Whamcloud - gitweb
LU-4941 lfsck: check LOV EA header 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) 2013, 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_linkea.h>
43 #include <lustre_fid.h>
44 #include <lustre_lib.h>
45 #include <lustre_net.h>
46 #include <lustre/lustre_user.h>
47 #include <md_object.h>
48 #include <obd_class.h>
49
50 #include "lfsck_internal.h"
51
52 #define LFSCK_LAYOUT_MAGIC              0xB173AE14
53
54 static const char lfsck_layout_name[] = "lfsck_layout";
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 };
76
77 struct lfsck_layout_slave_data {
78         /* list for lfsck_layout_seq */
79         struct list_head         llsd_seq_list;
80
81         /* list for the masters involve layout verification. */
82         struct list_head         llsd_master_list;
83         spinlock_t               llsd_lock;
84         __u64                    llsd_touch_gen;
85         struct dt_object        *llsd_rb_obj;
86         struct rb_root           llsd_rb_root;
87         rwlock_t                 llsd_rb_lock;
88         unsigned int             llsd_rbtree_valid:1;
89 };
90
91 struct lfsck_layout_object {
92         struct dt_object        *llo_obj;
93         struct lu_attr           llo_attr;
94         atomic_t                 llo_ref;
95         __u16                    llo_gen;
96 };
97
98 struct lfsck_layout_req {
99         struct list_head                 llr_list;
100         struct lfsck_layout_object      *llr_parent;
101         struct dt_object                *llr_child;
102         __u32                            llr_ost_idx;
103         __u32                            llr_lov_idx; /* offset in LOV EA */
104 };
105
106 struct lfsck_layout_master_data {
107         spinlock_t              llmd_lock;
108         struct list_head        llmd_req_list;
109
110         /* list for the ost targets involve layout verification. */
111         struct list_head        llmd_ost_list;
112
113         /* list for the ost targets in phase1 scanning. */
114         struct list_head        llmd_ost_phase1_list;
115
116         /* list for the ost targets in phase1 scanning. */
117         struct list_head        llmd_ost_phase2_list;
118
119         /* list for the mdt targets involve layout verification. */
120         struct list_head        llmd_mdt_list;
121
122         /* list for the mdt targets in phase1 scanning. */
123         struct list_head        llmd_mdt_phase1_list;
124
125         /* list for the mdt targets in phase1 scanning. */
126         struct list_head        llmd_mdt_phase2_list;
127
128         struct ptlrpc_thread    llmd_thread;
129         __u32                   llmd_touch_gen;
130         int                     llmd_prefetched;
131         int                     llmd_assistant_status;
132         int                     llmd_post_result;
133         unsigned int            llmd_to_post:1,
134                                 llmd_to_double_scan:1,
135                                 llmd_in_double_scan:1,
136                                 llmd_exit:1;
137 };
138
139 struct lfsck_layout_slave_async_args {
140         struct obd_export                *llsaa_exp;
141         struct lfsck_component           *llsaa_com;
142         struct lfsck_layout_slave_target *llsaa_llst;
143 };
144
145 static struct lfsck_layout_object *
146 lfsck_layout_object_init(const struct lu_env *env, struct dt_object *obj,
147                          __u16 gen)
148 {
149         struct lfsck_layout_object *llo;
150         int                         rc;
151
152         OBD_ALLOC_PTR(llo);
153         if (llo == NULL)
154                 return ERR_PTR(-ENOMEM);
155
156         rc = dt_attr_get(env, obj, &llo->llo_attr, BYPASS_CAPA);
157         if (rc != 0) {
158                 OBD_FREE_PTR(llo);
159
160                 return ERR_PTR(rc);
161         }
162
163         lu_object_get(&obj->do_lu);
164         llo->llo_obj = obj;
165         /* The gen can be used to check whether some others have changed the
166          * file layout after LFSCK pre-fetching but before real verification. */
167         llo->llo_gen = gen;
168         atomic_set(&llo->llo_ref, 1);
169
170         return llo;
171 }
172
173 static inline void
174 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
175 {
176         if (atomic_dec_and_test(&llst->llst_ref)) {
177                 LASSERT(list_empty(&llst->llst_list));
178
179                 OBD_FREE_PTR(llst);
180         }
181 }
182
183 static inline int
184 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
185 {
186         struct lfsck_layout_slave_target *llst;
187         struct lfsck_layout_slave_target *tmp;
188         int                               rc   = 0;
189
190         OBD_ALLOC_PTR(llst);
191         if (llst == NULL)
192                 return -ENOMEM;
193
194         INIT_LIST_HEAD(&llst->llst_list);
195         llst->llst_gen = 0;
196         llst->llst_index = index;
197         atomic_set(&llst->llst_ref, 1);
198
199         spin_lock(&llsd->llsd_lock);
200         list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
201                 if (tmp->llst_index == index) {
202                         rc = -EALREADY;
203                         break;
204                 }
205         }
206         if (rc == 0)
207                 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
208         spin_unlock(&llsd->llsd_lock);
209
210         if (rc != 0)
211                 OBD_FREE_PTR(llst);
212
213         return rc;
214 }
215
216 static inline void
217 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
218                       struct lfsck_layout_slave_target *llst)
219 {
220         bool del = false;
221
222         spin_lock(&llsd->llsd_lock);
223         if (!list_empty(&llst->llst_list)) {
224                 list_del_init(&llst->llst_list);
225                 del = true;
226         }
227         spin_unlock(&llsd->llsd_lock);
228
229         if (del)
230                 lfsck_layout_llst_put(llst);
231 }
232
233 static inline struct lfsck_layout_slave_target *
234 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
235                                __u32 index, bool unlink)
236 {
237         struct lfsck_layout_slave_target *llst;
238
239         spin_lock(&llsd->llsd_lock);
240         list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
241                 if (llst->llst_index == index) {
242                         if (unlink)
243                                 list_del_init(&llst->llst_list);
244                         else
245                                 atomic_inc(&llst->llst_ref);
246                         spin_unlock(&llsd->llsd_lock);
247
248                         return llst;
249                 }
250         }
251         spin_unlock(&llsd->llsd_lock);
252
253         return NULL;
254 }
255
256 static inline void lfsck_layout_object_put(const struct lu_env *env,
257                                            struct lfsck_layout_object *llo)
258 {
259         if (atomic_dec_and_test(&llo->llo_ref)) {
260                 lfsck_object_put(env, llo->llo_obj);
261                 OBD_FREE_PTR(llo);
262         }
263 }
264
265 static struct lfsck_layout_req *
266 lfsck_layout_req_init(struct lfsck_layout_object *parent,
267                       struct dt_object *child, __u32 ost_idx, __u32 lov_idx)
268 {
269         struct lfsck_layout_req *llr;
270
271         OBD_ALLOC_PTR(llr);
272         if (llr == NULL)
273                 return ERR_PTR(-ENOMEM);
274
275         INIT_LIST_HEAD(&llr->llr_list);
276         atomic_inc(&parent->llo_ref);
277         llr->llr_parent = parent;
278         llr->llr_child = child;
279         llr->llr_ost_idx = ost_idx;
280         llr->llr_lov_idx = lov_idx;
281
282         return llr;
283 }
284
285 static inline void lfsck_layout_req_fini(const struct lu_env *env,
286                                          struct lfsck_layout_req *llr)
287 {
288         lu_object_put(env, &llr->llr_child->do_lu);
289         lfsck_layout_object_put(env, llr->llr_parent);
290         OBD_FREE_PTR(llr);
291 }
292
293 static inline bool lfsck_layout_req_empty(struct lfsck_layout_master_data *llmd)
294 {
295         bool empty = false;
296
297         spin_lock(&llmd->llmd_lock);
298         if (list_empty(&llmd->llmd_req_list))
299                 empty = true;
300         spin_unlock(&llmd->llmd_lock);
301
302         return empty;
303 }
304
305 static int lfsck_layout_get_lovea(const struct lu_env *env,
306                                   struct dt_object *obj,
307                                   struct lu_buf *buf, ssize_t *buflen)
308 {
309         int rc;
310
311 again:
312         rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV, BYPASS_CAPA);
313         if (rc == -ERANGE) {
314                 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
315                                   BYPASS_CAPA);
316                 if (rc <= 0)
317                         return rc;
318
319                 lu_buf_realloc(buf, rc);
320                 if (buflen != NULL)
321                         *buflen = buf->lb_len;
322
323                 if (buf->lb_buf == NULL)
324                         return -ENOMEM;
325
326                 goto again;
327         }
328
329         if (rc == -ENODATA)
330                 rc = 0;
331
332         if (rc <= 0)
333                 return rc;
334
335         if (unlikely(buf->lb_buf == NULL)) {
336                 lu_buf_alloc(buf, rc);
337                 if (buflen != NULL)
338                         *buflen = buf->lb_len;
339
340                 if (buf->lb_buf == NULL)
341                         return -ENOMEM;
342
343                 goto again;
344         }
345
346         return rc;
347 }
348
349 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
350 {
351         __u32 magic;
352         __u32 pattern;
353
354         magic = le32_to_cpu(lmm->lmm_magic);
355         /* If magic crashed, keep it there. Sometime later, during OST-object
356          * orphan handling, if some OST-object(s) back-point to it, it can be
357          * verified and repaired. */
358         if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
359                 struct ost_id   oi;
360                 int             rc;
361
362                 lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
363                 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
364                         rc = -EOPNOTSUPP;
365                 else
366                         rc = -EINVAL;
367
368                 CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
369                        rc == -EINVAL ? "Unknown" : "Unsupported",
370                        magic, POSTID(&oi));
371
372                 return rc;
373         }
374
375         pattern = le32_to_cpu(lmm->lmm_pattern);
376         /* XXX: currently, we only support LOV_PATTERN_RAID0. */
377         if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
378                 struct ost_id oi;
379
380                 lmm_oi_cpu_to_le(&oi, &lmm->lmm_oi);
381                 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
382                        pattern, POSTID(&oi));
383
384                 return -EOPNOTSUPP;
385         }
386
387         return 0;
388 }
389
390 #define LFSCK_RBTREE_BITMAP_SIZE        PAGE_CACHE_SIZE
391 #define LFSCK_RBTREE_BITMAP_WIDTH       (LFSCK_RBTREE_BITMAP_SIZE << 3)
392 #define LFSCK_RBTREE_BITMAP_MASK        (LFSCK_RBTREE_BITMAP_WIDTH - 1)
393
394 struct lfsck_rbtree_node {
395         struct rb_node   lrn_node;
396         __u64            lrn_seq;
397         __u32            lrn_first_oid;
398         atomic_t         lrn_known_count;
399         atomic_t         lrn_accessed_count;
400         void            *lrn_known_bitmap;
401         void            *lrn_accessed_bitmap;
402 };
403
404 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
405                                    __u64 seq, __u32 oid)
406 {
407         if (seq < lrn->lrn_seq)
408                 return -1;
409
410         if (seq > lrn->lrn_seq)
411                 return 1;
412
413         if (oid < lrn->lrn_first_oid)
414                 return -1;
415
416         if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
417                 return 1;
418
419         return 0;
420 }
421
422 /* The caller should hold llsd->llsd_rb_lock. */
423 static struct lfsck_rbtree_node *
424 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
425                     const struct lu_fid *fid, bool *exact)
426 {
427         struct rb_node           *node  = llsd->llsd_rb_root.rb_node;
428         struct rb_node           *prev  = NULL;
429         struct lfsck_rbtree_node *lrn   = NULL;
430         int                       rc    = 0;
431
432         if (exact != NULL)
433                 *exact = true;
434
435         while (node != NULL) {
436                 prev = node;
437                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
438                 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
439                 if (rc < 0)
440                         node = node->rb_left;
441                 else if (rc > 0)
442                         node = node->rb_right;
443                 else
444                         return lrn;
445         }
446
447         if (exact == NULL)
448                 return NULL;
449
450         /* If there is no exactly matched one, then to the next valid one. */
451         *exact = false;
452
453         /* The rbtree is empty. */
454         if (rc == 0)
455                 return NULL;
456
457         if (rc < 0)
458                 return lrn;
459
460         node = rb_next(prev);
461
462         /* The end of the rbtree. */
463         if (node == NULL)
464                 return NULL;
465
466         lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
467
468         return lrn;
469 }
470
471 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
472                                                   const struct lu_fid *fid)
473 {
474         struct lfsck_rbtree_node *lrn;
475
476         OBD_ALLOC_PTR(lrn);
477         if (lrn == NULL)
478                 return ERR_PTR(-ENOMEM);
479
480         OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
481         if (lrn->lrn_known_bitmap == NULL) {
482                 OBD_FREE_PTR(lrn);
483
484                 return ERR_PTR(-ENOMEM);
485         }
486
487         OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
488         if (lrn->lrn_accessed_bitmap == NULL) {
489                 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
490                 OBD_FREE_PTR(lrn);
491
492                 return ERR_PTR(-ENOMEM);
493         }
494
495         rb_init_node(&lrn->lrn_node);
496         lrn->lrn_seq = fid_seq(fid);
497         lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
498         atomic_set(&lrn->lrn_known_count, 0);
499         atomic_set(&lrn->lrn_accessed_count, 0);
500
501         return lrn;
502 }
503
504 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
505 {
506         OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
507         OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
508         OBD_FREE_PTR(lrn);
509 }
510
511 /* The caller should hold lock. */
512 static struct lfsck_rbtree_node *
513 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
514                     struct lfsck_rbtree_node *lrn)
515 {
516         struct rb_node           **pos    = &llsd->llsd_rb_root.rb_node;
517         struct rb_node            *parent = NULL;
518         struct lfsck_rbtree_node  *tmp;
519         int                        rc;
520
521         while (*pos != NULL) {
522                 parent = *pos;
523                 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
524                 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
525                 if (rc < 0)
526                         pos = &(*pos)->rb_left;
527                 else if (rc > 0)
528                         pos = &(*pos)->rb_right;
529                 else
530                         return tmp;
531         }
532
533         rb_link_node(&lrn->lrn_node, parent, pos);
534         rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
535
536         return lrn;
537 }
538
539 extern const struct dt_index_operations lfsck_orphan_index_ops;
540
541 static int lfsck_rbtree_setup(const struct lu_env *env,
542                               struct lfsck_component *com)
543 {
544         struct lu_fid                   *fid    = &lfsck_env_info(env)->lti_fid;
545         struct lfsck_instance           *lfsck  = com->lc_lfsck;
546         struct dt_device                *dev    = lfsck->li_bottom;
547         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
548         struct dt_object                *obj;
549
550         fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
551         fid->f_oid = lfsck_dev_idx(dev);
552         fid->f_ver = 0;
553         obj = dt_locate(env, dev, fid);
554         if (IS_ERR(obj))
555                 RETURN(PTR_ERR(obj));
556
557         /* Generate an in-RAM object to stand for the layout rbtree.
558          * Scanning the layout rbtree will be via the iteration over
559          * the object. In the future, the rbtree may be written onto
560          * disk with the object.
561          *
562          * Mark the object to be as exist. */
563         obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
564         obj->do_index_ops = &lfsck_orphan_index_ops;
565         llsd->llsd_rb_obj = obj;
566         llsd->llsd_rbtree_valid = 1;
567         dev->dd_record_fid_accessed = 1;
568
569         return 0;
570 }
571
572 static void lfsck_rbtree_cleanup(const struct lu_env *env,
573                                  struct lfsck_component *com)
574 {
575         struct lfsck_instance           *lfsck = com->lc_lfsck;
576         struct lfsck_layout_slave_data  *llsd  = com->lc_data;
577         struct rb_node                  *node  = rb_first(&llsd->llsd_rb_root);
578         struct rb_node                  *next;
579         struct lfsck_rbtree_node        *lrn;
580
581         lfsck->li_bottom->dd_record_fid_accessed = 0;
582         /* Invalid the rbtree, then no others will use it. */
583         write_lock(&llsd->llsd_rb_lock);
584         llsd->llsd_rbtree_valid = 0;
585         write_unlock(&llsd->llsd_rb_lock);
586
587         while (node != NULL) {
588                 next = rb_next(node);
589                 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
590                 rb_erase(node, &llsd->llsd_rb_root);
591                 lfsck_rbtree_free(lrn);
592                 node = next;
593         }
594
595         if (llsd->llsd_rb_obj != NULL) {
596                 lu_object_put(env, &llsd->llsd_rb_obj->do_lu);
597                 llsd->llsd_rb_obj = NULL;
598         }
599 }
600
601 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
602                                        struct lfsck_component *com,
603                                        const struct lu_fid *fid,
604                                        bool accessed)
605 {
606         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
607         struct lfsck_rbtree_node        *lrn;
608         bool                             insert = false;
609         int                              idx;
610         int                              rc     = 0;
611         ENTRY;
612
613         CDEBUG(D_LFSCK, "%s: update bitmap for "DFID"\n",
614                lfsck_lfsck2name(com->lc_lfsck), PFID(fid));
615
616         if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
617                 RETURN_EXIT;
618
619         if (!fid_is_idif(fid) && !fid_is_norm(fid))
620                 RETURN_EXIT;
621
622         read_lock(&llsd->llsd_rb_lock);
623         if (!llsd->llsd_rbtree_valid)
624                 GOTO(unlock, rc = 0);
625
626         lrn = lfsck_rbtree_search(llsd, fid, NULL);
627         if (lrn == NULL) {
628                 struct lfsck_rbtree_node *tmp;
629
630                 LASSERT(!insert);
631
632                 read_unlock(&llsd->llsd_rb_lock);
633                 tmp = lfsck_rbtree_new(env, fid);
634                 if (IS_ERR(tmp))
635                         GOTO(out, rc = PTR_ERR(tmp));
636
637                 insert = true;
638                 write_lock(&llsd->llsd_rb_lock);
639                 if (!llsd->llsd_rbtree_valid) {
640                         lfsck_rbtree_free(tmp);
641                         GOTO(unlock, rc = 0);
642                 }
643
644                 lrn = lfsck_rbtree_insert(llsd, tmp);
645                 if (lrn != tmp)
646                         lfsck_rbtree_free(tmp);
647         }
648
649         idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
650         /* Any accessed object must be a known object. */
651         if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
652                 atomic_inc(&lrn->lrn_known_count);
653         if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
654                 atomic_inc(&lrn->lrn_accessed_count);
655
656         GOTO(unlock, rc = 0);
657
658 unlock:
659         if (insert)
660                 write_unlock(&llsd->llsd_rb_lock);
661         else
662                 read_unlock(&llsd->llsd_rb_lock);
663 out:
664         if (rc != 0 && accessed) {
665                 struct lfsck_layout *lo = com->lc_file_ram;
666
667                 CERROR("%s: Fail to update object accessed bitmap, will cause "
668                        "incorrect LFSCK OST-object handling, so disable it to "
669                        "cancel orphan handling for related device. rc = %d.\n",
670                        lfsck_lfsck2name(com->lc_lfsck), rc);
671                 lo->ll_flags |= LF_INCOMPLETE;
672                 lfsck_rbtree_cleanup(env, com);
673         }
674 }
675
676 static inline bool is_dummy_lov_ost_data(struct lov_ost_data_v1 *obj)
677 {
678         if (fid_is_zero(&obj->l_ost_oi.oi_fid) &&
679             obj->l_ost_gen == 0 && obj->l_ost_idx == 0)
680                 return true;
681
682         return false;
683 }
684
685 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
686                                    const struct lfsck_layout *src)
687 {
688         int i;
689
690         des->ll_magic = le32_to_cpu(src->ll_magic);
691         des->ll_status = le32_to_cpu(src->ll_status);
692         des->ll_flags = le32_to_cpu(src->ll_flags);
693         des->ll_success_count = le32_to_cpu(src->ll_success_count);
694         des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
695         des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
696         des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
697         des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
698         des->ll_time_last_checkpoint =
699                                 le64_to_cpu(src->ll_time_last_checkpoint);
700         des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
701         des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
702         des->ll_pos_first_inconsistent =
703                         le64_to_cpu(src->ll_pos_first_inconsistent);
704         des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
705         des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
706         des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
707         des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
708         for (i = 0; i < LLIT_MAX; i++)
709                 des->ll_objs_repaired[i] =
710                                 le64_to_cpu(src->ll_objs_repaired[i]);
711         des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
712 }
713
714 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
715                                    const struct lfsck_layout *src)
716 {
717         int i;
718
719         des->ll_magic = cpu_to_le32(src->ll_magic);
720         des->ll_status = cpu_to_le32(src->ll_status);
721         des->ll_flags = cpu_to_le32(src->ll_flags);
722         des->ll_success_count = cpu_to_le32(src->ll_success_count);
723         des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
724         des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
725         des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
726         des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
727         des->ll_time_last_checkpoint =
728                                 cpu_to_le64(src->ll_time_last_checkpoint);
729         des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
730         des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
731         des->ll_pos_first_inconsistent =
732                         cpu_to_le64(src->ll_pos_first_inconsistent);
733         des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
734         des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
735         des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
736         des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
737         for (i = 0; i < LLIT_MAX; i++)
738                 des->ll_objs_repaired[i] =
739                                 cpu_to_le64(src->ll_objs_repaired[i]);
740         des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
741 }
742
743 /**
744  * \retval +ve: the lfsck_layout is broken, the caller should reset it.
745  * \retval 0: succeed.
746  * \retval -ve: failed cases.
747  */
748 static int lfsck_layout_load(const struct lu_env *env,
749                              struct lfsck_component *com)
750 {
751         struct lfsck_layout             *lo     = com->lc_file_ram;
752         const struct dt_body_operations *dbo    = com->lc_obj->do_body_ops;
753         ssize_t                          size   = com->lc_file_size;
754         loff_t                           pos    = 0;
755         int                              rc;
756
757         rc = dbo->dbo_read(env, com->lc_obj,
758                            lfsck_buf_get(env, com->lc_file_disk, size), &pos,
759                            BYPASS_CAPA);
760         if (rc == 0) {
761                 return -ENOENT;
762         } else if (rc < 0) {
763                 CWARN("%s: failed to load lfsck_layout: rc = %d\n",
764                       lfsck_lfsck2name(com->lc_lfsck), rc);
765                 return rc;
766         } else if (rc != size) {
767                 CWARN("%s: crashed lfsck_layout, to be reset: rc = %d\n",
768                       lfsck_lfsck2name(com->lc_lfsck), rc);
769                 return 1;
770         }
771
772         lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
773         if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
774                 CWARN("%s: invalid lfsck_layout magic %#x != %#x, "
775                       "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
776                       lo->ll_magic, LFSCK_LAYOUT_MAGIC);
777                 return 1;
778         }
779
780         return 0;
781 }
782
783 static int lfsck_layout_store(const struct lu_env *env,
784                               struct lfsck_component *com)
785 {
786         struct dt_object         *obj           = com->lc_obj;
787         struct lfsck_instance    *lfsck         = com->lc_lfsck;
788         struct lfsck_layout      *lo            = com->lc_file_disk;
789         struct thandle           *handle;
790         ssize_t                   size          = com->lc_file_size;
791         loff_t                    pos           = 0;
792         int                       rc;
793         ENTRY;
794
795         lfsck_layout_cpu_to_le(lo, com->lc_file_ram);
796         handle = dt_trans_create(env, lfsck->li_bottom);
797         if (IS_ERR(handle)) {
798                 rc = PTR_ERR(handle);
799                 CERROR("%s: fail to create trans for storing lfsck_layout: "
800                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
801                 RETURN(rc);
802         }
803
804         rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
805                                      pos, handle);
806         if (rc != 0) {
807                 CERROR("%s: fail to declare trans for storing lfsck_layout(1): "
808                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
809                 GOTO(out, rc);
810         }
811
812         rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
813         if (rc != 0) {
814                 CERROR("%s: fail to start trans for storing lfsck_layout: "
815                        "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
816                 GOTO(out, rc);
817         }
818
819         rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos,
820                              handle);
821         if (rc != 0)
822                 CERROR("%s: fail to store lfsck_layout(1): size = %d, "
823                        "rc = %d\n", lfsck_lfsck2name(lfsck), (int)size, rc);
824
825         GOTO(out, rc);
826
827 out:
828         dt_trans_stop(env, lfsck->li_bottom, handle);
829
830         return rc;
831 }
832
833 static int lfsck_layout_init(const struct lu_env *env,
834                              struct lfsck_component *com)
835 {
836         struct lfsck_layout *lo = com->lc_file_ram;
837         int rc;
838
839         memset(lo, 0, com->lc_file_size);
840         lo->ll_magic = LFSCK_LAYOUT_MAGIC;
841         lo->ll_status = LS_INIT;
842         down_write(&com->lc_sem);
843         rc = lfsck_layout_store(env, com);
844         up_write(&com->lc_sem);
845
846         return rc;
847 }
848
849 static int fid_is_for_ostobj(const struct lu_env *env, struct dt_device *dt,
850                              struct dt_object *obj, const struct lu_fid *fid)
851 {
852         struct seq_server_site  *ss     = lu_site2seq(dt->dd_lu_dev.ld_site);
853         struct lu_seq_range      range  = { 0 };
854         struct lustre_mdt_attrs *lma;
855         int                      rc;
856
857         fld_range_set_any(&range);
858         rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), &range);
859         if (rc == 0) {
860                 if (fld_range_is_ost(&range))
861                         return 1;
862
863                 return 0;
864         }
865
866         lma = &lfsck_env_info(env)->lti_lma;
867         rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
868                           XATTR_NAME_LMA, BYPASS_CAPA);
869         if (rc == sizeof(*lma)) {
870                 lustre_lma_swab(lma);
871
872                 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
873         }
874
875         rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID, BYPASS_CAPA);
876
877         return rc > 0;
878 }
879
880 static struct lfsck_layout_seq *
881 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
882 {
883         struct lfsck_layout_seq *lls;
884
885         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
886                 if (lls->lls_seq == seq)
887                         return lls;
888
889                 if (lls->lls_seq > seq)
890                         return NULL;
891         }
892
893         return NULL;
894 }
895
896 static void
897 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
898                         struct lfsck_layout_seq *lls)
899 {
900         struct lfsck_layout_seq *tmp;
901         struct list_head        *pos = &llsd->llsd_seq_list;
902
903         list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
904                 if (lls->lls_seq < tmp->lls_seq) {
905                         pos = &tmp->lls_list;
906                         break;
907                 }
908         }
909         list_add_tail(&lls->lls_list, pos);
910 }
911
912 static int
913 lfsck_layout_lastid_create(const struct lu_env *env,
914                            struct lfsck_instance *lfsck,
915                            struct dt_object *obj)
916 {
917         struct lfsck_thread_info *info   = lfsck_env_info(env);
918         struct lu_attr           *la     = &info->lti_la;
919         struct dt_object_format  *dof    = &info->lti_dof;
920         struct lfsck_bookmark    *bk     = &lfsck->li_bookmark_ram;
921         struct dt_device         *dt     = lfsck->li_bottom;
922         struct thandle           *th;
923         __u64                     lastid = 0;
924         loff_t                    pos    = 0;
925         int                       rc;
926         ENTRY;
927
928         CDEBUG(D_LFSCK, "To create LAST_ID for <seq> "LPX64"\n",
929                fid_seq(lfsck_dto2fid(obj)));
930
931         if (bk->lb_param & LPF_DRYRUN)
932                 return 0;
933
934         memset(la, 0, sizeof(*la));
935         la->la_mode = S_IFREG |  S_IRUGO | S_IWUSR;
936         la->la_valid = LA_MODE | LA_UID | LA_GID;
937         dof->dof_type = dt_mode_to_dft(S_IFREG);
938
939         th = dt_trans_create(env, dt);
940         if (IS_ERR(th))
941                 RETURN(rc = PTR_ERR(th));
942
943         rc = dt_declare_create(env, obj, la, NULL, dof, th);
944         if (rc != 0)
945                 GOTO(stop, rc);
946
947         rc = dt_declare_record_write(env, obj,
948                                      lfsck_buf_get(env, &lastid,
949                                                    sizeof(lastid)),
950                                      pos, th);
951         if (rc != 0)
952                 GOTO(stop, rc);
953
954         rc = dt_trans_start_local(env, dt, th);
955         if (rc != 0)
956                 GOTO(stop, rc);
957
958         dt_write_lock(env, obj, 0);
959         if (likely(!dt_object_exists(obj))) {
960                 rc = dt_create(env, obj, la, NULL, dof, th);
961                 if (rc == 0)
962                         rc = dt_record_write(env, obj,
963                                 lfsck_buf_get(env, &lastid, sizeof(lastid)),
964                                 &pos, th);
965         }
966         dt_write_unlock(env, obj);
967
968         GOTO(stop, rc);
969
970 stop:
971         dt_trans_stop(env, dt, th);
972
973         return rc;
974 }
975
976 static int
977 lfsck_layout_lastid_reload(const struct lu_env *env,
978                            struct lfsck_component *com,
979                            struct lfsck_layout_seq *lls)
980 {
981         __u64   lastid;
982         loff_t  pos     = 0;
983         int     rc;
984
985         dt_read_lock(env, lls->lls_lastid_obj, 0);
986         rc = dt_record_read(env, lls->lls_lastid_obj,
987                             lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
988         dt_read_unlock(env, lls->lls_lastid_obj);
989         if (unlikely(rc != 0))
990                 return rc;
991
992         lastid = le64_to_cpu(lastid);
993         if (lastid < lls->lls_lastid_known) {
994                 struct lfsck_instance   *lfsck  = com->lc_lfsck;
995                 struct lfsck_layout     *lo     = com->lc_file_ram;
996
997                 lls->lls_lastid = lls->lls_lastid_known;
998                 lls->lls_dirty = 1;
999                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1000                         LASSERT(lfsck->li_out_notify != NULL);
1001
1002                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1003                                              LE_LASTID_REBUILDING);
1004                         lo->ll_flags |= LF_CRASHED_LASTID;
1005                 }
1006         } else if (lastid >= lls->lls_lastid) {
1007                 lls->lls_lastid = lastid;
1008                 lls->lls_dirty = 0;
1009         }
1010
1011         return 0;
1012 }
1013
1014 static int
1015 lfsck_layout_lastid_store(const struct lu_env *env,
1016                           struct lfsck_component *com)
1017 {
1018         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1019         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
1020         struct dt_device                *dt     = lfsck->li_bottom;
1021         struct lfsck_layout_slave_data  *llsd   = com->lc_data;
1022         struct lfsck_layout_seq         *lls;
1023         struct thandle                  *th;
1024         __u64                            lastid;
1025         int                              rc     = 0;
1026         int                              rc1    = 0;
1027
1028         list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1029                 loff_t pos = 0;
1030
1031                 /* XXX: Add the code back if we really found related
1032                  *      inconsistent cases in the future. */
1033 #if 0
1034                 if (!lls->lls_dirty) {
1035                         /* In OFD, before the pre-creation, the LAST_ID
1036                          * file will be updated firstly, which may hide
1037                          * some potential crashed cases. For example:
1038                          *
1039                          * The old obj1's ID is higher than old LAST_ID
1040                          * but lower than the new LAST_ID, but the LFSCK
1041                          * have not touch the obj1 until the OFD updated
1042                          * the LAST_ID. So the LFSCK does not regard it
1043                          * as crashed case. But when OFD does not create
1044                          * successfully, it will set the LAST_ID as the
1045                          * real created objects' ID, then LFSCK needs to
1046                          * found related inconsistency. */
1047                         rc = lfsck_layout_lastid_reload(env, com, lls);
1048                         if (likely(!lls->lls_dirty))
1049                                 continue;
1050                 }
1051 #endif
1052
1053                 CDEBUG(D_LFSCK, "To sync the LAST_ID for <seq> "LPX64
1054                        " as <oid> "LPU64"\n", lls->lls_seq, lls->lls_lastid);
1055
1056                 if (bk->lb_param & LPF_DRYRUN) {
1057                         lls->lls_dirty = 0;
1058                         continue;
1059                 }
1060
1061                 th = dt_trans_create(env, dt);
1062                 if (IS_ERR(th)) {
1063                         rc1 = PTR_ERR(th);
1064                         CERROR("%s: (1) failed to store "LPX64": rc = %d\n",
1065                                lfsck_lfsck2name(com->lc_lfsck),
1066                                lls->lls_seq, rc1);
1067                         continue;
1068                 }
1069
1070                 lastid = cpu_to_le64(lls->lls_lastid);
1071                 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1072                                              lfsck_buf_get(env, &lastid,
1073                                                            sizeof(lastid)),
1074                                              pos, th);
1075                 if (rc != 0)
1076                         goto stop;
1077
1078                 rc = dt_trans_start_local(env, dt, th);
1079                 if (rc != 0)
1080                         goto stop;
1081
1082                 dt_write_lock(env, lls->lls_lastid_obj, 0);
1083                 rc = dt_record_write(env, lls->lls_lastid_obj,
1084                                      lfsck_buf_get(env, &lastid,
1085                                      sizeof(lastid)), &pos, th);
1086                 dt_write_unlock(env, lls->lls_lastid_obj);
1087                 if (rc == 0)
1088                         lls->lls_dirty = 0;
1089
1090 stop:
1091                 dt_trans_stop(env, dt, th);
1092                 if (rc != 0) {
1093                         rc1 = rc;
1094                         CERROR("%s: (2) failed to store "LPX64": rc = %d\n",
1095                                lfsck_lfsck2name(com->lc_lfsck),
1096                                lls->lls_seq, rc1);
1097                 }
1098         }
1099
1100         return rc1;
1101 }
1102
1103 static int
1104 lfsck_layout_lastid_load(const struct lu_env *env,
1105                          struct lfsck_component *com,
1106                          struct lfsck_layout_seq *lls)
1107 {
1108         struct lfsck_instance   *lfsck  = com->lc_lfsck;
1109         struct lfsck_layout     *lo     = com->lc_file_ram;
1110         struct lu_fid           *fid    = &lfsck_env_info(env)->lti_fid;
1111         struct dt_object        *obj;
1112         loff_t                   pos    = 0;
1113         int                      rc;
1114         ENTRY;
1115
1116         lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck->li_bottom));
1117         obj = dt_locate(env, lfsck->li_bottom, fid);
1118         if (IS_ERR(obj))
1119                 RETURN(PTR_ERR(obj));
1120
1121         /* LAST_ID crashed, to be rebuilt */
1122         if (!dt_object_exists(obj)) {
1123                 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1124                         LASSERT(lfsck->li_out_notify != NULL);
1125
1126                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1127                                              LE_LASTID_REBUILDING);
1128                         lo->ll_flags |= LF_CRASHED_LASTID;
1129
1130                         if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1131                             cfs_fail_val > 0) {
1132                                 struct l_wait_info lwi = LWI_TIMEOUT(
1133                                                 cfs_time_seconds(cfs_fail_val),
1134                                                 NULL, NULL);
1135
1136                                 up_write(&com->lc_sem);
1137                                 l_wait_event(lfsck->li_thread.t_ctl_waitq,
1138                                              !thread_is_running(&lfsck->li_thread),
1139                                              &lwi);
1140                                 down_write(&com->lc_sem);
1141                         }
1142                 }
1143
1144                 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1145         } else {
1146                 dt_read_lock(env, obj, 0);
1147                 rc = dt_read(env, obj,
1148                         lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1149                         &pos);
1150                 dt_read_unlock(env, obj);
1151                 if (rc != 0 && rc != sizeof(__u64))
1152                         GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1153
1154                 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1155                         LASSERT(lfsck->li_out_notify != NULL);
1156
1157                         lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1158                                              LE_LASTID_REBUILDING);
1159                         lo->ll_flags |= LF_CRASHED_LASTID;
1160                 }
1161
1162                 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1163                 rc = 0;
1164         }
1165
1166         GOTO(out, rc);
1167
1168 out:
1169         if (rc != 0)
1170                 lfsck_object_put(env, obj);
1171         else
1172                 lls->lls_lastid_obj = obj;
1173
1174         return rc;
1175 }
1176
1177 static void lfsck_layout_record_failure(const struct lu_env *env,
1178                                                  struct lfsck_instance *lfsck,
1179                                                  struct lfsck_layout *lo)
1180 {
1181         lo->ll_objs_failed_phase1++;
1182         if (unlikely(lo->ll_pos_first_inconsistent == 0))
1183                 lo->ll_pos_first_inconsistent =
1184                         lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1185                                                         lfsck->li_di_oit);
1186 }
1187
1188 static int lfsck_layout_master_async_interpret(const struct lu_env *env,
1189                                                struct ptlrpc_request *req,
1190                                                void *args, int rc)
1191 {
1192         struct lfsck_async_interpret_args *laia = args;
1193         struct lfsck_component            *com  = laia->laia_com;
1194         struct lfsck_layout_master_data   *llmd = com->lc_data;
1195         struct lfsck_tgt_descs            *ltds = laia->laia_ltds;
1196         struct lfsck_tgt_desc             *ltd  = laia->laia_ltd;
1197         struct lfsck_request              *lr   = laia->laia_lr;
1198
1199         switch (lr->lr_event) {
1200         case LE_START:
1201                 if (rc != 0) {
1202                         struct lfsck_layout *lo = com->lc_file_ram;
1203
1204                         CERROR("%s: fail to notify %s %x for layout start: "
1205                                "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1206                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1207                                ltd->ltd_index, rc);
1208                         lo->ll_flags |= LF_INCOMPLETE;
1209                         break;
1210                 }
1211
1212                 spin_lock(&ltds->ltd_lock);
1213                 if (ltd->ltd_dead || ltd->ltd_layout_done) {
1214                         spin_unlock(&ltds->ltd_lock);
1215                         break;
1216                 }
1217
1218                 if (lr->lr_flags & LEF_TO_OST) {
1219                         if (list_empty(&ltd->ltd_layout_list))
1220                                 list_add_tail(&ltd->ltd_layout_list,
1221                                               &llmd->llmd_ost_list);
1222                         if (list_empty(&ltd->ltd_layout_phase_list))
1223                                 list_add_tail(&ltd->ltd_layout_phase_list,
1224                                               &llmd->llmd_ost_phase1_list);
1225                 } else {
1226                         if (list_empty(&ltd->ltd_layout_list))
1227                                 list_add_tail(&ltd->ltd_layout_list,
1228                                               &llmd->llmd_mdt_list);
1229                         if (list_empty(&ltd->ltd_layout_phase_list))
1230                                 list_add_tail(&ltd->ltd_layout_phase_list,
1231                                               &llmd->llmd_mdt_phase1_list);
1232                 }
1233                 spin_unlock(&ltds->ltd_lock);
1234                 break;
1235         case LE_STOP:
1236         case LE_PHASE1_DONE:
1237         case LE_PHASE2_DONE:
1238         case LE_PEER_EXIT:
1239                 if (rc != 0 && rc != -EALREADY)
1240                         CWARN("%s: fail to notify %s %x for layout: "
1241                               "event = %d, rc = %d\n",
1242                               lfsck_lfsck2name(com->lc_lfsck),
1243                               (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1244                               ltd->ltd_index, lr->lr_event, rc);
1245                 break;
1246         case LE_QUERY: {
1247                 struct lfsck_reply *reply;
1248
1249                 if (rc != 0) {
1250                         spin_lock(&ltds->ltd_lock);
1251                         list_del_init(&ltd->ltd_layout_phase_list);
1252                         list_del_init(&ltd->ltd_layout_list);
1253                         spin_unlock(&ltds->ltd_lock);
1254                         break;
1255                 }
1256
1257                 reply = req_capsule_server_get(&req->rq_pill,
1258                                                &RMF_LFSCK_REPLY);
1259                 if (reply == NULL) {
1260                         rc = -EPROTO;
1261                         CERROR("%s: invalid return value: rc = %d\n",
1262                                lfsck_lfsck2name(com->lc_lfsck), rc);
1263                         spin_lock(&ltds->ltd_lock);
1264                         list_del_init(&ltd->ltd_layout_phase_list);
1265                         list_del_init(&ltd->ltd_layout_list);
1266                         spin_unlock(&ltds->ltd_lock);
1267                         break;
1268                 }
1269
1270                 switch (reply->lr_status) {
1271                 case LS_SCANNING_PHASE1:
1272                         break;
1273                 case LS_SCANNING_PHASE2:
1274                         spin_lock(&ltds->ltd_lock);
1275                         list_del_init(&ltd->ltd_layout_phase_list);
1276                         if (ltd->ltd_dead || ltd->ltd_layout_done) {
1277                                 spin_unlock(&ltds->ltd_lock);
1278                                 break;
1279                         }
1280
1281                         if (lr->lr_flags & LEF_TO_OST)
1282                                 list_add_tail(&ltd->ltd_layout_phase_list,
1283                                               &llmd->llmd_ost_phase2_list);
1284                         else
1285                                 list_add_tail(&ltd->ltd_layout_phase_list,
1286                                               &llmd->llmd_mdt_phase2_list);
1287                         spin_unlock(&ltds->ltd_lock);
1288                         break;
1289                 default:
1290                         spin_lock(&ltds->ltd_lock);
1291                         list_del_init(&ltd->ltd_layout_phase_list);
1292                         list_del_init(&ltd->ltd_layout_list);
1293                         spin_unlock(&ltds->ltd_lock);
1294                         break;
1295                 }
1296                 break;
1297         }
1298         default:
1299                 CERROR("%s: unexpected event: rc = %d\n",
1300                        lfsck_lfsck2name(com->lc_lfsck), lr->lr_event);
1301                 break;
1302         }
1303
1304         if (!laia->laia_shared) {
1305                 lfsck_tgt_put(ltd);
1306                 lfsck_component_put(env, com);
1307         }
1308
1309         return 0;
1310 }
1311
1312 static int lfsck_layout_master_query_others(const struct lu_env *env,
1313                                             struct lfsck_component *com)
1314 {
1315         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1316         struct lfsck_request              *lr    = &info->lti_lr;
1317         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1318         struct lfsck_instance             *lfsck = com->lc_lfsck;
1319         struct lfsck_layout_master_data   *llmd  = com->lc_data;
1320         struct ptlrpc_request_set         *set;
1321         struct lfsck_tgt_descs            *ltds;
1322         struct lfsck_tgt_desc             *ltd;
1323         struct list_head                  *head;
1324         int                                rc    = 0;
1325         int                                rc1   = 0;
1326         ENTRY;
1327
1328         set = ptlrpc_prep_set();
1329         if (set == NULL)
1330                 RETURN(-ENOMEM);
1331
1332         llmd->llmd_touch_gen++;
1333         memset(lr, 0, sizeof(*lr));
1334         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1335         lr->lr_event = LE_QUERY;
1336         lr->lr_active = LT_LAYOUT;
1337         laia->laia_com = com;
1338         laia->laia_lr = lr;
1339         laia->laia_shared = 0;
1340
1341         if (!list_empty(&llmd->llmd_mdt_phase1_list)) {
1342                 ltds = &lfsck->li_mdt_descs;
1343                 lr->lr_flags = 0;
1344                 head = &llmd->llmd_mdt_phase1_list;
1345         } else {
1346
1347 again:
1348                 ltds = &lfsck->li_ost_descs;
1349                 lr->lr_flags = LEF_TO_OST;
1350                 head = &llmd->llmd_ost_phase1_list;
1351         }
1352
1353         laia->laia_ltds = ltds;
1354         spin_lock(&ltds->ltd_lock);
1355         while (!list_empty(head)) {
1356                 ltd = list_entry(head->next,
1357                                  struct lfsck_tgt_desc,
1358                                  ltd_layout_phase_list);
1359                 if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1360                         break;
1361
1362                 ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1363                 list_del(&ltd->ltd_layout_phase_list);
1364                 list_add_tail(&ltd->ltd_layout_phase_list, head);
1365                 atomic_inc(&ltd->ltd_ref);
1366                 laia->laia_ltd = ltd;
1367                 spin_unlock(&ltds->ltd_lock);
1368                 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1369                                          lfsck_layout_master_async_interpret,
1370                                          laia, LFSCK_QUERY);
1371                 if (rc != 0) {
1372                         CERROR("%s: fail to query %s %x for layout: rc = %d\n",
1373                                lfsck_lfsck2name(lfsck),
1374                                (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1375                                ltd->ltd_index, rc);
1376                         lfsck_tgt_put(ltd);
1377                         rc1 = rc;
1378                 }
1379                 spin_lock(&ltds->ltd_lock);
1380         }
1381         spin_unlock(&ltds->ltd_lock);
1382
1383         rc = ptlrpc_set_wait(set);
1384         if (rc < 0) {
1385                 ptlrpc_set_destroy(set);
1386                 RETURN(rc);
1387         }
1388
1389         if (!(lr->lr_flags & LEF_TO_OST) &&
1390             list_empty(&llmd->llmd_mdt_phase1_list))
1391                 goto again;
1392
1393         ptlrpc_set_destroy(set);
1394
1395         RETURN(rc1 != 0 ? rc1 : rc);
1396 }
1397
1398 static inline bool
1399 lfsck_layout_master_to_orphan(struct lfsck_layout_master_data *llmd)
1400 {
1401         return list_empty(&llmd->llmd_mdt_phase1_list) &&
1402                (!list_empty(&llmd->llmd_ost_phase2_list) ||
1403                 list_empty(&llmd->llmd_ost_phase1_list));
1404 }
1405
1406 static int lfsck_layout_master_notify_others(const struct lu_env *env,
1407                                              struct lfsck_component *com,
1408                                              struct lfsck_request *lr)
1409 {
1410         struct lfsck_thread_info          *info  = lfsck_env_info(env);
1411         struct lfsck_async_interpret_args *laia  = &info->lti_laia;
1412         struct lfsck_instance             *lfsck = com->lc_lfsck;
1413         struct lfsck_layout_master_data   *llmd  = com->lc_data;
1414         struct lfsck_layout               *lo    = com->lc_file_ram;
1415         struct lfsck_bookmark             *bk    = &lfsck->li_bookmark_ram;
1416         struct ptlrpc_request_set         *set;
1417         struct lfsck_tgt_descs            *ltds;
1418         struct lfsck_tgt_desc             *ltd;
1419         struct lfsck_tgt_desc             *next;
1420         struct list_head                  *head;
1421         __u32                              idx;
1422         int                                rc    = 0;
1423         ENTRY;
1424
1425         set = ptlrpc_prep_set();
1426         if (set == NULL)
1427                 RETURN(-ENOMEM);
1428
1429         lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1430         lr->lr_active = LT_LAYOUT;
1431         laia->laia_com = com;
1432         laia->laia_lr = lr;
1433         laia->laia_shared = 0;
1434         switch (lr->lr_event) {
1435         case LE_START:
1436                 /* Notify OSTs firstly, then handle other MDTs if needed. */
1437                 ltds = &lfsck->li_ost_descs;
1438                 laia->laia_ltds = ltds;
1439                 down_read(&ltds->ltd_rw_sem);
1440                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1441                         ltd = lfsck_tgt_get(ltds, idx);
1442                         LASSERT(ltd != NULL);
1443
1444                         laia->laia_ltd = ltd;
1445                         ltd->ltd_layout_done = 0;
1446                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1447                                         lfsck_layout_master_async_interpret,
1448                                         laia, LFSCK_NOTIFY);
1449                         if (rc != 0) {
1450                                 CERROR("%s: fail to notify %s %x for layout "
1451                                        "start: rc = %d\n",
1452                                        lfsck_lfsck2name(lfsck),
1453                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1454                                        "MDT", idx, rc);
1455                                 lfsck_tgt_put(ltd);
1456                                 lo->ll_flags |= LF_INCOMPLETE;
1457                         }
1458                 }
1459                 up_read(&ltds->ltd_rw_sem);
1460
1461                 /* Sync up */
1462                 rc = ptlrpc_set_wait(set);
1463                 if (rc < 0) {
1464                         ptlrpc_set_destroy(set);
1465                         RETURN(rc);
1466                 }
1467
1468                 if (!(bk->lb_param & LPF_ALL_TGT))
1469                         break;
1470
1471                 /* link other MDT targets locallly. */
1472                 spin_lock(&ltds->ltd_lock);
1473                 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1474                         ltd = LTD_TGT(ltds, idx);
1475                         LASSERT(ltd != NULL);
1476
1477                         if (!list_empty(&ltd->ltd_layout_list))
1478                                 continue;
1479
1480                         list_add_tail(&ltd->ltd_layout_list,
1481                                       &llmd->llmd_mdt_list);
1482                         list_add_tail(&ltd->ltd_layout_phase_list,
1483                                       &llmd->llmd_mdt_phase1_list);
1484                 }
1485                 spin_unlock(&ltds->ltd_lock);
1486                 break;
1487         case LE_STOP:
1488         case LE_PHASE2_DONE:
1489         case LE_PEER_EXIT: {
1490                 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1491                 if (bk->lb_param & LPF_ALL_TGT) {
1492                         head = &llmd->llmd_mdt_list;
1493                         ltds = &lfsck->li_mdt_descs;
1494                         if (lr->lr_event == LE_STOP) {
1495                                 /* unlink other MDT targets locallly. */
1496                                 spin_lock(&ltds->ltd_lock);
1497                                 list_for_each_entry_safe(ltd, next, head,
1498                                                          ltd_layout_list) {
1499                                         list_del_init(&ltd->ltd_layout_phase_list);
1500                                         list_del_init(&ltd->ltd_layout_list);
1501                                 }
1502                                 spin_unlock(&ltds->ltd_lock);
1503
1504                                 lr->lr_flags |= LEF_TO_OST;
1505                                 head = &llmd->llmd_ost_list;
1506                                 ltds = &lfsck->li_ost_descs;
1507                         } else {
1508                                 lr->lr_flags &= ~LEF_TO_OST;
1509                         }
1510                 } else {
1511                         lr->lr_flags |= LEF_TO_OST;
1512                         head = &llmd->llmd_ost_list;
1513                         ltds = &lfsck->li_ost_descs;
1514                 }
1515
1516 again:
1517                 laia->laia_ltds = ltds;
1518                 spin_lock(&ltds->ltd_lock);
1519                 while (!list_empty(head)) {
1520                         ltd = list_entry(head->next, struct lfsck_tgt_desc,
1521                                          ltd_layout_list);
1522                         if (!list_empty(&ltd->ltd_layout_phase_list))
1523                                 list_del_init(&ltd->ltd_layout_phase_list);
1524                         list_del_init(&ltd->ltd_layout_list);
1525                         atomic_inc(&ltd->ltd_ref);
1526                         laia->laia_ltd = ltd;
1527                         spin_unlock(&ltds->ltd_lock);
1528                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1529                                         lfsck_layout_master_async_interpret,
1530                                         laia, LFSCK_NOTIFY);
1531                         if (rc != 0) {
1532                                 CERROR("%s: fail to notify %s %x for layout "
1533                                        "stop/phase2: rc = %d\n",
1534                                        lfsck_lfsck2name(lfsck),
1535                                        (lr->lr_flags & LEF_TO_OST) ? "OST" :
1536                                        "MDT", ltd->ltd_index, rc);
1537                                 lfsck_tgt_put(ltd);
1538                         }
1539                         spin_lock(&ltds->ltd_lock);
1540                 }
1541                 spin_unlock(&ltds->ltd_lock);
1542
1543                 rc = ptlrpc_set_wait(set);
1544                 if (rc < 0) {
1545                         ptlrpc_set_destroy(set);
1546                         RETURN(rc);
1547                 }
1548
1549                 if (!(lr->lr_flags & LEF_TO_OST)) {
1550                         lr->lr_flags |= LEF_TO_OST;
1551                         head = &llmd->llmd_ost_list;
1552                         ltds = &lfsck->li_ost_descs;
1553                         goto again;
1554                 }
1555                 break;
1556         }
1557         case LE_PHASE1_DONE:
1558                 llmd->llmd_touch_gen++;
1559                 ltds = &lfsck->li_mdt_descs;
1560                 laia->laia_ltds = ltds;
1561                 spin_lock(&ltds->ltd_lock);
1562                 while (!list_empty(&llmd->llmd_mdt_phase1_list)) {
1563                         ltd = list_entry(llmd->llmd_mdt_phase1_list.next,
1564                                          struct lfsck_tgt_desc,
1565                                          ltd_layout_phase_list);
1566                         if (ltd->ltd_layout_gen == llmd->llmd_touch_gen)
1567                                 break;
1568
1569                         ltd->ltd_layout_gen = llmd->llmd_touch_gen;
1570                         list_del_init(&ltd->ltd_layout_phase_list);
1571                         list_add_tail(&ltd->ltd_layout_phase_list,
1572                                       &llmd->llmd_mdt_phase1_list);
1573                         atomic_inc(&ltd->ltd_ref);
1574                         laia->laia_ltd = ltd;
1575                         spin_unlock(&ltds->ltd_lock);
1576                         rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1577                                         lfsck_layout_master_async_interpret,
1578                                         laia, LFSCK_NOTIFY);
1579                         if (rc != 0) {
1580                                 CERROR("%s: fail to notify MDT %x for layout "
1581                                        "phase1 done: rc = %d\n",
1582                                        lfsck_lfsck2name(lfsck),
1583                                        ltd->ltd_index, rc);
1584                                 lfsck_tgt_put(ltd);
1585                         }
1586                         spin_lock(&ltds->ltd_lock);
1587                 }
1588                 spin_unlock(&ltds->ltd_lock);
1589                 break;
1590         default:
1591                 CERROR("%s: unexpected LFSCK event: rc = %d\n",
1592                        lfsck_lfsck2name(lfsck), lr->lr_event);
1593                 rc = -EINVAL;
1594                 break;
1595         }
1596
1597         rc = ptlrpc_set_wait(set);
1598         ptlrpc_set_destroy(set);
1599
1600         RETURN(rc);
1601 }
1602
1603 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1604                                            struct lfsck_component *com,
1605                                            int rc)
1606 {
1607         struct lfsck_instance   *lfsck = com->lc_lfsck;
1608         struct lfsck_layout     *lo    = com->lc_file_ram;
1609         struct lfsck_bookmark   *bk    = &lfsck->li_bookmark_ram;
1610
1611         down_write(&com->lc_sem);
1612         lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1613                                 HALF_SEC - lfsck->li_time_last_checkpoint);
1614         lo->ll_time_last_checkpoint = cfs_time_current_sec();
1615         lo->ll_objs_checked_phase2 += com->lc_new_checked;
1616
1617         if (rc > 0) {
1618                 com->lc_journal = 0;
1619                 if (lo->ll_flags & LF_INCOMPLETE)
1620                         lo->ll_status = LS_PARTIAL;
1621                 else
1622                         lo->ll_status = LS_COMPLETED;
1623                 if (!(bk->lb_param & LPF_DRYRUN))
1624                         lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1625                 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1626                 lo->ll_success_count++;
1627         } else if (rc == 0) {
1628                 lo->ll_status = lfsck->li_status;
1629                 if (lo->ll_status == 0)
1630                         lo->ll_status = LS_STOPPED;
1631         } else {
1632                 lo->ll_status = LS_FAILED;
1633         }
1634
1635         rc = lfsck_layout_store(env, com);
1636         up_write(&com->lc_sem);
1637
1638         return rc;
1639 }
1640
1641 static int lfsck_layout_lock(const struct lu_env *env,
1642                              struct lfsck_component *com,
1643                              struct dt_object *obj,
1644                              struct lustre_handle *lh, __u64 bits)
1645 {
1646         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1647         ldlm_policy_data_t              *policy = &info->lti_policy;
1648         struct ldlm_res_id              *resid  = &info->lti_resid;
1649         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1650         __u64                            flags  = LDLM_FL_ATOMIC_CB;
1651         int                              rc;
1652
1653         LASSERT(lfsck->li_namespace != NULL);
1654
1655         memset(policy, 0, sizeof(*policy));
1656         policy->l_inodebits.bits = bits;
1657         fid_build_reg_res_name(lfsck_dto2fid(obj), resid);
1658         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_IBITS,
1659                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
1660                                     ldlm_completion_ast, NULL, NULL, 0,
1661                                     LVB_T_NONE, NULL, lh);
1662         if (rc == ELDLM_OK) {
1663                 rc = 0;
1664         } else {
1665                 memset(lh, 0, sizeof(*lh));
1666                 rc = -EIO;
1667         }
1668
1669         return rc;
1670 }
1671
1672 static void lfsck_layout_unlock(struct lustre_handle *lh)
1673 {
1674         if (lustre_handle_is_used(lh)) {
1675                 ldlm_lock_decref(lh, LCK_EX);
1676                 memset(lh, 0, sizeof(*lh));
1677         }
1678 }
1679
1680 static int lfsck_layout_trans_stop(const struct lu_env *env,
1681                                    struct dt_device *dev,
1682                                    struct thandle *handle, int result)
1683 {
1684         int rc;
1685
1686         handle->th_result = result;
1687         rc = dt_trans_stop(env, dev, handle);
1688         if (rc > 0)
1689                 rc = 0;
1690         else if (rc == 0)
1691                 rc = 1;
1692
1693         return rc;
1694 }
1695
1696 /**
1697  * \retval       +1: repaired
1698  * \retval        0: did nothing
1699  * \retval      -ve: on error
1700  */
1701 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1702                                      struct thandle *handle,
1703                                      struct dt_object *parent,
1704                                      struct lu_fid *cfid,
1705                                      struct lu_buf *buf,
1706                                      struct lov_ost_data_v1 *slot,
1707                                      int fl, __u32 ost_idx)
1708 {
1709         struct ost_id   *oi     = &lfsck_env_info(env)->lti_oi;
1710         int              rc;
1711
1712         fid_to_ostid(cfid, oi);
1713         ostid_cpu_to_le(oi, &slot->l_ost_oi);
1714         slot->l_ost_gen = cpu_to_le32(0);
1715         slot->l_ost_idx = cpu_to_le32(ost_idx);
1716         rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV, fl, handle,
1717                           BYPASS_CAPA);
1718         if (rc == 0)
1719                 rc = 1;
1720
1721         return rc;
1722 }
1723
1724 /**
1725  * \retval       +1: repaired
1726  * \retval        0: did nothing
1727  * \retval      -ve: on error
1728  */
1729 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1730                                      struct thandle *handle,
1731                                      struct dt_object *parent,
1732                                      struct lu_fid *cfid,
1733                                      struct lu_buf *buf, int fl,
1734                                      __u32 ost_idx, __u32 ea_off, bool reset)
1735 {
1736         struct lov_mds_md_v1    *lmm    = buf->lb_buf;
1737         struct lov_ost_data_v1  *objs;
1738         int                      rc;
1739         ENTRY;
1740
1741         if (fl == LU_XATTR_CREATE || reset) {
1742                 LASSERT(buf->lb_len == lov_mds_md_size(ea_off + 1,
1743                                                        LOV_MAGIC_V1));
1744
1745                 memset(lmm, 0, buf->lb_len);
1746                 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1747                 /* XXX: currently, we only support LOV_PATTERN_RAID0. */
1748                 lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
1749                 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1750                 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1751                 /* XXX: We cannot know the stripe size,
1752                  *      then use the default value (1 MB). */
1753                 lmm->lmm_stripe_size =
1754                         cpu_to_le32(LOV_DESC_STRIPE_SIZE_DEFAULT);
1755                 objs = &(lmm->lmm_objects[ea_off]);
1756         } else {
1757                 __u16   count = le16_to_cpu(lmm->lmm_stripe_count);
1758                 int     gap   = ea_off - count;
1759                 __u32   magic = le32_to_cpu(lmm->lmm_magic);
1760
1761                 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3
1762                  * which has been verified in lfsck_layout_verify_header()
1763                  * already. If some new magic introduced in the future,
1764                  * then layout LFSCK needs to be updated also. */
1765                 if (magic == LOV_MAGIC_V1) {
1766                         objs = &(lmm->lmm_objects[count]);
1767                 } else {
1768                         LASSERT(magic == LOV_MAGIC_V3);
1769                         objs = &((struct lov_mds_md_v3 *)lmm)->
1770                                                         lmm_objects[count];
1771                 }
1772
1773                 if (gap > 0)
1774                         memset(objs, 0, gap * sizeof(*objs));
1775                 lmm->lmm_layout_gen =
1776                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1777                 objs += gap;
1778
1779                 LASSERT(buf->lb_len == lov_mds_md_size(ea_off + 1, magic));
1780         }
1781
1782         lmm->lmm_stripe_count = cpu_to_le16(ea_off + 1);
1783         rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1784                                        fl, ost_idx);
1785
1786         RETURN(rc);
1787 }
1788
1789 /**
1790  * \retval       +1: repaired
1791  * \retval        0: did nothing
1792  * \retval      -ve: on error
1793  */
1794 static int lfsck_layout_update_pfid(const struct lu_env *env,
1795                                     struct lfsck_component *com,
1796                                     struct dt_object *parent,
1797                                     struct lu_fid *cfid,
1798                                     struct dt_device *cdev, __u32 ea_off)
1799 {
1800         struct filter_fid       *pfid   = &lfsck_env_info(env)->lti_new_pfid;
1801         struct dt_object        *child;
1802         struct thandle          *handle;
1803         const struct lu_fid     *tfid   = lu_object_fid(&parent->do_lu);
1804         struct lu_buf           *buf;
1805         int                      rc     = 0;
1806         ENTRY;
1807
1808         child = lfsck_object_find_by_dev(env, cdev, cfid);
1809         if (IS_ERR(child))
1810                 RETURN(PTR_ERR(child));
1811
1812         handle = dt_trans_create(env, cdev);
1813         if (IS_ERR(handle))
1814                 GOTO(out, rc = PTR_ERR(handle));
1815
1816         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
1817         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
1818         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1819          * MDT-object's FID::f_ver, instead it is the OST-object index in its
1820          * parent MDT-object's layout EA. */
1821         pfid->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
1822         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
1823
1824         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
1825         if (rc != 0)
1826                 GOTO(stop, rc);
1827
1828         rc = dt_trans_start(env, cdev, handle);
1829         if (rc != 0)
1830                 GOTO(stop, rc);
1831
1832         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
1833                           BYPASS_CAPA);
1834
1835         GOTO(stop, rc = (rc == 0 ? 1 : rc));
1836
1837 stop:
1838         dt_trans_stop(env, cdev, handle);
1839
1840 out:
1841         lu_object_put(env, &child->do_lu);
1842
1843         return rc;
1844 }
1845
1846 /**
1847  * \retval       +1: repaired
1848  * \retval        0: did nothing
1849  * \retval      -ve: on error
1850  */
1851 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1852                                         struct lfsck_component *com,
1853                                         struct lfsck_tgt_desc *ltd,
1854                                         struct lu_orphan_rec *rec,
1855                                         struct lu_fid *cfid,
1856                                         const char *prefix,
1857                                         const char *postfix,
1858                                         __u32 ea_off)
1859 {
1860         struct lfsck_thread_info        *info   = lfsck_env_info(env);
1861         char                            *name   = info->lti_key;
1862         struct lu_attr                  *la     = &info->lti_la;
1863         struct dt_object_format         *dof    = &info->lti_dof;
1864         struct lfsck_instance           *lfsck  = com->lc_lfsck;
1865         struct lu_fid                   *pfid   = &rec->lor_fid;
1866         struct lu_fid                   *tfid   = &info->lti_fid3;
1867         struct dt_device                *next   = lfsck->li_next;
1868         struct dt_object                *pobj   = NULL;
1869         struct dt_object                *cobj   = NULL;
1870         struct thandle                  *th     = NULL;
1871         struct lu_buf                   *pbuf   = NULL;
1872         struct lu_buf                   *ea_buf = &info->lti_big_buf;
1873         struct lustre_handle             lh     = { 0 };
1874         int                              buflen = ea_buf->lb_len;
1875         int                              idx    = 0;
1876         int                              rc     = 0;
1877         ENTRY;
1878
1879         /* Create .lustre/lost+found/MDTxxxx when needed. */
1880         if (unlikely(lfsck->li_lpf_obj == NULL)) {
1881                 rc = lfsck_create_lpf(env, lfsck);
1882                 if (rc != 0)
1883                         RETURN(rc);
1884         }
1885
1886         if (fid_is_zero(pfid)) {
1887                 struct filter_fid *ff = &info->lti_new_pfid;
1888
1889                 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1890                 if (rc != 0)
1891                         RETURN(rc);
1892
1893                 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1894                 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1895                 /* Currently, the filter_fid::ff_parent::f_ver is not the
1896                  * real parent MDT-object's FID::f_ver, instead it is the
1897                  * OST-object index in its parent MDT-object's layout EA. */
1898                 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
1899                 pbuf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
1900                 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1901                 if (IS_ERR(cobj))
1902                         RETURN(PTR_ERR(cobj));
1903         }
1904
1905         CDEBUG(D_LFSCK, "Re-create the lost MDT-object: parent "
1906                DFID", child "DFID", OST-index %u, stripe-index %u, "
1907                "prefix %s, postfix %s\n",
1908                PFID(pfid), PFID(cfid), ltd->ltd_index, ea_off, prefix, postfix);
1909
1910         pobj = lfsck_object_find_by_dev(env, lfsck->li_bottom, pfid);
1911         if (IS_ERR(pobj))
1912                 GOTO(put, rc = PTR_ERR(pobj));
1913
1914         LASSERT(prefix != NULL);
1915         LASSERT(postfix != NULL);
1916
1917         /** name rules:
1918          *
1919          *  1. Use the MDT-object's FID as the name with prefix and postfix.
1920          *
1921          *  1.1 prefix "C-":    More than one OST-objects claim the same
1922          *                      MDT-object and the same slot in the layout EA.
1923          *                      It may be created for dangling referenced MDT
1924          *                      object or may be not.
1925          *  1.2 prefix "N-":    The orphan OST-object does not know which one
1926          *                      is the real parent, so the LFSCK assign a new
1927          *                      FID as its parent.
1928          *  1.3 prefix "R-":    The orphan OST-object know its parent FID but
1929          *                      does not know the position in the namespace.
1930          *
1931          *  2. If there is name conflict, append more index for new name. */
1932         sprintf(name, "%s"DFID"%s", prefix, PFID(pfid), postfix);
1933         do {
1934                 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1935                                (const struct dt_key *)name, BYPASS_CAPA);
1936                 if (rc != 0 && rc != -ENOENT)
1937                         GOTO(put, rc);
1938
1939                 if (unlikely(rc == 0)) {
1940                         CWARN("%s: The name %s under lost+found has been used "
1941                               "by the "DFID". Try to increase the FID version "
1942                               "for the new file name.\n",
1943                               lfsck_lfsck2name(lfsck), name, PFID(tfid));
1944                         sprintf(name, "%s"DFID"%s-%d", prefix, PFID(pfid),
1945                                 postfix, ++idx);
1946                 }
1947         } while (rc == 0);
1948
1949         memset(la, 0, sizeof(*la));
1950         la->la_uid = rec->lor_uid;
1951         la->la_gid = rec->lor_gid;
1952         la->la_mode = S_IFREG | S_IRUSR | S_IWUSR;
1953         la->la_valid = LA_MODE | LA_UID | LA_GID;
1954
1955         memset(dof, 0, sizeof(*dof));
1956         dof->dof_type = dt_mode_to_dft(S_IFREG);
1957
1958         rc = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1959         if (buflen < rc) {
1960                 lu_buf_realloc(ea_buf, rc);
1961                 buflen = ea_buf->lb_len;
1962                 if (ea_buf->lb_buf == NULL)
1963                         GOTO(put, rc = -ENOMEM);
1964         } else {
1965                 ea_buf->lb_len = rc;
1966         }
1967
1968         /* Hold update lock on the .lustre/lost+found/MDTxxxx/.
1969          *
1970          * XXX: Currently, we do not grab the PDO lock as normal create cases,
1971          *      because creating MDT-object for orphan OST-object is rare, we
1972          *      do not much care about the performance. It can be improved in
1973          *      the future when needed. */
1974         rc = lfsck_layout_lock(env, com, lfsck->li_lpf_obj, &lh,
1975                                MDS_INODELOCK_UPDATE);
1976         if (rc != 0)
1977                 GOTO(put, rc);
1978
1979         th = dt_trans_create(env, next);
1980         if (IS_ERR(th))
1981                 GOTO(unlock, rc = PTR_ERR(th));
1982
1983         /* 1a. Update OST-object's parent information remotely.
1984          *
1985          * If other subsequent modifications failed, then next LFSCK scanning
1986          * will process the OST-object as orphan again with known parent FID. */
1987         if (cobj != NULL) {
1988                 rc = dt_declare_xattr_set(env, cobj, pbuf, XATTR_NAME_FID, 0, th);
1989                 if (rc != 0)
1990                         GOTO(stop, rc);
1991         }
1992
1993         /* 2a. Create the MDT-object locally. */
1994         rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1995         if (rc != 0)
1996                 GOTO(stop, rc);
1997
1998         /* 3a. Add layout EA for the MDT-object. */
1999         rc = dt_declare_xattr_set(env, pobj, ea_buf, XATTR_NAME_LOV,
2000                                   LU_XATTR_CREATE, th);
2001         if (rc != 0)
2002                 GOTO(stop, rc);
2003
2004         /* 4a. Insert the MDT-object to .lustre/lost+found/MDTxxxx/ */
2005         rc = dt_declare_insert(env, lfsck->li_lpf_obj,
2006                                (const struct dt_rec *)pfid,
2007                                (const struct dt_key *)name, th);
2008         if (rc != 0)
2009                 GOTO(stop, rc);
2010
2011         rc = dt_trans_start(env, next, th);
2012         if (rc != 0)
2013                 GOTO(stop, rc);
2014
2015         /* 1b. Update OST-object's parent information remotely. */
2016         if (cobj != NULL) {
2017                 rc = dt_xattr_set(env, cobj, pbuf, XATTR_NAME_FID, 0, th,
2018                                   BYPASS_CAPA);
2019                 if (rc != 0)
2020                         GOTO(stop, rc);
2021         }
2022
2023         dt_write_lock(env, pobj, 0);
2024         /* 2b. Create the MDT-object locally. */
2025         rc = dt_create(env, pobj, la, NULL, dof, th);
2026         if (rc == 0)
2027                 /* 3b. Add layout EA for the MDT-object. */
2028                 rc = lfsck_layout_extend_lovea(env, th, pobj, cfid, ea_buf,
2029                                                LU_XATTR_CREATE, ltd->ltd_index,
2030                                                ea_off, false);
2031         dt_write_unlock(env, pobj);
2032         if (rc < 0)
2033                 GOTO(stop, rc);
2034
2035         /* 4b. Insert the MDT-object to .lustre/lost+found/MDTxxxx/ */
2036         rc = dt_insert(env, lfsck->li_lpf_obj,
2037                        (const struct dt_rec *)pfid,
2038                        (const struct dt_key *)name, th, BYPASS_CAPA, 1);
2039
2040         GOTO(stop, rc);
2041
2042 stop:
2043         dt_trans_stop(env, next, th);
2044
2045 unlock:
2046         lfsck_layout_unlock(&lh);
2047
2048 put:
2049         if (cobj != NULL && !IS_ERR(cobj))
2050                 lu_object_put(env, &cobj->do_lu);
2051         if (pobj != NULL && !IS_ERR(pobj))
2052                 lu_object_put(env, &pobj->do_lu);
2053         ea_buf->lb_len = buflen;
2054
2055         return rc >= 0 ? 1 : rc;
2056 }
2057
2058 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2059                                                    struct lfsck_component *com,
2060                                                    const struct lu_fid *fid,
2061                                                    __u32 index)
2062 {
2063         struct lfsck_thread_info *info  = lfsck_env_info(env);
2064         struct lfsck_request     *lr    = &info->lti_lr;
2065         struct lfsck_instance    *lfsck = com->lc_lfsck;
2066         struct lfsck_tgt_desc    *ltd;
2067         struct ptlrpc_request    *req;
2068         struct lfsck_request     *tmp;
2069         struct obd_export        *exp;
2070         int                       rc    = 0;
2071         ENTRY;
2072
2073         ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2074         if (unlikely(ltd == NULL))
2075                 RETURN(-ENXIO);
2076
2077         exp = ltd->ltd_exp;
2078         if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2079                 GOTO(put, rc = -EOPNOTSUPP);
2080
2081         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2082         if (req == NULL)
2083                 GOTO(put, rc = -ENOMEM);
2084
2085         rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2086         if (rc != 0) {
2087                 ptlrpc_request_free(req);
2088
2089                 GOTO(put, rc);
2090         }
2091
2092         memset(lr, 0, sizeof(*lr));
2093         lr->lr_event = LE_CONDITIONAL_DESTROY;
2094         lr->lr_active = LT_LAYOUT;
2095         lr->lr_fid = *fid;
2096
2097         tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2098         *tmp = *lr;
2099         ptlrpc_request_set_replen(req);
2100
2101         rc = ptlrpc_queue_wait(req);
2102         ptlrpc_req_finished(req);
2103
2104         GOTO(put, rc);
2105
2106 put:
2107         lfsck_tgt_put(ltd);
2108
2109         return rc;
2110 }
2111
2112 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2113                                                   struct lfsck_component *com,
2114                                                   struct lfsck_request *lr)
2115 {
2116         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2117         struct lu_attr                  *la     = &info->lti_la;
2118         ldlm_policy_data_t              *policy = &info->lti_policy;
2119         struct ldlm_res_id              *resid  = &info->lti_resid;
2120         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2121         struct dt_device                *dev    = lfsck->li_bottom;
2122         struct lu_fid                   *fid    = &lr->lr_fid;
2123         struct dt_object                *obj;
2124         struct thandle                  *th     = NULL;
2125         struct lustre_handle             lh     = { 0 };
2126         __u64                            flags  = 0;
2127         int                              rc     = 0;
2128         ENTRY;
2129
2130         obj = lfsck_object_find_by_dev(env, dev, fid);
2131         if (IS_ERR(obj))
2132                 RETURN(PTR_ERR(obj));
2133
2134         dt_read_lock(env, obj, 0);
2135         if (dt_object_exists(obj) == 0) {
2136                 dt_read_unlock(env, obj);
2137
2138                 GOTO(put, rc = -ENOENT);
2139         }
2140
2141         /* Get obj's attr without lock firstly. */
2142         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
2143         dt_read_unlock(env, obj);
2144         if (rc != 0)
2145                 GOTO(put, rc);
2146
2147         if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2148                 GOTO(put, rc = -ETXTBSY);
2149
2150         /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2151         LASSERT(lfsck->li_namespace != NULL);
2152
2153         memset(policy, 0, sizeof(*policy));
2154         policy->l_extent.end = OBD_OBJECT_EOF;
2155         ost_fid_build_resid(fid, resid);
2156         rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2157                                     policy, LCK_EX, &flags, ldlm_blocking_ast,
2158                                     ldlm_completion_ast, NULL, NULL, 0,
2159                                     LVB_T_NONE, NULL, &lh);
2160         if (rc != ELDLM_OK)
2161                 GOTO(put, rc = -EIO);
2162
2163         dt_write_lock(env, obj, 0);
2164         /* Get obj's attr within lock again. */
2165         rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
2166         if (rc != 0)
2167                 GOTO(unlock, rc);
2168
2169         if (la->la_ctime != 0)
2170                 GOTO(unlock, rc = -ETXTBSY);
2171
2172         th = dt_trans_create(env, dev);
2173         if (IS_ERR(th))
2174                 GOTO(unlock, rc = PTR_ERR(th));
2175
2176         rc = dt_declare_ref_del(env, obj, th);
2177         if (rc != 0)
2178                 GOTO(stop, rc);
2179
2180         rc = dt_declare_destroy(env, obj, th);
2181         if (rc != 0)
2182                 GOTO(stop, rc);
2183
2184         rc = dt_trans_start_local(env, dev, th);
2185         if (rc != 0)
2186                 GOTO(stop, rc);
2187
2188         rc = dt_ref_del(env, obj, th);
2189         if (rc != 0)
2190                 GOTO(stop, rc);
2191
2192         rc = dt_destroy(env, obj, th);
2193         if (rc == 0)
2194                 CDEBUG(D_LFSCK, "Destroy the empty OST-object "DFID" which "
2195                        "was created for reparing dangling referenced case. "
2196                        "But the original missed OST-object is found now.\n",
2197                        PFID(fid));
2198
2199         GOTO(stop, rc);
2200
2201 stop:
2202         dt_trans_stop(env, dev, th);
2203
2204 unlock:
2205         dt_write_unlock(env, obj);
2206         ldlm_lock_decref(&lh, LCK_EX);
2207
2208 put:
2209         lu_object_put(env, &obj->do_lu);
2210
2211         return rc;
2212 }
2213
2214 /**
2215  * Some OST-object has occupied the specified layout EA slot.
2216  * Such OST-object may be generated by the LFSCK when repair
2217  * dangling referenced MDT-object, which can be indicated by
2218  * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2219  * is true and such OST-object has not been modified yet, we
2220  * will replace it with the orphan OST-object; otherwise the
2221  * LFSCK will create new MDT-object to reference the orphan.
2222  *
2223  * \retval       +1: repaired
2224  * \retval        0: did nothing
2225  * \retval      -ve: on error
2226  */
2227 static int lfsck_layout_conflict_create(const struct lu_env *env,
2228                                         struct lfsck_component *com,
2229                                         struct lfsck_tgt_desc *ltd,
2230                                         struct lu_orphan_rec *rec,
2231                                         struct dt_object *parent,
2232                                         struct lu_fid *cfid,
2233                                         struct lu_buf *ea_buf,
2234                                         struct lov_ost_data_v1 *slot,
2235                                         __u32 ea_off, __u32 ori_len)
2236 {
2237         struct lfsck_thread_info *info          = lfsck_env_info(env);
2238         struct lu_fid            *cfid2         = &info->lti_fid2;
2239         struct ost_id            *oi            = &info->lti_oi;
2240         char                     *postfix       = info->lti_tmpbuf;
2241         struct lov_mds_md_v1     *lmm           = ea_buf->lb_buf;
2242         struct dt_device         *dev           = com->lc_lfsck->li_bottom;
2243         struct thandle           *th            = NULL;
2244         struct lustre_handle      lh            = { 0 };
2245         __u32                     ost_idx2      = le32_to_cpu(slot->l_ost_idx);
2246         int                       rc            = 0;
2247         ENTRY;
2248
2249         ostid_le_to_cpu(&slot->l_ost_oi, oi);
2250         ostid_to_fid(cfid2, oi, ost_idx2);
2251
2252         CDEBUG(D_LFSCK, "Handle layout EA conflict: parent "DFID
2253                ", cur-child "DFID" on the OST %u, orphan-child "
2254                DFID" on the OST %u, stripe-index %u\n",
2255                PFID(lfsck_dto2fid(parent)), PFID(cfid2), ost_idx2,
2256                PFID(cfid), ltd->ltd_index, ea_off);
2257
2258         /* Hold layout lock on the parent to prevent others to access. */
2259         rc = lfsck_layout_lock(env, com, parent, &lh,
2260                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2261         if (rc != 0)
2262                 GOTO(out, rc);
2263
2264         rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2265
2266         /* If the conflict OST-obejct is not created for fixing dangling
2267          * referenced MDT-object in former LFSCK check/repair, or it has
2268          * been modified by others, then we cannot destroy it. Re-create
2269          * a new MDT-object for the orphan OST-object. */
2270         if (rc == -ETXTBSY) {
2271                 /* No need the layout lock on the original parent. */
2272                 lfsck_layout_unlock(&lh);
2273                 ea_buf->lb_len = ori_len;
2274
2275                 fid_zero(&rec->lor_fid);
2276                 snprintf(postfix, LFSCK_TMPBUF_LEN, "-"DFID"-%x",
2277                          PFID(lu_object_fid(&parent->do_lu)), ea_off);
2278                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2279                                                   "C-", postfix, ea_off);
2280
2281                 RETURN(rc);
2282         }
2283
2284         if (rc != 0 && rc != -ENOENT)
2285                 GOTO(unlock, rc);
2286
2287         th = dt_trans_create(env, dev);
2288         if (IS_ERR(th))
2289                 GOTO(unlock, rc = PTR_ERR(th));
2290
2291         rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2292                                   LU_XATTR_REPLACE, th);
2293         if (rc != 0)
2294                 GOTO(stop, rc);
2295
2296         rc = dt_trans_start_local(env, dev, th);
2297         if (rc != 0)
2298                 GOTO(stop, rc);
2299
2300         dt_write_lock(env, parent, 0);
2301         lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2302         rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2303                                        LU_XATTR_REPLACE, ltd->ltd_index);
2304         dt_write_unlock(env, parent);
2305
2306         GOTO(stop, rc);
2307
2308 stop:
2309         dt_trans_stop(env, dev, th);
2310
2311 unlock:
2312         lfsck_layout_unlock(&lh);
2313
2314 out:
2315         ea_buf->lb_len = ori_len;
2316
2317         return rc >= 0 ? 1 : rc;
2318 }
2319
2320 /**
2321  * \retval       +1: repaired
2322  * \retval        0: did nothing
2323  * \retval      -ve: on error
2324  */
2325 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2326                                        struct lfsck_component *com,
2327                                        struct lfsck_tgt_desc *ltd,
2328                                        struct lu_orphan_rec *rec,
2329                                        struct dt_object *parent,
2330                                        struct lu_fid *cfid,
2331                                        __u32 ost_idx, __u32 ea_off)
2332 {
2333         struct lfsck_thread_info *info          = lfsck_env_info(env);
2334         struct lu_buf            *buf           = &info->lti_big_buf;
2335         struct lu_fid            *fid           = &info->lti_fid2;
2336         struct ost_id            *oi            = &info->lti_oi;
2337         struct lfsck_instance    *lfsck         = com->lc_lfsck;
2338         struct dt_device         *dt            = lfsck->li_bottom;
2339         struct lfsck_bookmark    *bk            = &lfsck->li_bookmark_ram;
2340         struct thandle            *handle       = NULL;
2341         size_t                    buflen        = buf->lb_len;
2342         struct lov_mds_md_v1     *lmm;
2343         struct lov_ost_data_v1   *objs;
2344         struct lustre_handle      lh            = { 0 };
2345         __u32                     magic;
2346         int                       fl            = 0;
2347         int                       rc            = 0;
2348         int                       rc1;
2349         int                       i;
2350         __u16                     count;
2351         bool                      locked        = false;
2352         ENTRY;
2353
2354         CDEBUG(D_LFSCK, "Re-create the crashed layout EA: parent "
2355                DFID", child "DFID", OST-index %u, stripe-index %u\n",
2356                PFID(lfsck_dto2fid(parent)), PFID(cfid), ost_idx, ea_off);
2357
2358         rc = lfsck_layout_lock(env, com, parent, &lh,
2359                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2360         if (rc != 0)
2361                 RETURN(rc);
2362
2363 again:
2364         if (locked) {
2365                 dt_write_unlock(env, parent);
2366                 locked = false;
2367         }
2368
2369         if (handle != NULL) {
2370                 dt_trans_stop(env, dt, handle);
2371                 handle = NULL;
2372         }
2373
2374         if (rc < 0)
2375                 GOTO(unlock_layout, rc);
2376
2377         if (buf->lb_len < rc) {
2378                 lu_buf_realloc(buf, rc);
2379                 buflen = buf->lb_len;
2380                 if (buf->lb_buf == NULL)
2381                         GOTO(unlock_layout, rc = -ENOMEM);
2382         }
2383
2384         if (!(bk->lb_param & LPF_DRYRUN)) {
2385                 handle = dt_trans_create(env, dt);
2386                 if (IS_ERR(handle))
2387                         GOTO(unlock_layout, rc = PTR_ERR(handle));
2388
2389                 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2390                                           fl, handle);
2391                 if (rc != 0)
2392                         GOTO(stop, rc);
2393
2394                 rc = dt_trans_start_local(env, dt, handle);
2395                 if (rc != 0)
2396                         GOTO(stop, rc);
2397         }
2398
2399         dt_write_lock(env, parent, 0);
2400         locked = true;
2401         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
2402         if (rc == -ERANGE) {
2403                 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV,
2404                                   BYPASS_CAPA);
2405                 LASSERT(rc != 0);
2406                 goto again;
2407         } else if (rc == -ENODATA || rc == 0) {
2408                 rc = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2409                 /* If the declared is not big enough, re-try. */
2410                 if (buf->lb_len < rc)
2411                         goto again;
2412
2413                 fl = LU_XATTR_CREATE;
2414         } else if (rc < 0) {
2415                 GOTO(unlock_parent, rc);
2416         } else if (unlikely(buf->lb_len == 0)) {
2417                 goto again;
2418         } else {
2419                 fl = LU_XATTR_REPLACE;
2420         }
2421
2422         if (fl == LU_XATTR_CREATE) {
2423                 if (bk->lb_param & LPF_DRYRUN)
2424                         GOTO(unlock_parent, rc = 1);
2425
2426                 LASSERT(buf->lb_len >= rc);
2427
2428                 buf->lb_len = rc;
2429                 rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
2430                                                fl, ost_idx, ea_off, false);
2431
2432                 GOTO(unlock_parent, rc);
2433         }
2434
2435         lmm = buf->lb_buf;
2436         rc1 = lfsck_layout_verify_header(lmm);
2437
2438         /* If the LOV EA crashed, the rebuild it. */
2439         if (rc1 == -EINVAL) {
2440                 if (bk->lb_param & LPF_DRYRUN)
2441                         GOTO(unlock_parent, rc = 1);
2442
2443                 LASSERT(buf->lb_len >= rc);
2444
2445                 buf->lb_len = rc;
2446                 memset(lmm, 0, buf->lb_len);
2447                 rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
2448                                                fl, ost_idx, ea_off, true);
2449
2450                 GOTO(unlock_parent, rc);
2451         }
2452
2453         /* For other unknown magic/pattern, keep the current LOV EA. */
2454         if (rc1 != 0)
2455                 GOTO(unlock_parent, rc = rc1);
2456
2457         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2458          * been verified in lfsck_layout_verify_header() already. If some
2459          * new magic introduced in the future, then layout LFSCK needs to
2460          * be updated also. */
2461         magic = le32_to_cpu(lmm->lmm_magic);
2462         if (magic == LOV_MAGIC_V1) {
2463                 objs = &(lmm->lmm_objects[0]);
2464         } else {
2465                 LASSERT(magic == LOV_MAGIC_V3);
2466                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2467         }
2468
2469         count = le16_to_cpu(lmm->lmm_stripe_count);
2470         if (count == 0)
2471                 GOTO(unlock_parent, rc = -EINVAL);
2472         LASSERT(count > 0);
2473
2474         /* Exceed the current end of MDT-object layout EA. Then extend it. */
2475         if (count <= ea_off) {
2476                 if (bk->lb_param & LPF_DRYRUN)
2477                         GOTO(unlock_parent, rc = 1);
2478
2479                 rc = lov_mds_md_size(ea_off + 1, magic);
2480                 /* If the declared is not big enough, re-try. */
2481                 if (buf->lb_len < rc)
2482                         goto again;
2483
2484                 buf->lb_len = rc;
2485                 rc = lfsck_layout_extend_lovea(env, handle, parent, cfid, buf,
2486                                                fl, ost_idx, ea_off, false);
2487                 GOTO(unlock_parent, rc);
2488         }
2489
2490         LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2491
2492         buf->lb_len = rc;
2493         for (i = 0; i < count; i++, objs++) {
2494                 /* The MDT-object was created via lfsck_layout_recover_create()
2495                  * by others before, and we fill the dummy layout EA. */
2496                 if (is_dummy_lov_ost_data(objs)) {
2497                         if (i != ea_off)
2498                                 continue;
2499
2500                         if (bk->lb_param & LPF_DRYRUN)
2501                                 GOTO(unlock_parent, rc = 1);
2502
2503                         lmm->lmm_layout_gen =
2504                             cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2505                         rc = lfsck_layout_refill_lovea(env, handle, parent,
2506                                                        cfid, buf, objs, fl,
2507                                                        ost_idx);
2508                         GOTO(unlock_parent, rc);
2509                 }
2510
2511                 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2512                 ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2513                 /* It should be rare case, the slot is there, but the LFSCK
2514                  * does not handle it during the first-phase cycle scanning. */
2515                 if (unlikely(lu_fid_eq(fid, cfid))) {
2516                         if (i == ea_off) {
2517                                 GOTO(unlock_parent, rc = 0);
2518                         } else {
2519                                 /* Rare case that the OST-object index
2520                                  * does not match the parent MDT-object
2521                                  * layout EA. We trust the later one. */
2522                                 if (bk->lb_param & LPF_DRYRUN)
2523                                         GOTO(unlock_parent, rc = 1);
2524
2525                                 dt_write_unlock(env, parent);
2526                                 if (handle != NULL)
2527                                         dt_trans_stop(env, dt, handle);
2528                                 lfsck_layout_unlock(&lh);
2529                                 buf->lb_len = buflen;
2530                                 rc = lfsck_layout_update_pfid(env, com, parent,
2531                                                         cfid, ltd->ltd_tgt, i);
2532
2533                                 RETURN(rc);
2534                         }
2535                 }
2536         }
2537
2538         /* The MDT-object exists, but related layout EA slot is occupied
2539          * by others. */
2540         if (bk->lb_param & LPF_DRYRUN)
2541                 GOTO(unlock_parent, rc = 1);
2542
2543         dt_write_unlock(env, parent);
2544         if (handle != NULL)
2545                 dt_trans_stop(env, dt, handle);
2546         lfsck_layout_unlock(&lh);
2547         if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2548                 objs = &(lmm->lmm_objects[ea_off]);
2549         else
2550                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2551         rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2552                                           buf, objs, ea_off, buflen);
2553
2554         RETURN(rc);
2555
2556 unlock_parent:
2557         if (locked)
2558                 dt_write_unlock(env, parent);
2559
2560 stop:
2561         if (handle != NULL)
2562                 dt_trans_stop(env, dt, handle);
2563
2564 unlock_layout:
2565         lfsck_layout_unlock(&lh);
2566         buf->lb_len = buflen;
2567
2568         return rc;
2569 }
2570
2571 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2572                                         struct lfsck_component *com,
2573                                         struct lfsck_tgt_desc *ltd,
2574                                         struct lu_orphan_rec *rec,
2575                                         struct lu_fid *cfid)
2576 {
2577         struct lfsck_layout     *lo     = com->lc_file_ram;
2578         struct lu_fid           *pfid   = &rec->lor_fid;
2579         struct dt_object        *parent = NULL;
2580         __u32                    ea_off = pfid->f_stripe_idx;
2581         int                      rc     = 0;
2582         ENTRY;
2583
2584         if (!fid_is_sane(cfid))
2585                 GOTO(out, rc = -EINVAL);
2586
2587         if (fid_is_zero(pfid)) {
2588                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2589                                                   "N-", "", ea_off);
2590                 GOTO(out, rc);
2591         }
2592
2593         pfid->f_ver = 0;
2594         if (!fid_is_sane(pfid))
2595                 GOTO(out, rc = -EINVAL);
2596
2597         parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2598         if (IS_ERR(parent))
2599                 GOTO(out, rc = PTR_ERR(parent));
2600
2601         if (unlikely(dt_object_remote(parent) != 0))
2602                 GOTO(put, rc = -EXDEV);
2603
2604         if (dt_object_exists(parent) == 0) {
2605                 lu_object_put(env, &parent->do_lu);
2606                 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2607                                                   "R-", "", ea_off);
2608                 GOTO(out, rc);
2609         }
2610
2611         if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2612                 GOTO(put, rc = -EISDIR);
2613
2614         rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2615                                          ltd->ltd_index, ea_off);
2616
2617         GOTO(put, rc);
2618
2619 put:
2620         if (rc <= 0)
2621                 lu_object_put(env, &parent->do_lu);
2622         else
2623                 /* The layout EA is changed, need to be reloaded next time. */
2624                 lu_object_put_nocache(env, &parent->do_lu);
2625
2626 out:
2627         down_write(&com->lc_sem);
2628         com->lc_new_scanned++;
2629         com->lc_new_checked++;
2630         if (rc > 0) {
2631                 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2632                 rc = 0;
2633         } else if (rc < 0) {
2634                 lo->ll_objs_failed_phase2++;
2635         }
2636         up_write(&com->lc_sem);
2637
2638         return rc;
2639 }
2640
2641 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2642                                     struct lfsck_component *com,
2643                                     struct lfsck_tgt_desc *ltd)
2644 {
2645         struct lfsck_layout             *lo     = com->lc_file_ram;
2646         struct lfsck_instance           *lfsck  = com->lc_lfsck;
2647         struct lfsck_bookmark           *bk     = &lfsck->li_bookmark_ram;
2648         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2649         struct ost_id                   *oi     = &info->lti_oi;
2650         struct lu_fid                   *fid    = &info->lti_fid;
2651         struct dt_object                *obj;
2652         const struct dt_it_ops          *iops;
2653         struct dt_it                    *di;
2654         int                              rc     = 0;
2655         ENTRY;
2656
2657         CDEBUG(D_LFSCK, "%s: start the orphan scanning for OST%04x\n",
2658                lfsck_lfsck2name(lfsck), ltd->ltd_index);
2659
2660         ostid_set_seq(oi, FID_SEQ_IDIF);
2661         ostid_set_id(oi, 0);
2662         ostid_to_fid(fid, oi, ltd->ltd_index);
2663         obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2664         if (unlikely(IS_ERR(obj)))
2665                 RETURN(PTR_ERR(obj));
2666
2667         rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_orphan_features);
2668         if (rc != 0)
2669                 GOTO(put, rc);
2670
2671         iops = &obj->do_index_ops->dio_it;
2672         di = iops->init(env, obj, 0, BYPASS_CAPA);
2673         if (IS_ERR(di))
2674                 GOTO(put, rc = PTR_ERR(di));
2675
2676         rc = iops->load(env, di, 0);
2677         if (rc == -ESRCH) {
2678                 /* -ESRCH means that the orphan OST-objects rbtree has been
2679                  * cleanup because of the OSS server restart or other errors. */
2680                 lo->ll_flags |= LF_INCOMPLETE;
2681                 GOTO(fini, rc);
2682         }
2683
2684         if (rc == 0)
2685                 rc = iops->next(env, di);
2686         else if (rc > 0)
2687                 rc = 0;
2688
2689         if (rc < 0)
2690                 GOTO(fini, rc);
2691
2692         if (rc > 0)
2693                 GOTO(fini, rc = 0);
2694
2695         do {
2696                 struct dt_key           *key;
2697                 struct lu_orphan_rec    *rec = &info->lti_rec;
2698
2699                 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
2700                     cfs_fail_val > 0) {
2701                         struct ptlrpc_thread    *thread = &lfsck->li_thread;
2702                         struct l_wait_info       lwi;
2703
2704                         lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
2705                                           NULL, NULL);
2706                         l_wait_event(thread->t_ctl_waitq,
2707                                      !thread_is_running(thread),
2708                                      &lwi);
2709                 }
2710
2711                 key = iops->key(env, di);
2712                 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2713                 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2714                 if (rc == 0)
2715                         rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2716                                         &com->lc_fid_latest_scanned_phase2);
2717                 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2718                         GOTO(fini, rc);
2719
2720                 lfsck_control_speed_by_self(com);
2721                 do {
2722                         rc = iops->next(env, di);
2723                 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2724         } while (rc == 0);
2725
2726         GOTO(fini, rc);
2727
2728 fini:
2729         iops->put(env, di);
2730         iops->fini(env, di);
2731 put:
2732         lu_object_put(env, &obj->do_lu);
2733
2734         CDEBUG(D_LFSCK, "%s: finish the orphan scanning for OST%04x, rc = %d\n",
2735                lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2736
2737         return rc > 0 ? 0 : rc;
2738 }
2739
2740 /* For the MDT-object with dangling reference, we need to re-create
2741  * the missed OST-object with the known FID/owner information. */
2742 static int lfsck_layout_recreate_ostobj(const struct lu_env *env,
2743                                         struct lfsck_component *com,
2744                                         struct lfsck_layout_req *llr,
2745                                         struct lu_attr *la)
2746 {
2747         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2748         struct filter_fid               *pfid   = &info->lti_new_pfid;
2749         struct dt_allocation_hint       *hint   = &info->lti_hint;
2750         struct dt_object                *parent = llr->llr_parent->llo_obj;
2751         struct dt_object                *child  = llr->llr_child;
2752         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
2753         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2754         struct thandle                  *handle;
2755         struct lu_buf                   *buf;
2756         struct lustre_handle             lh     = { 0 };
2757         int                              rc;
2758         ENTRY;
2759
2760         CDEBUG(D_LFSCK, "Repair dangling reference for: parent "DFID
2761                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
2762                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2763                llr->llr_ost_idx, llr->llr_lov_idx, la->la_uid, la->la_gid);
2764
2765         rc = lfsck_layout_lock(env, com, parent, &lh,
2766                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2767         if (rc != 0)
2768                 RETURN(rc);
2769
2770         handle = dt_trans_create(env, dev);
2771         if (IS_ERR(handle))
2772                 GOTO(unlock1, rc = PTR_ERR(handle));
2773
2774         hint->dah_parent = NULL;
2775         hint->dah_mode = 0;
2776         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2777         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2778         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2779          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2780          * parent MDT-object's layout EA. */
2781         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2782         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2783
2784         rc = dt_declare_create(env, child, la, hint, NULL, handle);
2785         if (rc != 0)
2786                 GOTO(stop, rc);
2787
2788         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2789                                   LU_XATTR_CREATE, handle);
2790         if (rc != 0)
2791                 GOTO(stop, rc);
2792
2793         rc = dt_trans_start(env, dev, handle);
2794         if (rc != 0)
2795                 GOTO(stop, rc);
2796
2797         dt_read_lock(env, parent, 0);
2798         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
2799                 GOTO(unlock2, rc = 1);
2800
2801         rc = dt_create(env, child, la, hint, NULL, handle);
2802         if (rc != 0)
2803                 GOTO(unlock2, rc);
2804
2805         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2806                           handle, BYPASS_CAPA);
2807
2808         GOTO(unlock2, rc);
2809
2810 unlock2:
2811         dt_read_unlock(env, parent);
2812
2813 stop:
2814         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2815
2816 unlock1:
2817         lfsck_layout_unlock(&lh);
2818
2819         return rc;
2820 }
2821
2822 /* If the OST-object does not recognize the MDT-object as its parent, and
2823  * there is no other MDT-object claims as its parent, then just trust the
2824  * given MDT-object as its parent. So update the OST-object filter_fid. */
2825 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2826                                               struct lfsck_component *com,
2827                                               struct lfsck_layout_req *llr,
2828                                               const struct lu_attr *pla)
2829 {
2830         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2831         struct filter_fid               *pfid   = &info->lti_new_pfid;
2832         struct lu_attr                  *tla    = &info->lti_la3;
2833         struct dt_object                *parent = llr->llr_parent->llo_obj;
2834         struct dt_object                *child  = llr->llr_child;
2835         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
2836         const struct lu_fid             *tfid   = lu_object_fid(&parent->do_lu);
2837         struct thandle                  *handle;
2838         struct lu_buf                   *buf;
2839         struct lustre_handle             lh     = { 0 };
2840         int                              rc;
2841         ENTRY;
2842
2843         CDEBUG(D_LFSCK, "Repair unmatched MDT-OST pair for: parent "DFID
2844                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
2845                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2846                llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid);
2847
2848         rc = lfsck_layout_lock(env, com, parent, &lh,
2849                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2850         if (rc != 0)
2851                 RETURN(rc);
2852
2853         handle = dt_trans_create(env, dev);
2854         if (IS_ERR(handle))
2855                 GOTO(unlock1, rc = PTR_ERR(handle));
2856
2857         pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2858         pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2859         /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2860          * MDT-object's FID::f_ver, instead it is the OST-object index in its
2861          * parent MDT-object's layout EA. */
2862         pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2863         buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2864
2865         rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2866         if (rc != 0)
2867                 GOTO(stop, rc);
2868
2869         tla->la_valid = LA_UID | LA_GID;
2870         tla->la_uid = pla->la_uid;
2871         tla->la_gid = pla->la_gid;
2872         rc = dt_declare_attr_set(env, child, tla, handle);
2873         if (rc != 0)
2874                 GOTO(stop, rc);
2875
2876         rc = dt_trans_start(env, dev, handle);
2877         if (rc != 0)
2878                 GOTO(stop, rc);
2879
2880         dt_write_lock(env, parent, 0);
2881         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
2882                 GOTO(unlock2, rc = 1);
2883
2884         rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle,
2885                           BYPASS_CAPA);
2886         if (rc != 0)
2887                 GOTO(unlock2, rc);
2888
2889         /* Get the latest parent's owner. */
2890         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
2891         if (rc != 0)
2892                 GOTO(unlock2, rc);
2893
2894         tla->la_valid = LA_UID | LA_GID;
2895         rc = dt_attr_set(env, child, tla, handle, BYPASS_CAPA);
2896
2897         GOTO(unlock2, rc);
2898
2899 unlock2:
2900         dt_write_unlock(env, parent);
2901
2902 stop:
2903         rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2904
2905 unlock1:
2906         lfsck_layout_unlock(&lh);
2907
2908         return rc;
2909 }
2910
2911 /* If there are more than one MDT-objects claim as the OST-object's parent,
2912  * and the OST-object only recognizes one of them, then we need to generate
2913  * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
2914 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
2915                                                    struct lfsck_component *com,
2916                                                    struct lfsck_layout_req *llr,
2917                                                    struct lu_attr *la,
2918                                                    struct lu_buf *buf)
2919 {
2920         struct lfsck_thread_info        *info   = lfsck_env_info(env);
2921         struct dt_allocation_hint       *hint   = &info->lti_hint;
2922         struct dt_object_format         *dof    = &info->lti_dof;
2923         struct dt_device                *pdev   = com->lc_lfsck->li_next;
2924         struct ost_id                   *oi     = &info->lti_oi;
2925         struct dt_object                *parent = llr->llr_parent->llo_obj;
2926         struct dt_device                *cdev   = lfsck_obj2dt_dev(llr->llr_child);
2927         struct dt_object                *child  = NULL;
2928         struct lu_device                *d      = &cdev->dd_lu_dev;
2929         struct lu_object                *o      = NULL;
2930         struct thandle                  *handle;
2931         struct lov_mds_md_v1            *lmm;
2932         struct lov_ost_data_v1          *objs;
2933         struct lustre_handle             lh     = { 0 };
2934         __u32                            magic;
2935         int                              rc;
2936         ENTRY;
2937
2938         CDEBUG(D_LFSCK, "Repair multiple references for: parent "DFID
2939                ", OST-index %u, stripe-index %u, owner %u:%u\n",
2940                PFID(lfsck_dto2fid(parent)), llr->llr_ost_idx,
2941                llr->llr_lov_idx, la->la_uid, la->la_gid);
2942
2943         rc = lfsck_layout_lock(env, com, parent, &lh,
2944                                MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR);
2945         if (rc != 0)
2946                 RETURN(rc);
2947
2948         handle = dt_trans_create(env, pdev);
2949         if (IS_ERR(handle))
2950                 GOTO(unlock1, rc = PTR_ERR(handle));
2951
2952         o = lu_object_anon(env, d, NULL);
2953         if (IS_ERR(o))
2954                 GOTO(stop, rc = PTR_ERR(o));
2955
2956         child = container_of(o, struct dt_object, do_lu);
2957         o = lu_object_locate(o->lo_header, d->ld_type);
2958         if (unlikely(o == NULL))
2959                 GOTO(stop, rc = -EINVAL);
2960
2961         child = container_of(o, struct dt_object, do_lu);
2962         la->la_valid = LA_UID | LA_GID;
2963         hint->dah_parent = NULL;
2964         hint->dah_mode = 0;
2965         dof->dof_type = DFT_REGULAR;
2966         rc = dt_declare_create(env, child, la, NULL, NULL, handle);
2967         if (rc != 0)
2968                 GOTO(stop, rc);
2969
2970         rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2971                                   LU_XATTR_REPLACE, handle);
2972         if (rc != 0)
2973                 GOTO(stop, rc);
2974
2975         rc = dt_trans_start(env, pdev, handle);
2976         if (rc != 0)
2977                 GOTO(stop, rc);
2978
2979         dt_write_lock(env, parent, 0);
2980         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
2981                 GOTO(unlock2, rc = 0);
2982
2983         rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV, BYPASS_CAPA);
2984         if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
2985                 GOTO(unlock2, rc = 0);
2986
2987         lmm = buf->lb_buf;
2988         /* Someone change layout during the LFSCK, no need to repair then. */
2989         if (le16_to_cpu(lmm->lmm_layout_gen) != llr->llr_parent->llo_gen)
2990                 GOTO(unlock2, rc = 0);
2991
2992         rc = dt_create(env, child, la, hint, dof, handle);
2993         if (rc != 0)
2994                 GOTO(unlock2, rc);
2995
2996         /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2997          * been verified in lfsck_layout_verify_header() already. If some
2998          * new magic introduced in the future, then layout LFSCK needs to
2999          * be updated also. */
3000         magic = le32_to_cpu(lmm->lmm_magic);
3001         if (magic == LOV_MAGIC_V1) {
3002                 objs = &(lmm->lmm_objects[0]);
3003         } else {
3004                 LASSERT(magic == LOV_MAGIC_V3);
3005                 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3006         }
3007
3008         lmm->lmm_layout_gen = cpu_to_le16(llr->llr_parent->llo_gen + 1);
3009         fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3010         ostid_cpu_to_le(oi, &objs[llr->llr_lov_idx].l_ost_oi);
3011         objs[llr->llr_lov_idx].l_ost_gen = cpu_to_le32(0);
3012         objs[llr->llr_lov_idx].l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3013         rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3014                           LU_XATTR_REPLACE, handle, BYPASS_CAPA);
3015
3016         GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
3017
3018 unlock2:
3019         dt_write_unlock(env, parent);
3020
3021 stop:
3022         if (child != NULL)
3023                 lu_object_put(env, &child->do_lu);
3024
3025         dt_trans_stop(env, pdev, handle);
3026
3027 unlock1:
3028         lfsck_layout_unlock(&lh);
3029
3030         return rc;
3031 }
3032
3033 /* If the MDT-object and the OST-object have different owner information,
3034  * then trust the MDT-object, because the normal chown/chgrp handle order
3035  * is from MDT to OST, and it is possible that some chown/chgrp operation
3036  * is partly done. */
3037 static int lfsck_layout_repair_owner(const struct lu_env *env,
3038                                      struct lfsck_component *com,
3039                                      struct lfsck_layout_req *llr,
3040                                      struct lu_attr *pla)
3041 {
3042         struct lfsck_thread_info        *info   = lfsck_env_info(env);
3043         struct lu_attr                  *tla    = &info->lti_la3;
3044         struct dt_object                *parent = llr->llr_parent->llo_obj;
3045         struct dt_object                *child  = llr->llr_child;
3046         struct dt_device                *dev    = lfsck_obj2dt_dev(child);
3047         struct thandle                  *handle;
3048         int                              rc;
3049         ENTRY;
3050
3051         CDEBUG(D_LFSCK, "Repair inconsistent file owner for: parent "DFID
3052                ", child "DFID", OST-index %u, stripe-index %u, owner %u:%u\n",
3053                PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3054                llr->llr_ost_idx, llr->llr_lov_idx, pla->la_uid, pla->la_gid);
3055
3056         handle = dt_trans_create(env, dev);
3057         if (IS_ERR(handle))
3058                 RETURN(PTR_ERR(handle));
3059
3060         tla->la_uid = pla->la_uid;
3061         tla->la_gid = pla->la_gid;
3062         tla->la_valid = LA_UID | LA_GID;
3063         rc = dt_declare_attr_set(env, child, tla, handle);
3064         if (rc != 0)
3065                 GOTO(stop, rc);
3066
3067         rc = dt_trans_start(env, dev, handle);
3068         if (rc != 0)
3069                 GOTO(stop, rc);
3070
3071         /* Use the dt_object lock to serialize with destroy and attr_set. */
3072         dt_read_lock(env, parent, 0);
3073         if (unlikely(lu_object_is_dying(parent->do_lu.lo_header)))
3074                 GOTO(unlock, rc = 1);
3075
3076         /* Get the latest parent's owner. */
3077         rc = dt_attr_get(env, parent, tla, BYPASS_CAPA);
3078         if (rc != 0) {
3079                 CWARN("%s: fail to get the latest parent's ("DFID") owner, "
3080                       "not sure whether some others chown/chgrp during the "
3081                       "LFSCK: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
3082                       PFID(lfsck_dto2fid(parent)), rc);
3083
3084                 GOTO(unlock, rc);
3085         }
3086
3087         /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3088         if (unlikely(tla->la_uid != pla->la_uid ||
3089                      tla->la_gid != pla->la_gid))
3090                 GOTO(unlock, rc = 1);
3091