl = LU_XATTR_CREATE;
2378         } else if (rc < 0) {
2379                 GOTO(unlock_parent, rc);
2380         } else if (unlikely(buf->lb_len == 0)) {
2381                 goto again;
2382         } else {
2383                 fl = LU_XATTR_REPLACE;
2384                 lovea_size = rc;
2385         }
2386
2387         if (fl == LU_XATTR_CREATE) {
2388                 if (bk->lb_param & LPF_DRYRUN)
2389                         GOTO(unlock_parent, rc = 1);
2390
2391                 LASSERT(buf->lb_len >= lovea_size);
2392
2393                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2394                                                buf, fl, ost_idx, ea_off, false);
2395
2396                 GOTO(unlock_parent, rc);
2397         }
2398
2399         lmm = buf->lb_buf;
2400         rc1 = lfsck_layout_verify_header(lmm);
2401
2402         /* If the LOV EA crashed, the rebuild it. */
2403         if (rc1 == -EINVAL) {
2404                 if (bk->lb_param & LPF_DRYRUN)
2405                         GOTO(unlock_parent, rc = 1);
2406
2407                 LASSERT(buf->lb_len >= lovea_size);
2408
2409                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2410                                                buf, fl, ost_idx, ea_off, true);
2411
2412                 GOTO(unlock_parent, rc);
2413         }
2414
2415         /* For other unknown magic/pattern, keep the current LOV EA. */
2416         if (rc1 != 0)
2417                 GOTO(unlock_parent, rc = rc1);
2418
2419         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2420          * been verified in lfsck_layout_verify_header() already. If some
2421          * new magic introduced in the future, then layout LFSCK needs to
2422          * be updated also. */
2423         magic = le32_to_cpu(lmm->lmm_magic);
2424         if (magic == LOV_MAGIC_V1) {
2425                 objs = &lmm->lmm_objects[0];
2426         } else {
2427                 LASSERT(magic == LOV_MAGIC_V3);
2428                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2429         }
2430
2431         count = le16_to_cpu(lmm->lmm_stripe_count);
2432         if (count == 0)
2433                 GOTO(unlock_parent, rc = -EINVAL);
2434         LASSERT(count > 0);
2435
2436         /* Exceed the current end of MDT-object layout EA. Then extend it. */
2437         if (count <= ea_off) {
2438                 if (bk->lb_param & LPF_DRYRUN)
2439                         GOTO(unlock_parent, rc = 1);
2440
2441                 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2442                 /* If the declared is not big enough, re-try. */
2443                 if (buf->lb_len < lovea_size) {
2444                         rc = lovea_size;
2445                         goto again;
2446                 }
2447
2448                 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2449                                                buf, fl, ost_idx, ea_off, false);
2450
2451                 GOTO(unlock_parent, rc);
2452         }
2453
2454         LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2455
2456         for (i = 0; i < count; i++, objs++) {
2457                 /* The MDT-object was created via lfsck_layout_recover_create()
2458                  * by others before, and we fill the dummy layout EA. */
2459                 if (lovea_slot_is_dummy(objs)) {
2460                         if (i != ea_off)
2461                                 continue;
2462
2463                         if (bk->lb_param & LPF_DRYRUN)
2464                                 GOTO(unlock_parent, rc = 1);
2465
2466                         lmm->lmm_layout_gen =
2467                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2468                         rc = lfsck_layout_refill_lovea(env, handle, parent,
2469                                                        cfid, buf, objs, fl,
2470                                                        ost_idx);
2471
2472                         CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2473                                "dummy layout slot for "DFID": parent "DFID
2474                                ", OST-index %u, stripe-index %u: rc = %d\n",
2475                                lfsck_lfsck2name(lfsck), PFID(cfid),
2476                                PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2477
2478                         GOTO(unlock_parent, rc);
2479                 }
2480
2481                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2482                 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2483                 if (rc != 0) {
2484                         CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2485                                "invalid layout EA at the slot %d, index %u\n",
2486                                lfsck_lfsck2name(lfsck),
2487                                PFID(lfsck_dto2fid(parent)), i,
2488                                le32_to_cpu(objs->l_ost_idx));
2489
2490                         GOTO(unlock_parent, rc);
2491                 }
2492
2493                 /* It should be rare case, the slot is there, but the LFSCK
2494                  * does not handle it during the first-phase cycle scanning. */
2495                 if (unlikely(lu_fid_eq(fid, cfid))) {
2496                         if (i == ea_off) {
2497                                 GOTO(unlock_parent, rc = 0);
2498                         } else {
2499                                 /* Rare case that the OST-object index
2500                                  * does not match the parent MDT-object
2501                                  * layout EA. We trust the later one. */
2502                                 if (bk->lb_param & LPF_DRYRUN)
2503                                         GOTO(unlock_parent, rc = 1);
2504
2505                                 dt_write_unlock(env, parent);
2506                                 if (handle != NULL)
2507                                         dt_trans_stop(env, dt, handle);
2508                                 lfsck_ibits_unlock(&lh, LCK_EX);
2509                                 rc = lfsck_layout_update_pfid(env, com, parent,
2510                                                         cfid, ltd->ltd_tgt, i);
2511
2512                                 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2513                                        "updated OST-object's pfid for "DFID
2514                                        ": parent "DFID", OST-index %u, "
2515                                        "stripe-index %u: rc = %d\n",
2516                                        lfsck_lfsck2name(lfsck), PFID(cfid),
2517                                        PFID(lfsck_dto2fid(parent)),
2518                                        ltd->ltd_index, i, rc);
2519
2520                                 RETURN(rc);
2521                         }
2522                 }
2523         }
2524
2525         /* The MDT-object exists, but related layout EA slot is occupied
2526          * by others. */
2527         if (bk->lb_param & LPF_DRYRUN)
2528                 GOTO(unlock_parent, rc = 1);
2529
2530         dt_write_unlock(env, parent);
2531         if (handle != NULL)
2532                 dt_trans_stop(env, dt, handle);
2533         lfsck_ibits_unlock(&lh, LCK_EX);
2534         if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2535                 objs = &lmm->lmm_objects[ea_off];
2536         else
2537                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2538         rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2539                                           buf, objs, ea_off);
2540
2541         RETURN(rc);
2542
2543 unlock_parent:
2544         if (locked)
2545                 dt_write_unlock(env, parent);
2546
2547 stop:
2548         if (handle != NULL)
2549                 dt_trans_stop(env, dt, handle);
2550
2551 unlock_layout:
2552         lfsck_ibits_unlock(&lh, LCK_EX);
2553
2554         return rc;
2555 }
2556
2557 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2558                                         struct lfsck_component *com,
2559                                         struct lfsck_tgt_desc *ltd,
2560                                         struct lu_orphan_rec *rec,
2561                                         struct lu_fid *cfid)
2562 {
2563         struct lfsck_layout     *lo     = com->lc_file_ram;
2564         struct lu_fid           *pfid   = &rec->lor_fid;
2565         struct dt_object        *parent = NULL;
2566         __u32                    ea_off = pfid->f_stripe_idx;
2567         int                      rc     = 0;
2568         ENTRY;
2569
2570         if (!fid_is_sane(cfid))
2571                 GOTO(out, rc = -EINVAL);
2572
2573         if (fid_is_zero(pfid)) {
2574                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2575                                                   "", "N", ea_off);
2576                 GOTO(out, rc);
2577         }
2578
2579         pfid->f_ver = 0;
2580         if (!fid_is_sane(pfid))
2581                 GOTO(out, rc = -EINVAL);
2582
2583         parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2584         if (IS_ERR(parent))
2585                 GOTO(out, rc = PTR_ERR(parent));
2586
2587         if (unlikely(dt_object_remote(parent) != 0))
2588                 GOTO(put, rc = -EXDEV);
2589
2590         if (dt_object_exists(parent) == 0) {
2591                 lfsck_object_put(env, parent);
2592                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2593                                                   "", "R", ea_off);
2594                 GOTO(out, rc);
2595         }
2596
2597         if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2598                 GOTO(put, rc = -EISDIR);
2599
2600         /* The orphan OST-object claims to be the parent's stripe, then
2601          * related dangling record in the trace file is meaningless. */
2602         rc = lfsck_layout_del_dangling_rec(env, com, pfid, ea_off);
2603         if (rc != 0 && rc != -ENOENT)
2604                 GOTO(put, rc);
2605
2606         rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2607                                          ltd->ltd_index, ea_off);
2608
2609         GOTO(put, rc);
2610
2611 put:
2612         if (rc <= 0)
2613                 lfsck_object_put(env, parent);
2614         else
2615                 /* The layout EA is changed, need to be reloaded next time. */
2616                 lu_object_put_nocache(env, &parent->do_lu);
2617
2618 out:
2619         down_write(&com->lc_sem);
2620         com->lc_new_scanned++;
2621         com->lc_new_checked++;
2622         if (rc > 0) {
2623                 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2624                 rc = 0;
2625         } else if (rc < 0) {
2626                 lo->ll_objs_failed_phase2++;
2627         }
2628         up_write(&com->lc_sem);
2629
2630         return rc;
2631 }
2632
2633 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2634                                     struct lfsck_component *com,
2635                                     struct lfsck_tgt_desc *ltd)
2636 {
2637         struct lfsck_assistant_data     *lad    = com->lc_data;
2638         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2639         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2640         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2641         struct ost_id                   *oi     = &info->lti_oi;
2642         struct lu_fid                   *fid    = &info->lti_fid;
2643         struct dt_object                *obj;
2644         const struct dt_it_ops          *iops;
2645         struct dt_it                    *di;
2646         int                              rc     = 0;
2647         ENTRY;
2648
2649         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
2650                "scanning for OST%04x\n",
2651                lfsck_lfsck2name(lfsck), ltd->ltd_index);
2652
2653         if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
2654                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
2655                        "scanning for OST%04x\n",
2656                        lfsck_lfsck2name(lfsck), ltd->ltd_index);
2657
2658                 RETURN(0);
2659         }
2660
2661         ostid_set_seq(oi, FID_SEQ_IDIF);
2662         ostid_set_id(oi, 0);
2663         rc = ostid_to_fid(fid, oi, ltd->ltd_index);
2664         if (rc != 0)
2665                 GOTO(log, rc);
2666
2667         obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2668         if (unlikely(IS_ERR(obj)))
2669                 GOTO(log, rc = PTR_ERR(obj));
2670
2671         rc = obj->do_ops->do_index_try(env, obj,
2672                                        &dt_lfsck_layout_orphan_features);
2673         if (rc != 0)
2674                 GOTO(put, rc);
2675
2676         iops = &obj->do_index_ops->dio_it;
2677         di = iops->init(env, obj, 0);
2678         if (IS_ERR(di))
2679                 GOTO(put, rc = PTR_ERR(di));
2680
2681         rc = iops->load(env, di, 0);
2682         if (rc == -ESRCH) {
2683                 /* -ESRCH means that the orphan OST-objects rbtree has been
2684                  * cleanup because of the OSS server restart or other errors. */
2685                 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
2686                 GOTO(fini, rc);
2687         }
2688
2689         if (rc == 0)
2690                 rc = iops->next(env, di);
2691         else if (rc > 0)
2692                 rc = 0;
2693
2694         if (rc < 0)
2695                 GOTO(fini, rc);
2696
2697         if (rc > 0)
2698                 GOTO(fini, rc = 0);
2699
2700         do {
2701                 struct dt_key           *key;
2702                 struct lu_orphan_rec    *rec = &info->lti_rec;
2703
2704                 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
2705                     unlikely(!thread_is_running(&lfsck->li_thread)))
2706                         break;
2707
2708                 key = iops->key(env, di);
2709                 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2710                 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2711                 if (rc == 0)
2712                         rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2713                                         &com->lc_fid_latest_scanned_phase2);
2714                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2715                         GOTO(fini, rc);
2716
2717                 lfsck_control_speed_by_self(com);
2718                 do {
2719                         rc = iops->next(env, di);
2720                 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2721         } while (rc == 0);
2722
2723         GOTO(fini, rc);
2724
2725 fini:
2726         iops->put(env, di);
2727         iops->fini(env, di);
2728 put:
2729         lfsck_object_put(env, obj);
2730
2731 log:
2732         CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
2733                "scanning for OST%04x: rc = %d\n",
2734                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2735
2736         return rc > 0 ? 0 : rc;
2737 }
2738
2739 /**
2740  * Repair the MDT-object with dangling LOV EA reference.
2741  *
2742  * we need to repair the inconsistency according to the users' requirement:
2743  *
2744  * 1) Keep the inconsistency there and report the inconsistency case,
2745  *    then give the chance to the application to find related issues,
2746  *    and the users can make the decision about how to handle it with
2747  *    more human knownledge. (by default)
2748  *
2749  * 2) Re-create the missing OST-object with the FID/owner information.
2750  *
2751  * \param[in] env       pointer to the thread context
2752  * \param[in] com       the layout LFSCK component
2753  * \param[in] parent    the MDT-object with dangling LOV EA reference
2754  * \param[in] child     the OST-object to be created
2755  * \param[in] ea_off    the offset of the OST-object in the LOV EA
2756  * \param[in] ost_idx   the index of OST on which the OST-object resides
2757  *
2758  * \retval              +1 for repair successfully
2759  * \retval              0 for did nothing
2760  * \retval              negative error number on failure
2761  */
2762 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
2763                                           struct lfsck_component *com,
2764                                           struct dt_object *parent,
2765                                           struct dt_object *child,
2766                                           __u32 ea_off, __u32 ost_idx, bool log)
2767 {
2768         struct lfsck_thread_info *info = lfsck_env_info(env);
2769         struct filter_fid *ff = &info->lti_new_pfid;
2770         struct dt_object_format *dof = &info->lti_dof;
2771         struct lu_attr *la = &info->lti_la;
2772         struct lfsck_instance *lfsck = com->lc_lfsck;
2773         struct dt_device *dev = lfsck_obj2dev(child);
2774         const struct lu_fid *pfid = lfsck_dto2fid(parent);
2775         const struct lu_fid *cfid = lfsck_dto2fid(child);
2776         struct thandle *handle;
2777         struct lu_buf *buf;
2778         struct lustre_handle lh = { 0 };
2779         int rc;
2780         ENTRY;
2781
2782         if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
2783                 GOTO(log, rc = 1);
2784
2785         rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2786                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2787                               LCK_EX);
2788         if (rc != 0)
2789                 GOTO(log, rc);
2790
2791         rc = dt_attr_get(env, parent, la);
2792         if (rc != 0)
2793                 GOTO(unlock1, rc);
2794
2795         la->la_mode = S_IFREG | 0666;
2796         la->la_atime = la->la_mtime = la->la_ctime = 0;
2797         la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2798                        LA_ATIME | LA_MTIME | LA_CTIME;
2799         memset(dof, 0, sizeof(*dof));
2800         ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2801         ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2802         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2803          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2804          * parent MDT-object's layout EA. */
2805         ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
2806         buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
2807
2808         handle = dt_trans_create(env, dev);
2809         if (IS_ERR(handle))
2810                 GOTO(unlock1, rc = PTR_ERR(handle));
2811
2812         rc = dt_declare_create(env, child, la, NULL, dof, handle);
2813         if (rc != 0)
2814                 GOTO(stop, rc);
2815
2816         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2817                                   LU_XATTR_CREATE, handle);
2818         if (rc != 0)
2819                 GOTO(stop, rc);
2820
2821         rc = dt_trans_start_local(env, dev, handle);
2822         if (rc != 0)
2823                 GOTO(stop, rc);
2824
2825         dt_read_lock(env, parent, 0);
2826         if (unlikely(lfsck_is_dead_obj(parent)))
2827                 GOTO(unlock2, rc = 0);
2828
2829         if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
2830                 struct ost_id *oi = &info->lti_oi;
2831                 struct lu_fid *tfid = &info->lti_fid2;
2832                 struct lu_buf *lovea = &info->lti_big_buf;
2833                 struct lov_mds_md_v1 *lmm;
2834                 struct lov_ost_data_v1 *objs;
2835                 __u32 magic;
2836                 int count;
2837                 int idx2;
2838
2839                 rc = lfsck_layout_get_lovea(env, parent, lovea);
2840                 if (rc <= 0)
2841                         GOTO(unlock2, rc);
2842
2843                 lmm = lovea->lb_buf;
2844                 rc = lfsck_layout_verify_header(lmm);
2845                 if (unlikely(rc != 0))
2846                         GOTO(unlock2, rc);
2847
2848                 count = le16_to_cpu(lmm->lmm_stripe_count);
2849                 /* Someone changed the LOV EA, do nothing. */
2850                 if (count <= ea_off)
2851                         GOTO(unlock2, rc = 0);
2852
2853                 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which
2854                  * has been verified in lfsck_layout_verify_header() already.
2855                  * If some new magic introduced in the future, then the layout
2856                  * LFSCK needs to be updated also. */
2857                 magic = le32_to_cpu(lmm->lmm_magic);
2858                 if (magic == LOV_MAGIC_V1) {
2859                         objs = &lmm->lmm_objects[ea_off];
2860                 } else {
2861                         LASSERT(magic == LOV_MAGIC_V3);
2862
2863                         objs = &((struct lov_mds_md_v3 *)lmm)->\
2864                                                         lmm_objects[ea_off];
2865                 }
2866
2867                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2868                 idx2 = le32_to_cpu(objs->l_ost_idx);
2869                 rc = ostid_to_fid(tfid, oi, idx2);
2870                 /* Someone changed the LOV EA, do nothing. */
2871                 if (rc != 0 || !lu_fid_eq(tfid, cfid))
2872                         GOTO(unlock2, rc);
2873         }
2874
2875         rc = dt_create(env, child, la, NULL, dof, handle);
2876         if (rc != 0)
2877                 GOTO(unlock2, rc);
2878
2879         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2880                           handle);
2881
2882         GOTO(unlock2, rc);
2883
2884 unlock2:
2885         dt_read_unlock(env, parent);
2886
2887 stop:
2888         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2889
2890 unlock1:
2891         lfsck_ibits_unlock(&lh, LCK_EX);
2892
2893 log:
2894         if (rc != 0 && log)
2895                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2896                        "dangling reference for: parent "DFID", child "
2897                        DFID", ea_off %u, ost_idx %u, %s: rc = %d\n",
2898                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2899                        ea_off, ost_idx,
2900                        (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
2901                                 "Create the lost OST-object as required" :
2902                                 "Keep the MDT-object there by default", rc);
2903
2904         return rc;
2905 }
2906
2907 /**
2908  * Repair the MDT-object with dangling LOV EA reference.
2909  *
2910  * Prepare parameters and call __lfsck_layout_repair_dangling()
2911  * to repair the dangling LOV EA reference.
2912  *
2913  * \param[in] env       pointer to the thread context
2914  * \param[in] com       the layout LFSCK component
2915  * \param[in] pfid      the MDT-object's FID
2916  * \param[in] cfid      the FID for the OST-object to be created
2917  * \param[in] ea_off    the offset of the OST-object in the LOV EA
2918  * \param[in] ost_idx   the index of OST on which the OST-object resides
2919  *
2920  * \retval              +1 for repair successfully
2921  * \retval              0 for did nothing
2922  * \retval              negative error number on failure
2923  */
2924 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2925                                         struct lfsck_component *com,
2926                                         const struct lu_fid *pfid,
2927                                         const struct lu_fid *cfid,
2928                                         __u32 ea_off, __u32 ost_idx)
2929 {
2930         struct lfsck_instance *lfsck = com->lc_lfsck;
2931         struct dt_object *parent = NULL;
2932         struct dt_object *child = NULL;
2933         struct lfsck_tgt_desc *ltd;
2934         int rc;
2935         ENTRY;
2936
2937         parent = lfsck_object_find_bottom(env, lfsck, pfid);
2938         if (IS_ERR(parent))
2939                 GOTO(log, rc = PTR_ERR(parent));
2940
2941         /* The MDT-object has been removed. */
2942         if (dt_object_exists(parent) == 0)
2943                 GOTO(log, rc = 0);
2944
2945         ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
2946         if (unlikely(ltd == NULL))
2947                 GOTO(log, rc = -ENODEV);
2948
2949         child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2950         if (IS_ERR(child))
2951                 GOTO(log, rc = PTR_ERR(child));
2952
2953         /* The OST-object has been created. */
2954         if (unlikely(dt_object_exists(child) != 0))
2955                 GOTO(log, rc = 0);
2956
2957         rc = __lfsck_layout_repair_dangling(env, com, parent, child,
2958                                             ea_off, ost_idx, false);
2959
2960         GOTO(log, rc);
2961
2962 log:
2963         if (child != NULL && !IS_ERR(child))
2964                 lfsck_object_put(env, child);
2965
2966         if (parent != NULL && !IS_ERR(parent))
2967                 lfsck_object_put(env, parent);
2968
2969         if (rc != 0)
2970                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2971                        "dangling reference for: parent "DFID", child "
2972                        DFID", ea_off %u, ost_idx %u, %s: rc = %d\n",
2973                        lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2974                        ea_off, ost_idx,
2975                        (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
2976                                 "Create the lost OST-object as required" :
2977                                 "Keep the MDT-object there by default", rc);
2978
2979         return rc;
2980 }
2981
2982 /* If the OST-object does not recognize the MDT-object as its parent, and
2983  * there is no other MDT-object claims as its parent, then just trust the
2984  * given MDT-object as its parent. So update the OST-object filter_fid. */
2985 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2986                                               struct lfsck_component *com,
2987                                               struct dt_object *parent,
2988                                               struct lfsck_layout_req *llr,
2989                                               struct lu_attr *la)
2990 {
2991         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2992         struct filter_fid               *pfid   = &info->lti_new_pfid;
2993         struct dt_object                *child  = llr->llr_child;
2994         struct dt_device                *dev    = lfsck_obj2dev(child);
2995         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2996         struct thandle                  *handle;
2997         struct lu_buf                   *buf;
2998         struct lustre_handle             lh     = { 0 };
2999         int                              rc;
3000         ENTRY;
3001
3002         rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3003                               MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3004                               LCK_EX);
3005         if (rc != 0)
3006                 GOTO(log, rc);
3007
3008         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3009         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3010         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3011          * MDT-object's FID::f_ver, instead it is the OST-object index in its
3012          * parent MDT-object's layout EA. */
3013         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3014         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
3015
3016         handle = dt_trans_create(env, dev);
3017         if (IS_ERR(handle))
3018                 GOTO(unlock1, rc = PTR_ERR(handle));
3019
3020         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3021         if (rc != 0)
3022                 GOTO(stop, rc);
3023
3024         rc = dt_attr_get(env, parent, la);
3025         if (rc != 0)
3026                 GOTO(stop, rc);
3027
3028         la->la_valid = LA_UID | LA_GID;
3029         rc = dt_declare_attr_set(env, child, la, handle);
3030         if (rc != 0)
3031                 GOTO(stop, rc);
3032
3033         rc = dt_trans_start_local(env, dev, handle);
3034         if (rc != 0)
3035                 GOTO(stop, rc);
3036
3037         dt_write_lock(env, parent, 0);
3038         if (unlikely(lfsck_is_dead_obj(parent)))
3039                 GOTO(unlock2, rc = 1);
3040
3041         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3042         if (rc != 0)
3043                 GOTO(unlock2, rc);
3044
3045         /* Get the latest parent's owner. */
3046         rc = dt_attr_get(env, parent, la);
3047         if (rc != 0)
3048                 GOTO(unlock2, rc);
3049
3050         la->la_valid = LA_UID | LA_GID;
3051         rc = dt_attr_set(env, child, la, handle);
3052
3053         GOTO(unlock2, rc);
3054
3055 unlock2:
3056         dt_write_unlock(env, parent);
3057
3058 stop:
3059         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3060
3061 unlock1:
3062         lfsck_ibits_unlock(&lh, LCK_EX);
3063
3064 log:
3065         if (rc != 0)
3066                 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3067                        "unmatched MDT-OST pair for: parent "DFID
3068                        ", child "DFID", OST-index %u, stripe-index %u, "
3069                        "owner %u/%u: rc = %d\n",
3070                        lfsck_lfsck2name(com->lc_lfsck),
3071                        PFID(lfsck_dto2fid(parent)),
3072                        PFID(lfsck_dto2fid(child)),
3073                        llr->llr_ost_idx, llr->llr_lov_idx,
3074                        la->la_uid, la->la_gid, rc);
3075
3076         return rc;
3077 }
3078
3079 /* If there are more than one MDT-objects claim as the OST-object's parent,
3080  * and the OST-object only recognizes one of them, then we need to generate
3081  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3082 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3083                                                    struct lfsck_component *com,
3084                                                    struct dt_object *parent,
3085                                                    struct lfsck_layout_req *llr,
3086                                                    struct lu_attr *la,
3087                                                    struct lu_buf *buf)
3088 {
3089         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3090         struct dt_allocation_hint       *hint   = &info->lti_hint;
3091         struct dt_object_format         *dof    = &info->lti_dof;
3092         struct ost_id                   *oi     = &info->lti_oi;
3093         struct lfsck_instance           *lfsck  = com->lc_lfsck;
3094         struct dt_device                *dev;
3095         struct lu_device                *d      =
3096                                 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3097         struct lu_object                *o;
3098         struct lu_object                *n;
3099         struct dt_object                *child  = NULL;
3100         struct thandle                  *handle = NULL;
3101         struct lov_mds_md_v1            *lmm;
3102         struct lov_ost_data_v1          *objs;
3103         const struct lu_fid             *pfid   = lfsck_dto2fid(parent);
3104         struct lu_fid                    tfid;
3105         struct lustre_handle             lh     = { 0 };
3106         struct lu_buf                    ea_buf;
3107         __u32                            magic;
3108         __u32                            index;
3109         int                              rc;
3110         ENTRY;
3111
3112         /* We use two separated transactions to repair the inconsistency.
3113          *
3114          * 1) create the child (OST-object).
3115          * 2) update the parent LOV EA according to the child's FID.
3116          *
3117          * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3118          * handled as orphan when the layout LFSCK run next time.
3119          *
3120          * If 1) failed, but 2) succeed, then such OST-object will be re-created
3121          * as dangling referened case when the layout LFSCK run next time. */
3122
3123         /* The 1st transaction. */
3124         o = lu_object_anon(env, d, NULL);
3125         if (IS_ERR(o))
3126                 GOTO(log, rc = PTR_ERR(o));
3127
3128