4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2014, 2016, Intel Corporation.
26 * lustre/lfsck/lfsck_layout.c
28 * Author: Fan, Yong <fan.yong@intel.com>
32 # define EXPORT_SYMTAB
34 #define DEBUG_SUBSYSTEM S_LFSCK
36 #include <linux/bitops.h>
37 #include <linux/rbtree.h>
39 #include <lustre/lustre_idl.h>
40 #include <lu_object.h>
41 #include <dt_object.h>
42 #include <lustre_fid.h>
43 #include <lustre_lib.h>
44 #include <lustre_net.h>
45 #include <lustre/lustre_user.h>
46 #include <md_object.h>
47 #include <obd_class.h>
49 #include "lfsck_internal.h"
51 #define LFSCK_LAYOUT_MAGIC_V1 0xB173AE14
52 #define LFSCK_LAYOUT_MAGIC_V2 0xB1734D76
54 #define LFSCK_LAYOUT_MAGIC LFSCK_LAYOUT_MAGIC_V2
56 struct lfsck_layout_seq {
57 struct list_head lls_list;
60 __u64 lls_lastid_known;
61 struct dt_object *lls_lastid_obj;
62 unsigned int lls_dirty:1;
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. */
75 /* How many times we have failed to get the master status. */
79 struct lfsck_layout_slave_data {
80 /* list for lfsck_layout_seq */
81 struct list_head llsd_seq_list;
83 /* list for the masters involve layout verification. */
84 struct list_head llsd_master_list;
87 struct dt_object *llsd_rb_obj;
88 struct rb_root llsd_rb_root;
89 rwlock_t llsd_rb_lock;
90 unsigned int llsd_rbtree_valid:1;
93 struct lfsck_layout_slave_async_args {
94 struct obd_export *llsaa_exp;
95 struct lfsck_component *llsaa_com;
96 struct lfsck_layout_slave_target *llsaa_llst;
99 static inline bool lfsck_comp_extent_aligned(__u64 size)
101 return (size & (LOV_MIN_STRIPE_SIZE - 1)) == 0;
105 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
107 if (atomic_dec_and_test(&llst->llst_ref)) {
108 LASSERT(list_empty(&llst->llst_list));
115 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
117 struct lfsck_layout_slave_target *llst;
118 struct lfsck_layout_slave_target *tmp;
125 INIT_LIST_HEAD(&llst->llst_list);
127 llst->llst_index = index;
128 atomic_set(&llst->llst_ref, 1);
130 spin_lock(&llsd->llsd_lock);
131 list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
132 if (tmp->llst_index == index) {
138 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
139 spin_unlock(&llsd->llsd_lock);
148 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
149 struct lfsck_layout_slave_target *llst)
153 spin_lock(&llsd->llsd_lock);
154 if (!list_empty(&llst->llst_list)) {
155 list_del_init(&llst->llst_list);
158 spin_unlock(&llsd->llsd_lock);
161 lfsck_layout_llst_put(llst);
164 static inline struct lfsck_layout_slave_target *
165 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
166 __u32 index, bool unlink)
168 struct lfsck_layout_slave_target *llst;
170 spin_lock(&llsd->llsd_lock);
171 list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
172 if (llst->llst_index == index) {
174 list_del_init(&llst->llst_list);
176 atomic_inc(&llst->llst_ref);
177 spin_unlock(&llsd->llsd_lock);
182 spin_unlock(&llsd->llsd_lock);
187 static struct lfsck_layout_req *
188 lfsck_layout_assistant_req_init(struct lfsck_assistant_object *lso,
189 struct dt_object *child, __u32 comp_id,
190 __u32 ost_idx, __u32 lov_idx)
192 struct lfsck_layout_req *llr;
196 return ERR_PTR(-ENOMEM);
198 INIT_LIST_HEAD(&llr->llr_lar.lar_list);
199 llr->llr_lar.lar_parent = lfsck_assistant_object_get(lso);
200 llr->llr_child = child;
201 llr->llr_comp_id = comp_id;
202 llr->llr_ost_idx = ost_idx;
203 llr->llr_lov_idx = lov_idx;
208 static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
209 struct lfsck_assistant_req *lar)
211 struct lfsck_layout_req *llr =
212 container_of0(lar, struct lfsck_layout_req, llr_lar);
214 lfsck_object_put(env, llr->llr_child);
215 lfsck_assistant_object_put(env, lar->lar_parent);
220 lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
221 struct ptlrpc_request *req,
225 struct lfsck_async_interpret_args *laia = args;
226 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
228 ltd->ltd_synced_failures = 1;
229 atomic_dec(laia->laia_count);
236 * Notify remote LFSCK instances about former failures.
238 * The local LFSCK instance has recorded which OSTs have ever failed to respond
239 * some LFSCK verification requests (maybe because of network issues or the OST
240 * itself trouble). During the respond gap, the OST may missed some OST-objects
241 * verification, then the OST cannot know whether related OST-objects have been
242 * referenced by related MDT-objects or not, then in the second-stage scanning,
243 * these OST-objects will be regarded as orphan, if the OST-object contains bad
244 * parent FID for back reference, then it will misguide the LFSCK to make wrong
245 * fixing for the fake orphan.
247 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
248 * it will scan the bitmap for the ever failed OSTs, and notify them that they
249 * have ever missed some OST-object verification and should skip the handling
250 * for orphan OST-objects on all MDTs that are in the layout LFSCK.
252 * \param[in] env pointer to the thread context
253 * \param[in] com pointer to the lfsck component
254 * \param[in] lr pointer to the lfsck request
256 static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
257 struct lfsck_component *com,
258 struct lfsck_request *lr)
260 struct lfsck_async_interpret_args *laia =
261 &lfsck_env_info(env)->lti_laia2;
262 struct lfsck_assistant_data *lad = com->lc_data;
263 struct lfsck_layout *lo = com->lc_file_ram;
264 struct lfsck_instance *lfsck = com->lc_lfsck;
265 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
266 struct lfsck_tgt_desc *ltd;
267 struct ptlrpc_request_set *set;
273 if (!lad->lad_incomplete)
276 /* If the MDT has ever failed to verfiy some OST-objects,
277 * then sync failures with them firstly. */
278 lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
280 atomic_set(&count, 0);
281 memset(laia, 0, sizeof(*laia));
282 laia->laia_count = &count;
283 set = ptlrpc_prep_set();
285 GOTO(out, rc = -ENOMEM);
287 down_read(<ds->ltd_rw_sem);
288 cfs_foreach_bit(lad->lad_bitmap, idx) {
289 ltd = lfsck_ltd2tgt(ltds, idx);
290 LASSERT(ltd != NULL);
292 laia->laia_ltd = ltd;
293 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
294 lfsck_layout_assistant_sync_failures_interpret,
297 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
298 "notify target %x for %s phase1 done: "
299 "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
300 ltd->ltd_index, lad->lad_name, rc);
307 up_read(<ds->ltd_rw_sem);
309 if (rc == 0 && atomic_read(&count) > 0)
310 rc = ptlrpc_set_wait(set);
312 ptlrpc_set_destroy(set);
314 if (rc == 0 && atomic_read(&count) > 0)
321 /* If failed to sync failures with the OSTs, then have to
322 * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
323 * subsequent orphan OST-object handling. */
324 lo->ll_flags |= LF_INCOMPLETE;
326 lr->lr_flags2 = lo->ll_flags;
329 static int lfsck_layout_verify_header_v1v3(struct dt_object *obj,
330 struct lov_mds_md_v1 *lmm)
335 magic = le32_to_cpu(lmm->lmm_magic);
336 /* If magic crashed, keep it there. Sometime later, during OST-object
337 * orphan handling, if some OST-object(s) back-point to it, it can be
338 * verified and repaired. */
339 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
342 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
347 CDEBUG(D_LFSCK, "%s LOV EA magic %u for the file "DFID"\n",
348 rc == -EINVAL ? "Unknown" : "Unsupported",
349 magic, PFID(lfsck_dto2fid(obj)));
354 pattern = le32_to_cpu(lmm->lmm_pattern);
355 /* XXX: currently, we only support LOV_PATTERN_RAID0. */
356 if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
357 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u for the file "
358 DFID"\n", pattern, PFID(lfsck_dto2fid(obj)));
366 static int lfsck_layout_verify_header(struct dt_object *obj,
367 struct lov_mds_md_v1 *lmm)
371 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1) {
372 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
374 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
376 if (unlikely(count == 0)) {
377 CDEBUG(D_LFSCK, "the PFL file "DFID" contains invalid "
378 "components count 0\n",
379 PFID(lfsck_dto2fid(obj)));
384 for (i = 0; i < count; i++) {
385 struct lov_comp_md_entry_v1 *lcme =
386 &lcm->lcm_entries[i];
387 __u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
388 __u64 end = le64_to_cpu(lcme->lcme_extent.e_end);
389 __u32 comp_id = le32_to_cpu(lcme->lcme_id);
391 if (unlikely(comp_id == 0 || comp_id > LCME_ID_MAX)) {
392 CDEBUG(D_LFSCK, "found invalid FPL ID %u "
393 "for the file "DFID" at idx %d\n",
394 comp_id, PFID(lfsck_dto2fid(obj)), i);
399 if (unlikely(start >= end ||
400 !lfsck_comp_extent_aligned(start) ||
401 (!lfsck_comp_extent_aligned(end) &&
402 end != LUSTRE_EOF))) {
403 CDEBUG(D_LFSCK, "found invalid FPL extent "
404 "range [%llu - %llu) for the file "
406 start, end, PFID(lfsck_dto2fid(obj)), i);
411 rc = lfsck_layout_verify_header_v1v3(obj,
412 (struct lov_mds_md_v1 *)((char *)lmm +
413 le32_to_cpu(lcme->lcme_offset)));
418 rc = lfsck_layout_verify_header_v1v3(obj, lmm);
424 static int lfsck_layout_get_lovea(const struct lu_env *env,
425 struct dt_object *obj, struct lu_buf *buf)
431 rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
433 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
437 lu_buf_realloc(buf, rc);
438 if (buf->lb_buf == NULL)
450 if (unlikely(buf->lb_buf == NULL)) {
451 lu_buf_alloc(buf, rc);
452 if (buf->lb_buf == NULL)
458 rc1 = lfsck_layout_verify_header(obj, buf->lb_buf);
460 return rc1 ? rc1 : rc;
463 #define LFSCK_RBTREE_BITMAP_SIZE PAGE_SIZE
464 #define LFSCK_RBTREE_BITMAP_WIDTH (LFSCK_RBTREE_BITMAP_SIZE << 3)
465 #define LFSCK_RBTREE_BITMAP_MASK (LFSCK_RBTREE_BITMAP_WIDTH - 1)
467 struct lfsck_rbtree_node {
468 struct rb_node lrn_node;
471 atomic_t lrn_known_count;
472 atomic_t lrn_accessed_count;
473 void *lrn_known_bitmap;
474 void *lrn_accessed_bitmap;
477 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
478 __u64 seq, __u32 oid)
480 if (seq < lrn->lrn_seq)
483 if (seq > lrn->lrn_seq)
486 if (oid < lrn->lrn_first_oid)
489 if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
495 /* The caller should hold llsd->llsd_rb_lock. */
496 static struct lfsck_rbtree_node *
497 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
498 const struct lu_fid *fid, bool *exact)
500 struct rb_node *node = llsd->llsd_rb_root.rb_node;
501 struct rb_node *prev = NULL;
502 struct lfsck_rbtree_node *lrn = NULL;
508 while (node != NULL) {
510 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
511 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
513 node = node->rb_left;
515 node = node->rb_right;
523 /* If there is no exactly matched one, then to the next valid one. */
526 /* The rbtree is empty. */
533 node = rb_next(prev);
535 /* The end of the rbtree. */
539 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
544 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
545 const struct lu_fid *fid)
547 struct lfsck_rbtree_node *lrn;
551 return ERR_PTR(-ENOMEM);
553 OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
554 if (lrn->lrn_known_bitmap == NULL) {
557 return ERR_PTR(-ENOMEM);
560 OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
561 if (lrn->lrn_accessed_bitmap == NULL) {
562 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
565 return ERR_PTR(-ENOMEM);
568 RB_CLEAR_NODE(&lrn->lrn_node);
569 lrn->lrn_seq = fid_seq(fid);
570 lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
571 atomic_set(&lrn->lrn_known_count, 0);
572 atomic_set(&lrn->lrn_accessed_count, 0);
577 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
579 OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
580 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
584 /* The caller should hold lock. */
585 static struct lfsck_rbtree_node *
586 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
587 struct lfsck_rbtree_node *lrn)
589 struct rb_node **pos = &llsd->llsd_rb_root.rb_node;
590 struct rb_node *parent = NULL;
591 struct lfsck_rbtree_node *tmp;
594 while (*pos != NULL) {
596 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
597 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
599 pos = &(*pos)->rb_left;
601 pos = &(*pos)->rb_right;
606 rb_link_node(&lrn->lrn_node, parent, pos);
607 rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
612 extern const struct dt_index_operations lfsck_orphan_index_ops;
614 static int lfsck_rbtree_setup(const struct lu_env *env,
615 struct lfsck_component *com)
617 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
618 struct lfsck_instance *lfsck = com->lc_lfsck;
619 struct dt_device *dev = lfsck->li_bottom;
620 struct lfsck_layout_slave_data *llsd = com->lc_data;
621 struct dt_object *obj;
623 fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
624 fid->f_oid = lfsck_dev_idx(lfsck);
626 obj = dt_locate(env, dev, fid);
628 RETURN(PTR_ERR(obj));
630 /* Generate an in-RAM object to stand for the layout rbtree.
631 * Scanning the layout rbtree will be via the iteration over
632 * the object. In the future, the rbtree may be written onto
633 * disk with the object.
635 * Mark the object to be as exist. */
636 obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
637 obj->do_index_ops = &lfsck_orphan_index_ops;
638 llsd->llsd_rb_obj = obj;
639 llsd->llsd_rbtree_valid = 1;
640 dev->dd_record_fid_accessed = 1;
642 CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
643 lfsck_lfsck2name(lfsck));
648 static void lfsck_rbtree_cleanup(const struct lu_env *env,
649 struct lfsck_component *com)
651 struct lfsck_instance *lfsck = com->lc_lfsck;
652 struct lfsck_layout_slave_data *llsd = com->lc_data;
653 struct rb_node *node = rb_first(&llsd->llsd_rb_root);
654 struct rb_node *next;
655 struct lfsck_rbtree_node *lrn;
657 lfsck->li_bottom->dd_record_fid_accessed = 0;
658 /* Invalid the rbtree, then no others will use it. */
659 write_lock(&llsd->llsd_rb_lock);
660 llsd->llsd_rbtree_valid = 0;
661 write_unlock(&llsd->llsd_rb_lock);
663 while (node != NULL) {
664 next = rb_next(node);
665 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
666 rb_erase(node, &llsd->llsd_rb_root);
667 lfsck_rbtree_free(lrn);
671 if (llsd->llsd_rb_obj != NULL) {
672 lfsck_object_put(env, llsd->llsd_rb_obj);
673 llsd->llsd_rb_obj = NULL;
676 CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
677 lfsck_lfsck2name(lfsck));
680 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
681 struct lfsck_component *com,
682 const struct lu_fid *fid,
685 struct lfsck_layout_slave_data *llsd = com->lc_data;
686 struct lfsck_rbtree_node *lrn;
692 if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
695 if (!fid_is_idif(fid) && !fid_is_norm(fid))
698 read_lock(&llsd->llsd_rb_lock);
699 if (!llsd->llsd_rbtree_valid)
700 GOTO(unlock, rc = 0);
702 lrn = lfsck_rbtree_search(llsd, fid, NULL);
704 struct lfsck_rbtree_node *tmp;
708 read_unlock(&llsd->llsd_rb_lock);
709 tmp = lfsck_rbtree_new(env, fid);
711 GOTO(out, rc = PTR_ERR(tmp));
714 write_lock(&llsd->llsd_rb_lock);
715 if (!llsd->llsd_rbtree_valid) {
716 lfsck_rbtree_free(tmp);
717 GOTO(unlock, rc = 0);
720 lrn = lfsck_rbtree_insert(llsd, tmp);
722 lfsck_rbtree_free(tmp);
725 idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
726 /* Any accessed object must be a known object. */
727 if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
728 atomic_inc(&lrn->lrn_known_count);
729 if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
730 atomic_inc(&lrn->lrn_accessed_count);
732 GOTO(unlock, rc = 0);
736 write_unlock(&llsd->llsd_rb_lock);
738 read_unlock(&llsd->llsd_rb_lock);
740 if (rc != 0 && accessed) {
741 struct lfsck_layout *lo = com->lc_file_ram;
743 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
744 "bitmap, and will cause incorrect LFSCK OST-object "
745 "handling, so disable it to cancel orphan handling "
746 "for related device. rc = %d\n",
747 lfsck_lfsck2name(com->lc_lfsck), rc);
749 lo->ll_flags |= LF_INCOMPLETE;
750 lfsck_rbtree_cleanup(env, com);
754 static inline void lldk_le_to_cpu(struct lfsck_layout_dangling_key *des,
755 const struct lfsck_layout_dangling_key *src)
757 fid_le_to_cpu(&des->lldk_fid, &src->lldk_fid);
758 des->lldk_comp_id = le32_to_cpu(src->lldk_comp_id);
759 des->lldk_ea_off = le32_to_cpu(src->lldk_ea_off);
762 static inline void lldk_cpu_to_le(struct lfsck_layout_dangling_key *des,
763 const struct lfsck_layout_dangling_key *src)
765 fid_cpu_to_le(&des->lldk_fid, &src->lldk_fid);
766 des->lldk_comp_id = cpu_to_le32(src->lldk_comp_id);
767 des->lldk_ea_off = cpu_to_le32(src->lldk_ea_off);
770 static inline void lldk_be_to_cpu(struct lfsck_layout_dangling_key *des,
771 const struct lfsck_layout_dangling_key *src)
773 fid_be_to_cpu(&des->lldk_fid, &src->lldk_fid);
774 des->lldk_comp_id = be32_to_cpu(src->lldk_comp_id);
775 des->lldk_ea_off = be32_to_cpu(src->lldk_ea_off);
778 static inline void lldk_cpu_to_be(struct lfsck_layout_dangling_key *des,
779 const struct lfsck_layout_dangling_key *src)
781 fid_cpu_to_be(&des->lldk_fid, &src->lldk_fid);
782 des->lldk_comp_id = cpu_to_be32(src->lldk_comp_id);
783 des->lldk_ea_off = cpu_to_be32(src->lldk_ea_off);
786 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
787 const struct lfsck_layout *src)
791 des->ll_magic = le32_to_cpu(src->ll_magic);
792 des->ll_status = le32_to_cpu(src->ll_status);
793 des->ll_flags = le32_to_cpu(src->ll_flags);
794 des->ll_success_count = le32_to_cpu(src->ll_success_count);
795 des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
796 des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
797 des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
798 des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
799 des->ll_time_last_checkpoint =
800 le64_to_cpu(src->ll_time_last_checkpoint);
801 des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
802 des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
803 des->ll_pos_first_inconsistent =
804 le64_to_cpu(src->ll_pos_first_inconsistent);
805 des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
806 des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
807 des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
808 des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
809 for (i = 0; i < LLIT_MAX; i++)
810 des->ll_objs_repaired[i] =
811 le64_to_cpu(src->ll_objs_repaired[i]);
812 des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
813 des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
814 lldk_le_to_cpu(&des->ll_lldk_latest_scanned_phase2,
815 &src->ll_lldk_latest_scanned_phase2);
818 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
819 const struct lfsck_layout *src)
823 des->ll_magic = cpu_to_le32(src->ll_magic);
824 des->ll_status = cpu_to_le32(src->ll_status);
825 des->ll_flags = cpu_to_le32(src->ll_flags);
826 des->ll_success_count = cpu_to_le32(src->ll_success_count);
827 des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
828 des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
829 des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
830 des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
831 des->ll_time_last_checkpoint =
832 cpu_to_le64(src->ll_time_last_checkpoint);
833 des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
834 des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
835 des->ll_pos_first_inconsistent =
836 cpu_to_le64(src->ll_pos_first_inconsistent);
837 des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
838 des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
839 des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
840 des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
841 for (i = 0; i < LLIT_MAX; i++)
842 des->ll_objs_repaired[i] =
843 cpu_to_le64(src->ll_objs_repaired[i]);
844 des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
845 des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
846 lldk_cpu_to_le(&des->ll_lldk_latest_scanned_phase2,
847 &src->ll_lldk_latest_scanned_phase2);
851 * Load the OST bitmap from the lfsck_layout trace file.
853 * \param[in] env pointer to the thread context
854 * \param[in] com pointer to the lfsck component
856 * \retval 0 for success
857 * \retval negative error number on failure or data corruption
859 static int lfsck_layout_load_bitmap(const struct lu_env *env,
860 struct lfsck_component *com)
862 struct dt_object *obj = com->lc_obj;
863 struct lfsck_assistant_data *lad = com->lc_data;
864 struct lfsck_layout *lo = com->lc_file_ram;
865 struct cfs_bitmap *bitmap = lad->lad_bitmap;
866 loff_t pos = com->lc_file_size;
872 if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
874 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
876 nbits = lo->ll_bitmap_size;
878 if (unlikely(nbits < BITS_PER_LONG))
879 nbits = BITS_PER_LONG;
881 if (nbits > bitmap->size) {
882 __u32 new_bits = bitmap->size;
883 struct cfs_bitmap *new_bitmap;
885 while (new_bits < nbits)
888 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
889 if (new_bitmap == NULL)
892 lad->lad_bitmap = new_bitmap;
893 CFS_FREE_BITMAP(bitmap);
897 if (lo->ll_bitmap_size == 0) {
898 lad->lad_incomplete = 0;
899 CFS_RESET_BITMAP(bitmap);
904 size = (lo->ll_bitmap_size + 7) >> 3;
905 rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
907 RETURN(rc >= 0 ? -EINVAL : rc);
909 if (cfs_bitmap_check_empty(bitmap))
910 lad->lad_incomplete = 0;
912 lad->lad_incomplete = 1;
918 * Load the layout LFSCK trace file from disk.
920 * The layout LFSCK trace file records the layout LFSCK status information
921 * and other statistics, such as how many objects have been scanned, and how
922 * many objects have been repaired, and etc. It also contains the bitmap for
923 * failed OSTs during the layout LFSCK. All these information will be loaded
924 * from disk to RAM when the layout LFSCK component setup.
926 * \param[in] env pointer to the thread context
927 * \param[in] com pointer to the lfsck component
929 * \retval positive number for file data corruption, the caller
930 * should reset the layout LFSCK trace file
931 * \retval 0 for success
932 * \retval negative error number on failure
934 static int lfsck_layout_load(const struct lu_env *env,
935 struct lfsck_component *com)
937 struct lfsck_layout *lo = com->lc_file_ram;
938 ssize_t size = com->lc_file_size;
942 rc = dt_read(env, com->lc_obj,
943 lfsck_buf_get(env, com->lc_file_disk, size), &pos);
947 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
948 lfsck_lfsck2name(com->lc_lfsck), rc);
950 } else if (rc != size) {
951 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
952 lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
956 lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
957 if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
958 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
959 "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
960 lo->ll_magic, LFSCK_LAYOUT_MAGIC);
968 * Store the layout LFSCK trace file on disk.
970 * The layout LFSCK trace file records the layout LFSCK status information
971 * and other statistics, such as how many objects have been scanned, and how
972 * many objects have been repaired, and etc. It also contains the bitmap for
973 * failed OSTs during the layout LFSCK. All these information will be synced
974 * from RAM to disk periodically.
976 * \param[in] env pointer to the thread context
977 * \param[in] com pointer to the lfsck component
979 * \retval 0 for success
980 * \retval negative error number on failure
982 static int lfsck_layout_store(const struct lu_env *env,
983 struct lfsck_component *com)
985 struct dt_object *obj = com->lc_obj;
986 struct lfsck_instance *lfsck = com->lc_lfsck;
987 struct lfsck_layout *lo_ram = com->lc_file_ram;
988 struct lfsck_layout *lo = com->lc_file_disk;
990 struct dt_device *dev = lfsck_obj2dev(obj);
991 struct cfs_bitmap *bitmap = NULL;
993 ssize_t size = com->lc_file_size;
998 if (lfsck->li_master) {
999 struct lfsck_assistant_data *lad = com->lc_data;
1001 bitmap = lad->lad_bitmap;
1002 nbits = bitmap->size;
1005 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
1008 lo_ram->ll_bitmap_size = nbits;
1009 lfsck_layout_cpu_to_le(lo, lo_ram);
1010 th = dt_trans_create(env, dev);
1012 GOTO(log, rc = PTR_ERR(th));
1014 rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
1019 if (bitmap != NULL) {
1020 rc = dt_declare_record_write(env, obj,
1021 lfsck_buf_get(env, bitmap->data, nbits >> 3),
1027 rc = dt_trans_start_local(env, dev, th);
1032 rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
1036 if (bitmap != NULL) {
1038 rc = dt_record_write(env, obj,
1039 lfsck_buf_get(env, bitmap->data, nbits >> 3),
1046 dt_trans_stop(env, dev, th);
1050 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
1051 lfsck_lfsck2name(lfsck), rc);
1056 static int lfsck_layout_init(const struct lu_env *env,
1057 struct lfsck_component *com)
1059 struct lfsck_layout *lo = com->lc_file_ram;
1062 memset(lo, 0, com->lc_file_size);
1063 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
1064 lo->ll_status = LS_INIT;
1065 down_write(&com->lc_sem);
1066 rc = lfsck_layout_store(env, com);
1067 if (rc == 0 && com->lc_lfsck->li_master)
1068 rc = lfsck_load_sub_trace_files(env, com,
1069 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
1070 up_write(&com->lc_sem);
1075 static int fid_is_for_ostobj(const struct lu_env *env,
1076 struct lfsck_instance *lfsck,
1077 struct dt_object *obj, const struct lu_fid *fid)
1079 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1080 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
1081 struct lustre_ost_attrs *loa;
1084 fld_range_set_any(range);
1085 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1087 if (fld_range_is_ost(range))
1093 loa = &lfsck_env_info(env)->lti_loa;
1094 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, loa, sizeof(*loa)),
1096 if (rc >= sizeof(struct lustre_mdt_attrs)) {
1097 lustre_lma_swab(&loa->loa_lma);
1099 return loa->loa_lma.lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1102 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1107 static struct lfsck_layout_seq *
1108 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1110 struct lfsck_layout_seq *lls;
1112 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1113 if (lls->lls_seq == seq)
1116 if (lls->lls_seq > seq)
1124 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1125 struct lfsck_layout_seq *lls)
1127 struct lfsck_layout_seq *tmp;
1128 struct list_head *pos = &llsd->llsd_seq_list;
1130 list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1131 if (lls->lls_seq < tmp->lls_seq) {
1132 pos = &tmp->lls_list;
1136 list_add_tail(&lls->lls_list, pos);
1140 lfsck_layout_lastid_create(const struct lu_env *env,
1141 struct lfsck_instance *lfsck,
1142 struct dt_object *obj)
1144 struct lfsck_thread_info *info = lfsck_env_info(env);
1145 struct lu_attr *la = &info->lti_la;
1146 struct dt_object_format *dof = &info->lti_dof;
1147 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1148 struct dt_device *dt = lfsck_obj2dev(obj);
1155 if (bk->lb_param & LPF_DRYRUN)
1158 memset(la, 0, sizeof(*la));
1159 la->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1160 la->la_valid = LA_MODE | LA_UID | LA_GID;
1161 memset(dof, 0, sizeof(*dof));
1162 dof->dof_type = dt_mode_to_dft(S_IFREG);
1164 th = dt_trans_create(env, dt);
1166 GOTO(log, rc = PTR_ERR(th));
1168 rc = dt_declare_create(env, obj, la, NULL, dof, th);
1172 rc = dt_declare_record_write(env, obj,
1173 lfsck_buf_get(env, &lastid,
1179 rc = dt_trans_start_local(env, dt, th);
1183 dt_write_lock(env, obj, 0);
1184 if (likely(dt_object_exists(obj) == 0)) {
1185 rc = dt_create(env, obj, la, NULL, dof, th);
1187 rc = dt_record_write(env, obj,
1188 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1191 dt_write_unlock(env, obj);
1196 dt_trans_stop(env, dt, th);
1199 CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1201 lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1207 lfsck_layout_lastid_reload(const struct lu_env *env,
1208 struct lfsck_component *com,
1209 struct lfsck_layout_seq *lls)
1215 dt_read_lock(env, lls->lls_lastid_obj, 0);
1216 rc = dt_record_read(env, lls->lls_lastid_obj,
1217 lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1218 dt_read_unlock(env, lls->lls_lastid_obj);
1219 if (unlikely(rc != 0))
1222 lastid = le64_to_cpu(lastid);
1223 if (lastid < lls->lls_lastid_known) {
1224 struct lfsck_instance *lfsck = com->lc_lfsck;
1225 struct lfsck_layout *lo = com->lc_file_ram;
1227 lls->lls_lastid = lls->lls_lastid_known;
1229 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1230 LASSERT(lfsck->li_out_notify != NULL);
1232 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1233 LE_LASTID_REBUILDING);
1234 lo->ll_flags |= LF_CRASHED_LASTID;
1236 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1237 "LAST_ID file (1) for the sequence %#llx"
1238 ", old value %llu, known value %llu\n",
1239 lfsck_lfsck2name(lfsck), lls->lls_seq,
1240 lastid, lls->lls_lastid);
1242 } else if (lastid >= lls->lls_lastid) {
1243 lls->lls_lastid = lastid;
1251 lfsck_layout_lastid_store(const struct lu_env *env,
1252 struct lfsck_component *com)
1254 struct lfsck_instance *lfsck = com->lc_lfsck;
1255 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1256 struct dt_device *dt = lfsck->li_bottom;
1257 struct lfsck_layout_slave_data *llsd = com->lc_data;
1258 struct lfsck_layout_seq *lls;
1264 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1267 if (!lls->lls_dirty)
1270 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1271 "<seq> %#llx as <oid> %llu\n",
1272 lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1274 if (bk->lb_param & LPF_DRYRUN) {
1279 th = dt_trans_create(env, dt);
1282 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1283 "the LAST_ID for <seq> %#llx(1): rc = %d\n",
1284 lfsck_lfsck2name(com->lc_lfsck),
1289 lastid = cpu_to_le64(lls->lls_lastid);
1290 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1291 lfsck_buf_get(env, &lastid,
1297 rc = dt_trans_start_local(env, dt, th);
1301 dt_write_lock(env, lls->lls_lastid_obj, 0);
1302 rc = dt_record_write(env, lls->lls_lastid_obj,
1303 lfsck_buf_get(env, &lastid,
1304 sizeof(lastid)), &pos, th);
1305 dt_write_unlock(env, lls->lls_lastid_obj);
1310 dt_trans_stop(env, dt, th);
1313 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1314 "the LAST_ID for <seq> %#llx(2): rc = %d\n",
1315 lfsck_lfsck2name(com->lc_lfsck),
1324 lfsck_layout_lastid_load(const struct lu_env *env,
1325 struct lfsck_component *com,
1326 struct lfsck_layout_seq *lls)
1328 struct lfsck_instance *lfsck = com->lc_lfsck;
1329 struct lfsck_layout *lo = com->lc_file_ram;
1330 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
1331 struct dt_object *obj;
1336 lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1337 obj = dt_locate(env, lfsck->li_bottom, fid);
1339 RETURN(PTR_ERR(obj));
1341 /* LAST_ID crashed, to be rebuilt */
1342 if (dt_object_exists(obj) == 0) {
1343 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1344 LASSERT(lfsck->li_out_notify != NULL);
1346 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1347 LE_LASTID_REBUILDING);
1348 lo->ll_flags |= LF_CRASHED_LASTID;
1350 CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1351 "LAST_ID file for sequence %#llx\n",
1352 lfsck_lfsck2name(lfsck), lls->lls_seq);
1354 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1356 struct l_wait_info lwi = LWI_TIMEOUT(
1357 cfs_time_seconds(cfs_fail_val),
1360 /* Some others may changed the cfs_fail_val
1361 * as zero after above check, re-check it for
1362 * sure to avoid falling into wait for ever. */
1363 if (likely(lwi.lwi_timeout > 0)) {
1364 struct ptlrpc_thread *thread =
1367 up_write(&com->lc_sem);
1368 l_wait_event(thread->t_ctl_waitq,
1369 !thread_is_running(thread),
1371 down_write(&com->lc_sem);
1376 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1378 dt_read_lock(env, obj, 0);
1379 rc = dt_read(env, obj,
1380 lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1382 dt_read_unlock(env, obj);
1383 if (rc != 0 && rc != sizeof(__u64))
1384 GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1386 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1387 LASSERT(lfsck->li_out_notify != NULL);
1389 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1390 LE_LASTID_REBUILDING);
1391 lo->ll_flags |= LF_CRASHED_LASTID;
1393 CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1394 "LAST_ID file for the sequence %#llx"
1396 lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1399 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1407 lfsck_object_put(env, obj);
1409 lls->lls_lastid_obj = obj;
1414 static void lfsck_layout_record_failure(const struct lu_env *env,
1415 struct lfsck_instance *lfsck,
1416 struct lfsck_layout *lo)
1420 lo->ll_objs_failed_phase1++;
1421 cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1423 if (lo->ll_pos_first_inconsistent == 0 ||
1424 lo->ll_pos_first_inconsistent < cookie) {
1425 lo->ll_pos_first_inconsistent = cookie;
1427 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1428 "inconsistency at the pos [%llu]\n",
1429 lfsck_lfsck2name(lfsck),
1430 lo->ll_pos_first_inconsistent);
1434 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1435 struct lfsck_component *com,
1438 struct lfsck_instance *lfsck = com->lc_lfsck;
1439 struct lfsck_layout *lo = com->lc_file_ram;
1441 down_write(&com->lc_sem);
1442 lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1443 HALF_SEC - com->lc_time_last_checkpoint);
1444 lo->ll_time_last_checkpoint = cfs_time_current_sec();
1445 lo->ll_objs_checked_phase2 += com->lc_new_checked;
1448 if (lo->ll_flags & LF_INCOMPLETE) {
1449 lo->ll_status = LS_PARTIAL;
1451 if (lfsck->li_master) {
1452 struct lfsck_assistant_data *lad = com->lc_data;
1454 if (lad->lad_incomplete)
1455 lo->ll_status = LS_PARTIAL;
1457 lo->ll_status = LS_COMPLETED;
1459 lo->ll_status = LS_COMPLETED;
1462 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1463 lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1464 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1465 lo->ll_success_count++;
1466 } else if (rc == 0) {
1467 if (lfsck->li_status != 0)
1468 lo->ll_status = lfsck->li_status;
1470 lo->ll_status = LS_STOPPED;
1472 lo->ll_status = LS_FAILED;
1475 rc = lfsck_layout_store(env, com);
1476 up_write(&com->lc_sem);
1481 static int lfsck_layout_trans_stop(const struct lu_env *env,
1482 struct dt_device *dev,
1483 struct thandle *handle, int result)
1487 /* XXX: If there is something worng or it needs to repair nothing,
1488 * then notify the lower to stop the modification. Currently,
1489 * we use th_result for such purpose, that may be replaced by
1490 * some rollback mechanism in the future. */
1491 handle->th_result = result;
1492 rc = dt_trans_stop(env, dev, handle);
1494 return result > 0 ? 0 : result;
1496 return rc == 0 ? 1 : rc;
1499 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1500 struct lfsck_component *com,
1501 const struct lu_fid *pfid,
1502 const struct lu_fid *cfid,
1503 __u32 comp_id, __u32 ea_off,
1506 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1507 struct lu_fid *rec = &lfsck_env_info(env)->lti_fid3;
1508 struct dt_device *dev;
1509 struct dt_object *obj;
1510 struct thandle *th = NULL;
1515 idx = lfsck_sub_trace_file_fid2idx(pfid);
1516 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1517 dev = lfsck_obj2dev(obj);
1519 fid_cpu_to_be(&key->lldk_fid, pfid);
1520 key->lldk_comp_id = cpu_to_be32(comp_id);
1521 key->lldk_ea_off = cpu_to_be32(ea_off);
1523 fid_cpu_to_be(rec, cfid);
1524 rec->f_ver = cpu_to_be32(ost_idx);
1526 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1528 th = dt_trans_create(env, dev);
1530 GOTO(unlock, rc = PTR_ERR(th));
1532 rc = dt_declare_insert(env, obj,
1533 (const struct dt_rec *)rec,
1534 (const struct dt_key *)key, th);
1538 rc = dt_trans_start_local(env, dev, th);
1542 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1543 (const struct dt_key *)key, th, 1);
1548 if (th && !IS_ERR(th))
1549 dt_trans_stop(env, dev, th);
1551 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1553 CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", comp_id = %u, "
1554 "ea_off = %u, ost_idx = %u, into the trace file for further "
1555 "dangling check: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1556 PFID(pfid), PFID(cfid), comp_id, ea_off, ost_idx, rc);
1561 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1562 struct lfsck_component *com,
1563 const struct lu_fid *fid,
1564 __u32 comp_id, __u32 ea_off)
1566 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1567 struct dt_device *dev;
1568 struct dt_object *obj;
1569 struct thandle *th = NULL;
1574 idx = lfsck_sub_trace_file_fid2idx(fid);
1575 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1576 dev = lfsck_obj2dev(obj);
1578 fid_cpu_to_be(&key->lldk_fid, fid);
1579 key->lldk_comp_id = cpu_to_be32(comp_id);
1580 key->lldk_ea_off = cpu_to_be32(ea_off);
1582 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1584 th = dt_trans_create(env, dev);
1586 GOTO(unlock, rc = PTR_ERR(th));
1588 rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1592 rc = dt_trans_start_local(env, dev, th);
1596 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1601 if (th && !IS_ERR(th))
1602 dt_trans_stop(env, dev, th);
1604 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1606 CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1607 ", comp_id = %u, ea_off = %u from the trace file: rc = %d\n",
1608 lfsck_lfsck2name(com->lc_lfsck), PFID(fid), comp_id, ea_off, rc);
1614 * Get the system default stripe size.
1616 * \param[in] env pointer to the thread context
1617 * \param[in] lfsck pointer to the lfsck instance
1618 * \param[out] size pointer to the default stripe size
1620 * \retval 0 for success
1621 * \retval negative error number on failure
1623 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1624 struct lfsck_instance *lfsck,
1627 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1628 struct dt_object *root;
1631 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1633 return PTR_ERR(root);
1635 /* Get the default stripe size via xattr_get on the backend root. */
1636 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1639 /* The lum->lmm_stripe_size is LE mode. The *size also
1640 * should be LE mode. So it is unnecessary to convert. */
1641 *size = lum->lmm_stripe_size;
1643 } else if (unlikely(rc == 0)) {
1647 lfsck_object_put(env, root);
1653 * \retval +1: repaired
1654 * \retval 0: did nothing
1655 * \retval -ve: on error
1657 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1658 struct lfsck_instance *lfsck,
1659 struct thandle *handle,
1660 struct dt_object *parent,
1661 const struct lu_fid *cfid,
1663 struct lov_mds_md_v1 *lmm,
1664 struct lov_ost_data_v1 *slot,
1665 int fl, __u32 ost_idx, int size)
1667 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1668 struct lu_buf ea_buf;
1675 magic = le32_to_cpu(lmm->lmm_magic);
1676 pattern = le32_to_cpu(lmm->lmm_pattern);
1677 count = le16_to_cpu(lmm->lmm_stripe_count);
1679 fid_to_ostid(cfid, oi);
1680 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1681 slot->l_ost_gen = cpu_to_le32(0);
1682 slot->l_ost_idx = cpu_to_le32(ost_idx);
1684 if (pattern & LOV_PATTERN_F_HOLE) {
1685 struct lov_ost_data_v1 *objs;
1688 if (magic == LOV_MAGIC_V1)
1689 objs = &lmm->lmm_objects[0];
1691 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1692 for (i = 0; i < count; i++, objs++) {
1693 if (lovea_slot_is_dummy(objs))
1697 /* If the @slot is the last dummy slot to be refilled,
1698 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1701 cpu_to_le32(pattern & ~LOV_PATTERN_F_HOLE);
1703 CDEBUG(D_LFSCK, "%s: remove layout HOLE for "DFID
1704 ": parent "DFID"\n", lfsck_lfsck2name(lfsck),
1705 PFID(cfid), PFID(lfsck_dto2fid(parent)));
1709 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
1710 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1717 static struct lov_ost_data_v1 *
1718 __lfsck_layout_new_v1_lovea(struct lov_mds_md_v1 *lmm,
1719 const struct lu_fid *pfid,
1720 __u32 stripe_size, __u32 ea_off,
1721 __u32 pattern, __u16 count)
1723 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1724 lmm->lmm_pattern = cpu_to_le32(pattern);
1725 fid_to_lmm_oi(pfid, &lmm->lmm_oi);
1726 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1727 lmm->lmm_stripe_size = cpu_to_le32(stripe_size);
1728 lmm->lmm_stripe_count = cpu_to_le16(count);
1729 lmm->lmm_layout_gen = cpu_to_le16(1);
1730 memset(&lmm->lmm_objects[0], 0,
1731 sizeof(struct lov_ost_data_v1) * count);
1733 return &lmm->lmm_objects[ea_off];
1736 static int lfsck_layout_new_v1_lovea(const struct lu_env *env,
1737 struct lfsck_instance *lfsck,
1738 struct ost_layout *ol,
1739 struct dt_object *parent,
1740 struct lu_buf *buf, __u32 ea_off,
1741 struct lov_mds_md_v1 **lmm,
1742 struct lov_ost_data_v1 **objs)
1745 __u32 stripe_size = ol->ol_stripe_size;
1746 __u32 pattern = LOV_PATTERN_RAID0;
1749 if (ol->ol_stripe_count != 0)
1750 count = ol->ol_stripe_count;
1754 size = lov_mds_md_size(count, LOV_MAGIC_V1);
1755 LASSERTF(buf->lb_len >= size,
1756 "buffer len %d is less than real size %d\n",
1757 (int)buf->lb_len, size);
1759 if (stripe_size == 0) {
1762 rc = lfsck_layout_get_def_stripesize(env, lfsck, &stripe_size);
1768 if (ol->ol_stripe_count > 1 ||
1769 (ol->ol_stripe_count == 0 && ea_off != 0)) {
1770 pattern |= LOV_PATTERN_F_HOLE;
1771 memset(&(*lmm)->lmm_objects[0], 0,
1772 count * sizeof(struct lov_ost_data_v1));
1775 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1776 stripe_size, ea_off, pattern, count);
1781 static int lfsck_layout_new_comp_lovea(const struct lu_env *env,
1782 struct ost_layout *ol,
1783 struct dt_object *parent,
1784 struct lu_buf *buf, __u32 ea_off,
1785 struct lov_mds_md_v1 **lmm,
1786 struct lov_ost_data_v1 **objs)
1788 struct lov_comp_md_v1 *lcm;
1789 struct lov_comp_md_entry_v1 *lcme;
1790 __u32 pattern = LOV_PATTERN_RAID0;
1791 __u32 offset = sizeof(*lcm) + sizeof(*lcme);
1792 int lcme_size = lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1793 int size = offset + lcme_size;
1795 LASSERTF(buf->lb_len >= size,
1796 "buffer len %d is less than real size %d\n",
1797 (int)buf->lb_len, size);
1800 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1801 lcm->lcm_size = cpu_to_le32(size);
1802 lcm->lcm_layout_gen = cpu_to_le32(1);
1804 lcm->lcm_entry_count = cpu_to_le16(1);
1806 lcme = &lcm->lcm_entries[0];
1807 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1808 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1809 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1810 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1811 lcme->lcme_offset = cpu_to_le32(offset);
1812 lcme->lcme_size = cpu_to_le32(lcme_size);
1813 if (ol->ol_stripe_count > 1)
1814 pattern |= LOV_PATTERN_F_HOLE;
1816 *lmm = buf->lb_buf + offset;
1817 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1818 ol->ol_stripe_size, ea_off,
1819 pattern, ol->ol_stripe_count);
1824 static int lfsck_layout_add_comp_comp(const struct lu_env *env,
1825 struct lfsck_instance *lfsck,
1826 struct thandle *handle,
1827 struct ost_layout *ol,
1828 struct dt_object *parent,
1829 const struct lu_fid *cfid,
1830 struct lu_buf *buf, __u32 ost_idx,
1831 __u32 ea_off, int pos)
1833 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1834 struct lov_comp_md_entry_v1 *lcme;
1835 struct lov_mds_md_v1 *lmm;
1836 struct lov_ost_data_v1 *objs;
1837 int added = sizeof(*lcme) +
1838 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1839 int size = le32_to_cpu(lcm->lcm_size) + added;
1843 __u32 pattern = LOV_PATTERN_RAID0;
1844 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1847 lu_buf_check_and_grow(buf, size);
1848 /* set the lcm again because lu_buf_check_and_grow() may
1849 * have reallocated the buf. */
1851 lcm->lcm_size = cpu_to_le32(size);
1852 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1853 lcm->lcm_entry_count = cpu_to_le16(count + 1);
1855 /* 1. Move the component bodies from [pos, count-1] to [pos+1, count]
1856 * with distance of 'added'. */
1859 for (i = pos; i < count; i++) {
1860 lcme = &lcm->lcm_entries[i];
1861 size += le32_to_cpu(lcme->lcme_size);
1864 offset = le32_to_cpu(lcm->lcm_entries[pos].lcme_offset);
1865 memmove(buf->lb_buf + offset + added,
1866 buf->lb_buf + offset, size);
1870 /* 2. Move the component header [0, pos-1] to [0, pos-1] with distance
1871 * of 'sizeof(struct lov_comp_md_entry_v1)' */
1873 for (i = 0; i < pos; i++) {
1874 lcme = &lcm->lcm_entries[i];
1875 size += le32_to_cpu(lcme->lcme_size);
1878 offset = le32_to_cpu(lcm->lcm_entries[0].lcme_offset);
1879 memmove(buf->lb_buf + offset + sizeof(*lcme),
1880 buf->lb_buf + offset, size);
1883 /* 3. Recalculate the enter offset for the component [pos, count-1] */
1884 for (i = count - 1; i >= pos; i--) {
1885 lcm->lcm_entries[i + 1] = lcm->lcm_entries[i];
1886 lcm->lcm_entries[i + 1].lcme_offset =
1887 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i + 1].
1888 lcme_offset) + added);
1891 /* 4. Recalculate the enter offset for the component [0, pos) */
1892 for (i = 0; i < pos; i++) {
1893 lcm->lcm_entries[i].lcme_offset =
1894 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i].
1895 lcme_offset) + sizeof(*lcme));
1898 offset = sizeof(*lcm) + sizeof(*lcme) * (count + 1) + size;
1899 /* 4. Insert the new component header (entry) at the slot 'pos'. */
1900 lcme = &lcm->lcm_entries[pos];
1901 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1902 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1903 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1904 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1905 lcme->lcme_offset = cpu_to_le32(offset);
1906 lcme->lcme_size = cpu_to_le32(lov_mds_md_size(ol->ol_stripe_count,
1909 if (ol->ol_stripe_count > 1)
1910 pattern |= LOV_PATTERN_F_HOLE;
1912 lmm = buf->lb_buf + offset;
1913 /* 5. Insert teh new component body at the 'offset'. */
1914 objs = __lfsck_layout_new_v1_lovea(lmm, lfsck_dto2fid(parent),
1915 ol->ol_stripe_size, ea_off,
1916 pattern, ol->ol_stripe_count);
1918 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1919 lmm, objs, LU_XATTR_REPLACE, ost_idx,
1920 le32_to_cpu(lcm->lcm_size));
1922 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant add new COMP for "
1923 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1924 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
1925 "comp_end %llu, %s LOV EA hole: rc = %d\n",
1926 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1927 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
1928 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
1929 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
1930 "with" : "without", rc);
1935 static int lfsck_layout_extend_v1v3_lovea(const struct lu_env *env,
1936 struct lfsck_instance *lfsck,
1937 struct thandle *handle,
1938 struct ost_layout *ol,
1939 struct dt_object *parent,
1940 const struct lu_fid *cfid,
1941 struct lu_buf *buf, __u32 ost_idx,
1944 struct lov_mds_md_v1 *lmm = buf->lb_buf;
1945 struct lov_ost_data_v1 *objs;
1946 __u16 count = le16_to_cpu(lmm->lmm_stripe_count);
1947 __u32 magic = le32_to_cpu(lmm->lmm_magic);
1953 /* The original LOVEA maybe re-generated via old filter_fid, at
1954 * that time, we do not know the stripe count and stripe size. */
1955 if (ol->ol_stripe_count > count)
1956 count = ol->ol_stripe_count;
1957 if (ol->ol_stripe_size != 0 &&
1958 ol->ol_stripe_size != le32_to_cpu(lmm->lmm_stripe_size))
1959 lmm->lmm_stripe_size = cpu_to_le32(ol->ol_stripe_size);
1961 if (magic == LOV_MAGIC_V1)
1962 objs = &lmm->lmm_objects[count];
1964 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[count];
1966 gap = ea_off - count;
1970 size = lov_mds_md_size(count, magic);
1971 LASSERTF(buf->lb_len >= size,
1972 "buffer len %d is less than real size %d\n",
1973 (int)buf->lb_len, size);
1976 memset(objs, 0, gap * sizeof(*objs));
1977 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1980 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1981 lmm->lmm_stripe_count = cpu_to_le16(count);
1984 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1985 lmm, objs, LU_XATTR_REPLACE, ost_idx, size);
1987 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1988 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1989 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
1990 "comp_end %llu, %s LOV EA hole: rc = %d\n",
1991 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1992 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
1993 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
1994 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
1995 "with" : "without", rc);
2001 * \retval +1: repaired
2002 * \retval 0: did nothing
2003 * \retval -ve: on error
2005 static int lfsck_layout_update_lovea(const struct lu_env *env,
2006 struct lfsck_instance *lfsck,
2007 struct thandle *handle,
2008 struct ost_layout *ol,
2009 struct dt_object *parent,
2010 const struct lu_fid *cfid,
2011 struct lu_buf *buf, int fl,
2012 __u32 ost_idx, __u32 ea_off)
2014 struct lov_mds_md_v1 *lmm = NULL;
2015 struct lov_ost_data_v1 *objs = NULL;
2019 if (ol->ol_comp_id != 0)
2020 rc = lfsck_layout_new_comp_lovea(env, ol, parent, buf, ea_off,
2023 rc = lfsck_layout_new_v1_lovea(env, lfsck, ol, parent, buf,
2024 ea_off, &lmm, &objs);
2027 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid,
2028 buf, lmm, objs, fl, ost_idx, rc);
2030 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant created layout EA for "
2031 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2032 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2033 "comp_end %llu, fl %d, %s LOV EA hole: rc = %d\n",
2034 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2035 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2036 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end, fl,
2037 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2038 "with" : "without", rc);
2043 static int __lfsck_layout_update_pfid(const struct lu_env *env,
2044 struct dt_object *child,
2045 const struct lu_fid *pfid,
2046 const struct ost_layout *ol, __u32 offset)
2048 struct dt_device *dev = lfsck_obj2dev(child);
2049 struct filter_fid *ff = &lfsck_env_info(env)->lti_ff;
2050 struct thandle *handle;
2051 struct lu_buf buf = { NULL };
2054 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2055 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2056 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2057 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2058 * parent MDT-object's layout EA. */
2059 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
2060 ost_layout_cpu_to_le(&ff->ff_layout, ol);
2061 lfsck_buf_init(&buf, ff, sizeof(*ff));
2063 handle = dt_trans_create(env, dev);
2065 RETURN(PTR_ERR(handle));
2067 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2071 rc = dt_trans_start_local(env, dev, handle);
2075 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2080 dt_trans_stop(env, dev, handle);
2086 * \retval +1: repaired
2087 * \retval 0: did nothing
2088 * \retval -ve: on error
2090 static int lfsck_layout_update_pfid(const struct lu_env *env,
2091 struct lfsck_component *com,
2092 struct dt_object *parent,
2093 struct lu_fid *cfid,
2094 struct dt_device *cdev,
2095 struct ost_layout *ol, __u32 ea_off)
2097 struct dt_object *child;
2101 child = lfsck_object_find_by_dev(env, cdev, cfid);
2103 RETURN(PTR_ERR(child));
2105 rc = __lfsck_layout_update_pfid(env, child,
2106 lu_object_fid(&parent->do_lu),
2108 lfsck_object_put(env, child);
2110 RETURN(rc == 0 ? 1 : rc);
2113 static int lfsck_lovea_size(struct ost_layout *ol, __u32 ea_off)
2115 if (ol->ol_comp_id != 0)
2116 return sizeof(struct lov_comp_md_v1) +
2117 sizeof(struct lov_comp_md_entry_v1) +
2118 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2120 if (ol->ol_stripe_count != 0)
2121 return lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2123 return lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2127 * This function will create the MDT-object with the given (partial) LOV EA.
2129 * Under some data corruption cases, the MDT-object of the file may be lost,
2130 * but its OST-objects, or some of them are there. The layout LFSCK needs to
2131 * re-create the MDT-object with the orphan OST-object(s) information.
2133 * On the other hand, the LFSCK may has created some OST-object for repairing
2134 * dangling LOV EA reference, but as the LFSCK processing, it may find that
2135 * the old OST-object is there and should replace the former new created OST
2136 * object. Unfortunately, some others have modified such newly created object.
2137 * To keep the data (both new and old), the LFSCK will create MDT-object with
2138 * new FID to reference the original OST-object.
2140 * \param[in] env pointer to the thread context
2141 * \param[in] com pointer to the lfsck component
2142 * \param[in] ltd pointer to target device descriptor
2143 * \param[in] rec pointer to the record for the orphan OST-object
2144 * \param[in] cfid pointer to FID for the orphan OST-object
2145 * \param[in] infix additional information, such as the FID for original
2146 * MDT-object and the stripe offset in the LOV EA
2147 * \param[in] type the type for describing why the orphan MDT-object is
2148 * created. The rules are as following:
2150 * type "C": Multiple OST-objects claim the same MDT-object and the
2151 * same slot in the layout EA. Then the LFSCK will create
2152 * new MDT-object(s) to hold the conflict OST-object(s).
2154 * type "N": The orphan OST-object does not know which one was the
2155 * real parent MDT-object, so the LFSCK uses new FID for
2156 * its parent MDT-object.
2158 * type "R": The orphan OST-object knows its parent MDT-object FID,
2159 * but does not know the position (the file name) in the
2162 * type "D": The MDT-object is a directory, it may knows its parent
2163 * but because there is no valid linkEA, the LFSCK cannot
2164 * know where to put it back to the namespace.
2165 * type "O": The MDT-object has no linkEA, and there is no name
2166 * entry that references the MDT-object.
2168 * type "P": The orphan object to be created was a parent directory
2169 * of some MDT-object which linkEA shows that the @orphan
2170 * object is missing.
2172 * The orphan name will be like:
2173 * ${FID}-${infix}-${type}-${conflict_version}
2175 * \param[in] ea_off the stripe offset in the LOV EA
2177 * \retval positive on repaired something
2178 * \retval 0 if needs to repair nothing
2179 * \retval negative error number on failure
2181 static int lfsck_layout_recreate_parent(const struct lu_env *env,
2182 struct lfsck_component *com,
2183 struct lfsck_tgt_desc *ltd,
2184 struct lu_orphan_rec_v2 *rec,
2185 struct lu_fid *cfid,
2190 struct lfsck_thread_info *info = lfsck_env_info(env);
2191 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
2192 char *name = info->lti_key;
2193 struct lu_attr *la = &info->lti_la2;
2194 struct dt_object_format *dof = &info->lti_dof;
2195 struct lfsck_instance *lfsck = com->lc_lfsck;
2196 struct ost_layout *ol = &rec->lor_layout;
2197 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
2198 struct lu_fid *tfid = &info->lti_fid3;
2199 struct dt_device *dev = lfsck->li_bottom;
2200 struct dt_object *lpf = lfsck->li_lpf_obj;
2201 struct dt_object *pobj = NULL;
2202 struct dt_object *cobj = NULL;
2203 struct thandle *th = NULL;
2204 struct lu_buf *ea_buf = &info->lti_big_buf;
2205 struct lu_buf lov_buf;
2206 struct lfsck_lock_handle *llh = &info->lti_llh;
2207 struct linkea_data ldata = { NULL };
2208 struct lu_buf linkea_buf;
2209 const struct lu_name *pname;
2215 if (unlikely(lpf == NULL))
2216 GOTO(log, rc = -ENXIO);
2218 /* We use two separated transactions to repair the inconsistency.
2220 * 1) create the MDT-object locally.
2221 * 2) update the OST-object's PFID EA if necessary.
2223 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
2224 * updated when the layout LFSCK run next time.
2226 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
2227 * when the layout LFSCK run next time. */
2229 if (fid_is_zero(pfid)) {
2230 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
2234 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2236 GOTO(log, rc = PTR_ERR(cobj));
2239 pobj = lfsck_object_find_by_dev(env, dev, pfid);
2241 GOTO(log, rc = PTR_ERR(pobj));
2243 LASSERT(infix != NULL);
2244 LASSERT(type != NULL);
2246 memset(la, 0, sizeof(*la));
2247 la->la_uid = rec->lor_rec.lor_uid;
2248 la->la_gid = rec->lor_rec.lor_gid;
2249 la->la_mode = S_IFREG | S_IRUSR;
2250 la->la_valid = LA_MODE | LA_UID | LA_GID;
2252 memset(dof, 0, sizeof(*dof));
2253 dof->dof_type = dt_mode_to_dft(S_IFREG);
2254 /* Because the dof->dof_reg.striped = 0, the LOD will not create
2255 * the stripe(s). The LFSCK will specify the LOV EA via
2256 * lfsck_layout_update_lovea(). */
2258 size = lfsck_lovea_size(ol, ea_off);
2259 if (ea_buf->lb_len < size) {
2260 lu_buf_realloc(ea_buf, size);
2261 if (ea_buf->lb_buf == NULL)
2262 GOTO(log, rc = -ENOMEM);
2267 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
2269 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2270 (const struct dt_key *)name);
2271 if (rc != 0 && rc != -ENOENT)
2275 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
2276 MDS_INODELOCK_UPDATE, LCK_PW);
2280 /* Re-check whether the name conflict with othrs after taken
2282 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2283 (const struct dt_key *)name);
2284 if (unlikely(rc == 0)) {
2292 pname = lfsck_name_get_const(env, name, strlen(name));
2293 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
2294 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
2298 /* The 1st transaction. */
2299 th = dt_trans_create(env, dev);
2301 GOTO(unlock, rc = PTR_ERR(th));
2303 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
2307 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
2308 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
2309 LU_XATTR_CREATE, th);
2313 dtrec->rec_fid = pfid;
2314 dtrec->rec_type = S_IFREG;
2315 rc = dt_declare_insert(env, lpf,
2316 (const struct dt_rec *)dtrec,
2317 (const struct dt_key *)name, th);
2321 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2322 ldata.ld_leh->leh_len);
2323 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
2324 XATTR_NAME_LINK, 0, th);
2328 rc = dt_trans_start_local(env, dev, th);
2332 dt_write_lock(env, pobj, 0);
2333 rc = dt_create(env, pobj, la, NULL, dof, th);
2335 rc = lfsck_layout_update_lovea(env, lfsck, th, ol, pobj, cfid,
2336 &lov_buf, LU_XATTR_CREATE, ltd->ltd_index, ea_off);
2337 dt_write_unlock(env, pobj);
2341 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
2342 (const struct dt_key *)name, th, 1);
2346 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
2347 if (rc == 0 && cobj != NULL) {
2348 dt_trans_stop(env, dev, th);
2351 /* The 2nd transaction. */
2352 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ol, ea_off);
2359 dt_trans_stop(env, dev, th);
2365 if (cobj != NULL && !IS_ERR(cobj))
2366 lfsck_object_put(env, cobj);
2367 if (pobj != NULL && !IS_ERR(pobj))
2368 lfsck_object_put(env, pobj);
2371 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2372 "recreate the lost MDT-object: parent "DFID
2373 ", child "DFID", OST-index %u, stripe-index %u, "
2374 "infix %s, type %s: rc = %d\n",
2375 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2376 ltd->ltd_index, ea_off, infix, type, rc);
2378 return rc >= 0 ? 1 : rc;
2381 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2382 struct lfsck_component *com,
2383 const struct lu_fid *fid,
2386 struct lfsck_thread_info *info = lfsck_env_info(env);
2387 struct lfsck_request *lr = &info->lti_lr;
2388 struct lfsck_instance *lfsck = com->lc_lfsck;
2389 struct lfsck_tgt_desc *ltd;
2390 struct ptlrpc_request *req;
2391 struct lfsck_request *tmp;
2392 struct obd_export *exp;
2396 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2397 if (unlikely(ltd == NULL))
2401 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2402 GOTO(put, rc = -EOPNOTSUPP);
2404 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2406 GOTO(put, rc = -ENOMEM);
2408 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2410 ptlrpc_request_free(req);
2415 memset(lr, 0, sizeof(*lr));
2416 lr->lr_event = LE_CONDITIONAL_DESTROY;
2417 lr->lr_active = LFSCK_TYPE_LAYOUT;
2420 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2422 ptlrpc_request_set_replen(req);
2424 rc = ptlrpc_queue_wait(req);
2425 ptlrpc_req_finished(req);
2435 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2436 struct lfsck_component *com,
2437 struct lfsck_request *lr)
2439 struct lfsck_thread_info *info = lfsck_env_info(env);
2440 struct lu_attr *la = &info->lti_la;
2441 union ldlm_policy_data *policy = &info->lti_policy;
2442 struct ldlm_res_id *resid = &info->lti_resid;
2443 struct lfsck_instance *lfsck = com->lc_lfsck;
2444 struct dt_device *dev = lfsck->li_bottom;
2445 struct lu_fid *fid = &lr->lr_fid;
2446 struct dt_object *obj;
2447 struct thandle *th = NULL;
2448 struct lustre_handle lh = { 0 };
2453 obj = lfsck_object_find_by_dev(env, dev, fid);
2455 RETURN(PTR_ERR(obj));
2457 dt_read_lock(env, obj, 0);
2458 if (dt_object_exists(obj) == 0 ||
2459 lfsck_is_dead_obj(obj)) {
2460 dt_read_unlock(env, obj);
2462 GOTO(put, rc = -ENOENT);
2465 /* Get obj's attr without lock firstly. */
2466 rc = dt_attr_get(env, obj, la);
2467 dt_read_unlock(env, obj);
2471 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2472 GOTO(put, rc = -ETXTBSY);
2474 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2475 LASSERT(lfsck->li_namespace != NULL);
2477 memset(policy, 0, sizeof(*policy));
2478 policy->l_extent.end = OBD_OBJECT_EOF;
2479 ost_fid_build_resid(fid, resid);
2480 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2481 policy, LCK_EX, &flags, ldlm_blocking_ast,
2482 ldlm_completion_ast, NULL, NULL, 0,
2483 LVB_T_NONE, NULL, &lh);
2485 GOTO(put, rc = -EIO);
2487 dt_write_lock(env, obj, 0);
2488 /* Get obj's attr within lock again. */
2489 rc = dt_attr_get(env, obj, la);
2493 if (la->la_ctime != 0)
2494 GOTO(unlock, rc = -ETXTBSY);
2496 th = dt_trans_create(env, dev);
2498 GOTO(unlock, rc = PTR_ERR(th));
2500 rc = dt_declare_ref_del(env, obj, th);
2504 rc = dt_declare_destroy(env, obj, th);
2508 rc = dt_trans_start_local(env, dev, th);
2512 rc = dt_ref_del(env, obj, th);
2516 rc = dt_destroy(env, obj, th);
2518 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2519 "OST-object "DFID" that was created for reparing "
2520 "dangling referenced case. But the original missing "
2521 "OST-object is found now.\n",
2522 lfsck_lfsck2name(lfsck), PFID(fid));
2527 dt_trans_stop(env, dev, th);
2530 dt_write_unlock(env, obj);
2531 ldlm_lock_decref(&lh, LCK_EX);
2534 lfsck_object_put(env, obj);
2540 * Some OST-object has occupied the specified layout EA slot.
2541 * Such OST-object may be generated by the LFSCK when repair
2542 * dangling referenced MDT-object, which can be indicated by
2543 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2544 * is true and such OST-object has not been modified yet, we
2545 * will replace it with the orphan OST-object; otherwise the
2546 * LFSCK will create new MDT-object to reference the orphan.
2548 * \retval +1: repaired
2549 * \retval 0: did nothing
2550 * \retval -ve: on error
2552 static int lfsck_layout_conflict_create(const struct lu_env *env,
2553 struct lfsck_component *com,
2554 struct lfsck_tgt_desc *ltd,
2555 struct lu_orphan_rec_v2 *rec,
2556 struct dt_object *parent,
2557 struct lu_fid *cfid,
2558 struct lu_buf *ea_buf,
2559 struct lov_mds_md_v1 *lmm,
2560 struct lov_ost_data_v1 *slot,
2561 __u32 ea_off, int lovea_size)
2563 struct lfsck_thread_info *info = lfsck_env_info(env);
2564 struct lu_fid *cfid2 = &info->lti_fid2;
2565 struct ost_id *oi = &info->lti_oi;
2566 struct dt_device *dev = lfsck_obj2dev(parent);
2567 struct thandle *th = NULL;
2568 struct lustre_handle lh = { 0 };
2569 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2573 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2574 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2578 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2579 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2583 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2584 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2589 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2591 /* If the conflict OST-obejct is not created for fixing dangling
2592 * referenced MDT-object in former LFSCK check/repair, or it has
2593 * been modified by others, then we cannot destroy it. Re-create
2594 * a new MDT-object for the orphan OST-object. */
2595 if (rc == -ETXTBSY) {
2596 /* No need the layout lock on the original parent. */
2597 lfsck_ibits_unlock(&lh, LCK_EX);
2599 fid_zero(&rec->lor_rec.lor_fid);
2600 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2601 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2603 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2604 info->lti_tmpbuf, "C", ea_off);
2609 if (rc != 0 && rc != -ENOENT)
2612 th = dt_trans_create(env, dev);
2614 GOTO(unlock, rc = PTR_ERR(th));
2616 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2617 LU_XATTR_REPLACE, th);
2621 rc = dt_trans_start_local(env, dev, th);
2625 dt_write_lock(env, parent, 0);
2626 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2627 rc = lfsck_layout_refill_lovea(env, com->lc_lfsck, th, parent, cfid,
2628 ea_buf, lmm, slot, LU_XATTR_REPLACE,
2629 ltd->ltd_index, lovea_size);
2630 dt_write_unlock(env, parent);
2635 dt_trans_stop(env, dev, th);
2638 lfsck_ibits_unlock(&lh, LCK_EX);
2641 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2642 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2643 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2644 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2645 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2648 return rc >= 0 ? 1 : rc;
2652 * \retval +1: repaired
2653 * \retval 0: did nothing
2654 * \retval -ve: on error
2656 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2657 struct lfsck_component *com,
2658 struct lfsck_tgt_desc *ltd,
2659 struct lu_orphan_rec_v2 *rec,
2660 struct dt_object *parent,
2661 struct lu_fid *cfid,
2662 __u32 ost_idx, __u32 ea_off)
2664 struct lfsck_thread_info *info = lfsck_env_info(env);
2665 struct lu_buf *buf = &info->lti_big_buf;
2666 struct lu_fid *fid = &info->lti_fid2;
2667 struct ost_id *oi = &info->lti_oi;
2668 struct lfsck_instance *lfsck = com->lc_lfsck;
2669 struct dt_device *dt = lfsck_obj2dev(parent);
2670 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2671 struct ost_layout *ol = &rec->lor_layout;
2672 struct lov_comp_md_v1 *lcm = NULL;
2673 struct lov_comp_md_entry_v1 *lcme = NULL;
2674 struct thandle *handle = NULL;
2676 struct lov_mds_md_v1 *lmm;
2677 struct lov_ost_data_v1 *objs;
2678 struct lustre_handle lh = { 0 };
2686 bool locked = false;
2689 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2690 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2693 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2694 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2695 "stripe-index %u, comp_id %u, comp_start %llu, "
2696 "comp_end %llu: rc = %d\n",
2697 lfsck_lfsck2name(lfsck), PFID(cfid),
2698 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off,
2699 ol->ol_comp_id, ol->ol_comp_start,
2700 ol->ol_comp_end, rc);
2707 dt_write_unlock(env, parent);
2711 if (handle != NULL) {
2712 dt_trans_stop(env, dt, handle);
2717 GOTO(unlock_layout, rc);
2720 if (buf->lb_len < lovea_size) {
2721 lu_buf_realloc(buf, lovea_size);
2722 if (buf->lb_buf == NULL)
2723 GOTO(unlock_layout, rc = -ENOMEM);
2726 if (!(bk->lb_param & LPF_DRYRUN)) {
2727 handle = dt_trans_create(env, dt);
2729 GOTO(unlock_layout, rc = PTR_ERR(handle));
2731 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2736 rc = dt_trans_start_local(env, dt, handle);
2741 dt_write_lock(env, parent, 0);
2743 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2744 if (rc == -ERANGE) {
2745 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2748 } else if (rc == -ENODATA || rc == 0) {
2749 lovea_size = lfsck_lovea_size(ol, ea_off);
2750 /* If the declared is not big enough, re-try. */
2751 if (buf->lb_len < lovea_size) {
2755 fl = LU_XATTR_CREATE;
2756 } else if (rc < 0) {
2757 GOTO(unlock_parent, rc);
2758 } else if (unlikely(buf->lb_len == 0)) {
2761 fl = LU_XATTR_REPLACE;
2765 if (fl == LU_XATTR_CREATE) {
2766 if (bk->lb_param & LPF_DRYRUN)
2767 GOTO(unlock_parent, rc = 1);
2769 LASSERT(buf->lb_len >= lovea_size);
2771 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2772 cfid, buf, fl, ost_idx, ea_off);
2774 GOTO(unlock_parent, rc);
2778 rc1 = lfsck_layout_verify_header(parent, lmm);
2780 /* If the LOV EA crashed, the rebuild it. */
2781 if (rc1 == -EINVAL) {
2782 if (bk->lb_param & LPF_DRYRUN)
2783 GOTO(unlock_parent, rc = 1);
2785 LASSERT(buf->lb_len >= lovea_size);
2787 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2788 cfid, buf, fl, ost_idx, ea_off);
2790 GOTO(unlock_parent, rc);
2793 /* For other unknown magic/pattern, keep the current LOV EA. */
2795 GOTO(unlock_parent, rc = rc1);
2797 magic = le32_to_cpu(lmm->lmm_magic);
2798 if (magic == LOV_MAGIC_COMP_V1) {
2803 count = le16_to_cpu(lcm->lcm_entry_count);
2804 for (i = 0; i < count; i++) {
2805 lcme = &lcm->lcm_entries[i];
2806 start = le64_to_cpu(lcme->lcme_extent.e_start);
2807 end = le64_to_cpu(lcme->lcme_extent.e_end);
2809 if (end <= ol->ol_comp_start)
2812 if (start >= ol->ol_comp_end)
2815 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
2816 magic = le32_to_cpu(lmm->lmm_magic);
2817 flags = le32_to_cpu(lcme->lcme_flags);
2821 rc = lfsck_layout_add_comp_comp(env, lfsck, handle, ol, parent,
2822 cfid, buf, ost_idx, ea_off, i);
2824 GOTO(unlock_parent, rc);
2828 count = le16_to_cpu(lmm->lmm_stripe_count);
2830 GOTO(unlock_parent, rc = -EINVAL);
2833 /* Exceed the current end of MDT-object layout EA. Then extend it. */
2834 if (count <= ea_off) {
2835 if (bk->lb_param & LPF_DRYRUN)
2836 GOTO(unlock_parent, rc = 1);
2838 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2839 /* If the declared is not big enough, re-try. */
2840 if (buf->lb_len < lovea_size) {
2845 if (lcme && !(flags & LCME_FL_INIT))
2846 lcme->lcme_flags = cpu_to_le32(flags | LCME_FL_INIT);
2848 rc = lfsck_layout_extend_v1v3_lovea(env, lfsck, handle, ol,
2849 parent, cfid, buf, ost_idx, ea_off);
2851 GOTO(unlock_parent, rc);
2854 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2856 if (magic == LOV_MAGIC_V1) {
2857 objs = &lmm->lmm_objects[0];
2859 LASSERT(magic == LOV_MAGIC_V3);
2860 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2863 for (i = 0; i < count; i++, objs++) {
2864 /* The MDT-object was created via lfsck_layout_recover_create()
2865 * by others before, and we fill the dummy layout EA. */
2866 if ((lcme && !(flags & LCME_FL_INIT)) ||
2867 lovea_slot_is_dummy(objs)) {
2871 if (bk->lb_param & LPF_DRYRUN)
2872 GOTO(unlock_parent, rc = 1);
2874 lmm->lmm_layout_gen =
2875 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2879 if (le32_to_cpu(lmm->lmm_stripe_size) !=
2880 ol->ol_stripe_size ||
2881 le16_to_cpu(lmm->lmm_stripe_count) !=
2882 ol->ol_stripe_count ||
2883 le64_to_cpu(lcme->lcme_extent.e_start) !=
2884 ol->ol_comp_start ||
2885 le64_to_cpu(lcme->lcme_extent.e_end) !=
2887 CDEBUG(D_LFSCK, "%s: found invalid "
2888 "component for "DFID ": parent "DFID
2889 ", stripe-index %u, stripe_size %u, "
2890 "stripe_count %u, comp_id %u, "
2891 "comp_start %llu, comp_end %llu, "
2892 "cur_stripe_size %u, "
2893 "cur_stripe_count %u, "
2894 "cur_comp_start %llu, "
2895 "cur_comp_end %llu\n",
2896 lfsck_lfsck2name(lfsck), PFID(cfid),
2897 PFID(lfsck_dto2fid(parent)), ea_off,
2899 ol->ol_stripe_count, ol->ol_comp_id,
2900 ol->ol_comp_start, ol->ol_comp_end,
2901 le32_to_cpu(lmm->lmm_stripe_size),
2902 le16_to_cpu(lmm->lmm_stripe_count),
2903 le64_to_cpu(lcme->lcme_extent.e_start),
2904 le64_to_cpu(lcme->lcme_extent.e_end));
2906 GOTO(unlock_parent, rc = -EINVAL);
2909 le32_add_cpu(&lcm->lcm_layout_gen, 1);
2910 lovea_size = le32_to_cpu(lcm->lcm_size);
2911 if (!(flags & LCME_FL_INIT))
2912 lcme->lcme_flags = cpu_to_le32(flags |
2916 LASSERTF(buf->lb_len >= lovea_size,
2917 "buffer len %d is less than real size %d\n",
2918 (int)buf->lb_len, (int)lovea_size);
2920 rc = lfsck_layout_refill_lovea(env, lfsck, handle,
2921 parent, cfid, buf, lmm, objs,
2922 fl, ost_idx, lovea_size);
2924 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2925 "dummy layout slot for "DFID": parent "DFID
2926 ", OST-index %u, stripe-index %u: rc = %d\n",
2927 lfsck_lfsck2name(lfsck), PFID(cfid),
2928 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2930 GOTO(unlock_parent, rc);
2933 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2934 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2936 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2937 "invalid layout EA at the slot %d, index %u\n",
2938 lfsck_lfsck2name(lfsck),
2939 PFID(lfsck_dto2fid(parent)), i,
2940 le32_to_cpu(objs->l_ost_idx));
2942 GOTO(unlock_parent, rc);
2945 /* It should be rare case, the slot is there, but the LFSCK
2946 * does not handle it during the first-phase cycle scanning. */
2947 if (unlikely(lu_fid_eq(fid, cfid))) {
2949 GOTO(unlock_parent, rc = 0);
2951 /* Rare case that the OST-object index
2952 * does not match the parent MDT-object
2953 * layout EA. We trust the later one. */
2954 if (bk->lb_param & LPF_DRYRUN)
2955 GOTO(unlock_parent, rc = 1);
2957 dt_write_unlock(env, parent);
2959 dt_trans_stop(env, dt, handle);
2960 lfsck_ibits_unlock(&lh, LCK_EX);
2961 rc = lfsck_layout_update_pfid(env, com, parent,
2965 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2966 "updated OST-object's pfid for "DFID
2967 ": parent "DFID", OST-index %u, "
2968 "stripe-index %u: rc = %d\n",
2969 lfsck_lfsck2name(lfsck), PFID(cfid),
2970 PFID(lfsck_dto2fid(parent)),
2971 ltd->ltd_index, i, rc);
2978 /* The MDT-object exists, but related layout EA slot is occupied
2980 if (bk->lb_param & LPF_DRYRUN)
2981 GOTO(unlock_parent, rc = 1);
2983 dt_write_unlock(env, parent);
2985 dt_trans_stop(env, dt, handle);
2986 lfsck_ibits_unlock(&lh, LCK_EX);
2987 if (magic == LOV_MAGIC_V1)
2988 objs = &lmm->lmm_objects[ea_off];
2990 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2991 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2992 buf, lmm, objs, ea_off, lovea_size);
2998 dt_write_unlock(env, parent);
3002 dt_trans_stop(env, dt, handle);
3005 lfsck_ibits_unlock(&lh, LCK_EX);
3010 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
3011 struct lfsck_component *com,
3012 struct lfsck_tgt_desc *ltd,
3013 struct lu_orphan_rec_v2 *rec,
3014 struct lu_fid *cfid)
3016 struct lfsck_layout *lo = com->lc_file_ram;
3017 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
3018 struct dt_object *parent = NULL;
3019 __u32 ea_off = pfid->f_stripe_idx;
3023 if (!fid_is_sane(cfid))
3024 GOTO(out, rc = -EINVAL);
3027 if (fid_is_zero(pfid)) {
3028 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3033 if (!fid_is_sane(pfid))
3034 GOTO(out, rc = -EINVAL);
3036 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
3038 GOTO(out, rc = PTR_ERR(parent));
3040 if (unlikely(dt_object_remote(parent) != 0))
3041 GOTO(put, rc = -EXDEV);
3043 if (dt_object_exists(parent) == 0) {
3044 lfsck_object_put(env, parent);
3045 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3050 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
3051 GOTO(put, rc = -EISDIR);
3053 /* The orphan OST-object claims to be the parent's stripe, then
3054 * related dangling record in the trace file is meaningless. */
3055 rc = lfsck_layout_del_dangling_rec(env, com, pfid,
3056 rec->lor_layout.ol_comp_id, ea_off);
3057 if (rc && rc != -ENOENT)
3060 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
3061 ltd->ltd_index, ea_off);
3067 lfsck_object_put(env, parent);
3069 /* The layout EA is changed, need to be reloaded next time. */
3070 dt_object_put_nocache(env, parent);
3073 down_write(&com->lc_sem);
3074 com->lc_new_scanned++;
3075 com->lc_new_checked++;
3077 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
3079 } else if (rc < 0) {
3080 lo->ll_objs_failed_phase2++;
3082 up_write(&com->lc_sem);
3087 static int lfsck_layout_scan_orphan(const struct lu_env *env,
3088 struct lfsck_component *com,
3089 struct lfsck_tgt_desc *ltd)
3091 struct lfsck_assistant_data *lad = com->lc_data;
3092 struct lfsck_instance *lfsck = com->lc_lfsck;
3093 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3094 struct lfsck_thread_info *info = lfsck_env_info(env);
3095 struct ost_id *oi = &info->lti_oi;
3096 struct lu_fid *fid = &info->lti_fid;
3097 struct dt_object *obj;
3098 const struct dt_it_ops *iops;
3103 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
3104 "scanning for OST%04x\n",
3105 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3107 if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
3108 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
3109 "scanning for OST%04x\n",
3110 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3115 ostid_set_seq(oi, FID_SEQ_IDIF);
3116 ostid_set_id(oi, 0);
3117 rc = ostid_to_fid(fid, oi, ltd->ltd_index);
3121 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
3122 if (unlikely(IS_ERR(obj)))
3123 GOTO(log, rc = PTR_ERR(obj));
3125 rc = obj->do_ops->do_index_try(env, obj,
3126 &dt_lfsck_layout_orphan_features);
3130 iops = &obj->do_index_ops->dio_it;
3131 di = iops->init(env, obj, 0);
3133 GOTO(put, rc = PTR_ERR(di));
3135 rc = iops->load(env, di, 0);
3137 /* -ESRCH means that the orphan OST-objects rbtree has been
3138 * cleanup because of the OSS server restart or other errors. */
3139 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
3144 rc = iops->next(env, di);
3156 struct lu_orphan_rec_v2 *rec = &info->lti_rec;
3158 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3159 unlikely(!thread_is_running(&lfsck->li_thread)))
3162 key = iops->key(env, di);
3163 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
3164 /* Remote target OST may be runnning old LFSCK */
3165 memset(rec, 0, sizeof(*rec));
3166 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
3168 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
3169 &com->lc_fid_latest_scanned_phase2);
3170 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3173 lfsck_control_speed_by_self(com);
3175 rc = iops->next(env, di);
3176 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
3183 iops->fini(env, di);
3185 lfsck_object_put(env, obj);
3188 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
3189 "scanning for OST%04x: rc = %d\n",
3190 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
3192 return rc > 0 ? 0 : rc;
3195 static int lfsck_lmm2layout(struct lov_mds_md_v1 *lmm, struct ost_layout *ol,
3198 __u32 magic = le32_to_cpu(lmm->lmm_magic);
3202 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3) {
3203 ol->ol_stripe_size = lmm->lmm_stripe_size;
3204 ol->ol_stripe_count = lmm->lmm_stripe_count;
3205 ol->ol_comp_start = 0;
3206 ol->ol_comp_end = 0;
3208 } else if (magic == LOV_MAGIC_COMP_V1) {
3209 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
3210 struct lov_comp_md_entry_v1 *lcme;
3211 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3214 for (i = 0; i < count; i++) {
3215 lcme = &lcm->lcm_entries[i];
3216 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3217 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3224 /* The comp has been removed, do nothing. */
3228 lmm = (void *)lmm + le32_to_cpu(lcme->lcme_offset);
3229 ol->ol_stripe_size = lmm->lmm_stripe_size;
3230 ol->ol_stripe_count = lmm->lmm_stripe_count;
3231 ol->ol_comp_start = lcme->lcme_extent.e_start;
3232 ol->ol_comp_end = lcme->lcme_extent.e_end;
3233 ol->ol_comp_id = lcme->lcme_id;
3235 GOTO(out, rc = -EINVAL);
3245 * Repair the MDT-object with dangling LOV EA reference.
3247 * we need to repair the inconsistency according to the users' requirement:
3249 * 1) Keep the inconsistency there and report the inconsistency case,
3250 * then give the chance to the application to find related issues,
3251 * and the users can make the decision about how to handle it with
3252 * more human knownledge. (by default)
3254 * 2) Re-create the missing OST-object with the FID/owner information.
3256 * \param[in] env pointer to the thread context
3257 * \param[in] com the layout LFSCK component
3258 * \param[in] parent the MDT-object with dangling LOV EA reference
3259 * \param[in] child the OST-object to be created
3260 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3261 * \param[in] ea_off the offset of the OST-object in the LOV EA
3262 * \param[in] ost_idx the index of OST on which the OST-object resides
3264 * \retval +1 for repair successfully
3265 * \retval 0 for did nothing
3266 * \retval negative error number on failure
3268 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
3269 struct lfsck_component *com,
3270 struct dt_object *parent,
3271 struct dt_object *child,
3272 __u32 comp_id, __u32 ea_off,
3273 __u32 ost_idx, bool log)
3275 struct lfsck_thread_info *info = lfsck_env_info(env);
3276 struct filter_fid *ff = &info->lti_ff;
3277 struct ost_layout *ol = &ff->ff_layout;
3278 struct dt_object_format *dof = &info->lti_dof;
3279 struct lu_attr *la = &info->lti_la;
3280 struct lfsck_instance *lfsck = com->lc_lfsck;
3281 struct dt_device *dev = lfsck_obj2dev(child);
3282 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3283 const struct lu_fid *cfid = lfsck_dto2fid(child);
3284 struct lu_buf *tbuf = &info->lti_big_buf;
3285 struct thandle *handle;
3287 struct lustre_handle lh = { 0 };
3291 if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
3294 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3295 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3300 rc = dt_attr_get(env, parent, la);
3304 la->la_mode = S_IFREG | 0666;
3305 la->la_atime = la->la_mtime = la->la_ctime = 0;
3306 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
3307 LA_ATIME | LA_MTIME | LA_CTIME;
3308 memset(dof, 0, sizeof(*dof));
3309 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
3310 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
3311 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3312 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3313 * parent MDT-object's layout EA. */
3314 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
3316 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3320 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, comp_id);
3324 buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
3325 handle = dt_trans_create(env, dev);
3327 GOTO(unlock1, rc = PTR_ERR(handle));
3329 rc = dt_declare_create(env, child, la, NULL, dof, handle);
3333 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
3334 LU_XATTR_CREATE, handle);
3338 rc = dt_trans_start_local(env, dev, handle);
3342 dt_read_lock(env, parent, 0);
3343 if (unlikely(lfsck_is_dead_obj(parent)))
3344 GOTO(unlock2, rc = 0);
3346 if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
3347 struct ost_id *oi = &info->lti_oi;
3348 struct lu_fid *tfid = &info->lti_fid2;
3349 struct lu_buf *lovea = &info->lti_big_buf;
3350 struct lov_mds_md_v1 *lmm;
3351 struct lov_ost_data_v1 *objs;
3356 rc = lfsck_layout_get_lovea(env, parent, lovea);
3360 lmm = lovea->lb_buf;
3361 magic = le32_to_cpu(lmm->lmm_magic);
3362 if (magic == LOV_MAGIC_COMP_V1) {
3363 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3364 struct lov_comp_md_entry_v1 *lcme;
3365 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3368 for (i = 0; i < count; i++) {
3369 lcme = &lcm->lcm_entries[i];
3370 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3371 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3374 lmm = lovea->lb_buf +
3375 le32_to_cpu(lcme->lcme_offset);
3376 magic = le32_to_cpu(lmm->lmm_magic);
3381 /* Someone removed the component, do nothing. */
3382 GOTO(unlock2, rc = 0);
3386 count = le16_to_cpu(lmm->lmm_stripe_count);
3387 /* Someone changed the LOV EA, do nothing. */
3388 if (count <= ea_off)
3389 GOTO(unlock2, rc = 0);
3391 if (magic == LOV_MAGIC_V1) {
3392 objs = &lmm->lmm_objects[ea_off];
3394 LASSERT(magic == LOV_MAGIC_V3);
3396 objs = &((struct lov_mds_md_v3 *)lmm)->\
3397 lmm_objects[ea_off];
3400 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3401 idx2 = le32_to_cpu(objs->l_ost_idx);
3402 rc = ostid_to_fid(tfid, oi, idx2);
3403 /* Someone changed the LOV EA, do nothing. */
3404 if (rc != 0 || !lu_fid_eq(tfid, cfid))
3408 rc = dt_create(env, child, la, NULL, dof, handle);
3412 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
3418 dt_read_unlock(env, parent);
3421 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3424 lfsck_ibits_unlock(&lh, LCK_EX);
3428 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3429 "dangling reference for: parent "DFID", child "
3430 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: "
3432 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3433 comp_id, ea_off, ost_idx,
3434 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3435 "Create the lost OST-object as required" :
3436 "Keep the MDT-object there by default", rc);
3442 * Repair the MDT-object with dangling LOV EA reference.
3444 * Prepare parameters and call __lfsck_layout_repair_dangling()
3445 * to repair the dangling LOV EA reference.
3447 * \param[in] env pointer to the thread context
3448 * \param[in] com the layout LFSCK component
3449 * \param[in] pfid the MDT-object's FID
3450 * \param[in] cfid the FID for the OST-object to be created
3451 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3452 * \param[in] ea_off the offset of the OST-object in the LOV EA
3453 * \param[in] ost_idx the index of OST on which the OST-object resides
3455 * \retval +1 for repair successfully
3456 * \retval 0 for did nothing
3457 * \retval negative error number on failure
3459 static int lfsck_layout_repair_dangling(const struct lu_env *env,
3460 struct lfsck_component *com,
3461 const struct lu_fid *pfid,
3462 const struct lu_fid *cfid,
3463 __u32 comp_id, __u32 ea_off,
3466 struct lfsck_instance *lfsck = com->lc_lfsck;
3467 struct dt_object *parent = NULL;
3468 struct dt_object *child = NULL;
3469 struct lfsck_tgt_desc *ltd;
3473 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3475 GOTO(log, rc = PTR_ERR(parent));
3477 /* The MDT-object has been removed. */
3478 if (dt_object_exists(parent) == 0)
3481 ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
3482 if (unlikely(ltd == NULL))
3483 GOTO(log, rc = -ENODEV);
3485 child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
3487 GOTO(log, rc = PTR_ERR(child));
3489 /* The OST-object has been created. */
3490 if (unlikely(dt_object_exists(child) != 0))
3493 rc = __lfsck_layout_repair_dangling(env, com, parent, child,
3494 comp_id, ea_off, ost_idx, false);
3499 if (child != NULL && !IS_ERR(child))
3500 lfsck_object_put(env, child);
3502 if (parent != NULL && !IS_ERR(parent))
3503 lfsck_object_put(env, parent);
3506 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3507 "dangling reference for: parent "DFID", child "
3508 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: rc = %d\n",
3509 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3510 comp_id, ea_off, ost_idx,
3511 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3512 "Create the lost OST-object as required" :
3513 "Keep the MDT-object there by default", rc);
3518 /* If the OST-object does not recognize the MDT-object as its parent, and
3519 * there is no other MDT-object claims as its parent, then just trust the
3520 * given MDT-object as its parent. So update the OST-object filter_fid. */
3521 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
3522 struct lfsck_component *com,
3523 struct dt_object *parent,
3524 struct lfsck_layout_req *llr,
3527 struct lfsck_thread_info *info = lfsck_env_info(env);
3528 struct filter_fid *ff = &info->lti_ff;
3529 struct ost_layout *ol = &ff->ff_layout;
3530 struct dt_object *child = llr->llr_child;
3531 struct dt_device *dev = lfsck_obj2dev(child);
3532 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
3533 struct lu_buf *tbuf = &info->lti_big_buf;
3534 struct thandle *handle;
3536 struct lustre_handle lh = { 0 };
3540 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3541 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3546 ff->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3547 ff->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3548 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3549 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3550 * parent MDT-object's layout EA. */
3551 ff->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3553 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3557 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, llr->llr_comp_id);
3561 buf = lfsck_buf_get(env, ff, sizeof(*ff));
3563 handle = dt_trans_create(env, dev);
3565 GOTO(unlock1, rc = PTR_ERR(handle));
3567 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3571 rc = dt_attr_get(env, parent, la);
3575 la->la_valid = LA_UID | LA_GID;
3576 rc = dt_declare_attr_set(env, child, la, handle);
3580 rc = dt_trans_start_local(env, dev, handle);
3584 dt_write_lock(env, parent, 0);
3585 if (unlikely(lfsck_is_dead_obj(parent)))
3586 GOTO(unlock2, rc = 1);
3588 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3592 /* Get the latest parent's owner. */
3593 rc = dt_attr_get(env, parent, la);
3597 la->la_valid = LA_UID | LA_GID;
3598 rc = dt_attr_set(env, child, la, handle);
3603 dt_write_unlock(env, parent);
3606 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3609 lfsck_ibits_unlock(&lh, LCK_EX);
3613 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3614 "unmatched MDT-OST pair for: parent "DFID
3615 ", child "DFID", comp_id %u, OST-index %u, "
3616 "stripe-index %u, owner %u/%u: rc = %d\n",
3617 lfsck_lfsck2name(com->lc_lfsck),
3618 PFID(lfsck_dto2fid(parent)),
3619 PFID(lfsck_dto2fid(child)),
3620 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3621 la->la_uid, la->la_gid, rc);
3626 /* If there are more than one MDT-objects claim as the OST-object's parent,
3627 * and the OST-object only recognizes one of them, then we need to generate
3628 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3629 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3630 struct lfsck_component *com,
3631 struct dt_object *parent,
3632 struct lfsck_layout_req *llr,
3635 struct lfsck_thread_info *info = lfsck_env_info(env);
3636 struct dt_allocation_hint *hint = &info->lti_hint;
3637 struct dt_object_format *dof = &info->lti_dof;
3638 struct ost_id *oi = &info->lti_oi;
3639 struct lu_buf *buf = &info->lti_big_buf;
3640 struct lfsck_instance *lfsck = com->lc_lfsck;
3641 struct dt_device *dev;
3642 struct lu_device *d =
3643 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3644 struct lu_object *o;
3645 struct lu_object *n;
3646 struct dt_object *child = NULL;
3647 struct thandle *handle = NULL;
3648 struct lov_mds_md_v1 *lmm;
3649 struct lov_ost_data_v1 *objs;
3650 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3652 struct lustre_handle lh = { 0 };
3658 /* We use two separated transactions to repair the inconsistency.
3660 * 1) create the child (OST-object).
3661 * 2) update the parent LOV EA according to the child's FID.
3663 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3664 * handled as orphan when the layout LFSCK run next time.
3666 * If 1) failed, but 2) succeed, then such OST-object will be re-created
3667 * as dangling referened case when the layout LFSCK run next time. */
3669 /* The 1st transaction. */
3670 o = lu_object_anon(env, d, NULL);
3672 GOTO(log, rc = PTR_ERR(o));
3674 n = lu_object_locate(o->lo_header, d->ld_type);
3675 if (unlikely(n == NULL)) {
3676 lu_object_put_nocache(env, o);
3678 GOTO(log, rc = -EINVAL);
3681 child = container_of(n, struct dt_object, do_lu);
3682 memset(hint, 0, sizeof(*hint));
3683 rc = dt_attr_get(env, parent, la);
3687 la->la_valid = LA_UID | LA_GID;
3688 memset(dof, 0, sizeof(*dof));
3690 dev = lfsck_obj2dev(child);
3691 handle = dt_trans_create(env, dev);
3693 GOTO(log, rc = PTR_ERR(handle));
3695 rc = dt_declare_create(env, child, la, hint, dof, handle);
3699 rc = dt_trans_start_local(env, dev, handle);
3703 rc = dt_create(env, child, la, hint, dof, handle);
3704 dt_trans_stop(env, dev, handle);
3709 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3710 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3715 /* The 2nd transaction. */
3717 /* XXX: Generally, we should use bottom device (OSD) to update parent
3718 * LOV EA. But because the LOD-object still references the wrong
3719 * OSP-object that should be detached after the parent's LOV EA
3720 * refreshed. Unfortunately, there is no suitable API for that.
3721 * So we have to make the LOD to re-load the OSP-object(s) via
3722 * replacing the LOV EA against the LOD-object.
3724 * Once the DNE2 patches have been landed, we can replace the
3725 * LOD device with the OSD device. LU-6230. */
3727 dev = lfsck->li_next;
3728 parent = lfsck_object_locate(dev, parent);
3730 GOTO(log, rc = PTR_ERR(parent));
3732 handle = dt_trans_create(env, dev);
3734 GOTO(log, rc = PTR_ERR(handle));
3736 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3737 LU_XATTR_REPLACE, handle);
3741 rc = dt_trans_start_local(env, dev, handle);
3745 dt_write_lock(env, parent, 0);
3746 if (unlikely(lfsck_is_dead_obj(parent)))
3747 GOTO(unlock, rc = 0);
3749 rc = lfsck_layout_get_lovea(env, parent, buf);
3750 if (unlikely(!rc || rc == -ENODATA))
3751 GOTO(unlock, rc = 0);
3754 magic = le32_to_cpu(lmm->lmm_magic);
3755 if (magic == LOV_MAGIC_COMP_V1) {
3756 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3757 struct lov_comp_md_entry_v1 *lcme;
3758 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3761 LASSERT(llr->llr_comp_id != 0);
3763 for (i = 0; i < count; i++) {
3764 lcme = &lcm->lcm_entries[i];
3765 if (le32_to_cpu(lcme->lcme_id) == llr->llr_comp_id) {
3766 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3769 le32_add_cpu(&lcm->lcm_layout_gen, 1);
3771 le32_to_cpu(lcme->lcme_offset);
3772 magic = le32_to_cpu(lmm->lmm_magic);
3777 GOTO(unlock, rc = 0);
3781 if (magic == LOV_MAGIC_V1) {
3782 objs = &lmm->lmm_objects[llr->llr_lov_idx];
3784 LASSERT(magic == LOV_MAGIC_V3);
3786 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
3789 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3790 index = le32_to_cpu(objs->l_ost_idx);
3791 rc = ostid_to_fid(&tfid, oi, index);
3792 /* Someone changed layout during the LFSCK, no need to repair then. */
3793 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
3794 GOTO(unlock, rc = 0);
3796 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3797 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3798 ostid_cpu_to_le(oi, &objs->l_ost_oi);
3799 objs->l_ost_gen = cpu_to_le32(0);
3800 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3801 rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3802 LU_XATTR_REPLACE, handle);
3804 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3807 dt_write_unlock(env, parent);
3811 dt_trans_stop(env, dev, handle);
3814 lfsck_ibits_unlock(&lh, LCK_EX);
3816 lfsck_object_put(env, child);
3819 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3820 "multiple references for: parent "DFID", comp_id %u, "
3821 "OST-index %u, stripe-index %u, owner %u/%u: rc = %d\n",
3822 lfsck_lfsck2name(lfsck), PFID(pfid),
3823 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3824 la->la_uid, la->la_gid, rc);
3829 /* If the MDT-object and the OST-object have different owner information,
3830 * then trust the MDT-object, because the normal chown/chgrp handle order
3831 * is from MDT to OST, and it is possible that some chown/chgrp operation
3832 * is partly done. */
3833 static int lfsck_layout_repair_owner(const struct lu_env *env,
3834 struct lfsck_component *com,
3835 struct dt_object *parent,
3836 struct lfsck_layout_req *llr,
3837 struct lu_attr *pla,
3838 const struct lu_attr *cla)
3840 struct lfsck_thread_info *info = lfsck_env_info(env);
3841 struct lu_attr *tla = &info->lti_la2;
3842 struct dt_object *child = llr->llr_child;
3843 struct dt_device *dev = lfsck_obj2dev(child);
3844 struct thandle *handle;
3848 tla->la_uid = pla->la_uid;
3849 tla->la_gid = pla->la_gid;
3850 tla->la_valid = LA_UID | LA_GID;
3851 handle = dt_trans_create(env, dev);
3853 GOTO(log, rc = PTR_ERR(handle));
3855 rc = dt_declare_attr_set(env, child, tla, handle);
3859 rc = dt_trans_start_local(env, dev, handle);
3863 /* Use the dt_object lock to serialize with destroy and attr_set. */
3864 dt_read_lock(env, parent, 0);
3865 if (unlikely(lfsck_is_dead_obj(parent)))
3866 GOTO(unlock, rc = 1);
3868 /* Get the latest parent's owner. */
3869 rc = dt_attr_get(env, parent, pla);
3873 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3874 if (unlikely(tla->la_uid != pla->la_uid ||
3875 tla->la_gid != pla->la_gid))
3878 rc = dt_attr_set(env, child, tla, handle);
3883 dt_read_unlock(env, parent);
3886 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3890 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3891 "inconsistent file owner for: parent "DFID", child "DFID
3892 ", OST-index %u, stripe-index %u, old owner %u/%u, "
3893 "new owner %u/%u: rc = %d\n",
3894 lfsck_lfsck2name(com->lc_lfsck),
3895 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3896 llr->llr_ost_idx, llr->llr_lov_idx,
3897 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3902 /* Check whether the OST-object correctly back points to the
3903 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3904 static int lfsck_layout_check_parent(const struct lu_env *env,
3905 struct lfsck_component *com,
3906 struct lfsck_assistant_object *lso,
3907 struct filter_fid *ff,
3908 const struct lu_fid *cfid,
3909 const struct lu_attr *cla,
3910 struct lfsck_layout_req *llr)
3912 struct lfsck_thread_info *info = lfsck_env_info(env);
3913 struct lu_buf *buf = &info->lti_big_buf;
3914 struct lu_fid *pfid = &info->lti_fid;
3915 struct dt_object *tobj;
3916 struct lov_mds_md_v1 *lmm;
3917 struct lov_ost_data_v1 *objs;
3918 struct lustre_handle lh = { 0 };
3926 *pfid = ff->ff_parent;
3927 idx = pfid->f_stripe_idx;
3930 if (unlikely(!fid_is_sane(pfid)))
3931 RETURN(LLIT_UNMATCHED_PAIR);
3933 if (lu_fid_eq(pfid, &lso->lso_fid)) {
3934 if (likely(llr->llr_lov_idx == idx))
3937 RETURN(LLIT_UNMATCHED_PAIR);
3940 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3942 RETURN(PTR_ERR(tobj));
3944 if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3945 !S_ISREG(lfsck_object_type(tobj)))
3946 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3948 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3949 * remote one on another MDT. Then check whether the given OST-object
3950 * is in such layout. If yes, it is multiple referenced, otherwise it
3951 * is unmatched referenced case. */
3952 rc = lfsck_layout_get_lovea(env, tobj, buf);
3953 if (rc == 0 || rc == -ENOENT)
3954 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3960 magic = le32_to_cpu(lmm->lmm_magic);
3961 if (magic == LOV_MAGIC_COMP_V1) {
3962 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3963 struct lov_comp_md_entry_v1 *lcme;
3965 if (ff->ff_layout.ol_comp_id == 0)
3966 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3968 count = le16_to_cpu(lcm->lcm_entry_count);
3969 for (i = 0; i < count; i++) {
3970 lcme = &lcm->lcm_entries[i];
3971 if (le32_to_cpu(lcme->lcme_id) ==
3972 ff->ff_layout.ol_comp_id) {
3974 le32_to_cpu(lcme->lcme_offset);
3975 magic = le32_to_cpu(lmm->lmm_magic);
3976 if (!(le32_to_cpu(lcme->lcme_flags) &
3978 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3984 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3988 if (magic == LOV_MAGIC_V1) {
3989 objs = &lmm->lmm_objects[0];
3991 LASSERT(magic == LOV_MAGIC_V3);
3992 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3995 count = le16_to_cpu(lmm->lmm_stripe_count);
3996 for (i = 0; i < count; i++, objs++) {
3997 struct lu_fid *tfid = &info->lti_fid2;
3998 struct ost_id *oi = &info->lti_oi;
4001 if (lovea_slot_is_dummy(objs))
4004 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4005 idx2 = le32_to_cpu(objs->l_ost_idx);
4006 rc = ostid_to_fid(tfid, oi, idx2);
4008 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
4009 "invalid layout EA at the slot %d, index %u\n",
4010 lfsck_lfsck2name(com->lc_lfsck),
4011 PFID(pfid), i, idx2);
4013 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4016 if (lu_fid_eq(cfid, tfid)) {
4017 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
4018 MDS_INODELOCK_UPDATE |
4019 MDS_INODELOCK_LAYOUT |
4020 MDS_INODELOCK_XATTR,
4025 dt_read_lock(env, tobj, 0);
4027 /* For local MDT-object, re-check existence
4028 * after taken the lock. */
4029 if (!dt_object_remote(tobj)) {
4030 if (dt_object_exists(tobj) == 0 ||
4031 lfsck_is_dead_obj(tobj))
4032 rc = LLIT_UNMATCHED_PAIR;
4034 rc = LLIT_MULTIPLE_REFERENCED;
4039 /* For migration case, the new MDT-object and old
4040 * MDT-object may reference the same OST-object at
4041 * some migration internal time.
4043 * For remote MDT-object, the local MDT may not know
4044 * whether it has been removed or not. Try checking
4045 * for a non-existent xattr to check if this object
4046 * has been been removed or not. */
4047 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
4049 if (unlikely(rc == -ENOENT || rc >= 0))
4050 rc = LLIT_UNMATCHED_PAIR;
4051 else if (rc == -ENODATA)
4052 rc = LLIT_MULTIPLE_REFERENCED;
4058 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4061 if (lustre_handle_is_used(&lh)) {
4062 dt_read_unlock(env, tobj);
4063 lfsck_ibits_unlock(&lh, LCK_EX);
4067 lfsck_object_put(env, tobj);
4072 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
4073 struct lfsck_component *com,
4074 struct lfsck_assistant_req *lar)
4076 struct lfsck_layout_req *llr =
4077 container_of0(lar, struct lfsck_layout_req, llr_lar);
4078 struct lfsck_assistant_object *lso = lar->lar_parent;
4079 struct lfsck_layout *lo = com->lc_file_ram;
4080 struct lfsck_thread_info *info = lfsck_env_info(env);
4081 struct filter_fid *ff = &info->lti_ff;
4082 struct lu_buf buf = { .lb_buf = ff,
4083 .lb_len = sizeof(*ff) };
4084 struct dt_object *parent = NULL;
4085 struct dt_object *child = llr->llr_child;
4086 struct lu_attr *pla = &lso->lso_attr;
4087 struct lu_attr *cla = &info->lti_la;
4088 struct lfsck_instance *lfsck = com->lc_lfsck;
4089 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4090 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
4097 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ASSISTANT_DIRECT, cfs_fail_val);
4099 rc = dt_attr_get(env, child, cla);
4100 if (rc == -ENOENT) {
4101 parent = lfsck_assistant_object_load(env, lfsck, lso);
4102 if (IS_ERR(parent)) {
4103 rc = PTR_ERR(parent);
4105 RETURN(rc == -ENOENT ? 0 : rc);
4108 type = LLIT_DANGLING;
4115 lfsck_buf_init(&buf, ff, sizeof(*ff));
4116 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
4117 if (unlikely(rc > 0 && rc < sizeof(struct lu_fid))) {
4118 type = LLIT_UNMATCHED_PAIR;
4122 if (rc < 0 && rc != -ENODATA)
4125 if (rc == 0 || rc == -ENODATA)
4126 GOTO(check_owner, rc = 0);
4128 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
4129 rc = lfsck_layout_check_parent(env, com, lso, ff,
4130 lu_object_fid(&child->do_lu), cla, llr);
4140 /* Someone may has changed the owner after the parent attr pre-loaded.
4141 * It can be handled later inside the lfsck_layout_repair_owner(). */
4142 if (unlikely(cla->la_uid != pla->la_uid ||
4143 cla->la_gid != pla->la_gid)) {
4144 type = LLIT_INCONSISTENT_OWNER;
4149 if (type == LLIT_NONE)
4152 if (bk->lb_param & LPF_DRYRUN)
4155 if (parent == NULL) {
4156 parent = lfsck_assistant_object_load(env, lfsck, lso);
4157 if (IS_ERR(parent)) {
4158 rc = PTR_ERR(parent);
4169 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
4170 rc = lfsck_layout_ins_dangling_rec(env, com,
4171 lfsck_dto2fid(parent), lfsck_dto2fid(child),
4172 llr->llr_comp_id, llr->llr_lov_idx,
4175 rc = __lfsck_layout_repair_dangling(env, com, parent,
4182 case LLIT_UNMATCHED_PAIR:
4183 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
4186 case LLIT_MULTIPLE_REFERENCED:
4187 rc = lfsck_layout_repair_multiple_references(env, com, parent,
4190 case LLIT_INCONSISTENT_OWNER:
4191 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
4201 down_write(&com->lc_sem);
4203 struct lfsck_assistant_data *lad = com->lc_data;
4205 if (unlikely(lad->lad_exit)) {
4207 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
4208 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
4209 rc == -EHOSTUNREACH) {
4210 /* If cannot touch the target server,
4211 * mark the LFSCK as INCOMPLETE. */
4212 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
4213 "talk with OST %x: rc = %d\n",
4214 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
4215 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
4216 lo->ll_objs_skipped++;
4219 lfsck_layout_record_failure(env, lfsck, lo);
4221 } else if (rc > 0 && (type != LLIT_DANGLING ||
4222 !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
4223 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
4224 "unknown type = %d\n", type);
4226 lo->ll_objs_repaired[type - 1]++;
4227 if (bk->lb_param & LPF_DRYRUN &&
4228 unlikely(lo->ll_pos_first_inconsistent == 0))
4229 lo->ll_pos_first_inconsistent =
4230 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
4233 up_write(&com->lc_sem);
4235 if (parent != NULL && !IS_ERR(parent))
4236 lfsck_object_put(env, parent);
4242 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
4243 struct lfsck_component *com,
4244 struct dt_object *obj, bool first)
4246 struct lfsck_instance *lfsck = com->lc_lfsck;
4247 struct ptlrpc_thread *thread = &lfsck->li_thread;
4248 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4249 struct lfsck_layout *lo = com->lc_file_ram;
4250 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
4253 struct lfsck_layout_dangling_key *parent =
4254 &lfsck_env_info(env)->lti_lldk;
4255 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
4260 di = iops->init(env, obj, 0);
4262 RETURN(PTR_ERR(di));
4265 lldk_cpu_to_be(parent, &lo->ll_lldk_latest_scanned_phase2);
4267 memset(parent, 0, sizeof(*parent));
4268 rc = iops->get(env, di, (const struct dt_key *)parent);
4273 /* The start one either has been processed or does not exist,
4275 rc = iops->next(env, di);
4281 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
4282 unlikely(!thread_is_running(thread)))
4285 key = iops->key(env, di);
4294 lldk_be_to_cpu(parent,
4295 (const struct lfsck_layout_dangling_key *)key);
4296 if (!fid_is_sane(&parent->lldk_fid)) {
4301 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
4303 fid_be_to_cpu(cfid, cfid);
4304 ost_idx = cfid->f_ver;
4306 if (!fid_is_sane(cfid)) {
4311 rc = lfsck_layout_repair_dangling(env, com,
4312 &parent->lldk_fid, cfid,
4313 parent->lldk_comp_id,
4314 parent->lldk_ea_off, ost_idx);
4318 down_write(&com->lc_sem);
4319 com->lc_new_checked++;
4320 com->lc_new_scanned++;
4322 lo->ll_lldk_latest_scanned_phase2 = *parent;
4325 lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
4327 lo->ll_objs_failed_phase2++;
4328 up_write(&com->lc_sem);
4330 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4333 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
4334 cfs_time_current())) &&
4335 com->lc_new_checked != 0) {
4336 down_write(&com->lc_sem);
4337 lo->ll_run_time_phase2 +=
4338 cfs_duration_sec(cfs_time_current() +
4339 HALF_SEC - com->lc_time_last_checkpoint);
4340 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4341 lo->ll_objs_checked_phase2 += com->lc_new_checked;
4342 com->lc_new_checked = 0;
4343 lfsck_layout_store(env, com);
4344 up_write(&com->lc_sem);
4346 com->lc_time_last_checkpoint = cfs_time_current();
4347 com->lc_time_next_checkpoint =
4348 com->lc_time_last_checkpoint +
4349 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4352 lfsck_control_speed_by_self(com);
4353 if (unlikely(!thread_is_running(thread)))
4356 rc = iops->next(env, di);
4365 iops->fini(env, di);
4370 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
4371 struct lfsck_component *com)
4373 struct lfsck_assistant_data *lad = com->lc_data;
4374 struct lfsck_instance *lfsck = com->lc_lfsck;
4375 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4376 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4377 struct lfsck_tgt_desc *ltd;
4381 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
4382 lfsck_lfsck2name(lfsck));
4384 spin_lock(<ds->ltd_lock);
4385 while (!list_empty(&lad->lad_ost_phase2_list)) {
4386 ltd = list_entry(lad->lad_ost_phase2_list.next,
4387 struct lfsck_tgt_desc,
4388 ltd_layout_phase_list);
4389 list_del_init(<d->ltd_layout_phase_list);
4390 if (bk->lb_param & LPF_OST_ORPHAN) {
4391 spin_unlock(<ds->ltd_lock);
4392 rc = lfsck_layout_scan_orphan(env, com, ltd);
4393 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
4396 if (unlikely(lad->lad_exit ||
4397 !thread_is_running(&lfsck->li_thread)))
4399 spin_lock(<ds->ltd_lock);
4403 if (list_empty(&lad->lad_ost_phase1_list))
4407 spin_unlock(<ds->ltd_lock);
4409 if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
4410 struct lfsck_layout *lo = com->lc_file_ram;
4413 com->lc_new_checked = 0;
4414 com->lc_new_scanned = 0;
4415 com->lc_time_last_checkpoint = cfs_time_current();
4416 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4417 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4419 i = lfsck_sub_trace_file_fid2idx(
4420 &lo->ll_lldk_latest_scanned_phase2.lldk_fid);
4421 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4422 com->lc_sub_trace_objs[i].lsto_obj, true);
4423 while (rc > 0 && ++i < LFSCK_STF_COUNT)
4424 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4425 com->lc_sub_trace_objs[i].lsto_obj, false);
4427 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
4428 "at the No. %d trace file: rc = %d\n",
4429 lfsck_lfsck2name(lfsck), i, rc);
4432 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
4433 lfsck_lfsck2name(lfsck), rc);
4439 lfsck_layout_slave_async_interpret(const struct lu_env *env,
4440 struct ptlrpc_request *req,
4443 struct lfsck_layout_slave_async_args *llsaa = args;
4444 struct obd_export *exp = llsaa->llsaa_exp;
4445 struct lfsck_component *com = llsaa->llsaa_com;
4446 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
4447 struct lfsck_layout_slave_data *llsd = com->lc_data;
4448 struct lfsck_reply *lr = NULL;
4452 /* It is probably caused by network trouble, or target crash,
4453 * it will try several times (depends on the obd_timeout, and
4454 * will not less than 3 times). But to make the LFSCK can go
4455 * ahead, we should not try for ever. After some try but still
4456 * hit failure, it will assume that the target exit the LFSCK
4457 * prcoessing and stop try. */
4458 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
4459 int max_try = max_t(int, obd_timeout / 30, 3);
4461 if (++(llst->llst_failures) > max_try)
4467 llst->llst_failures = 0;
4468 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
4469 if (lr->lr_status != LS_SCANNING_PHASE1 &&
4470 lr->lr_status != LS_SCANNING_PHASE2)
4475 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
4476 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
4477 llst->llst_index, lr != NULL ? lr->lr_status : rc,
4478 llst->llst_failures);
4480 lfsck_layout_llst_del(llsd, llst);
4483 lfsck_layout_llst_put(llst);
4484 lfsck_component_put(env, com);
4485 class_export_put(exp);
4490 static int lfsck_layout_async_query(const struct lu_env *env,
4491 struct lfsck_component *com,
4492 struct obd_export *exp,
4493 struct lfsck_layout_slave_target *llst,
4494 struct lfsck_request *lr,
4495 struct ptlrpc_request_set *set)
4497 struct lfsck_layout_slave_async_args *llsaa;
4498 struct ptlrpc_request *req;
4499 struct lfsck_request *tmp;
4503 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
4507 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
4509 ptlrpc_request_free(req);
4513 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4515 ptlrpc_request_set_replen(req);
4517 llsaa = ptlrpc_req_async_args(req);
4518 llsaa->llsaa_exp = exp;
4519 llsaa->llsaa_com = lfsck_component_get(com);
4520 llsaa->llsaa_llst = llst;
4521 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
4522 req->rq_allow_intr = 1;
4523 ptlrpc_set_add_req(set, req);
4528 static int lfsck_layout_async_notify(const struct lu_env *env,
4529 struct obd_export *exp,
4530 struct lfsck_request *lr,
4531 struct ptlrpc_request_set *set)
4533 struct ptlrpc_request *req;
4534 struct lfsck_request *tmp;
4538 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4542 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4544 ptlrpc_request_free(req);
4548 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4550 ptlrpc_request_set_replen(req);
4551 req->rq_allow_intr = 1;
4552 ptlrpc_set_add_req(set, req);
4558 lfsck_layout_slave_query_master(const struct lu_env *env,
4559 struct lfsck_component *com)
4561 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4562 struct lfsck_instance *lfsck = com->lc_lfsck;
4563 struct lfsck_layout_slave_data *llsd = com->lc_data;
4564 struct lfsck_layout_slave_target *llst;
4565 struct obd_export *exp;
4566 struct ptlrpc_request_set *set;
4571 set = ptlrpc_prep_set();
4573 GOTO(log, rc = -ENOMEM);
4575 memset(lr, 0, sizeof(*lr));
4576 lr->lr_event = LE_QUERY;
4577 lr->lr_active = LFSCK_TYPE_LAYOUT;
4579 llsd->llsd_touch_gen++;
4580 spin_lock(&llsd->llsd_lock);
4581 while (!list_empty(&llsd->llsd_master_list)) {
4582 llst = list_entry(llsd->llsd_master_list.next,
4583 struct lfsck_layout_slave_target,
4585 if (llst->llst_gen == llsd->llsd_touch_gen)
4588 llst->llst_gen = llsd->llsd_touch_gen;
4589 list_move_tail(&llst->llst_list,
4590 &llsd->llsd_master_list);
4591 atomic_inc(&llst->llst_ref);
4592 spin_unlock(&llsd->llsd_lock);
4594 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4597 lfsck_layout_llst_del(llsd, llst);
4598 lfsck_layout_llst_put(llst);
4599 spin_lock(&llsd->llsd_lock);
4603 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
4605 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4606 "query %s for layout: rc = %d\n",
4607 lfsck_lfsck2name(lfsck),
4608 exp->exp_obd->obd_name, rc);
4611 lfsck_layout_llst_put(llst);
4612 class_export_put(exp);
4614 spin_lock(&llsd->llsd_lock);
4616 spin_unlock(&llsd->llsd_lock);
4618 rc = ptlrpc_set_wait(set);
4619 ptlrpc_set_destroy(set);
4621 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4624 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4625 lfsck_lfsck2name(com->lc_lfsck), rc);
4631 lfsck_layout_slave_notify_master(const struct lu_env *env,
4632 struct lfsck_component *com,
4633 enum lfsck_events event, int result)
4635 struct lfsck_layout *lo = com->lc_file_ram;
4636 struct lfsck_instance *lfsck = com->lc_lfsck;
4637 struct lfsck_layout_slave_data *llsd = com->lc_data;
4638 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4639 struct lfsck_layout_slave_target *llst;
4640 struct obd_export *exp;
4641 struct ptlrpc_request_set *set;
4645 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4646 lfsck_lfsck2name(com->lc_lfsck));
4648 set = ptlrpc_prep_set();
4652 memset(lr, 0, sizeof(*lr));
4653 lr->lr_event = event;
4654 lr->lr_flags = LEF_FROM_OST;
4655 lr->lr_status = result;
4656 lr->lr_index = lfsck_dev_idx(lfsck);
4657 lr->lr_active = LFSCK_TYPE_LAYOUT;
4658 lr->lr_flags2 = lo->ll_flags;
4659 llsd->llsd_touch_gen++;
4660 spin_lock(&llsd->llsd_lock);
4661 while (!list_empty(&llsd->llsd_master_list)) {
4662 llst = list_entry(llsd->llsd_master_list.next,
4663 struct lfsck_layout_slave_target,
4665 if (llst->llst_gen == llsd->llsd_touch_gen)
4668 llst->llst_gen = llsd->llsd_touch_gen;
4669 list_move_tail(&llst->llst_list,
4670 &llsd->llsd_master_list);
4671 atomic_inc(&llst->llst_ref);
4672 spin_unlock(&llsd->llsd_lock);
4674 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4677 lfsck_layout_llst_del(llsd, llst);
4678 lfsck_layout_llst_put(llst);
4679 spin_lock(&llsd->llsd_lock);
4683 rc = lfsck_layout_async_notify(env, exp, lr, set);
4685 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4686 "notify %s for layout: rc = %d\n",
4687 lfsck_lfsck2name(lfsck),
4688 exp->exp_obd->obd_name, rc);
4690 lfsck_layout_llst_put(llst);
4691 class_export_put(exp);
4692 spin_lock(&llsd->llsd_lock);
4694 spin_unlock(&llsd->llsd_lock);
4696 ptlrpc_set_wait(set);
4697 ptlrpc_set_destroy(set);
4703 * \ret -ENODATA: unrecognized stripe
4704 * \ret = 0 : recognized stripe
4705 * \ret < 0 : other failures
4707 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4708 struct lfsck_component *com,
4709 struct lu_fid *cfid,
4710 struct lu_fid *pfid, __u32 comp_id)
4712 struct lfsck_thread_info *info = lfsck_env_info(env);
4713 struct lu_buf *buf = &info->lti_big_buf;
4714 struct ost_id *oi = &info->lti_oi;
4715 struct dt_object *obj;
4716 struct lov_mds_md_v1 *lmm;
4717 struct lov_ost_data_v1 *objs;
4718 __u32 idx = pfid->f_stripe_idx;
4726 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4728 RETURN(PTR_ERR(obj));
4730 dt_read_lock(env, obj, 0);
4731 if (unlikely(dt_object_exists(obj) == 0 ||
4732 lfsck_is_dead_obj(obj)))
4733 GOTO(unlock, rc = -ENOENT);
4735 if (!S_ISREG(lfsck_object_type(obj)))
4736 GOTO(unlock, rc = -ENODATA);
4738 rc = lfsck_layout_get_lovea(env, obj, buf);
4743 GOTO(unlock, rc = -ENODATA);
4746 magic = le32_to_cpu(lmm->lmm_magic);
4747 if (magic == LOV_MAGIC_COMP_V1) {
4748 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4749 struct lov_comp_md_entry_v1 *lcme;
4752 GOTO(unlock, rc = -ENODATA);
4754 count = le16_to_cpu(lcm->lcm_entry_count);
4755 for (i = 0; i < count; i++) {
4756 lcme = &lcm->lcm_entries[i];
4757 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
4759 le32_to_cpu(lcme->lcme_offset);
4760 magic = le32_to_cpu(lmm->lmm_magic);
4761 if (!(le32_to_cpu(lcme->lcme_flags) &
4763 GOTO(unlock, rc = -ENODATA);
4769 GOTO(unlock, rc = -ENODATA);
4773 if (magic == LOV_MAGIC_V1) {
4774 objs = &lmm->lmm_objects[0];
4776 LASSERT(magic == LOV_MAGIC_V3);
4777 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4780 fid_to_ostid(cfid, oi);
4781 count = le16_to_cpu(lmm->lmm_stripe_count);
4782 for (i = 0; i < count; i++, objs++) {
4785 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
4786 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
4787 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
4790 GOTO(unlock, rc = -ENODATA);
4793 dt_read_unlock(env, obj);
4794 lfsck_object_put(env, obj);
4800 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
4801 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
4802 * layout EA from MDT to OST. On one hand, the OST no need to understand
4803 * the layout EA structure; on the other hand, it may cause trouble when
4804 * transfer large layout EA from MDT to OST via normal OUT RPC.
4806 * \ret > 0: unrecognized stripe
4807 * \ret = 0: recognized stripe
4808 * \ret < 0: other failures
4810 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
4811 struct lfsck_component *com,
4812 struct lu_fid *cfid,
4813 struct lu_fid *pfid, __u32 comp_id)
4815 struct lfsck_instance *lfsck = com->lc_lfsck;
4816 struct obd_device *obd = lfsck->li_obd;
4817 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4818 struct obd_export *exp = NULL;
4819 struct ptlrpc_request *req = NULL;
4820 struct lfsck_request *lr;
4821 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
4825 if (unlikely(fid_is_idif(pfid)))
4828 fld_range_set_any(range);
4829 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
4831 RETURN(rc == -ENOENT ? 1 : rc);
4833 if (unlikely(!fld_range_is_mdt(range)))
4836 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
4837 if (unlikely(exp == NULL))
4840 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
4841 GOTO(out, rc = -EOPNOTSUPP);
4843 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4845 GOTO(out, rc = -ENOMEM);
4847 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4849 ptlrpc_request_free(req);
4854 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4855 memset(lr, 0, sizeof(*lr));
4856 lr->lr_event = LE_PAIRS_VERIFY;
4857 lr->lr_active = LFSCK_TYPE_LAYOUT;
4858 lr->lr_fid = *cfid; /* OST-object itself FID. */
4859 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
4860 lr->lr_comp_id = comp_id;
4862 ptlrpc_request_set_replen(req);
4863 rc = ptlrpc_queue_wait(req);
4864 ptlrpc_req_finished(req);
4866 if (rc == -ENOENT || rc == -ENODATA)
4873 class_export_put(exp);
4878 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
4879 struct lfsck_component *com,
4880 struct lfsck_req_local *lrl)
4882 struct dt_object *obj;
4886 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lrl->lrl_fid);
4888 GOTO(log, rc = PTR_ERR(obj));
4890 dt_write_lock(env, obj, 0);
4891 if (unlikely(dt_object_exists(obj) == 0 ||
4892 lfsck_is_dead_obj(obj)))
4893 GOTO(unlock, rc = 0);
4895 rc = __lfsck_layout_update_pfid(env, obj, &lrl->lrl_ff_client.ff_parent,
4896 &lrl->lrl_ff_client.ff_layout,
4897 lrl->lrl_ff_client.ff_parent.f_ver);
4902 dt_write_unlock(env, obj);
4903 lfsck_object_put(env, obj);
4906 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
4907 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
4908 PFID(&lrl->lrl_fid), PFID(&lrl->lrl_ff_client.ff_parent), rc);
4915 static void lfsck_layout_slave_quit(const struct lu_env *env,
4916 struct lfsck_component *com);
4918 static int lfsck_layout_reset(const struct lu_env *env,
4919 struct lfsck_component *com, bool init)
4921 struct lfsck_layout *lo = com->lc_file_ram;
4924 down_write(&com->lc_sem);
4926 memset(lo, 0, com->lc_file_size);
4928 __u32 count = lo->ll_success_count;
4929 __u64 last_time = lo->ll_time_last_complete;
4931 memset(lo, 0, com->lc_file_size);
4932 lo->ll_success_count = count;
4933 lo->ll_time_last_complete = last_time;
4936 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
4937 lo->ll_status = LS_INIT;
4939 if (com->lc_lfsck->li_master) {
4940 struct lfsck_assistant_data *lad = com->lc_data;
4942 lad->lad_incomplete = 0;
4943 CFS_RESET_BITMAP(lad->lad_bitmap);
4946 rc = lfsck_layout_store(env, com);
4947 if (rc == 0 && com->lc_lfsck->li_master)
4948 rc = lfsck_load_sub_trace_files(env, com,
4949 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
4950 up_write(&com->lc_sem);
4952 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
4953 lfsck_lfsck2name(com->lc_lfsck), rc);
4958 static void lfsck_layout_fail(const struct lu_env *env,
4959 struct lfsck_component *com, bool new_checked)
4961 struct lfsck_layout *lo = com->lc_file_ram;
4963 down_write(&com->lc_sem);
4965 com->lc_new_checked++;
4966 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
4967 up_write(&com->lc_sem);
4970 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
4971 struct lfsck_component *com, bool init)
4973 struct lfsck_instance *lfsck = com->lc_lfsck;
4974 struct lfsck_layout *lo = com->lc_file_ram;
4978 rc = lfsck_checkpoint_generic(env, com);
4980 return rc > 0 ? 0 : rc;
4983 down_write(&com->lc_sem);
4985 lo->ll_pos_latest_start =
4986 lfsck->li_pos_checkpoint.lp_oit_cookie;
4988 lo->ll_pos_last_checkpoint =
4989 lfsck->li_pos_checkpoint.lp_oit_cookie;
4990 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4991 HALF_SEC - lfsck->li_time_last_checkpoint);
4992 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4993 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4994 com->lc_new_checked = 0;
4997 rc = lfsck_layout_store(env, com);
4998 up_write(&com->lc_sem);
5000 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
5001 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5002 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5007 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
5008 struct lfsck_component *com, bool init)
5010 struct lfsck_instance *lfsck = com->lc_lfsck;
5011 struct lfsck_layout *lo = com->lc_file_ram;
5014 if (com->lc_new_checked == 0 && !init)
5017 down_write(&com->lc_sem);
5019 lo->ll_pos_latest_start =
5020 lfsck->li_pos_checkpoint.lp_oit_cookie;
5022 lo->ll_pos_last_checkpoint =
5023 lfsck->li_pos_checkpoint.lp_oit_cookie;
5024 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5025 HALF_SEC - lfsck->li_time_last_checkpoint);
5026 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5027 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5028 com->lc_new_checked = 0;
5031 rc = lfsck_layout_store(env, com);
5032 up_write(&com->lc_sem);
5034 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
5035 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5036 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5041 static int lfsck_layout_prep(const struct lu_env *env,
5042 struct lfsck_component *com,
5043 struct lfsck_start *start)
5045 struct lfsck_instance *lfsck = com->lc_lfsck;
5046 struct lfsck_layout *lo = com->lc_file_ram;
5047 struct lfsck_position *pos = &com->lc_pos_start;
5049 fid_zero(&pos->lp_dir_parent);
5050 pos->lp_dir_cookie = 0;
5051 if (lo->ll_status == LS_COMPLETED ||
5052 lo->ll_status == LS_PARTIAL ||
5053 /* To handle orphan, must scan from the beginning. */
5054 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
5057 rc = lfsck_layout_reset(env, com, false);
5059 rc = lfsck_set_param(env, lfsck, start, true);
5062 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
5063 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
5069 down_write(&com->lc_sem);
5070 lo->ll_time_latest_start = cfs_time_current_sec();
5071 spin_lock(&lfsck->li_lock);
5072 if (lo->ll_flags & LF_SCANNED_ONCE) {
5073 if (!lfsck->li_drop_dryrun ||
5074 lo->ll_pos_first_inconsistent == 0) {
5075 lo->ll_status = LS_SCANNING_PHASE2;
5076 list_move_tail(&com->lc_link,
5077 &lfsck->li_list_double_scan);
5078 pos->lp_oit_cookie = 0;
5082 lo->ll_status = LS_SCANNING_PHASE1;
5083 lo->ll_run_time_phase1 = 0;
5084 lo->ll_run_time_phase2 = 0;
5085 lo->ll_objs_checked_phase1 = 0;
5086 lo->ll_objs_checked_phase2 = 0;
5087 lo->ll_objs_failed_phase1 = 0;
5088 lo->ll_objs_failed_phase2 = 0;
5089 for (i = 0; i < LLIT_MAX; i++)
5090 lo->ll_objs_repaired[i] = 0;
5092 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5093 fid_zero(&com->lc_fid_latest_scanned_phase2);
5096 lo->ll_status = LS_SCANNING_PHASE1;
5097 if (!lfsck->li_drop_dryrun ||
5098 lo->ll_pos_first_inconsistent == 0)
5099 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
5101 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5103 spin_unlock(&lfsck->li_lock);
5104 up_write(&com->lc_sem);
5109 static int lfsck_layout_slave_prep(const struct lu_env *env,
5110 struct lfsck_component *com,
5111 struct lfsck_start_param *lsp)
5113 struct lfsck_layout_slave_data *llsd = com->lc_data;
5114 struct lfsck_instance *lfsck = com->lc_lfsck;
5115 struct lfsck_layout *lo = com->lc_file_ram;
5116 struct lfsck_start *start = lsp->lsp_start;
5119 rc = lfsck_layout_prep(env, com, start);
5123 if (lo->ll_flags & LF_CRASHED_LASTID &&
5124 list_empty(&llsd->llsd_master_list)) {
5125 LASSERT(lfsck->li_out_notify != NULL);
5127 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5128 LE_LASTID_REBUILDING);
5131 if (!lsp->lsp_index_valid)
5134 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5135 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
5136 LASSERT(!llsd->llsd_rbtree_valid);
5138 write_lock(&llsd->llsd_rb_lock);
5139 rc = lfsck_rbtree_setup(env, com);
5140 write_unlock(&llsd->llsd_rb_lock);
5143 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
5144 "%llu]\n", lfsck_lfsck2name(lfsck),
5145 com->lc_pos_start.lp_oit_cookie);
5150 static int lfsck_layout_master_prep(const struct lu_env *env,
5151 struct lfsck_component *com,
5152 struct lfsck_start_param *lsp)
5157 rc = lfsck_layout_load_bitmap(env, com);
5159 rc = lfsck_layout_reset(env, com, false);
5161 rc = lfsck_set_param(env, com->lc_lfsck,
5162 lsp->lsp_start, true);
5168 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
5172 rc = lfsck_start_assistant(env, com, lsp);
5177 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
5178 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
5179 com->lc_pos_start.lp_oit_cookie);
5184 /* Pre-fetch the attribute for each stripe in the given layout EA. */
5185 static int lfsck_layout_scan_stripes(const struct lu_env *env,
5186 struct lfsck_component *com,
5187 struct dt_object *parent,
5188 struct lov_mds_md_v1 *lmm, __u32 comp_id)
5190 struct lfsck_thread_info *info = lfsck_env_info(env);
5191 struct lfsck_instance *lfsck = com->lc_lfsck;
5192 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5193 struct lfsck_layout *lo = com->lc_file_ram;
5194 struct lfsck_assistant_data *lad = com->lc_data;
5195 struct lfsck_assistant_object *lso = NULL;
5196 struct lov_ost_data_v1 *objs;
5197 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
5198 struct ptlrpc_thread *mthread = &lfsck->li_thread;
5199 struct ptlrpc_thread *athread = &lad->lad_thread;
5200 struct l_wait_info lwi = { 0 };
5208 lfsck_buf_init(&buf, &info->lti_ff, sizeof(struct filter_fid));
5209 magic = le32_to_cpu(lmm->lmm_magic);
5210 if (magic == LOV_MAGIC_V1) {
5211 objs = &lmm->lmm_objects[0];
5213 LASSERT(magic == LOV_MAGIC_V3);
5214 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5217 count = le16_to_cpu(lmm->lmm_stripe_count);
5218 for (i = 0; i < count; i++, objs++) {
5219 struct lu_fid *fid = &info->lti_fid;
5220 struct ost_id *oi = &info->lti_oi;
5221 struct lfsck_layout_req *llr;
5222 struct lfsck_tgt_desc *tgt = NULL;
5223 struct dt_object *cobj = NULL;
5225 bool wakeup = false;
5227 if (unlikely(lovea_slot_is_dummy(objs)))
5230 l_wait_event(mthread->t_ctl_waitq,
5231 lad->lad_prefetched < bk->lb_async_windows ||
5232 !thread_is_running(mthread) ||
5233 thread_is_stopped(athread),
5236 if (unlikely(!thread_is_running(mthread)) ||
5237 thread_is_stopped(athread))
5240 if (unlikely(lfsck_is_dead_obj(parent)))
5243 ostid_le_to_cpu(&objs->l_ost_oi, oi);
5244 index = le32_to_cpu(objs->l_ost_idx);
5245 rc = ostid_to_fid(fid, oi, index);
5247 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
5248 ": "DOSTID", idx %u, comp_id %u\n",
5249 lfsck_lfsck2name(lfsck),
5250 PFID(lfsck_dto2fid(parent)), POSTID(oi),
5255 tgt = lfsck_tgt_get(ltds, index);
5256 if (unlikely(tgt == NULL)) {
5257 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
5258 "did not join the layout LFSCK, comp_id %u\n",
5259 lfsck_lfsck2name(lfsck), index, comp_id);
5260 lfsck_lad_set_bitmap(env, com, index);
5264 /* There is potential deadlock race condition between object
5265 * destroy and layout LFSCK. Consider the following scenario:
5267 * 1) The LFSCK thread obtained the parent object firstly, at
5268 * that time, the parent object has not been destroyed yet.
5270 * 2) One RPC service thread destroyed the parent and all its
5271 * children objects. Because the LFSCK is referencing the
5272 * parent object, then the parent object will be marked as
5273 * dying in RAM. On the other hand, the parent object is
5274 * referencing all its children objects, then all children
5275 * objects will be marked as dying in RAM also.
5277 * 3) The LFSCK thread tries to find some child object with
5278 * the parent object referenced. Then it will find that the
5279 * child object is dying. According to the object visibility
5280 * rules: the object with dying flag cannot be returned to
5281 * others. So the LFSCK thread has to wait until the dying
5282 * object has been purged from RAM, then it can allocate a
5283 * new object (with the same FID) in RAM. Unfortunately, the
5284 * LFSCK thread itself is referencing the parent object, and
5285 * cause the parent object cannot be purged, then cause the
5286 * child object cannot be purged also. So the LFSCK thread
5287 * will fall into deadlock.
5289 * We introduce non-blocked version lu_object_find() to allow
5290 * the LFSCK thread to return failure immediately (instead of
5291 * wait) when it finds dying (child) object, then the LFSCK
5292 * thread can check whether the parent object is dying or not.
5293 * So avoid above deadlock. LU-5395 */
5294 cobj = lfsck_object_find_by_dev_nowait(env, tgt->ltd_tgt, fid);
5296 if (lfsck_is_dead_obj(parent)) {
5306 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_ASSISTANT_DIRECT)) {
5307 rc = dt_declare_attr_get(env, cobj);
5311 rc = dt_declare_xattr_get(env, cobj, &buf,
5318 struct lu_attr *attr = &info->lti_la;
5320 rc = dt_attr_get(env, parent, attr);
5324 lso = lfsck_assistant_object_init(env,
5325 lfsck_dto2fid(parent), attr,
5326 lfsck->li_pos_current.lp_oit_cookie, false);
5335 llr = lfsck_layout_assistant_req_init(lso, cobj, comp_id,
5343 spin_lock(&lad->lad_lock);
5344 if (lad->lad_assistant_status < 0) {
5345 spin_unlock(&lad->lad_lock);
5346 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
5348 RETURN(lad->lad_assistant_status);
5351 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
5352 if (lad->lad_prefetched == 0)
5355 lad->lad_prefetched++;
5356 spin_unlock(&lad->lad_lock);
5358 wake_up_all(&athread->t_ctl_waitq);
5361 down_write(&com->lc_sem);
5362 com->lc_new_checked++;
5364 lfsck_layout_record_failure(env, lfsck, lo);
5365 up_write(&com->lc_sem);
5367 if (cobj != NULL && !IS_ERR(cobj))
5368 lfsck_object_put(env, cobj);
5370 if (likely(tgt != NULL))
5373 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5381 lfsck_assistant_object_put(env, lso);
5386 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
5387 * the OST-object's attribute and generate an structure lfsck_layout_req on the
5388 * list ::lad_req_list.
5390 * For each request on above list, the lfsck_layout_assistant thread compares
5391 * the OST side attribute with local attribute, if inconsistent, then repair it.
5393 * All above processing is async mode with pipeline. */
5394 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
5395 struct lfsck_component *com,
5396 struct dt_object *obj)
5398 struct lfsck_thread_info *info = lfsck_env_info(env);
5399 struct ost_id *oi = &info->lti_oi;
5400 struct lfsck_layout *lo = com->lc_file_ram;
5401 struct lfsck_assistant_data *lad = com->lc_data;
5402 struct lfsck_instance *lfsck = com->lc_lfsck;
5403 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5404 struct thandle *handle = NULL;
5405 struct lu_buf *buf = &info->lti_big_buf;
5406 struct lov_mds_md_v1 *lmm = NULL;
5407 struct dt_device *dev = lfsck_obj2dev(obj);
5408 struct lustre_handle lh = { 0 };
5409 struct lu_buf ea_buf = { NULL };
5410 struct lov_comp_md_v1 *lcm = NULL;
5411 struct lov_comp_md_entry_v1 *lcme = NULL;
5416 bool locked = false;
5417 bool stripe = false;
5418 bool bad_oi = false;
5421 if (!S_ISREG(lfsck_object_type(obj)))
5424 if (lad->lad_assistant_status < 0)
5425 GOTO(out, rc = -ESRCH);
5427 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
5428 lmm_oi_cpu_to_le(oi, oi);
5429 dt_read_lock(env, obj, 0);
5434 if (dt_object_exists(obj) == 0 ||
5435 lfsck_is_dead_obj(obj))
5438 rc = lfsck_layout_get_lovea(env, obj, buf);
5440 /* Skip bad lov EA during the 1st cycle scanning, and
5441 * try to recover it via orphan in the 2nd scanning. */
5442 GOTO(out, rc = (rc == -EINVAL ? 0 : rc));
5446 magic = le32_to_cpu(lmm->lmm_magic);
5447 if (magic == LOV_MAGIC_COMP_V1) {
5451 count = le16_to_cpu(lcm->lcm_entry_count);
5452 for (i = 0; i < count; i++) {
5453 lcme = &lcm->lcm_entries[i];
5454 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5455 if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) != 0)
5459 GOTO(out, stripe = true);
5460 } else if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0) {
5461 GOTO(out, stripe = true);
5465 /* Inconsistent lmm_oi, should be repaired. */
5468 if (bk->lb_param & LPF_DRYRUN) {
5469 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5471 GOTO(out, stripe = true);
5474 if (!lustre_handle_is_used(&lh)) {
5475 dt_read_unlock(env, obj);
5477 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5478 MDS_INODELOCK_LAYOUT |
5479 MDS_INODELOCK_XATTR, LCK_EX);
5483 handle = dt_trans_create(env, dev);
5485 GOTO(out, rc = PTR_ERR(handle));
5487 lfsck_buf_init(&ea_buf, lmm, size);
5488 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5489 LU_XATTR_REPLACE, handle);
5493 rc = dt_trans_start_local(env, dev, handle);
5497 dt_write_lock(env, obj, 0);
5503 if (magic == LOV_MAGIC_COMP_V1) {
5506 for (i = 0; i < count; i++) {
5507 lcme = &lcm->lcm_entries[i];
5508 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5515 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5516 LU_XATTR_REPLACE, handle);
5520 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5522 GOTO(out, stripe = true);
5526 if (lustre_handle_is_used(&lh))
5527 dt_write_unlock(env, obj);
5529 dt_read_unlock(env, obj);
5532 if (handle != NULL && !IS_ERR(handle))
5533 dt_trans_stop(env, dev, handle);
5535 lfsck_ibits_unlock(&lh, LCK_EX);
5538 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
5539 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
5540 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
5541 PFID(lfsck_dto2fid(obj)), rc);
5544 if (magic == LOV_MAGIC_COMP_V1) {
5547 for (i = 0; i < count; i++) {
5548 lcme = &lcm->lcm_entries[i];
5549 if (!(le32_to_cpu(lcme->lcme_flags) &
5553 rc = lfsck_layout_scan_stripes(env, com, obj,
5554 (struct lov_mds_md_v1 *)(buf->lb_buf +
5555 le32_to_cpu(lcme->lcme_offset)),
5556 le32_to_cpu(lcme->lcme_id));
5559 rc = lfsck_layout_scan_stripes(env, com, obj, lmm, 0);
5562 down_write(&com->lc_sem);
5563 com->lc_new_checked++;
5565 lfsck_layout_record_failure(env, lfsck, lo);
5566 up_write(&com->lc_sem);
5572 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
5573 struct lfsck_component *com,
5574 struct dt_object *obj)
5576 struct lfsck_instance *lfsck = com->lc_lfsck;
5577 struct lfsck_layout *lo = com->lc_file_ram;
5578 const struct lu_fid *fid = lfsck_dto2fid(obj);
5579 struct lfsck_layout_slave_data *llsd = com->lc_data;
5580 struct lfsck_layout_seq *lls;
5586 LASSERT(llsd != NULL);
5588 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
5589 cfs_fail_val == lfsck_dev_idx(lfsck)) {
5590 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(1),
5592 struct ptlrpc_thread *thread = &lfsck->li_thread;
5594 l_wait_event(thread->t_ctl_waitq,
5595 !thread_is_running(thread),
5599 lfsck_rbtree_update_bitmap(env, com, fid, false);
5601 down_write(&com->lc_sem);
5602 if (fid_is_idif(fid))
5604 else if (!fid_is_norm(fid) ||
5605 !fid_is_for_ostobj(env, lfsck, obj, fid))
5606 GOTO(unlock, rc = 0);
5609 com->lc_new_checked++;
5611 lls = lfsck_layout_seq_lookup(llsd, seq);
5614 if (unlikely(lls == NULL))
5615 GOTO(unlock, rc = -ENOMEM);
5617 INIT_LIST_HEAD(&lls->lls_list);
5619 rc = lfsck_layout_lastid_load(env, com, lls);
5621 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5622 "load LAST_ID for %#llx: rc = %d\n",
5623 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
5624 lo->ll_objs_failed_phase1++;
5629 lfsck_layout_seq_insert(llsd, lls);
5632 if (unlikely(fid_is_last_id(fid)))
5633 GOTO(unlock, rc = 0);
5635 if (fid_is_idif(fid))
5636 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
5640 if (oid > lls->lls_lastid_known)
5641 lls->lls_lastid_known = oid;
5643 if (oid > lls->lls_lastid) {
5644 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
5645 /* OFD may create new objects during LFSCK scanning. */
5646 rc = lfsck_layout_lastid_reload(env, com, lls);
5647 if (unlikely(rc != 0)) {
5648 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5649 "reload LAST_ID for %#llx: rc = %d\n",
5650 lfsck_lfsck2name(com->lc_lfsck),
5656 if (oid <= lls->lls_lastid ||
5657 lo->ll_flags & LF_CRASHED_LASTID)
5658 GOTO(unlock, rc = 0);
5660 LASSERT(lfsck->li_out_notify != NULL);
5662 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5663 LE_LASTID_REBUILDING);
5664 lo->ll_flags |= LF_CRASHED_LASTID;
5666 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
5667 "LAST_ID file (2) for the sequence %#llx"
5668 ", old value %llu, known value %llu\n",
5669 lfsck_lfsck2name(lfsck), lls->lls_seq,
5670 lls->lls_lastid, oid);
5673 lls->lls_lastid = oid;
5677 GOTO(unlock, rc = 0);
5680 up_write(&com->lc_sem);
5685 static int lfsck_layout_exec_dir(const struct lu_env *env,
5686 struct lfsck_component *com,
5687 struct lfsck_assistant_object *lso,
5688 struct lu_dirent *ent, __u16 type)
5693 static int lfsck_layout_master_post(const struct lu_env *env,
5694 struct lfsck_component *com,
5695 int result, bool init)
5697 struct lfsck_instance *lfsck = com->lc_lfsck;
5698 struct lfsck_layout *lo = com->lc_file_ram;
5702 lfsck_post_generic(env, com, &result);
5704 down_write(&com->lc_sem);
5705 spin_lock(&lfsck->li_lock);
5707 lo->ll_pos_last_checkpoint =
5708 lfsck->li_pos_checkpoint.lp_oit_cookie;
5711 if (lo->ll_flags & LF_INCOMPLETE)
5712 lo->ll_status = LS_PARTIAL;
5714 lo->ll_status = LS_SCANNING_PHASE2;
5715 lo->ll_flags |= LF_SCANNED_ONCE;
5716 lo->ll_flags &= ~LF_UPGRADE;
5717 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5718 } else if (result == 0) {
5719 if (lfsck->li_status != 0)
5720 lo->ll_status = lfsck->li_status;
5722 lo->ll_status = LS_STOPPED;
5723 if (lo->ll_status != LS_PAUSED)
5724 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5726 lo->ll_status = LS_FAILED;
5727 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5729 spin_unlock(&lfsck->li_lock);
5732 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5733 HALF_SEC - lfsck->li_time_last_checkpoint);
5734 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5735 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5736 com->lc_new_checked = 0;
5739 rc = lfsck_layout_store(env, com);
5740 up_write(&com->lc_sem);
5742 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
5743 lfsck_lfsck2name(lfsck), rc);
5748 static int lfsck_layout_slave_post(const struct lu_env *env,
5749 struct lfsck_component *com,
5750 int result, bool init)
5752 struct lfsck_instance *lfsck = com->lc_lfsck;
5753 struct lfsck_layout *lo = com->lc_file_ram;
5757 down_write(&com->lc_sem);
5758 rc = lfsck_layout_lastid_store(env, com);
5762 LASSERT(lfsck->li_out_notify != NULL);
5764 spin_lock(&lfsck->li_lock);
5766 lo->ll_pos_last_checkpoint =
5767 lfsck->li_pos_checkpoint.lp_oit_cookie;
5770 lo->ll_status = LS_SCANNING_PHASE2;
5771 lo->ll_flags |= LF_SCANNED_ONCE;
5772 if (lo->ll_flags & LF_CRASHED_LASTID) {
5774 lo->ll_flags &= ~LF_CRASHED_LASTID;
5776 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
5777 "crashed LAST_ID files successfully\n",
5778 lfsck_lfsck2name(lfsck));
5780 lo->ll_flags &= ~LF_UPGRADE;
5781 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5782 } else if (result == 0) {
5783 if (lfsck->li_status != 0)
5784 lo->ll_status = lfsck->li_status;
5786 lo->ll_status = LS_STOPPED;
5787 if (lo->ll_status != LS_PAUSED)
5788 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5790 lo->ll_status = LS_FAILED;
5791 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5793 spin_unlock(&lfsck->li_lock);
5796 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5800 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5801 HALF_SEC - lfsck->li_time_last_checkpoint);
5802 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5803 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5804 com->lc_new_checked = 0;
5807 rc = lfsck_layout_store(env, com);
5808 up_write(&com->lc_sem);
5810 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
5812 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
5813 lfsck_lfsck2name(lfsck), rc);
5818 static void lfsck_layout_dump(const struct lu_env *env,
5819 struct lfsck_component *com, struct seq_file *m)
5821 struct lfsck_instance *lfsck = com->lc_lfsck;
5822 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5823 struct lfsck_layout *lo = com->lc_file_ram;
5825 down_read(&com->lc_sem);
5826 seq_printf(m, "name: lfsck_layout\n"
5832 lfsck_status2name(lo->ll_status));
5834 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
5836 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
5838 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
5840 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
5842 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
5844 seq_printf(m, "latest_start_position: %llu\n"
5845 "last_checkpoint_position: %llu\n"
5846 "first_failure_position: %llu\n",
5847 lo->ll_pos_latest_start,
5848 lo->ll_pos_last_checkpoint,
5849 lo->ll_pos_first_inconsistent);
5851 seq_printf(m, "success_count: %u\n"
5852 "repaired_dangling: %llu\n"
5853 "repaired_unmatched_pair: %llu\n"
5854 "repaired_multiple_referenced: %llu\n"
5855 "repaired_orphan: %llu\n"
5856 "repaired_inconsistent_owner: %llu\n"
5857 "repaired_others: %llu\n"
5859 "failed_phase1: %llu\n"
5860 "failed_phase2: %llu\n",
5861 lo->ll_success_count,
5862 lo->ll_objs_repaired[LLIT_DANGLING - 1],
5863 lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
5864 lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
5865 lo->ll_objs_repaired[LLIT_ORPHAN - 1],
5866 lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
5867 lo->ll_objs_repaired[LLIT_OTHERS - 1],
5868 lo->ll_objs_skipped,
5869 lo->ll_objs_failed_phase1,
5870 lo->ll_objs_failed_phase2);
5872 if (lo->ll_status == LS_SCANNING_PHASE1) {
5874 cfs_duration_t duration = cfs_time_current() -
5875 lfsck->li_time_last_checkpoint;
5876 __u64 checked = lo->ll_objs_checked_phase1 +
5877 com->lc_new_checked;
5878 __u64 speed = checked;
5879 __u64 new_checked = com->lc_new_checked *
5880 msecs_to_jiffies(MSEC_PER_SEC);
5881 __u32 rtime = lo->ll_run_time_phase1 +
5882 cfs_duration_sec(duration + HALF_SEC);
5885 do_div(new_checked, duration);
5887 do_div(speed, rtime);
5888 seq_printf(m, "checked_phase1: %llu\n"
5889 "checked_phase2: %llu\n"
5890 "run_time_phase1: %u seconds\n"
5891 "run_time_phase2: %u seconds\n"
5892 "average_speed_phase1: %llu items/sec\n"
5893 "average_speed_phase2: N/A\n"
5894 "real-time_speed_phase1: %llu items/sec\n"
5895 "real-time_speed_phase2: N/A\n",
5897 lo->ll_objs_checked_phase2,
5899 lo->ll_run_time_phase2,
5903 if (likely(lfsck->li_di_oit)) {
5904 const struct dt_it_ops *iops =
5905 &lfsck->li_obj_oit->do_index_ops->dio_it;
5907 /* The low layer otable-based iteration position may NOT
5908 * exactly match the layout-based directory traversal
5909 * cookie. Generally, it is not a serious issue. But the
5910 * caller should NOT make assumption on that. */
5911 pos = iops->store(env, lfsck->li_di_oit);
5912 if (!lfsck->li_current_oit_processed)
5915 pos = lo->ll_pos_last_checkpoint;
5918 seq_printf(m, "current_position: %llu\n", pos);
5919 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
5920 cfs_duration_t duration = cfs_time_current() -
5921 com->lc_time_last_checkpoint;
5922 __u64 checked = lo->ll_objs_checked_phase2 +
5923 com->lc_new_checked;
5924 __u64 speed1 = lo->ll_objs_checked_phase1;
5925 __u64 speed2 = checked;
5926 __u64 new_checked = com->lc_new_checked *
5927 msecs_to_jiffies(MSEC_PER_SEC);
5928 __u32 rtime = lo->ll_run_time_phase2 +
5929 cfs_duration_sec(duration + HALF_SEC);
5932 do_div(new_checked, duration);
5933 if (lo->ll_run_time_phase1 != 0)
5934 do_div(speed1, lo->ll_run_time_phase1);
5936 do_div(speed2, rtime);
5937 seq_printf(m, "checked_phase1: %llu\n"
5938 "checked_phase2: %llu\n"
5939 "run_time_phase1: %u seconds\n"
5940 "run_time_phase2: %u seconds\n"
5941 "average_speed_phase1: %llu items/sec\n"
5942 "average_speed_phase2: %llu items/sec\n"
5943 "real-time_speed_phase1: N/A\n"
5944 "real-time_speed_phase2: %llu items/sec\n"
5945 "current_position: "DFID"\n",
5946 lo->ll_objs_checked_phase1,
5948 lo->ll_run_time_phase1,
5953 PFID(&com->lc_fid_latest_scanned_phase2));
5955 __u64 speed1 = lo->ll_objs_checked_phase1;
5956 __u64 speed2 = lo->ll_objs_checked_phase2;
5958 if (lo->ll_run_time_phase1 != 0)
5959 do_div(speed1, lo->ll_run_time_phase1);
5960 if (lo->ll_run_time_phase2 != 0)
5961 do_div(speed2, lo->ll_run_time_phase2);
5962 seq_printf(m, "checked_phase1: %llu\n"
5963 "checked_phase2: %llu\n"
5964 "run_time_phase1: %u seconds\n"
5965 "run_time_phase2: %u seconds\n"
5966 "average_speed_phase1: %llu items/sec\n"
5967 "average_speed_phase2: %llu objs/sec\n"
5968 "real-time_speed_phase1: N/A\n"
5969 "real-time_speed_phase2: N/A\n"
5970 "current_position: N/A\n",
5971 lo->ll_objs_checked_phase1,
5972 lo->ll_objs_checked_phase2,
5973 lo->ll_run_time_phase1,
5974 lo->ll_run_time_phase2,
5979 up_read(&com->lc_sem);
5982 static int lfsck_layout_master_double_scan(const struct lu_env *env,
5983 struct lfsck_component *com)
5985 struct lfsck_layout *lo = com->lc_file_ram;
5986 struct lfsck_assistant_data *lad = com->lc_data;
5987 struct lfsck_instance *lfsck = com->lc_lfsck;
5988 struct lfsck_tgt_descs *ltds;
5989 struct lfsck_tgt_desc *ltd;
5990 struct lfsck_tgt_desc *next;
5993 rc = lfsck_double_scan_generic(env, com, lo->ll_status);
5995 if (thread_is_stopped(&lad->lad_thread)) {
5996 LASSERT(list_empty(&lad->lad_req_list));
5997 LASSERT(list_empty(&lad->lad_ost_phase1_list));
5998 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
6000 ltds = &lfsck->li_ost_descs;
6001 spin_lock(<ds->ltd_lock);
6002 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6003 ltd_layout_phase_list) {
6004 list_del_init(<d->ltd_layout_phase_list);
6006 spin_unlock(<ds->ltd_lock);
6008 ltds = &lfsck->li_mdt_descs;
6009 spin_lock(<ds->ltd_lock);
6010 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6011 ltd_layout_phase_list) {
6012 list_del_init(<d->ltd_layout_phase_list);
6014 spin_unlock(<ds->ltd_lock);
6020 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
6021 struct lfsck_component *com)
6023 struct lfsck_instance *lfsck = com->lc_lfsck;
6024 struct lfsck_layout_slave_data *llsd = com->lc_data;
6025 struct lfsck_layout *lo = com->lc_file_ram;
6026 struct ptlrpc_thread *thread = &lfsck->li_thread;
6030 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
6031 lfsck_lfsck2name(lfsck));
6033 atomic_inc(&lfsck->li_double_scan_count);
6035 if (lo->ll_flags & LF_INCOMPLETE)
6038 com->lc_new_checked = 0;
6039 com->lc_new_scanned = 0;
6040 com->lc_time_last_checkpoint = cfs_time_current();
6041 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6042 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6045 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
6048 rc = lfsck_layout_slave_query_master(env, com);
6049 if (list_empty(&llsd->llsd_master_list)) {
6050 if (unlikely(!thread_is_running(thread)))
6061 rc = l_wait_event(thread->t_ctl_waitq,
6062 !thread_is_running(thread) ||
6063 lo->ll_flags & LF_INCOMPLETE ||
6064 list_empty(&llsd->llsd_master_list),
6066 if (unlikely(!thread_is_running(thread)))
6069 if (lo->ll_flags & LF_INCOMPLETE)
6072 if (rc == -ETIMEDOUT)
6075 GOTO(done, rc = (rc < 0 ? rc : 1));
6079 rc = lfsck_layout_double_scan_result(env, com, rc);
6080 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
6081 (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
6082 lfsck_layout_slave_quit(env, com);
6083 if (atomic_dec_and_test(&lfsck->li_double_scan_count))
6084 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6086 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
6087 "status %d: rc = %d\n",
6088 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
6093 static void lfsck_layout_master_data_release(const struct lu_env *env,
6094 struct lfsck_component *com)
6096 struct lfsck_assistant_data *lad = com->lc_data;
6097 struct lfsck_instance *lfsck = com->lc_lfsck;
6098 struct lfsck_tgt_descs *ltds;
6099 struct lfsck_tgt_desc *ltd;
6100 struct lfsck_tgt_desc *next;
6102 LASSERT(lad != NULL);
6103 LASSERT(thread_is_init(&lad->lad_thread) ||
6104 thread_is_stopped(&lad->lad_thread));
6105 LASSERT(list_empty(&lad->lad_req_list));
6107 com->lc_data = NULL;
6109 ltds = &lfsck->li_ost_descs;
6110 spin_lock(<ds->ltd_lock);
6111 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6112 ltd_layout_phase_list) {
6113 list_del_init(<d->ltd_layout_phase_list);
6115 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6116 ltd_layout_phase_list) {
6117 list_del_init(<d->ltd_layout_phase_list);
6119 list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
6121 list_del_init(<d->ltd_layout_list);
6123 spin_unlock(<ds->ltd_lock);
6125 ltds = &lfsck->li_mdt_descs;
6126 spin_lock(<ds->ltd_lock);
6127 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6128 ltd_layout_phase_list) {
6129 list_del_init(<d->ltd_layout_phase_list);
6131 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6132 ltd_layout_phase_list) {
6133 list_del_init(<d->ltd_layout_phase_list);
6135 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
6137 list_del_init(<d->ltd_layout_list);
6139 spin_unlock(<ds->ltd_lock);
6141 if (likely(lad->lad_bitmap != NULL))
6142 CFS_FREE_BITMAP(lad->lad_bitmap);
6147 static void lfsck_layout_slave_data_release(const struct lu_env *env,
6148 struct lfsck_component *com)
6150 struct lfsck_layout_slave_data *llsd = com->lc_data;
6152 lfsck_layout_slave_quit(env, com);
6153 com->lc_data = NULL;
6157 static void lfsck_layout_master_quit(const struct lu_env *env,
6158 struct lfsck_component *com)
6160 struct lfsck_assistant_data *lad = com->lc_data;
6161 struct lfsck_instance *lfsck = com->lc_lfsck;
6162 struct lfsck_tgt_descs *ltds;
6163 struct lfsck_tgt_desc *ltd;
6164 struct lfsck_tgt_desc *next;
6166 LASSERT(lad != NULL);
6168 lfsck_quit_generic(env, com);
6170 LASSERT(thread_is_init(&lad->lad_thread) ||
6171 thread_is_stopped(&lad->lad_thread));
6172 LASSERT(list_empty(&lad->lad_req_list));
6174 ltds = &lfsck->li_ost_descs;
6175 spin_lock(<ds->ltd_lock);
6176 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6177 ltd_layout_phase_list) {
6178 list_del_init(<d->ltd_layout_phase_list);
6180 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6181 ltd_layout_phase_list) {
6182 list_del_init(<d->ltd_layout_phase_list);
6184 spin_unlock(<ds->ltd_lock);
6186 ltds = &lfsck->li_mdt_descs;
6187 spin_lock(<ds->ltd_lock);
6188 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6189 ltd_layout_phase_list) {
6190 list_del_init(<d->ltd_layout_phase_list);
6192 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6193 ltd_layout_phase_list) {
6194 list_del_init(<d->ltd_layout_phase_list);
6196 spin_unlock(<ds->ltd_lock);
6199 static void lfsck_layout_slave_quit(const struct lu_env *env,
6200 struct lfsck_component *com)
6202 struct lfsck_layout_slave_data *llsd = com->lc_data;
6203 struct lfsck_layout_seq *lls;
6204 struct lfsck_layout_seq *next;
6205 struct lfsck_layout_slave_target *llst;
6207 LASSERT(llsd != NULL);
6209 down_write(&com->lc_sem);
6210 list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
6212 list_del_init(&lls->lls_list);
6213 lfsck_object_put(env, lls->lls_lastid_obj);
6216 up_write(&com->lc_sem);
6218 spin_lock(&llsd->llsd_lock);
6219 while (!list_empty(&llsd->llsd_master_list)) {
6220 llst = list_entry(llsd->llsd_master_list.next,
6221 struct lfsck_layout_slave_target, llst_list);
6222 list_del_init(&llst->llst_list);
6223 spin_unlock(&llsd->llsd_lock);
6224 lfsck_layout_llst_put(llst);
6225 spin_lock(&llsd->llsd_lock);
6227 spin_unlock(&llsd->llsd_lock);
6229 lfsck_rbtree_cleanup(env, com);
6232 static int lfsck_layout_master_in_notify(const struct lu_env *env,
6233 struct lfsck_component *com,
6234 struct lfsck_request *lr)
6236 struct lfsck_instance *lfsck = com->lc_lfsck;
6237 struct lfsck_layout *lo = com->lc_file_ram;
6238 struct lfsck_assistant_data *lad = com->lc_data;
6239 struct lfsck_tgt_descs *ltds;
6240 struct lfsck_tgt_desc *ltd;
6244 if (lr->lr_event == LE_PAIRS_VERIFY) {
6247 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
6254 CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
6255 "from %s %x, status %d, flags %x, flags2 %x\n",
6256 lfsck_lfsck2name(lfsck), lr->lr_event,
6257 (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
6258 lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
6260 if (lr->lr_event != LE_PHASE1_DONE &&
6261 lr->lr_event != LE_PHASE2_DONE &&
6262 lr->lr_event != LE_PEER_EXIT)
6265 if (lr->lr_flags & LEF_FROM_OST)
6266 ltds = &lfsck->li_ost_descs;
6268 ltds = &lfsck->li_mdt_descs;
6269 spin_lock(<ds->ltd_lock);
6270 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
6272 spin_unlock(<ds->ltd_lock);
6277 list_del_init(<d->ltd_layout_phase_list);
6278 switch (lr->lr_event) {
6279 case LE_PHASE1_DONE:
6280 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
6281 if (lr->lr_flags2 & LF_INCOMPLETE) {
6282 if (lr->lr_flags & LEF_FROM_OST)
6283 lfsck_lad_set_bitmap(env, com,
6286 lo->ll_flags |= LF_INCOMPLETE;
6288 ltd->ltd_layout_done = 1;
6289 list_del_init(<d->ltd_layout_list);
6294 if (lr->lr_flags & LEF_FROM_OST) {
6295 if (list_empty(<d->ltd_layout_list))
6296 list_add_tail(<d->ltd_layout_list,
6297 &lad->lad_ost_list);
6298 list_add_tail(<d->ltd_layout_phase_list,
6299 &lad->lad_ost_phase2_list);
6301 if (list_empty(<d->ltd_layout_list))
6302 list_add_tail(<d->ltd_layout_list,
6303 &lad->lad_mdt_list);
6304 list_add_tail(<d->ltd_layout_phase_list,
6305 &lad->lad_mdt_phase2_list);
6308 case LE_PHASE2_DONE:
6309 ltd->ltd_layout_done = 1;
6310 if (!list_empty(<d->ltd_layout_list)) {
6311 list_del_init(<d->ltd_layout_list);
6312 if (lr->lr_flags2 & LF_INCOMPLETE) {
6313 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
6321 ltd->ltd_layout_done = 1;
6322 list_del_init(<d->ltd_layout_list);
6323 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
6324 !(lr->lr_flags & LEF_FROM_OST))
6325 lo->ll_flags |= LF_INCOMPLETE;
6330 spin_unlock(<ds->ltd_lock);
6332 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
6333 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6335 memset(stop, 0, sizeof(*stop));
6336 stop->ls_status = lr->lr_status;
6337 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6338 lfsck_stop(env, lfsck->li_bottom, stop);
6339 } else if (lfsck_phase2_next_ready(lad)) {
6340 wake_up_all(&lad->lad_thread.t_ctl_waitq);
6346 static int lfsck_layout_slave_in_notify_local(const struct lu_env *env,
6347 struct lfsck_component *com,
6348 struct lfsck_req_local *lrl,
6353 switch (lrl->lrl_event) {
6354 case LEL_FID_ACCESSED:
6355 lfsck_rbtree_update_bitmap(env, com, &lrl->lrl_fid, true);
6357 case LEL_PAIRS_VERIFY_LOCAL: {
6360 lrl->lrl_status = LPVS_INIT;
6361 /* Firstly, if the MDT-object which is claimed via OST-object
6362 * local stored PFID xattr recognizes the OST-object, then it
6363 * must be that the client given PFID is wrong. */
6364 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6365 &lrl->lrl_ff_local.ff_parent,
6366 lrl->lrl_ff_local.ff_layout.ol_comp_id);
6370 lrl->lrl_status = LPVS_INCONSISTENT;
6371 /* The OST-object local stored PFID xattr is stale. We need to
6372 * check whether the MDT-object that is claimed via the client
6373 * given PFID information recognizes the OST-object or not. If
6374 * matches, then need to update the OST-object's PFID xattr. */
6375 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6376 &lrl->lrl_ff_client.ff_parent,
6377 lrl->lrl_ff_client.ff_layout.ol_comp_id);
6379 * We are not sure whether the client given PFID information
6380 * is correct or not, do nothing to avoid improper fixing.
6383 * The client given PFID information is also invalid, we can
6384 * NOT fix the OST-object inconsistency.
6387 lrl->lrl_status = LPVS_INCONSISTENT_TOFIX;
6388 rc = lfsck_layout_slave_repair_pfid(env, com, lrl);
6397 RETURN(-EOPNOTSUPP);
6400 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
6401 struct lfsck_component *com,
6402 struct lfsck_request *lr)
6404 struct lfsck_instance *lfsck = com->lc_lfsck;
6405 struct lfsck_layout_slave_data *llsd = com->lc_data;
6406 struct lfsck_layout_slave_target *llst;
6410 switch (lr->lr_event) {
6411 case LE_CONDITIONAL_DESTROY:
6412 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
6414 case LE_PHASE1_DONE: {
6415 if (lr->lr_flags2 & LF_INCOMPLETE) {
6416 struct lfsck_layout *lo = com->lc_file_ram;
6418 lo->ll_flags |= LF_INCOMPLETE;
6419 llst = lfsck_layout_llst_find_and_del(llsd,
6423 lfsck_layout_llst_put(llst);
6424 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6430 case LE_PHASE2_DONE:
6432 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
6433 "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
6434 lr->lr_event, lr->lr_index, lr->lr_status);
6440 llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
6444 lfsck_layout_llst_put(llst);
6445 if (list_empty(&llsd->llsd_master_list))
6446 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6448 if (lr->lr_event == LE_PEER_EXIT &&
6449 (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
6450 (list_empty(&llsd->llsd_master_list) &&
6451 (lr->lr_status == LS_STOPPED ||
6452 lr->lr_status == LS_CO_STOPPED)))) {
6453 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6455 memset(stop, 0, sizeof(*stop));
6456 stop->ls_status = lr->lr_status;
6457 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6458 lfsck_stop(env, lfsck->li_bottom, stop);
6464 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
6468 for (i = 0; i < LLIT_MAX; i++)
6469 *count += lo->ll_objs_repaired[i];
6472 static int lfsck_layout_query_all(const struct lu_env *env,
6473 struct lfsck_component *com,
6474 __u32 *mdts_count, __u32 *osts_count,
6477 struct lfsck_layout *lo = com->lc_file_ram;
6478 struct lfsck_tgt_descs *ltds;
6479 struct lfsck_tgt_desc *ltd;
6484 rc = lfsck_query_all(env, com);
6488 ltds = &com->lc_lfsck->li_mdt_descs;
6489 down_read(<ds->ltd_rw_sem);
6490 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
6491 ltd = lfsck_ltd2tgt(ltds, idx);
6492 LASSERT(ltd != NULL);
6494 mdts_count[ltd->ltd_layout_status]++;
6495 *repaired += ltd->ltd_layout_repaired;
6497 up_read(<ds->ltd_rw_sem);
6499 ltds = &com->lc_lfsck->li_ost_descs;
6500 down_read(<ds->ltd_rw_sem);
6501 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
6502 ltd = lfsck_ltd2tgt(ltds, idx);
6503 LASSERT(ltd != NULL);
6505 osts_count[ltd->ltd_layout_status]++;
6506 *repaired += ltd->ltd_layout_repaired;
6508 up_read(<ds->ltd_rw_sem);
6510 down_read(&com->lc_sem);
6511 mdts_count[lo->ll_status]++;
6512 lfsck_layout_repaired(lo, repaired);
6513 up_read(&com->lc_sem);
6518 static int lfsck_layout_query(const struct lu_env *env,
6519 struct lfsck_component *com,
6520 struct lfsck_request *req,
6521 struct lfsck_reply *rep,
6522 struct lfsck_query *que, int idx)
6524 struct lfsck_layout *lo = com->lc_file_ram;
6528 LASSERT(com->lc_lfsck->li_master);
6530 rc = lfsck_layout_query_all(env, com,
6531 que->lu_mdts_count[idx],
6532 que->lu_osts_count[idx],
6533 &que->lu_repaired[idx]);
6535 down_read(&com->lc_sem);
6536 rep->lr_status = lo->ll_status;
6537 if (req->lr_flags & LEF_QUERY_ALL)
6538 lfsck_layout_repaired(lo, &rep->lr_repaired);
6539 up_read(&com->lc_sem);
6545 /* with lfsck::li_lock held */
6546 static int lfsck_layout_slave_join(const struct lu_env *env,
6547 struct lfsck_component *com,
6548 struct lfsck_start_param *lsp)
6550 struct lfsck_instance *lfsck = com->lc_lfsck;
6551 struct lfsck_layout_slave_data *llsd = com->lc_data;
6552 struct lfsck_layout_slave_target *llst;
6553 struct lfsck_start *start = lsp->lsp_start;
6557 if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
6560 if (!lsp->lsp_index_valid)
6563 /* If someone is running the LFSCK without orphan handling,
6564 * it will not maintain the object accessing rbtree. So we
6565 * cannot join it for orphan handling. */
6566 if (!llsd->llsd_rbtree_valid)
6569 spin_unlock(&lfsck->li_lock);
6570 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
6571 spin_lock(&lfsck->li_lock);
6572 if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
6573 spin_unlock(&lfsck->li_lock);
6574 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
6577 lfsck_layout_llst_put(llst);
6578 spin_lock(&lfsck->li_lock);
6585 static struct lfsck_operations lfsck_layout_master_ops = {
6586 .lfsck_reset = lfsck_layout_reset,
6587 .lfsck_fail = lfsck_layout_fail,
6588 .lfsck_checkpoint = lfsck_layout_master_checkpoint,
6589 .lfsck_prep = lfsck_layout_master_prep,
6590 .lfsck_exec_oit = lfsck_layout_master_exec_oit,
6591 .lfsck_exec_dir = lfsck_layout_exec_dir,
6592 .lfsck_post = lfsck_layout_master_post,
6593 .lfsck_dump = lfsck_layout_dump,
6594 .lfsck_double_scan = lfsck_layout_master_double_scan,
6595 .lfsck_data_release = lfsck_layout_master_data_release,
6596 .lfsck_quit = lfsck_layout_master_quit,
6597 .lfsck_in_notify = lfsck_layout_master_in_notify,
6598 .lfsck_query = lfsck_layout_query,
6601 static struct lfsck_operations lfsck_layout_slave_ops = {
6602 .lfsck_reset = lfsck_layout_reset,
6603 .lfsck_fail = lfsck_layout_fail,
6604 .lfsck_checkpoint = lfsck_layout_slave_checkpoint,
6605 .lfsck_prep = lfsck_layout_slave_prep,
6606 .lfsck_exec_oit = lfsck_layout_slave_exec_oit,
6607 .lfsck_exec_dir = lfsck_layout_exec_dir,
6608 .lfsck_post = lfsck_layout_slave_post,
6609 .lfsck_dump = lfsck_layout_dump,
6610 .lfsck_double_scan = lfsck_layout_slave_double_scan,
6611 .lfsck_data_release = lfsck_layout_slave_data_release,
6612 .lfsck_quit = lfsck_layout_slave_quit,
6613 .lfsck_in_notify_local = lfsck_layout_slave_in_notify_local,
6614 .lfsck_in_notify = lfsck_layout_slave_in_notify,
6615 .lfsck_query = lfsck_layout_query,
6616 .lfsck_join = lfsck_layout_slave_join,
6619 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
6620 struct lfsck_component *com,
6621 struct lfsck_position *pos)
6623 struct lfsck_assistant_data *lad = com->lc_data;
6624 struct lfsck_layout_req *llr;
6626 if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
6630 if (list_empty(&lad->lad_req_list))
6633 llr = list_entry(lad->lad_req_list.next,
6634 struct lfsck_layout_req,
6636 pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
6639 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
6640 .la_handler_p1 = lfsck_layout_assistant_handler_p1,
6641 .la_handler_p2 = lfsck_layout_assistant_handler_p2,
6642 .la_fill_pos = lfsck_layout_assistant_fill_pos,
6643 .la_double_scan_result = lfsck_layout_double_scan_result,
6644 .la_req_fini = lfsck_layout_assistant_req_fini,
6645 .la_sync_failures = lfsck_layout_assistant_sync_failures,
6648 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
6650 struct lfsck_component *com;
6651 struct lfsck_layout *lo;
6652 struct dt_object *root = NULL;
6653 struct dt_object *obj;
6662 INIT_LIST_HEAD(&com->lc_link);
6663 INIT_LIST_HEAD(&com->lc_link_dir);
6664 init_rwsem(&com->lc_sem);
6665 atomic_set(&com->lc_ref, 1);
6666 com->lc_lfsck = lfsck;
6667 com->lc_type = LFSCK_TYPE_LAYOUT;
6668 if (lfsck->li_master) {
6669 com->lc_ops = &lfsck_layout_master_ops;
6670 com->lc_data = lfsck_assistant_data_init(
6671 &lfsck_layout_assistant_ops,
6673 if (com->lc_data == NULL)
6674 GOTO(out, rc = -ENOMEM);
6676 for (i = 0; i < LFSCK_STF_COUNT; i++)
6677 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6679 struct lfsck_layout_slave_data *llsd;
6681 com->lc_ops = &lfsck_layout_slave_ops;
6682 OBD_ALLOC_PTR(llsd);
6684 GOTO(out, rc = -ENOMEM);
6686 INIT_LIST_HEAD(&llsd->llsd_seq_list);
6687 INIT_LIST_HEAD(&llsd->llsd_master_list);
6688 spin_lock_init(&llsd->llsd_lock);
6689 llsd->llsd_rb_root = RB_ROOT;
6690 rwlock_init(&llsd->llsd_rb_lock);
6691 com->lc_data = llsd;
6693 com->lc_file_size = sizeof(*lo);
6694 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6695 if (com->lc_file_ram == NULL)
6696 GOTO(out, rc = -ENOMEM);
6698 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6699 if (com->lc_file_disk == NULL)
6700 GOTO(out, rc = -ENOMEM);
6702 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6704 GOTO(out, rc = PTR_ERR(root));
6706 if (unlikely(!dt_try_as_dir(env, root)))
6707 GOTO(out, rc = -ENOTDIR);
6709 obj = local_file_find_or_create(env, lfsck->li_los, root,
6711 S_IFREG | S_IRUGO | S_IWUSR);
6713 GOTO(out, rc = PTR_ERR(obj));
6716 rc = lfsck_layout_load(env, com);
6718 rc = lfsck_layout_reset(env, com, true);
6719 else if (rc == -ENOENT)
6720 rc = lfsck_layout_init(env, com);
6721 else if (lfsck->li_master)
6722 rc = lfsck_load_sub_trace_files(env, com,
6723 &dt_lfsck_layout_dangling_features,
6724 LFSCK_LAYOUT, false);
6729 lo = com->lc_file_ram;
6730 switch (lo->ll_status) {
6736 spin_lock(&lfsck->li_lock);
6737 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6738 spin_unlock(&lfsck->li_lock);
6741 CERROR("%s: unknown lfsck_layout status %d\n",
6742 lfsck_lfsck2name(lfsck), lo->ll_status);
6744 case LS_SCANNING_PHASE1:
6745 case LS_SCANNING_PHASE2:
6746 /* No need to store the status to disk right now.
6747 * If the system crashed before the status stored,
6748 * it will be loaded back when next time. */
6749 lo->ll_status = LS_CRASHED;
6750 if (!lfsck->li_master)
6751 lo->ll_flags |= LF_INCOMPLETE;
6758 spin_lock(&lfsck->li_lock);
6759 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6760 spin_unlock(&lfsck->li_lock);
6764 if (lo->ll_flags & LF_CRASHED_LASTID) {
6765 LASSERT(lfsck->li_out_notify != NULL);
6767 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6768 LE_LASTID_REBUILDING);
6774 if (root != NULL && !IS_ERR(root))
6775 lfsck_object_put(env, root);
6778 lfsck_component_cleanup(env, com);
6779 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
6780 lfsck_lfsck2name(lfsck), rc);
6786 struct lfsck_orphan_it {
6787 struct lfsck_component *loi_com;
6788 struct lfsck_rbtree_node *loi_lrn;
6789 struct lfsck_layout_slave_target *loi_llst;
6790 struct lu_fid loi_key;
6791 struct lu_orphan_rec_v2 loi_rec;
6793 unsigned int loi_over:1;
6796 static int lfsck_fid_match_idx(const struct lu_env *env,
6797 struct lfsck_instance *lfsck,
6798 const struct lu_fid *fid, int idx)
6800 struct seq_server_site *ss;
6801 struct lu_server_fld *sf;
6802 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
6805 /* All abnormal cases will be returned to MDT0. */
6806 if (!fid_is_norm(fid)) {
6813 ss = lfsck_dev_site(lfsck);
6814 if (unlikely(ss == NULL))
6817 sf = ss->ss_server_fld;
6818 LASSERT(sf != NULL);
6820 fld_range_set_any(range);
6821 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
6825 if (!fld_range_is_mdt(range))
6828 if (range->lsr_index == idx)
6834 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
6835 struct dt_object *obj)
6837 struct dt_device *dev = lfsck_obj2dev(obj);
6838 struct thandle *handle;
6842 handle = dt_trans_create(env, dev);
6846 rc = dt_declare_ref_del(env, obj, handle);
6850 rc = dt_declare_destroy(env, obj, handle);
6854 rc = dt_trans_start_local(env, dev, handle);
6858 dt_write_lock(env, obj, 0);
6859 rc = dt_ref_del(env, obj, handle);
6861 rc = dt_destroy(env, obj, handle);
6862 dt_write_unlock(env, obj);
6867 dt_trans_stop(env, dev, handle);
6869 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
6870 PFID(lfsck_dto2fid(obj)), rc);
6875 static int lfsck_orphan_index_lookup(const struct lu_env *env,
6876 struct dt_object *dt,
6878 const struct dt_key *key)
6883 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
6884 struct dt_object *dt,
6885 const struct dt_rec *rec,
6886 const struct dt_key *key,
6887 struct thandle *handle)
6892 static int lfsck_orphan_index_insert(const struct lu_env *env,
6893 struct dt_object *dt,
6894 const struct dt_rec *rec,
6895 const struct dt_key *key,
6896 struct thandle *handle,
6902 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
6903 struct dt_object *dt,
6904 const struct dt_key *key,
6905 struct thandle *handle)
6910 static int lfsck_orphan_index_delete(const struct lu_env *env,
6911 struct dt_object *dt,
6912 const struct dt_key *key,
6913 struct thandle *handle)
6918 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
6919 struct dt_object *dt,
6922 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
6923 struct lfsck_instance *lfsck;
6924 struct lfsck_component *com = NULL;
6925 struct lfsck_layout_slave_data *llsd;
6926 struct lfsck_orphan_it *it = NULL;
6927 struct lfsck_layout *lo;
6931 lfsck = lfsck_instance_find(dev, true, false);
6932 if (unlikely(lfsck == NULL))
6933 RETURN(ERR_PTR(-ENXIO));
6935 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
6936 if (unlikely(com == NULL))
6937 GOTO(out, rc = -ENOENT);
6939 lo = com->lc_file_ram;
6940 if (lo->ll_flags & LF_INCOMPLETE)
6941 GOTO(out, rc = -ESRCH);
6943 llsd = com->lc_data;
6944 if (!llsd->llsd_rbtree_valid)
6945 GOTO(out, rc = -ESRCH);
6949 GOTO(out, rc = -ENOMEM);
6951 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
6952 if (it->loi_llst == NULL)
6953 GOTO(out, rc = -ENXIO);
6955 if (dev->dd_record_fid_accessed) {
6956 /* The first iteration against the rbtree, scan the whole rbtree
6957 * to remove the nodes which do NOT need to be handled. */
6958 write_lock(&llsd->llsd_rb_lock);
6959 if (dev->dd_record_fid_accessed) {
6960 struct rb_node *node;
6961 struct rb_node *next;
6962 struct lfsck_rbtree_node *lrn;
6964 /* No need to record the fid accessing anymore. */
6965 dev->dd_record_fid_accessed = 0;
6967 node = rb_first(&llsd->llsd_rb_root);
6968 while (node != NULL) {
6969 next = rb_next(node);
6970 lrn = rb_entry(node, struct lfsck_rbtree_node,
6972 if (atomic_read(&lrn->lrn_known_count) <=
6973 atomic_read(&lrn->lrn_accessed_count)) {
6974 rb_erase(node, &llsd->llsd_rb_root);
6975 lfsck_rbtree_free(lrn);
6980 write_unlock(&llsd->llsd_rb_lock);
6983 /* read lock the rbtree when init, and unlock when fini */
6984 read_lock(&llsd->llsd_rb_lock);
6992 lfsck_component_put(env, com);
6994 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
6995 lfsck_lfsck2name(lfsck), rc);
6997 lfsck_instance_put(env, lfsck);
7002 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
7005 return (struct dt_it *)it;
7008 static void lfsck_orphan_it_fini(const struct lu_env *env,
7011 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7012 struct lfsck_component *com = it->loi_com;
7013 struct lfsck_layout_slave_data *llsd;
7014 struct lfsck_layout_slave_target *llst;
7017 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
7018 lfsck_lfsck2name(com->lc_lfsck));
7020 llsd = com->lc_data;
7021 read_unlock(&llsd->llsd_rb_lock);
7022 llst = it->loi_llst;
7023 LASSERT(llst != NULL);
7025 /* Save the key and hash for iterate next. */
7026 llst->llst_fid = it->loi_key;
7027 llst->llst_hash = it->loi_hash;
7028 lfsck_layout_llst_put(llst);
7029 lfsck_component_put(env, com);
7035 * \retval +1: the iteration finished
7036 * \retval 0: on success, not finished
7037 * \retval -ve: on error
7039 static int lfsck_orphan_it_next(const struct lu_env *env,
7042 struct lfsck_thread_info *info = lfsck_env_info(env);
7043 struct filter_fid *ff = &info->lti_ff;
7044 struct lu_attr *la = &info->lti_la;
7045 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7046 struct lu_fid *key = &it->loi_key;
7047 struct lu_orphan_rec_v2 *rec = &it->loi_rec;
7048 struct ost_layout *ol = &rec->lor_layout;
7049 struct lfsck_component *com = it->loi_com;
7050 struct lfsck_instance *lfsck = com->lc_lfsck;
7051 struct lfsck_layout_slave_data *llsd = com->lc_data;
7052 struct dt_object *obj;
7053 struct lfsck_rbtree_node *lrn;
7057 __u32 idx = it->loi_llst->llst_index;
7067 lrn = lfsck_rbtree_search(llsd, key, &exact);
7075 key->f_seq = lrn->lrn_seq;
7076 key->f_oid = lrn->lrn_first_oid;
7081 if (unlikely(key->f_oid == 0)) {
7088 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
7094 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
7095 atomic_read(&lrn->lrn_accessed_count))) {
7096 struct rb_node *next = rb_next(&lrn->lrn_node);
7098 while (next != NULL) {
7099 lrn = rb_entry(next, struct lfsck_rbtree_node,
7101 if (atomic_read(&lrn->lrn_known_count) >
7102 atomic_read(&lrn->lrn_accessed_count))
7104 next = rb_next(next);
7113 key->f_seq = lrn->lrn_seq;
7114 key->f_oid = lrn->lrn_first_oid;
7118 pos = key->f_oid - lrn->lrn_first_oid;
7121 pos = find_next_bit(lrn->lrn_known_bitmap,
7122 LFSCK_RBTREE_BITMAP_WIDTH, pos);
7123 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
7124 key->f_oid = lrn->lrn_first_oid + pos;
7125 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
7133 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
7138 key->f_oid = lrn->lrn_first_oid + pos;
7139 obj = lfsck_object_find_bottom(env, lfsck, key);
7142 if (rc == -ENOENT) {
7149 dt_read_lock(env, obj, 0);
7150 if (dt_object_exists(obj) == 0 ||
7151 lfsck_is_dead_obj(obj)) {
7152 dt_read_unlock(env, obj);
7153 lfsck_object_put(env, obj);
7158 rc = dt_attr_get(env, obj, la);
7162 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, ff, sizeof(*ff)),
7164 if (rc == -ENODATA) {
7165 /* For the pre-created OST-object, update the bitmap to avoid
7166 * others LFSCK (second phase) iteration to touch it again. */
7167 if (la->la_ctime == 0) {
7168 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
7169 atomic_inc(&lrn->lrn_accessed_count);
7171 /* For the race between repairing dangling referenced
7172 * MDT-object and unlink the file, it may left orphan
7173 * OST-object there. Destroy it now! */
7174 if (unlikely(!(la->la_mode & S_ISUID))) {
7175 dt_read_unlock(env, obj);
7176 lfsck_layout_destroy_orphan(env, obj);
7177 lfsck_object_put(env, obj);
7181 } else if (idx == 0) {
7182 /* If the orphan OST-object has no parent information,
7183 * regard it as referenced by the MDT-object on MDT0. */
7184 fid_zero(&rec->lor_rec.lor_fid);
7185 rec->lor_rec.lor_uid = la->la_uid;
7186 rec->lor_rec.lor_gid = la->la_gid;
7187 memset(ol, 0, sizeof(*ol));
7192 dt_read_unlock(env, obj);
7193 lfsck_object_put(env, obj);
7198 if (rc < sizeof(struct lu_fid))
7199 GOTO(out, rc = (rc < 0 ? rc : -EINVAL));
7201 fid_le_to_cpu(&rec->lor_rec.lor_fid, &ff->ff_parent);
7202 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
7203 * MDT-object's FID::f_ver, instead it is the OST-object index in its
7204 * parent MDT-object's layout EA. */
7205 save = rec->lor_rec.lor_fid.f_stripe_idx;
7206 rec->lor_rec.lor_fid.f_ver = 0;
7207 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_rec.lor_fid, idx);
7208 /* If the orphan OST-object does not claim the MDT, then next.
7210 * If we do not know whether it matches or not, then return it
7211 * to the MDT for further check. */
7213 dt_read_unlock(env, obj);
7214 lfsck_object_put(env, obj);
7219 rec->lor_rec.lor_fid.f_stripe_idx = save;
7220 rec->lor_rec.lor_uid = la->la_uid;
7221 rec->lor_rec.lor_gid = la->la_gid;
7222 ost_layout_le_to_cpu(ol, &ff->ff_layout);
7224 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u, "
7225 "stripe size %u, stripe count %u, COMP id %u, COMP start %llu, "
7226 "COMP end %llu\n", lfsck_lfsck2name(com->lc_lfsck), PFID(key),
7227 PFID(&rec->lor_rec.lor_fid), rec->lor_rec.lor_uid,
7228 rec->lor_rec.lor_gid, ol->ol_stripe_size, ol->ol_stripe_count,
7229 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end);
7234 dt_read_unlock(env, obj);
7235 lfsck_object_put(env, obj);
7243 * \retval +1: locate to the exactly position
7244 * \retval 0: cannot locate to the exactly position,
7245 * call next() to move to a valid position.
7246 * \retval -ve: on error
7248 static int lfsck_orphan_it_get(const struct lu_env *env,
7250 const struct dt_key *key)
7252 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7255 it->loi_key = *(struct lu_fid *)key;
7256 rc = lfsck_orphan_it_next(env, di);
7266 static void lfsck_orphan_it_put(const struct lu_env *env,
7271 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
7272 const struct dt_it *di)
7274 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7276 return (struct dt_key *)&it->loi_key;
7279 static int lfsck_orphan_it_key_size(const struct lu_env *env,
7280 const struct dt_it *di)
7282 return sizeof(struct lu_fid);
7285 static int lfsck_orphan_it_rec(const struct lu_env *env,
7286 const struct dt_it *di,
7290 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7292 *(struct lu_orphan_rec_v2 *)rec = it->loi_rec;
7297 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
7298 const struct dt_it *di)
7300 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7302 return it->loi_hash;
7306 * \retval +1: locate to the exactly position
7307 * \retval 0: cannot locate to the exactly position,
7308 * call next() to move to a valid position.
7309 * \retval -ve: on error
7311 static int lfsck_orphan_it_load(const struct lu_env *env,
7312 const struct dt_it *di,
7315 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7316 struct lfsck_layout_slave_target *llst = it->loi_llst;
7319 LASSERT(llst != NULL);
7321 if (hash != llst->llst_hash) {
7322 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
7323 "iteration does not match the one when fini "
7324 "%llu, to be reset.\n",
7325 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
7327 fid_zero(&llst->llst_fid);
7328 llst->llst_hash = 0;
7331 it->loi_key = llst->llst_fid;
7332 it->loi_hash = llst->llst_hash;
7333 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
7343 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
7344 const struct dt_it *di,
7350 const struct dt_index_operations lfsck_orphan_index_ops = {
7351 .dio_lookup = lfsck_orphan_index_lookup,
7352 .dio_declare_insert = lfsck_orphan_index_declare_insert,
7353 .dio_insert = lfsck_orphan_index_insert,
7354 .dio_declare_delete = lfsck_orphan_index_declare_delete,
7355 .dio_delete = lfsck_orphan_index_delete,
7357 .init = lfsck_orphan_it_init,
7358 .fini = lfsck_orphan_it_fini,
7359 .get = lfsck_orphan_it_get,
7360 .put = lfsck_orphan_it_put,
7361 .next = lfsck_orphan_it_next,
7362 .key = lfsck_orphan_it_key,
7363 .key_size = lfsck_orphan_it_key_size,
7364 .rec = lfsck_orphan_it_rec,
7365 .store = lfsck_orphan_it_store,
7366 .load = lfsck_orphan_it_load,
7367 .key_rec = lfsck_orphan_it_key_rec,