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, 2017, 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 <lu_object.h>
40 #include <dt_object.h>
41 #include <lustre_fid.h>
42 #include <lustre_lib.h>
43 #include <lustre_net.h>
44 #include <md_object.h>
45 #include <obd_class.h>
47 #include "lfsck_internal.h"
49 #define LFSCK_LAYOUT_MAGIC_V1 0xB173AE14
50 #define LFSCK_LAYOUT_MAGIC_V2 0xB1734D76
51 #define LFSCK_LAYOUT_MAGIC_V3 0xB17371B9
52 #define LFSCK_LAYOUT_MAGIC_V4 0xB1732FED
54 #define LFSCK_LAYOUT_MAGIC LFSCK_LAYOUT_MAGIC_V4
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);
293 laia->laia_ltd = ltd;
294 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
295 lfsck_layout_assistant_sync_failures_interpret,
298 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
299 "notify target %x for %s phase1 done: "
300 "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
301 ltd->ltd_index, lad->lad_name, rc);
308 up_read(<ds->ltd_rw_sem);
310 if (rc == 0 && atomic_read(&count) > 0)
311 rc = ptlrpc_set_wait(set);
313 ptlrpc_set_destroy(set);
315 if (rc == 0 && atomic_read(&count) > 0)
322 /* If failed to sync failures with the OSTs, then have to
323 * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
324 * subsequent orphan OST-object handling. */
325 lo->ll_flags |= LF_INCOMPLETE;
327 lr->lr_flags2 = lo->ll_flags;
330 static int lfsck_layout_verify_header_v1v3(struct dt_object *obj,
331 struct lov_mds_md_v1 *lmm)
336 magic = le32_to_cpu(lmm->lmm_magic);
337 /* If magic crashed, keep it there. Sometime later, during OST-object
338 * orphan handling, if some OST-object(s) back-point to it, it can be
339 * verified and repaired. */
340 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
343 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
348 CDEBUG(D_LFSCK, "%s LOV EA magic %u for the file "DFID"\n",
349 rc == -EINVAL ? "Unknown" : "Unsupported",
350 magic, PFID(lfsck_dto2fid(obj)));
355 pattern = le32_to_cpu(lmm->lmm_pattern);
356 /* XXX: currently, we only support LOV_PATTERN_RAID0. */
357 if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
358 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u for the file "
359 DFID"\n", pattern, PFID(lfsck_dto2fid(obj)));
367 static int lfsck_layout_verify_header(struct dt_object *obj,
368 struct lov_mds_md_v1 *lmm)
372 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1) {
373 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
375 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
377 if (unlikely(count == 0)) {
378 CDEBUG(D_LFSCK, "the PFL file "DFID" contains invalid "
379 "components count 0\n",
380 PFID(lfsck_dto2fid(obj)));
385 for (i = 0; i < count; i++) {
386 struct lov_comp_md_entry_v1 *lcme =
387 &lcm->lcm_entries[i];
388 __u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
389 __u64 end = le64_to_cpu(lcme->lcme_extent.e_end);
390 __u32 comp_id = le32_to_cpu(lcme->lcme_id);
392 if (unlikely(comp_id == LCME_ID_INVAL ||
393 comp_id > LCME_ID_MAX)) {
394 CDEBUG(D_LFSCK, "found invalid FPL ID %u "
395 "for the file "DFID" at idx %d\n",
396 comp_id, PFID(lfsck_dto2fid(obj)), i);
401 if (unlikely(start >= end ||
402 !lfsck_comp_extent_aligned(start) ||
403 (!lfsck_comp_extent_aligned(end) &&
404 end != LUSTRE_EOF))) {
405 CDEBUG(D_LFSCK, "found invalid FPL extent "
406 "range [%llu - %llu) for the file "
408 start, end, PFID(lfsck_dto2fid(obj)), i);
413 rc = lfsck_layout_verify_header_v1v3(obj,
414 (struct lov_mds_md_v1 *)((char *)lmm +
415 le32_to_cpu(lcme->lcme_offset)));
420 rc = lfsck_layout_verify_header_v1v3(obj, lmm);
426 static int lfsck_layout_get_lovea(const struct lu_env *env,
427 struct dt_object *obj, struct lu_buf *buf)
433 rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
435 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
439 lu_buf_realloc(buf, rc);
440 if (buf->lb_buf == NULL)
452 if (unlikely(buf->lb_buf == NULL)) {
453 lu_buf_alloc(buf, rc);
454 if (buf->lb_buf == NULL)
460 rc1 = lfsck_layout_verify_header(obj, buf->lb_buf);
462 return rc1 ? rc1 : rc;
465 #define LFSCK_RBTREE_BITMAP_SIZE PAGE_SIZE
466 #define LFSCK_RBTREE_BITMAP_WIDTH (LFSCK_RBTREE_BITMAP_SIZE << 3)
467 #define LFSCK_RBTREE_BITMAP_MASK (LFSCK_RBTREE_BITMAP_WIDTH - 1)
469 struct lfsck_rbtree_node {
470 struct rb_node lrn_node;
473 atomic_t lrn_known_count;
474 atomic_t lrn_accessed_count;
475 void *lrn_known_bitmap;
476 void *lrn_accessed_bitmap;
479 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
480 __u64 seq, __u32 oid)
482 if (seq < lrn->lrn_seq)
485 if (seq > lrn->lrn_seq)
488 if (oid < lrn->lrn_first_oid)
491 if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
497 /* The caller should hold llsd->llsd_rb_lock. */
498 static struct lfsck_rbtree_node *
499 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
500 const struct lu_fid *fid, bool *exact)
502 struct rb_node *node = llsd->llsd_rb_root.rb_node;
503 struct rb_node *prev = NULL;
504 struct lfsck_rbtree_node *lrn = NULL;
510 while (node != NULL) {
512 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
513 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
515 node = node->rb_left;
517 node = node->rb_right;
525 /* If there is no exactly matched one, then to the next valid one. */
528 /* The rbtree is empty. */
535 node = rb_next(prev);
537 /* The end of the rbtree. */
541 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
546 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
547 const struct lu_fid *fid)
549 struct lfsck_rbtree_node *lrn;
553 return ERR_PTR(-ENOMEM);
555 OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
556 if (lrn->lrn_known_bitmap == NULL) {
559 return ERR_PTR(-ENOMEM);
562 OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
563 if (lrn->lrn_accessed_bitmap == NULL) {
564 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
567 return ERR_PTR(-ENOMEM);
570 RB_CLEAR_NODE(&lrn->lrn_node);
571 lrn->lrn_seq = fid_seq(fid);
572 lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
573 atomic_set(&lrn->lrn_known_count, 0);
574 atomic_set(&lrn->lrn_accessed_count, 0);
579 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
581 OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
582 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
586 /* The caller should hold lock. */
587 static struct lfsck_rbtree_node *
588 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
589 struct lfsck_rbtree_node *lrn)
591 struct rb_node **pos = &llsd->llsd_rb_root.rb_node;
592 struct rb_node *parent = NULL;
593 struct lfsck_rbtree_node *tmp;
596 while (*pos != NULL) {
598 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
599 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
601 pos = &(*pos)->rb_left;
603 pos = &(*pos)->rb_right;
608 rb_link_node(&lrn->lrn_node, parent, pos);
609 rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
614 extern const struct dt_index_operations lfsck_orphan_index_ops;
616 static int lfsck_rbtree_setup(const struct lu_env *env,
617 struct lfsck_component *com)
619 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
620 struct lfsck_instance *lfsck = com->lc_lfsck;
621 struct dt_device *dev = lfsck->li_bottom;
622 struct lfsck_layout_slave_data *llsd = com->lc_data;
623 struct dt_object *obj;
625 fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
626 fid->f_oid = lfsck_dev_idx(lfsck);
628 obj = dt_locate(env, dev, fid);
630 RETURN(PTR_ERR(obj));
632 /* Generate an in-RAM object to stand for the layout rbtree.
633 * Scanning the layout rbtree will be via the iteration over
634 * the object. In the future, the rbtree may be written onto
635 * disk with the object.
637 * Mark the object to be as exist. */
638 obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
639 obj->do_index_ops = &lfsck_orphan_index_ops;
640 llsd->llsd_rb_obj = obj;
641 llsd->llsd_rbtree_valid = 1;
642 dev->dd_record_fid_accessed = 1;
644 CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
645 lfsck_lfsck2name(lfsck));
650 static void lfsck_rbtree_cleanup(const struct lu_env *env,
651 struct lfsck_component *com)
653 struct lfsck_instance *lfsck = com->lc_lfsck;
654 struct lfsck_layout_slave_data *llsd = com->lc_data;
655 struct rb_node *node = rb_first(&llsd->llsd_rb_root);
656 struct rb_node *next;
657 struct lfsck_rbtree_node *lrn;
659 lfsck->li_bottom->dd_record_fid_accessed = 0;
660 /* Invalid the rbtree, then no others will use it. */
661 write_lock(&llsd->llsd_rb_lock);
662 llsd->llsd_rbtree_valid = 0;
663 write_unlock(&llsd->llsd_rb_lock);
665 while (node != NULL) {
666 next = rb_next(node);
667 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
668 rb_erase(node, &llsd->llsd_rb_root);
669 lfsck_rbtree_free(lrn);
673 if (llsd->llsd_rb_obj != NULL) {
674 lfsck_object_put(env, llsd->llsd_rb_obj);
675 llsd->llsd_rb_obj = NULL;
678 CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
679 lfsck_lfsck2name(lfsck));
682 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
683 struct lfsck_component *com,
684 const struct lu_fid *fid,
687 struct lfsck_layout_slave_data *llsd = com->lc_data;
688 struct lfsck_rbtree_node *lrn;
694 if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
697 if (!fid_is_idif(fid) && !fid_is_norm(fid))
700 read_lock(&llsd->llsd_rb_lock);
701 if (!llsd->llsd_rbtree_valid)
702 GOTO(unlock, rc = 0);
704 lrn = lfsck_rbtree_search(llsd, fid, NULL);
706 struct lfsck_rbtree_node *tmp;
710 read_unlock(&llsd->llsd_rb_lock);
711 tmp = lfsck_rbtree_new(env, fid);
713 GOTO(out, rc = PTR_ERR(tmp));
716 write_lock(&llsd->llsd_rb_lock);
717 if (!llsd->llsd_rbtree_valid) {
718 lfsck_rbtree_free(tmp);
719 GOTO(unlock, rc = 0);
722 lrn = lfsck_rbtree_insert(llsd, tmp);
724 lfsck_rbtree_free(tmp);
727 idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
728 /* Any accessed object must be a known object. */
729 if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
730 atomic_inc(&lrn->lrn_known_count);
731 if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
732 atomic_inc(&lrn->lrn_accessed_count);
734 GOTO(unlock, rc = 0);
738 write_unlock(&llsd->llsd_rb_lock);
740 read_unlock(&llsd->llsd_rb_lock);
742 if (rc != 0 && accessed) {
743 struct lfsck_layout *lo = com->lc_file_ram;
745 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
746 "bitmap, and will cause incorrect LFSCK OST-object "
747 "handling, so disable it to cancel orphan handling "
748 "for related device. rc = %d\n",
749 lfsck_lfsck2name(com->lc_lfsck), rc);
751 lo->ll_flags |= LF_INCOMPLETE;
752 lfsck_rbtree_cleanup(env, com);
756 static inline void lldk_le_to_cpu(struct lfsck_layout_dangling_key *des,
757 const struct lfsck_layout_dangling_key *src)
759 fid_le_to_cpu(&des->lldk_fid, &src->lldk_fid);
760 des->lldk_comp_id = le32_to_cpu(src->lldk_comp_id);
761 des->lldk_ea_off = le32_to_cpu(src->lldk_ea_off);
764 static inline void lldk_cpu_to_le(struct lfsck_layout_dangling_key *des,
765 const struct lfsck_layout_dangling_key *src)
767 fid_cpu_to_le(&des->lldk_fid, &src->lldk_fid);
768 des->lldk_comp_id = cpu_to_le32(src->lldk_comp_id);
769 des->lldk_ea_off = cpu_to_le32(src->lldk_ea_off);
772 static inline void lldk_be_to_cpu(struct lfsck_layout_dangling_key *des,
773 const struct lfsck_layout_dangling_key *src)
775 fid_be_to_cpu(&des->lldk_fid, &src->lldk_fid);
776 des->lldk_comp_id = be32_to_cpu(src->lldk_comp_id);
777 des->lldk_ea_off = be32_to_cpu(src->lldk_ea_off);
780 static inline void lldk_cpu_to_be(struct lfsck_layout_dangling_key *des,
781 const struct lfsck_layout_dangling_key *src)
783 fid_cpu_to_be(&des->lldk_fid, &src->lldk_fid);
784 des->lldk_comp_id = cpu_to_be32(src->lldk_comp_id);
785 des->lldk_ea_off = cpu_to_be32(src->lldk_ea_off);
788 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
789 const struct lfsck_layout *src)
793 des->ll_magic = le32_to_cpu(src->ll_magic);
794 des->ll_status = le32_to_cpu(src->ll_status);
795 des->ll_flags = le32_to_cpu(src->ll_flags);
796 des->ll_success_count = le32_to_cpu(src->ll_success_count);
797 des->ll_run_time_phase1 = le64_to_cpu(src->ll_run_time_phase1);
798 des->ll_run_time_phase2 = le64_to_cpu(src->ll_run_time_phase2);
799 des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
800 des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
801 des->ll_time_last_checkpoint =
802 le64_to_cpu(src->ll_time_last_checkpoint);
803 des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
804 des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
805 des->ll_pos_first_inconsistent =
806 le64_to_cpu(src->ll_pos_first_inconsistent);
807 des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
808 des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
809 des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
810 des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
811 for (i = 0; i < LLIT_MAX; i++)
812 des->ll_objs_repaired[i] =
813 le64_to_cpu(src->ll_objs_repaired[i]);
814 des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
815 des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
816 lldk_le_to_cpu(&des->ll_lldk_latest_scanned_phase2,
817 &src->ll_lldk_latest_scanned_phase2);
820 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
821 const struct lfsck_layout *src)
825 des->ll_magic = cpu_to_le32(src->ll_magic);
826 des->ll_status = cpu_to_le32(src->ll_status);
827 des->ll_flags = cpu_to_le32(src->ll_flags);
828 des->ll_success_count = cpu_to_le32(src->ll_success_count);
829 des->ll_run_time_phase1 = cpu_to_le64(src->ll_run_time_phase1);
830 des->ll_run_time_phase2 = cpu_to_le64(src->ll_run_time_phase2);
831 des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
832 des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
833 des->ll_time_last_checkpoint =
834 cpu_to_le64(src->ll_time_last_checkpoint);
835 des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
836 des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
837 des->ll_pos_first_inconsistent =
838 cpu_to_le64(src->ll_pos_first_inconsistent);
839 des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
840 des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
841 des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
842 des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
843 for (i = 0; i < LLIT_MAX; i++)
844 des->ll_objs_repaired[i] =
845 cpu_to_le64(src->ll_objs_repaired[i]);
846 des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
847 des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
848 lldk_cpu_to_le(&des->ll_lldk_latest_scanned_phase2,
849 &src->ll_lldk_latest_scanned_phase2);
853 * Load the OST bitmap from the lfsck_layout trace file.
855 * \param[in] env pointer to the thread context
856 * \param[in] com pointer to the lfsck component
858 * \retval 0 for success
859 * \retval negative error number on failure or data corruption
861 static int lfsck_layout_load_bitmap(const struct lu_env *env,
862 struct lfsck_component *com)
864 struct dt_object *obj = com->lc_obj;
865 struct lfsck_assistant_data *lad = com->lc_data;
866 struct lfsck_layout *lo = com->lc_file_ram;
867 struct cfs_bitmap *bitmap = lad->lad_bitmap;
868 loff_t pos = com->lc_file_size;
874 if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
876 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
878 nbits = lo->ll_bitmap_size;
880 if (unlikely(nbits < BITS_PER_LONG))
881 nbits = BITS_PER_LONG;
883 if (nbits > bitmap->size) {
884 __u32 new_bits = bitmap->size;
885 struct cfs_bitmap *new_bitmap;
887 while (new_bits < nbits)
890 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
891 if (new_bitmap == NULL)
894 lad->lad_bitmap = new_bitmap;
895 CFS_FREE_BITMAP(bitmap);
899 if (lo->ll_bitmap_size == 0) {
900 lad->lad_incomplete = 0;
901 CFS_RESET_BITMAP(bitmap);
906 size = (lo->ll_bitmap_size + 7) >> 3;
907 rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
909 RETURN(rc >= 0 ? -EINVAL : rc);
911 if (cfs_bitmap_check_empty(bitmap))
912 lad->lad_incomplete = 0;
914 lad->lad_incomplete = 1;
920 * Load the layout LFSCK trace file from disk.
922 * The layout LFSCK trace file records the layout LFSCK status information
923 * and other statistics, such as how many objects have been scanned, and how
924 * many objects have been repaired, and etc. It also contains the bitmap for
925 * failed OSTs during the layout LFSCK. All these information will be loaded
926 * from disk to RAM when the layout LFSCK component setup.
928 * \param[in] env pointer to the thread context
929 * \param[in] com pointer to the lfsck component
931 * \retval positive number for file data corruption, the caller
932 * should reset the layout LFSCK trace file
933 * \retval 0 for success
934 * \retval negative error number on failure
936 static int lfsck_layout_load(const struct lu_env *env,
937 struct lfsck_component *com)
939 struct lfsck_layout *lo = com->lc_file_ram;
940 ssize_t size = com->lc_file_size;
944 rc = dt_read(env, com->lc_obj,
945 lfsck_buf_get(env, com->lc_file_disk, size), &pos);
949 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
950 lfsck_lfsck2name(com->lc_lfsck), rc);
952 } else if (rc != size) {
953 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
954 lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
958 lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
959 if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
960 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
961 "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
962 lo->ll_magic, LFSCK_LAYOUT_MAGIC);
970 * Store the layout LFSCK trace file on disk.
972 * The layout LFSCK trace file records the layout LFSCK status information
973 * and other statistics, such as how many objects have been scanned, and how
974 * many objects have been repaired, and etc. It also contains the bitmap for
975 * failed OSTs during the layout LFSCK. All these information will be synced
976 * from RAM to disk periodically.
978 * \param[in] env pointer to the thread context
979 * \param[in] com pointer to the lfsck component
981 * \retval 0 for success
982 * \retval negative error number on failure
984 static int lfsck_layout_store(const struct lu_env *env,
985 struct lfsck_component *com)
987 struct dt_object *obj = com->lc_obj;
988 struct lfsck_instance *lfsck = com->lc_lfsck;
989 struct lfsck_layout *lo_ram = com->lc_file_ram;
990 struct lfsck_layout *lo = com->lc_file_disk;
992 struct dt_device *dev = lfsck_obj2dev(obj);
993 struct cfs_bitmap *bitmap = NULL;
995 ssize_t size = com->lc_file_size;
1000 if (lfsck->li_master) {
1001 struct lfsck_assistant_data *lad = com->lc_data;
1003 bitmap = lad->lad_bitmap;
1004 nbits = bitmap->size;
1007 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
1010 lo_ram->ll_bitmap_size = nbits;
1011 lfsck_layout_cpu_to_le(lo, lo_ram);
1012 th = dt_trans_create(env, dev);
1014 GOTO(log, rc = PTR_ERR(th));
1016 rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
1021 if (bitmap != NULL) {
1022 rc = dt_declare_record_write(env, obj,
1023 lfsck_buf_get(env, bitmap->data, nbits >> 3),
1029 rc = dt_trans_start_local(env, dev, th);
1034 rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
1038 if (bitmap != NULL) {
1040 rc = dt_record_write(env, obj,
1041 lfsck_buf_get(env, bitmap->data, nbits >> 3),
1048 dt_trans_stop(env, dev, th);
1052 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
1053 lfsck_lfsck2name(lfsck), rc);
1058 static int lfsck_layout_init(const struct lu_env *env,
1059 struct lfsck_component *com)
1061 struct lfsck_layout *lo = com->lc_file_ram;
1064 memset(lo, 0, com->lc_file_size);
1065 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
1066 lo->ll_status = LS_INIT;
1067 down_write(&com->lc_sem);
1068 rc = lfsck_layout_store(env, com);
1069 if (rc == 0 && com->lc_lfsck->li_master)
1070 rc = lfsck_load_sub_trace_files(env, com,
1071 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
1072 up_write(&com->lc_sem);
1077 static int fid_is_for_ostobj(const struct lu_env *env,
1078 struct lfsck_instance *lfsck,
1079 struct dt_object *obj, const struct lu_fid *fid)
1081 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1082 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
1083 struct lustre_ost_attrs *loa;
1086 fld_range_set_any(range);
1087 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1089 if (fld_range_is_ost(range))
1095 loa = &lfsck_env_info(env)->lti_loa;
1096 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, loa, sizeof(*loa)),
1098 if (rc >= sizeof(struct lustre_mdt_attrs)) {
1099 lustre_lma_swab(&loa->loa_lma);
1101 return loa->loa_lma.lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1104 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1109 static struct lfsck_layout_seq *
1110 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1112 struct lfsck_layout_seq *lls;
1114 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1115 if (lls->lls_seq == seq)
1118 if (lls->lls_seq > seq)
1126 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1127 struct lfsck_layout_seq *lls)
1129 struct lfsck_layout_seq *tmp;
1130 struct list_head *pos = &llsd->llsd_seq_list;
1132 list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1133 if (lls->lls_seq < tmp->lls_seq) {
1134 pos = &tmp->lls_list;
1138 list_add_tail(&lls->lls_list, pos);
1142 lfsck_layout_lastid_create(const struct lu_env *env,
1143 struct lfsck_instance *lfsck,
1144 struct dt_object *obj)
1146 struct lfsck_thread_info *info = lfsck_env_info(env);
1147 struct lu_attr *la = &info->lti_la;
1148 struct dt_object_format *dof = &info->lti_dof;
1149 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1150 struct dt_device *dt = lfsck_obj2dev(obj);
1157 if (bk->lb_param & LPF_DRYRUN)
1160 memset(la, 0, sizeof(*la));
1161 la->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1162 la->la_valid = LA_MODE | LA_UID | LA_GID;
1163 memset(dof, 0, sizeof(*dof));
1164 dof->dof_type = dt_mode_to_dft(S_IFREG);
1166 th = dt_trans_create(env, dt);
1168 GOTO(log, rc = PTR_ERR(th));
1170 rc = dt_declare_create(env, obj, la, NULL, dof, th);
1174 rc = dt_declare_record_write(env, obj,
1175 lfsck_buf_get(env, &lastid,
1181 rc = dt_trans_start_local(env, dt, th);
1185 dt_write_lock(env, obj, 0);
1186 if (likely(dt_object_exists(obj) == 0)) {
1187 rc = dt_create(env, obj, la, NULL, dof, th);
1189 rc = dt_record_write(env, obj,
1190 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1193 dt_write_unlock(env, obj);
1198 dt_trans_stop(env, dt, th);
1201 CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1203 lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1209 lfsck_layout_lastid_reload(const struct lu_env *env,
1210 struct lfsck_component *com,
1211 struct lfsck_layout_seq *lls)
1217 dt_read_lock(env, lls->lls_lastid_obj, 0);
1218 rc = dt_record_read(env, lls->lls_lastid_obj,
1219 lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1220 dt_read_unlock(env, lls->lls_lastid_obj);
1221 if (unlikely(rc != 0))
1224 lastid = le64_to_cpu(lastid);
1225 if (lastid < lls->lls_lastid_known) {
1226 struct lfsck_instance *lfsck = com->lc_lfsck;
1227 struct lfsck_layout *lo = com->lc_file_ram;
1229 lls->lls_lastid = lls->lls_lastid_known;
1231 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1232 LASSERT(lfsck->li_out_notify != NULL);
1234 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1235 LE_LASTID_REBUILDING);
1236 lo->ll_flags |= LF_CRASHED_LASTID;
1238 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1239 "LAST_ID file (1) for the sequence %#llx"
1240 ", old value %llu, known value %llu\n",
1241 lfsck_lfsck2name(lfsck), lls->lls_seq,
1242 lastid, lls->lls_lastid);
1244 } else if (lastid >= lls->lls_lastid) {
1245 lls->lls_lastid = lastid;
1253 lfsck_layout_lastid_store(const struct lu_env *env,
1254 struct lfsck_component *com)
1256 struct lfsck_instance *lfsck = com->lc_lfsck;
1257 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1258 struct dt_device *dt = lfsck->li_bottom;
1259 struct lfsck_layout_slave_data *llsd = com->lc_data;
1260 struct lfsck_layout_seq *lls;
1266 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1269 if (!lls->lls_dirty)
1272 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1273 "<seq> %#llx as <oid> %llu\n",
1274 lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1276 if (bk->lb_param & LPF_DRYRUN) {
1281 th = dt_trans_create(env, dt);
1284 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1285 "the LAST_ID for <seq> %#llx(1): rc = %d\n",
1286 lfsck_lfsck2name(com->lc_lfsck),
1291 lastid = cpu_to_le64(lls->lls_lastid);
1292 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1293 lfsck_buf_get(env, &lastid,
1299 rc = dt_trans_start_local(env, dt, th);
1303 dt_write_lock(env, lls->lls_lastid_obj, 0);
1304 rc = dt_record_write(env, lls->lls_lastid_obj,
1305 lfsck_buf_get(env, &lastid,
1306 sizeof(lastid)), &pos, th);
1307 dt_write_unlock(env, lls->lls_lastid_obj);
1312 dt_trans_stop(env, dt, th);
1315 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1316 "the LAST_ID for <seq> %#llx(2): rc = %d\n",
1317 lfsck_lfsck2name(com->lc_lfsck),
1326 lfsck_layout_lastid_load(const struct lu_env *env,
1327 struct lfsck_component *com,
1328 struct lfsck_layout_seq *lls)
1330 struct lfsck_instance *lfsck = com->lc_lfsck;
1331 struct lfsck_layout *lo = com->lc_file_ram;
1332 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
1333 struct dt_object *obj;
1338 lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1339 obj = dt_locate(env, lfsck->li_bottom, fid);
1341 RETURN(PTR_ERR(obj));
1343 /* LAST_ID crashed, to be rebuilt */
1344 if (dt_object_exists(obj) == 0) {
1345 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1346 LASSERT(lfsck->li_out_notify != NULL);
1348 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1349 LE_LASTID_REBUILDING);
1350 lo->ll_flags |= LF_CRASHED_LASTID;
1352 CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1353 "LAST_ID file for sequence %#llx\n",
1354 lfsck_lfsck2name(lfsck), lls->lls_seq);
1356 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1358 struct l_wait_info lwi = LWI_TIMEOUT(
1359 cfs_time_seconds(cfs_fail_val),
1362 /* Some others may changed the cfs_fail_val
1363 * as zero after above check, re-check it for
1364 * sure to avoid falling into wait for ever. */
1365 if (likely(lwi.lwi_timeout > 0)) {
1366 struct ptlrpc_thread *thread =
1369 up_write(&com->lc_sem);
1370 l_wait_event(thread->t_ctl_waitq,
1371 !thread_is_running(thread),
1373 down_write(&com->lc_sem);
1378 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1380 dt_read_lock(env, obj, 0);
1381 rc = dt_read(env, obj,
1382 lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1384 dt_read_unlock(env, obj);
1385 if (rc != 0 && rc != sizeof(__u64))
1386 GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1388 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1389 LASSERT(lfsck->li_out_notify != NULL);
1391 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1392 LE_LASTID_REBUILDING);
1393 lo->ll_flags |= LF_CRASHED_LASTID;
1395 CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1396 "LAST_ID file for the sequence %#llx"
1398 lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1401 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1409 lfsck_object_put(env, obj);
1411 lls->lls_lastid_obj = obj;
1416 static void lfsck_layout_record_failure(const struct lu_env *env,
1417 struct lfsck_instance *lfsck,
1418 struct lfsck_layout *lo)
1422 lo->ll_objs_failed_phase1++;
1423 cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1425 if (lo->ll_pos_first_inconsistent == 0 ||
1426 lo->ll_pos_first_inconsistent < cookie) {
1427 lo->ll_pos_first_inconsistent = cookie;
1429 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1430 "inconsistency at the pos [%llu]\n",
1431 lfsck_lfsck2name(lfsck),
1432 lo->ll_pos_first_inconsistent);
1436 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1437 struct lfsck_component *com,
1440 struct lfsck_instance *lfsck = com->lc_lfsck;
1441 struct lfsck_layout *lo = com->lc_file_ram;
1443 CDEBUG(D_LFSCK, "%s: layout LFSCK double scan: rc = %d\n",
1444 lfsck_lfsck2name(lfsck), rc);
1446 down_write(&com->lc_sem);
1447 lo->ll_run_time_phase2 += ktime_get_seconds() -
1448 com->lc_time_last_checkpoint;
1449 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
1450 lo->ll_objs_checked_phase2 += com->lc_new_checked;
1453 if (lo->ll_flags & LF_INCOMPLETE) {
1454 lo->ll_status = LS_PARTIAL;
1456 if (lfsck->li_master) {
1457 struct lfsck_assistant_data *lad = com->lc_data;
1459 if (lad->lad_incomplete)
1460 lo->ll_status = LS_PARTIAL;
1462 lo->ll_status = LS_COMPLETED;
1464 lo->ll_status = LS_COMPLETED;
1467 lo->ll_flags &= ~LF_SCANNED_ONCE;
1468 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1469 lo->ll_flags &= ~LF_INCONSISTENT;
1470 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1471 lo->ll_success_count++;
1472 } else if (rc == 0) {
1473 if (lfsck->li_status != 0)
1474 lo->ll_status = lfsck->li_status;
1476 lo->ll_status = LS_STOPPED;
1478 lo->ll_status = LS_FAILED;
1481 rc = lfsck_layout_store(env, com);
1482 up_write(&com->lc_sem);
1484 CDEBUG(D_LFSCK, "%s: layout LFSCK double scan result %u: rc = %d\n",
1485 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
1490 static int lfsck_layout_trans_stop(const struct lu_env *env,
1491 struct dt_device *dev,
1492 struct thandle *handle, int result)
1496 /* XXX: If there is something worng or it needs to repair nothing,
1497 * then notify the lower to stop the modification. Currently,
1498 * we use th_result for such purpose, that may be replaced by
1499 * some rollback mechanism in the future. */
1500 handle->th_result = result;
1501 rc = dt_trans_stop(env, dev, handle);
1503 return result > 0 ? 0 : result;
1505 return rc == 0 ? 1 : rc;
1508 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1509 struct lfsck_component *com,
1510 const struct lu_fid *pfid,
1511 const struct lu_fid *cfid,
1512 __u32 comp_id, __u32 ea_off,
1515 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1516 struct lu_fid *rec = &lfsck_env_info(env)->lti_fid3;
1517 struct dt_device *dev;
1518 struct dt_object *obj;
1519 struct thandle *th = NULL;
1524 idx = lfsck_sub_trace_file_fid2idx(pfid);
1525 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1526 dev = lfsck_obj2dev(obj);
1528 fid_cpu_to_be(&key->lldk_fid, pfid);
1529 key->lldk_comp_id = cpu_to_be32(comp_id);
1530 key->lldk_ea_off = cpu_to_be32(ea_off);
1532 fid_cpu_to_be(rec, cfid);
1533 rec->f_ver = cpu_to_be32(ost_idx);
1535 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1537 th = dt_trans_create(env, dev);
1539 GOTO(unlock, rc = PTR_ERR(th));
1541 rc = dt_declare_insert(env, obj,
1542 (const struct dt_rec *)rec,
1543 (const struct dt_key *)key, th);
1547 rc = dt_trans_start_local(env, dev, th);
1551 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1552 (const struct dt_key *)key, th, 1);
1557 if (th && !IS_ERR(th))
1558 dt_trans_stop(env, dev, th);
1560 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1562 CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", comp_id = %u, "
1563 "ea_off = %u, ost_idx = %u, into the trace file for further "
1564 "dangling check: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1565 PFID(pfid), PFID(cfid), comp_id, ea_off, ost_idx, rc);
1570 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1571 struct lfsck_component *com,
1572 const struct lu_fid *fid,
1573 __u32 comp_id, __u32 ea_off)
1575 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1576 struct dt_device *dev;
1577 struct dt_object *obj;
1578 struct thandle *th = NULL;
1583 idx = lfsck_sub_trace_file_fid2idx(fid);
1584 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1585 dev = lfsck_obj2dev(obj);
1587 fid_cpu_to_be(&key->lldk_fid, fid);
1588 key->lldk_comp_id = cpu_to_be32(comp_id);
1589 key->lldk_ea_off = cpu_to_be32(ea_off);
1591 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1593 th = dt_trans_create(env, dev);
1595 GOTO(unlock, rc = PTR_ERR(th));
1597 rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1601 rc = dt_trans_start_local(env, dev, th);
1605 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1610 if (th && !IS_ERR(th))
1611 dt_trans_stop(env, dev, th);
1613 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1615 CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1616 ", comp_id = %u, ea_off = %u from the trace file: rc = %d\n",
1617 lfsck_lfsck2name(com->lc_lfsck), PFID(fid), comp_id, ea_off, rc);
1623 * Get the system default stripe size.
1625 * \param[in] env pointer to the thread context
1626 * \param[in] lfsck pointer to the lfsck instance
1627 * \param[out] size pointer to the default stripe size
1629 * \retval 0 for success
1630 * \retval negative error number on failure
1632 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1633 struct lfsck_instance *lfsck,
1636 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1637 struct dt_object *root;
1640 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1642 return PTR_ERR(root);
1644 /* Get the default stripe size via xattr_get on the backend root. */
1645 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1648 /* The lum->lmm_stripe_size is LE mode. The *size also
1649 * should be LE mode. So it is unnecessary to convert. */
1650 *size = lum->lmm_stripe_size;
1652 } else if (unlikely(rc == 0)) {
1656 lfsck_object_put(env, root);
1662 * \retval +1: repaired
1663 * \retval 0: did nothing
1664 * \retval -ve: on error
1666 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1667 struct lfsck_instance *lfsck,
1668 struct thandle *handle,
1669 struct dt_object *parent,
1670 const struct lu_fid *cfid,
1672 struct lov_mds_md_v1 *lmm,
1673 struct lov_ost_data_v1 *slot,
1674 int fl, __u32 ost_idx, int size)
1676 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1677 struct lu_buf ea_buf;
1684 magic = le32_to_cpu(lmm->lmm_magic);
1685 pattern = le32_to_cpu(lmm->lmm_pattern);
1686 count = le16_to_cpu(lmm->lmm_stripe_count);
1688 fid_to_ostid(cfid, oi);
1689 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1690 slot->l_ost_gen = cpu_to_le32(0);
1691 slot->l_ost_idx = cpu_to_le32(ost_idx);
1693 if (pattern & LOV_PATTERN_F_HOLE) {
1694 struct lov_ost_data_v1 *objs;
1697 if (magic == LOV_MAGIC_V1)
1698 objs = &lmm->lmm_objects[0];
1700 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1701 for (i = 0; i < count; i++, objs++) {
1702 if (lovea_slot_is_dummy(objs))
1706 /* If the @slot is the last dummy slot to be refilled,
1707 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1710 cpu_to_le32(pattern & ~LOV_PATTERN_F_HOLE);
1712 CDEBUG(D_LFSCK, "%s: remove layout HOLE for "DFID
1713 ": parent "DFID"\n", lfsck_lfsck2name(lfsck),
1714 PFID(cfid), PFID(lfsck_dto2fid(parent)));
1718 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
1719 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1726 static struct lov_ost_data_v1 *
1727 __lfsck_layout_new_v1_lovea(struct lov_mds_md_v1 *lmm,
1728 const struct lu_fid *pfid,
1729 __u32 stripe_size, __u32 ea_off,
1730 __u32 pattern, __u16 count)
1732 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1733 lmm->lmm_pattern = cpu_to_le32(pattern);
1734 fid_to_lmm_oi(pfid, &lmm->lmm_oi);
1735 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1736 lmm->lmm_stripe_size = cpu_to_le32(stripe_size);
1737 lmm->lmm_stripe_count = cpu_to_le16(count);
1738 lmm->lmm_layout_gen = cpu_to_le16(1);
1739 memset(&lmm->lmm_objects[0], 0,
1740 sizeof(struct lov_ost_data_v1) * count);
1742 return &lmm->lmm_objects[ea_off];
1745 static int lfsck_layout_new_v1_lovea(const struct lu_env *env,
1746 struct lfsck_instance *lfsck,
1747 struct ost_layout *ol,
1748 struct dt_object *parent,
1749 struct lu_buf *buf, __u32 ea_off,
1750 struct lov_mds_md_v1 **lmm,
1751 struct lov_ost_data_v1 **objs)
1754 __u32 stripe_size = ol->ol_stripe_size;
1755 __u32 pattern = LOV_PATTERN_RAID0;
1758 if (ol->ol_stripe_count != 0)
1759 count = ol->ol_stripe_count;
1763 size = lov_mds_md_size(count, LOV_MAGIC_V1);
1764 LASSERTF(buf->lb_len >= size,
1765 "buffer len %d is less than real size %d\n",
1766 (int)buf->lb_len, size);
1768 if (stripe_size == 0) {
1771 rc = lfsck_layout_get_def_stripesize(env, lfsck, &stripe_size);
1777 if (ol->ol_stripe_count > 1 ||
1778 (ol->ol_stripe_count == 0 && ea_off != 0)) {
1779 pattern |= LOV_PATTERN_F_HOLE;
1780 memset(&(*lmm)->lmm_objects[0], 0,
1781 count * sizeof(struct lov_ost_data_v1));
1784 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1785 stripe_size, ea_off, pattern, count);
1790 static int lfsck_layout_new_comp_lovea(const struct lu_env *env,
1791 struct ost_layout *ol,
1792 struct dt_object *parent,
1793 struct lu_buf *buf, __u32 ea_off,
1794 struct lov_mds_md_v1 **lmm,
1795 struct lov_ost_data_v1 **objs)
1797 struct lov_comp_md_v1 *lcm;
1798 struct lov_comp_md_entry_v1 *lcme;
1799 __u32 pattern = LOV_PATTERN_RAID0;
1800 __u32 offset = sizeof(*lcm) + sizeof(*lcme);
1801 int lcme_size = lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1802 int size = offset + lcme_size;
1804 LASSERTF(buf->lb_len >= size,
1805 "buffer len %d is less than real size %d\n",
1806 (int)buf->lb_len, size);
1809 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1810 lcm->lcm_size = cpu_to_le32(size);
1811 lcm->lcm_layout_gen = cpu_to_le32(1);
1813 lcm->lcm_entry_count = cpu_to_le16(1);
1815 lcme = &lcm->lcm_entries[0];
1816 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1817 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1818 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1819 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1820 lcme->lcme_offset = cpu_to_le32(offset);
1821 lcme->lcme_size = cpu_to_le32(lcme_size);
1822 if (ol->ol_stripe_count > 1)
1823 pattern |= LOV_PATTERN_F_HOLE;
1825 *lmm = buf->lb_buf + offset;
1826 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1827 ol->ol_stripe_size, ea_off,
1828 pattern, ol->ol_stripe_count);
1833 static int lfsck_layout_add_comp_comp(const struct lu_env *env,
1834 struct lfsck_instance *lfsck,
1835 struct thandle *handle,
1836 struct ost_layout *ol,
1837 struct dt_object *parent,
1838 const struct lu_fid *cfid,
1839 struct lu_buf *buf, __u32 ost_idx,
1840 __u32 ea_off, int pos)
1842 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1843 struct lov_comp_md_entry_v1 *lcme;
1844 struct lov_mds_md_v1 *lmm;
1845 struct lov_ost_data_v1 *objs;
1846 int added = sizeof(*lcme) +
1847 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1848 int size = le32_to_cpu(lcm->lcm_size) + added;
1852 __u32 pattern = LOV_PATTERN_RAID0;
1853 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1856 lu_buf_check_and_grow(buf, size);
1857 /* set the lcm again because lu_buf_check_and_grow() may
1858 * have reallocated the buf. */
1860 lcm->lcm_size = cpu_to_le32(size);
1861 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1862 lcm->lcm_entry_count = cpu_to_le16(count + 1);
1864 /* 1. Move the component bodies from [pos, count-1] to [pos+1, count]
1865 * with distance of 'added'. */
1868 for (i = pos; i < count; i++) {
1869 lcme = &lcm->lcm_entries[i];
1870 size += le32_to_cpu(lcme->lcme_size);
1873 offset = le32_to_cpu(lcm->lcm_entries[pos].lcme_offset);
1874 memmove(buf->lb_buf + offset + added,
1875 buf->lb_buf + offset, size);
1879 /* 2. Move the component header [0, pos-1] to [0, pos-1] with distance
1880 * of 'sizeof(struct lov_comp_md_entry_v1)' */
1882 for (i = 0; i < pos; i++) {
1883 lcme = &lcm->lcm_entries[i];
1884 size += le32_to_cpu(lcme->lcme_size);
1887 offset = le32_to_cpu(lcm->lcm_entries[0].lcme_offset);
1888 memmove(buf->lb_buf + offset + sizeof(*lcme),
1889 buf->lb_buf + offset, size);
1892 /* 3. Recalculate the enter offset for the component [pos, count-1] */
1893 for (i = count - 1; i >= pos; i--) {
1894 lcm->lcm_entries[i + 1] = lcm->lcm_entries[i];
1895 lcm->lcm_entries[i + 1].lcme_offset =
1896 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i + 1].
1897 lcme_offset) + added);
1900 /* 4. Recalculate the enter offset for the component [0, pos) */
1901 for (i = 0; i < pos; i++) {
1902 lcm->lcm_entries[i].lcme_offset =
1903 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i].
1904 lcme_offset) + sizeof(*lcme));
1907 offset = sizeof(*lcm) + sizeof(*lcme) * (count + 1) + size;
1908 /* 4. Insert the new component header (entry) at the slot 'pos'. */
1909 lcme = &lcm->lcm_entries[pos];
1910 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1911 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1912 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1913 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1914 lcme->lcme_offset = cpu_to_le32(offset);
1915 lcme->lcme_size = cpu_to_le32(lov_mds_md_size(ol->ol_stripe_count,
1918 if (ol->ol_stripe_count > 1)
1919 pattern |= LOV_PATTERN_F_HOLE;
1921 lmm = buf->lb_buf + offset;
1922 /* 5. Insert teh new component body at the 'offset'. */
1923 objs = __lfsck_layout_new_v1_lovea(lmm, lfsck_dto2fid(parent),
1924 ol->ol_stripe_size, ea_off,
1925 pattern, ol->ol_stripe_count);
1927 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1928 lmm, objs, LU_XATTR_REPLACE, ost_idx,
1929 le32_to_cpu(lcm->lcm_size));
1931 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant add new COMP for "
1932 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1933 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
1934 "comp_end %llu, %s LOV EA hole: rc = %d\n",
1935 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1936 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
1937 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
1938 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
1939 "with" : "without", rc);
1944 static int lfsck_layout_extend_v1v3_lovea(const struct lu_env *env,
1945 struct lfsck_instance *lfsck,
1946 struct thandle *handle,
1947 struct ost_layout *ol,
1948 struct dt_object *parent,
1949 const struct lu_fid *cfid,
1950 struct lu_buf *buf, __u32 ost_idx,
1953 struct lov_mds_md_v1 *lmm = buf->lb_buf;
1954 struct lov_ost_data_v1 *objs;
1955 __u16 count = le16_to_cpu(lmm->lmm_stripe_count);
1956 __u32 magic = le32_to_cpu(lmm->lmm_magic);
1962 /* The original LOVEA maybe re-generated via old filter_fid, at
1963 * that time, we do not know the stripe count and stripe size. */
1964 if (ol->ol_stripe_count > count)
1965 count = ol->ol_stripe_count;
1966 if (ol->ol_stripe_size != 0 &&
1967 ol->ol_stripe_size != le32_to_cpu(lmm->lmm_stripe_size))
1968 lmm->lmm_stripe_size = cpu_to_le32(ol->ol_stripe_size);
1970 if (magic == LOV_MAGIC_V1)
1971 objs = &lmm->lmm_objects[count];
1973 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[count];
1975 gap = ea_off - count;
1979 size = lov_mds_md_size(count, magic);
1980 LASSERTF(buf->lb_len >= size,
1981 "buffer len %d is less than real size %d\n",
1982 (int)buf->lb_len, size);
1985 memset(objs, 0, gap * sizeof(*objs));
1986 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1989 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1990 lmm->lmm_stripe_count = cpu_to_le16(count);
1993 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1994 lmm, objs, LU_XATTR_REPLACE, ost_idx, size);
1996 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1997 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1998 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
1999 "comp_end %llu, %s LOV EA hole: rc = %d\n",
2000 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2001 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2002 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2003 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2004 "with" : "without", rc);
2010 * \retval +1: repaired
2011 * \retval 0: did nothing
2012 * \retval -ve: on error
2014 static int lfsck_layout_update_lovea(const struct lu_env *env,
2015 struct lfsck_instance *lfsck,
2016 struct thandle *handle,
2017 struct ost_layout *ol,
2018 struct dt_object *parent,
2019 const struct lu_fid *cfid,
2020 struct lu_buf *buf, int fl,
2021 __u32 ost_idx, __u32 ea_off)
2023 struct lov_mds_md_v1 *lmm = NULL;
2024 struct lov_ost_data_v1 *objs = NULL;
2028 if (ol->ol_comp_id != 0)
2029 rc = lfsck_layout_new_comp_lovea(env, ol, parent, buf, ea_off,
2032 rc = lfsck_layout_new_v1_lovea(env, lfsck, ol, parent, buf,
2033 ea_off, &lmm, &objs);
2036 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid,
2037 buf, lmm, objs, fl, ost_idx, rc);
2039 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant created layout EA for "
2040 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2041 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2042 "comp_end %llu, fl %d, %s LOV EA hole: rc = %d\n",
2043 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2044 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2045 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end, fl,
2046 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2047 "with" : "without", rc);
2052 static int __lfsck_layout_update_pfid(const struct lu_env *env,
2053 struct dt_object *child,
2054 const struct lu_fid *pfid,
2055 const struct ost_layout *ol, __u32 offset)
2057 struct dt_device *dev = lfsck_obj2dev(child);
2058 struct filter_fid *ff = &lfsck_env_info(env)->lti_ff;
2059 struct thandle *handle;
2060 struct lu_buf buf = { NULL };
2063 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2064 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2065 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2066 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2067 * parent MDT-object's layout EA. */
2068 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
2069 ost_layout_cpu_to_le(&ff->ff_layout, ol);
2070 lfsck_buf_init(&buf, ff, sizeof(*ff));
2072 handle = dt_trans_create(env, dev);
2074 RETURN(PTR_ERR(handle));
2076 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2080 rc = dt_trans_start_local(env, dev, handle);
2084 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2089 dt_trans_stop(env, dev, handle);
2095 * \retval +1: repaired
2096 * \retval 0: did nothing
2097 * \retval -ve: on error
2099 static int lfsck_layout_update_pfid(const struct lu_env *env,
2100 struct lfsck_component *com,
2101 struct dt_object *parent,
2102 struct lu_fid *cfid,
2103 struct dt_device *cdev,
2104 struct ost_layout *ol, __u32 ea_off)
2106 struct dt_object *child;
2110 child = lfsck_object_find_by_dev(env, cdev, cfid);
2112 RETURN(PTR_ERR(child));
2114 rc = __lfsck_layout_update_pfid(env, child,
2115 lu_object_fid(&parent->do_lu),
2117 lfsck_object_put(env, child);
2119 RETURN(rc == 0 ? 1 : rc);
2122 static int lfsck_lovea_size(struct ost_layout *ol, __u32 ea_off)
2124 if (ol->ol_comp_id != 0)
2125 return sizeof(struct lov_comp_md_v1) +
2126 sizeof(struct lov_comp_md_entry_v1) +
2127 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2129 if (ol->ol_stripe_count != 0)
2130 return lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2132 return lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2136 * This function will create the MDT-object with the given (partial) LOV EA.
2138 * Under some data corruption cases, the MDT-object of the file may be lost,
2139 * but its OST-objects, or some of them are there. The layout LFSCK needs to
2140 * re-create the MDT-object with the orphan OST-object(s) information.
2142 * On the other hand, the LFSCK may has created some OST-object for repairing
2143 * dangling LOV EA reference, but as the LFSCK processing, it may find that
2144 * the old OST-object is there and should replace the former new created OST
2145 * object. Unfortunately, some others have modified such newly created object.
2146 * To keep the data (both new and old), the LFSCK will create MDT-object with
2147 * new FID to reference the original OST-object.
2149 * \param[in] env pointer to the thread context
2150 * \param[in] com pointer to the lfsck component
2151 * \param[in] ltd pointer to target device descriptor
2152 * \param[in] rec pointer to the record for the orphan OST-object
2153 * \param[in] cfid pointer to FID for the orphan OST-object
2154 * \param[in] infix additional information, such as the FID for original
2155 * MDT-object and the stripe offset in the LOV EA
2156 * \param[in] type the type for describing why the orphan MDT-object is
2157 * created. The rules are as following:
2159 * type "C": Multiple OST-objects claim the same MDT-object and the
2160 * same slot in the layout EA. Then the LFSCK will create
2161 * new MDT-object(s) to hold the conflict OST-object(s).
2163 * type "N": The orphan OST-object does not know which one was the
2164 * real parent MDT-object, so the LFSCK uses new FID for
2165 * its parent MDT-object.
2167 * type "R": The orphan OST-object knows its parent MDT-object FID,
2168 * but does not know the position (the file name) in the
2171 * type "D": The MDT-object is a directory, it may knows its parent
2172 * but because there is no valid linkEA, the LFSCK cannot
2173 * know where to put it back to the namespace.
2174 * type "O": The MDT-object has no linkEA, and there is no name
2175 * entry that references the MDT-object.
2177 * type "P": The orphan object to be created was a parent directory
2178 * of some MDT-object which linkEA shows that the @orphan
2179 * object is missing.
2181 * The orphan name will be like:
2182 * ${FID}-${infix}-${type}-${conflict_version}
2184 * \param[in] ea_off the stripe offset in the LOV EA
2186 * \retval positive on repaired something
2187 * \retval 0 if needs to repair nothing
2188 * \retval negative error number on failure
2190 static int lfsck_layout_recreate_parent(const struct lu_env *env,
2191 struct lfsck_component *com,
2192 struct lfsck_tgt_desc *ltd,
2193 struct lu_orphan_rec_v2 *rec,
2194 struct lu_fid *cfid,
2199 struct lfsck_thread_info *info = lfsck_env_info(env);
2200 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
2201 char *name = info->lti_key;
2202 struct lu_attr *la = &info->lti_la2;
2203 struct dt_object_format *dof = &info->lti_dof;
2204 struct lfsck_instance *lfsck = com->lc_lfsck;
2205 struct ost_layout *ol = &rec->lor_layout;
2206 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
2207 struct lu_fid *tfid = &info->lti_fid3;
2208 struct dt_device *dev = lfsck->li_bottom;
2209 struct dt_object *lpf = lfsck->li_lpf_obj;
2210 struct dt_object *pobj = NULL;
2211 struct dt_object *cobj = NULL;
2212 struct thandle *th = NULL;
2213 struct lu_buf *ea_buf = &info->lti_big_buf;
2214 struct lu_buf lov_buf;
2215 struct lfsck_lock_handle *llh = &info->lti_llh;
2216 struct linkea_data ldata = { NULL };
2217 struct lu_buf linkea_buf;
2218 const struct lu_name *pname;
2224 if (unlikely(lpf == NULL))
2225 GOTO(log, rc = -ENXIO);
2227 /* We use two separated transactions to repair the inconsistency.
2229 * 1) create the MDT-object locally.
2230 * 2) update the OST-object's PFID EA if necessary.
2232 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
2233 * updated when the layout LFSCK run next time.
2235 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
2236 * when the layout LFSCK run next time. */
2238 if (fid_is_zero(pfid)) {
2239 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
2243 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2245 GOTO(log, rc = PTR_ERR(cobj));
2248 pobj = lfsck_object_find_by_dev(env, dev, pfid);
2250 GOTO(log, rc = PTR_ERR(pobj));
2252 LASSERT(infix != NULL);
2253 LASSERT(type != NULL);
2255 memset(la, 0, sizeof(*la));
2256 la->la_uid = rec->lor_rec.lor_uid;
2257 la->la_gid = rec->lor_rec.lor_gid;
2258 la->la_mode = S_IFREG | S_IRUSR;
2259 la->la_valid = LA_MODE | LA_UID | LA_GID;
2261 memset(dof, 0, sizeof(*dof));
2262 dof->dof_type = dt_mode_to_dft(S_IFREG);
2263 /* Because the dof->dof_reg.striped = 0, the LOD will not create
2264 * the stripe(s). The LFSCK will specify the LOV EA via
2265 * lfsck_layout_update_lovea(). */
2267 size = lfsck_lovea_size(ol, ea_off);
2268 if (ea_buf->lb_len < size) {
2269 lu_buf_realloc(ea_buf, size);
2270 if (ea_buf->lb_buf == NULL)
2271 GOTO(log, rc = -ENOMEM);
2276 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
2278 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2279 (const struct dt_key *)name);
2280 if (rc != 0 && rc != -ENOENT)
2284 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
2285 MDS_INODELOCK_UPDATE, LCK_PW);
2289 /* Re-check whether the name conflict with othrs after taken
2291 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2292 (const struct dt_key *)name);
2293 if (unlikely(rc == 0)) {
2301 pname = lfsck_name_get_const(env, name, strlen(name));
2302 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
2303 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
2307 /* The 1st transaction. */
2308 th = dt_trans_create(env, dev);
2310 GOTO(unlock, rc = PTR_ERR(th));
2312 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
2316 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
2317 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
2318 LU_XATTR_CREATE, th);
2322 dtrec->rec_fid = pfid;
2323 dtrec->rec_type = S_IFREG;
2324 rc = dt_declare_insert(env, lpf,
2325 (const struct dt_rec *)dtrec,
2326 (const struct dt_key *)name, th);
2330 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2331 ldata.ld_leh->leh_len);
2332 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
2333 XATTR_NAME_LINK, 0, th);
2337 rc = dt_trans_start_local(env, dev, th);
2341 dt_write_lock(env, pobj, 0);
2342 rc = dt_create(env, pobj, la, NULL, dof, th);
2344 rc = lfsck_layout_update_lovea(env, lfsck, th, ol, pobj, cfid,
2345 &lov_buf, LU_XATTR_CREATE, ltd->ltd_index, ea_off);
2346 dt_write_unlock(env, pobj);
2350 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
2351 (const struct dt_key *)name, th, 1);
2355 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
2356 if (rc == 0 && cobj != NULL) {
2357 dt_trans_stop(env, dev, th);
2360 /* The 2nd transaction. */
2361 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ol, ea_off);
2368 dt_trans_stop(env, dev, th);
2374 if (cobj != NULL && !IS_ERR(cobj))
2375 lfsck_object_put(env, cobj);
2376 if (pobj != NULL && !IS_ERR(pobj))
2377 lfsck_object_put(env, pobj);
2380 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2381 "recreate the lost MDT-object: parent "DFID
2382 ", child "DFID", OST-index %u, stripe-index %u, "
2383 "infix %s, type %s: rc = %d\n",
2384 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2385 ltd->ltd_index, ea_off, infix, type, rc);
2387 return rc >= 0 ? 1 : rc;
2390 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2391 struct lfsck_component *com,
2392 const struct lu_fid *fid,
2395 struct lfsck_thread_info *info = lfsck_env_info(env);
2396 struct lfsck_request *lr = &info->lti_lr;
2397 struct lfsck_instance *lfsck = com->lc_lfsck;
2398 struct lfsck_tgt_desc *ltd;
2399 struct ptlrpc_request *req;
2400 struct lfsck_request *tmp;
2401 struct obd_export *exp;
2405 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2406 if (unlikely(ltd == NULL))
2410 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2411 GOTO(put, rc = -EOPNOTSUPP);
2413 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2415 GOTO(put, rc = -ENOMEM);
2417 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2419 ptlrpc_request_free(req);
2424 memset(lr, 0, sizeof(*lr));
2425 lr->lr_event = LE_CONDITIONAL_DESTROY;
2426 lr->lr_active = LFSCK_TYPE_LAYOUT;
2429 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2431 ptlrpc_request_set_replen(req);
2433 rc = ptlrpc_queue_wait(req);
2434 ptlrpc_req_finished(req);
2444 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2445 struct lfsck_component *com,
2446 struct lfsck_request *lr)
2448 struct lfsck_thread_info *info = lfsck_env_info(env);
2449 struct lu_attr *la = &info->lti_la;
2450 union ldlm_policy_data *policy = &info->lti_policy;
2451 struct ldlm_res_id *resid = &info->lti_resid;
2452 struct lfsck_instance *lfsck = com->lc_lfsck;
2453 struct dt_device *dev = lfsck->li_bottom;
2454 struct lu_fid *fid = &lr->lr_fid;
2455 struct dt_object *obj;
2456 struct thandle *th = NULL;
2457 struct lustre_handle lh = { 0 };
2462 obj = lfsck_object_find_by_dev(env, dev, fid);
2464 RETURN(PTR_ERR(obj));
2466 dt_read_lock(env, obj, 0);
2467 if (dt_object_exists(obj) == 0 ||
2468 lfsck_is_dead_obj(obj)) {
2469 dt_read_unlock(env, obj);
2471 GOTO(put, rc = -ENOENT);
2474 /* Get obj's attr without lock firstly. */
2475 rc = dt_attr_get(env, obj, la);
2476 dt_read_unlock(env, obj);
2480 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2481 GOTO(put, rc = -ETXTBSY);
2483 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2484 LASSERT(lfsck->li_namespace != NULL);
2486 memset(policy, 0, sizeof(*policy));
2487 policy->l_extent.end = OBD_OBJECT_EOF;
2488 ost_fid_build_resid(fid, resid);
2489 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2490 policy, LCK_EX, &flags, ldlm_blocking_ast,
2491 ldlm_completion_ast, NULL, NULL, 0,
2492 LVB_T_NONE, NULL, &lh);
2494 GOTO(put, rc = -EIO);
2496 dt_write_lock(env, obj, 0);
2497 /* Get obj's attr within lock again. */
2498 rc = dt_attr_get(env, obj, la);
2502 if (la->la_ctime != 0)
2503 GOTO(unlock, rc = -ETXTBSY);
2505 th = dt_trans_create(env, dev);
2507 GOTO(unlock, rc = PTR_ERR(th));
2509 rc = dt_declare_ref_del(env, obj, th);
2513 rc = dt_declare_destroy(env, obj, th);
2517 rc = dt_trans_start_local(env, dev, th);
2521 rc = dt_ref_del(env, obj, th);
2525 rc = dt_destroy(env, obj, th);
2527 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2528 "OST-object "DFID" that was created for reparing "
2529 "dangling referenced case. But the original missing "
2530 "OST-object is found now.\n",
2531 lfsck_lfsck2name(lfsck), PFID(fid));
2536 dt_trans_stop(env, dev, th);
2539 dt_write_unlock(env, obj);
2540 ldlm_lock_decref(&lh, LCK_EX);
2543 lfsck_object_put(env, obj);
2549 * Some OST-object has occupied the specified layout EA slot.
2550 * Such OST-object may be generated by the LFSCK when repair
2551 * dangling referenced MDT-object, which can be indicated by
2552 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2553 * is true and such OST-object has not been modified yet, we
2554 * will replace it with the orphan OST-object; otherwise the
2555 * LFSCK will create new MDT-object to reference the orphan.
2557 * \retval +1: repaired
2558 * \retval 0: did nothing
2559 * \retval -ve: on error
2561 static int lfsck_layout_conflict_create(const struct lu_env *env,
2562 struct lfsck_component *com,
2563 struct lfsck_tgt_desc *ltd,
2564 struct lu_orphan_rec_v2 *rec,
2565 struct dt_object *parent,
2566 struct lu_fid *cfid,
2567 struct lu_buf *ea_buf,
2568 struct lov_mds_md_v1 *lmm,
2569 struct lov_ost_data_v1 *slot,
2570 __u32 ea_off, int lovea_size)
2572 struct lfsck_thread_info *info = lfsck_env_info(env);
2573 struct lu_fid *cfid2 = &info->lti_fid2;
2574 struct ost_id *oi = &info->lti_oi;
2575 struct dt_device *dev = lfsck_obj2dev(parent);
2576 struct thandle *th = NULL;
2577 struct lustre_handle lh = { 0 };
2578 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2582 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2583 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2587 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2588 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2592 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2593 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2598 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2600 /* If the conflict OST-obejct is not created for fixing dangling
2601 * referenced MDT-object in former LFSCK check/repair, or it has
2602 * been modified by others, then we cannot destroy it. Re-create
2603 * a new MDT-object for the orphan OST-object. */
2604 if (rc == -ETXTBSY) {
2605 /* No need the layout lock on the original parent. */
2606 lfsck_ibits_unlock(&lh, LCK_EX);
2608 fid_zero(&rec->lor_rec.lor_fid);
2609 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2610 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2612 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2613 info->lti_tmpbuf, "C", ea_off);
2618 if (rc != 0 && rc != -ENOENT)
2621 th = dt_trans_create(env, dev);
2623 GOTO(unlock, rc = PTR_ERR(th));
2625 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2626 LU_XATTR_REPLACE, th);
2630 rc = dt_trans_start_local(env, dev, th);
2634 dt_write_lock(env, parent, 0);
2635 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2636 rc = lfsck_layout_refill_lovea(env, com->lc_lfsck, th, parent, cfid,
2637 ea_buf, lmm, slot, LU_XATTR_REPLACE,
2638 ltd->ltd_index, lovea_size);
2639 dt_write_unlock(env, parent);
2644 dt_trans_stop(env, dev, th);
2647 lfsck_ibits_unlock(&lh, LCK_EX);
2650 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2651 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2652 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2653 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2654 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2657 return rc >= 0 ? 1 : rc;
2661 * \retval +1: repaired
2662 * \retval 0: did nothing
2663 * \retval -ve: on error
2665 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2666 struct lfsck_component *com,
2667 struct lfsck_tgt_desc *ltd,
2668 struct lu_orphan_rec_v2 *rec,
2669 struct dt_object *parent,
2670 struct lu_fid *cfid,
2671 __u32 ost_idx, __u32 ea_off)
2673 struct lfsck_thread_info *info = lfsck_env_info(env);
2674 struct lu_buf *buf = &info->lti_big_buf;
2675 struct lu_fid *fid = &info->lti_fid2;
2676 struct ost_id *oi = &info->lti_oi;
2677 struct lfsck_instance *lfsck = com->lc_lfsck;
2678 struct dt_device *dt = lfsck_obj2dev(parent);
2679 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2680 struct ost_layout *ol = &rec->lor_layout;
2681 struct lov_comp_md_v1 *lcm = NULL;
2682 struct lov_comp_md_entry_v1 *lcme = NULL;
2683 struct thandle *handle = NULL;
2685 struct lov_mds_md_v1 *lmm;
2686 struct lov_ost_data_v1 *objs;
2687 struct lustre_handle lh = { 0 };
2695 bool locked = false;
2698 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2699 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2702 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2703 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2704 "stripe-index %u, comp_id %u, comp_start %llu, "
2705 "comp_end %llu: rc = %d\n",
2706 lfsck_lfsck2name(lfsck), PFID(cfid),
2707 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off,
2708 ol->ol_comp_id, ol->ol_comp_start,
2709 ol->ol_comp_end, rc);
2716 dt_write_unlock(env, parent);
2720 if (handle != NULL) {
2721 dt_trans_stop(env, dt, handle);
2726 GOTO(unlock_layout, rc);
2729 if (buf->lb_len < lovea_size) {
2730 lu_buf_realloc(buf, lovea_size);
2731 if (buf->lb_buf == NULL)
2732 GOTO(unlock_layout, rc = -ENOMEM);
2735 if (!(bk->lb_param & LPF_DRYRUN)) {
2736 handle = dt_trans_create(env, dt);
2738 GOTO(unlock_layout, rc = PTR_ERR(handle));
2740 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2745 rc = dt_trans_start_local(env, dt, handle);
2750 dt_write_lock(env, parent, 0);
2752 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2753 if (rc == -ERANGE) {
2754 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2757 } else if (rc == -ENODATA || rc == 0) {
2758 lovea_size = lfsck_lovea_size(ol, ea_off);
2759 /* If the declared is not big enough, re-try. */
2760 if (buf->lb_len < lovea_size) {
2764 fl = LU_XATTR_CREATE;
2765 } else if (rc < 0) {
2766 GOTO(unlock_parent, rc);
2767 } else if (unlikely(buf->lb_len == 0)) {
2770 fl = LU_XATTR_REPLACE;
2774 if (fl == LU_XATTR_CREATE) {
2775 if (bk->lb_param & LPF_DRYRUN)
2776 GOTO(unlock_parent, rc = 1);
2778 LASSERT(buf->lb_len >= lovea_size);
2780 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2781 cfid, buf, fl, ost_idx, ea_off);
2783 GOTO(unlock_parent, rc);
2787 rc1 = lfsck_layout_verify_header(parent, lmm);
2789 /* If the LOV EA crashed, the rebuild it. */
2790 if (rc1 == -EINVAL) {
2791 if (bk->lb_param & LPF_DRYRUN)
2792 GOTO(unlock_parent, rc = 1);
2794 LASSERT(buf->lb_len >= lovea_size);
2796 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2797 cfid, buf, fl, ost_idx, ea_off);
2799 GOTO(unlock_parent, rc);
2802 /* For other unknown magic/pattern, keep the current LOV EA. */
2804 GOTO(unlock_parent, rc = rc1);
2806 magic = le32_to_cpu(lmm->lmm_magic);
2807 if (magic == LOV_MAGIC_COMP_V1) {
2812 count = le16_to_cpu(lcm->lcm_entry_count);
2813 for (i = 0; i < count; i++) {
2814 lcme = &lcm->lcm_entries[i];
2815 start = le64_to_cpu(lcme->lcme_extent.e_start);
2816 end = le64_to_cpu(lcme->lcme_extent.e_end);
2818 if (end <= ol->ol_comp_start)
2821 if (start >= ol->ol_comp_end)
2824 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
2825 magic = le32_to_cpu(lmm->lmm_magic);
2826 flags = le32_to_cpu(lcme->lcme_flags);
2830 rc = lfsck_layout_add_comp_comp(env, lfsck, handle, ol, parent,
2831 cfid, buf, ost_idx, ea_off, i);
2833 GOTO(unlock_parent, rc);
2837 count = le16_to_cpu(lmm->lmm_stripe_count);
2839 GOTO(unlock_parent, rc = -EINVAL);
2842 /* Exceed the current end of MDT-object layout EA. Then extend it. */
2843 if (count <= ea_off) {
2844 if (bk->lb_param & LPF_DRYRUN)
2845 GOTO(unlock_parent, rc = 1);
2847 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2848 /* If the declared is not big enough, re-try. */
2849 if (buf->lb_len < lovea_size) {
2854 if (lcme && !(flags & LCME_FL_INIT))
2855 lcme->lcme_flags = cpu_to_le32(flags | LCME_FL_INIT);
2857 rc = lfsck_layout_extend_v1v3_lovea(env, lfsck, handle, ol,
2858 parent, cfid, buf, ost_idx, ea_off);
2860 GOTO(unlock_parent, rc);
2863 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2865 if (magic == LOV_MAGIC_V1) {
2866 objs = &lmm->lmm_objects[0];
2868 LASSERT(magic == LOV_MAGIC_V3);
2869 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2872 for (i = 0; i < count; i++, objs++) {
2873 /* The MDT-object was created via lfsck_layout_recover_create()
2874 * by others before, and we fill the dummy layout EA. */
2875 if ((lcme && !(flags & LCME_FL_INIT)) ||
2876 lovea_slot_is_dummy(objs)) {
2880 if (bk->lb_param & LPF_DRYRUN)
2881 GOTO(unlock_parent, rc = 1);
2883 lmm->lmm_layout_gen =
2884 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2888 if (le32_to_cpu(lmm->lmm_stripe_size) !=
2889 ol->ol_stripe_size ||
2890 le16_to_cpu(lmm->lmm_stripe_count) !=
2891 ol->ol_stripe_count ||
2892 le64_to_cpu(lcme->lcme_extent.e_start) !=
2893 ol->ol_comp_start ||
2894 le64_to_cpu(lcme->lcme_extent.e_end) !=
2896 CDEBUG(D_LFSCK, "%s: found invalid "
2897 "component for "DFID ": parent "DFID
2898 ", stripe-index %u, stripe_size %u, "
2899 "stripe_count %u, comp_id %u, "
2900 "comp_start %llu, comp_end %llu, "
2901 "cur_stripe_size %u, "
2902 "cur_stripe_count %u, "
2903 "cur_comp_start %llu, "
2904 "cur_comp_end %llu\n",
2905 lfsck_lfsck2name(lfsck), PFID(cfid),
2906 PFID(lfsck_dto2fid(parent)), ea_off,
2908 ol->ol_stripe_count, ol->ol_comp_id,
2909 ol->ol_comp_start, ol->ol_comp_end,
2910 le32_to_cpu(lmm->lmm_stripe_size),
2911 le16_to_cpu(lmm->lmm_stripe_count),
2912 le64_to_cpu(lcme->lcme_extent.e_start),
2913 le64_to_cpu(lcme->lcme_extent.e_end));
2915 GOTO(unlock_parent, rc = -EINVAL);
2918 le32_add_cpu(&lcm->lcm_layout_gen, 1);
2919 lovea_size = le32_to_cpu(lcm->lcm_size);
2920 if (!(flags & LCME_FL_INIT))
2921 lcme->lcme_flags = cpu_to_le32(flags |
2925 LASSERTF(buf->lb_len >= lovea_size,
2926 "buffer len %d is less than real size %d\n",
2927 (int)buf->lb_len, (int)lovea_size);
2929 rc = lfsck_layout_refill_lovea(env, lfsck, handle,
2930 parent, cfid, buf, lmm, objs,
2931 fl, ost_idx, lovea_size);
2933 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2934 "dummy layout slot for "DFID": parent "DFID
2935 ", OST-index %u, stripe-index %u: rc = %d\n",
2936 lfsck_lfsck2name(lfsck), PFID(cfid),
2937 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2939 GOTO(unlock_parent, rc);
2942 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2943 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2945 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2946 "invalid layout EA at the slot %d, index %u\n",
2947 lfsck_lfsck2name(lfsck),
2948 PFID(lfsck_dto2fid(parent)), i,
2949 le32_to_cpu(objs->l_ost_idx));
2951 GOTO(unlock_parent, rc);
2954 /* It should be rare case, the slot is there, but the LFSCK
2955 * does not handle it during the first-phase cycle scanning. */
2956 if (unlikely(lu_fid_eq(fid, cfid))) {
2958 GOTO(unlock_parent, rc = 0);
2960 /* Rare case that the OST-object index
2961 * does not match the parent MDT-object
2962 * layout EA. We trust the later one. */
2963 if (bk->lb_param & LPF_DRYRUN)
2964 GOTO(unlock_parent, rc = 1);
2966 dt_write_unlock(env, parent);
2968 dt_trans_stop(env, dt, handle);
2969 lfsck_ibits_unlock(&lh, LCK_EX);
2970 rc = lfsck_layout_update_pfid(env, com, parent,
2974 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2975 "updated OST-object's pfid for "DFID
2976 ": parent "DFID", OST-index %u, "
2977 "stripe-index %u: rc = %d\n",
2978 lfsck_lfsck2name(lfsck), PFID(cfid),
2979 PFID(lfsck_dto2fid(parent)),
2980 ltd->ltd_index, i, rc);
2987 /* The MDT-object exists, but related layout EA slot is occupied
2989 if (bk->lb_param & LPF_DRYRUN)
2990 GOTO(unlock_parent, rc = 1);
2992 dt_write_unlock(env, parent);
2994 dt_trans_stop(env, dt, handle);
2995 lfsck_ibits_unlock(&lh, LCK_EX);
2996 if (magic == LOV_MAGIC_V1)
2997 objs = &lmm->lmm_objects[ea_off];
2999 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
3000 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
3001 buf, lmm, objs, ea_off, lovea_size);
3007 dt_write_unlock(env, parent);
3011 dt_trans_stop(env, dt, handle);
3014 lfsck_ibits_unlock(&lh, LCK_EX);
3019 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
3020 struct lfsck_component *com,
3021 struct lfsck_tgt_desc *ltd,
3022 struct lu_orphan_rec_v2 *rec,
3023 struct lu_fid *cfid)
3025 struct lfsck_layout *lo = com->lc_file_ram;
3026 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
3027 struct dt_object *parent = NULL;
3028 __u32 ea_off = pfid->f_stripe_idx;
3032 if (!fid_is_sane(cfid))
3033 GOTO(out, rc = -EINVAL);
3036 if (fid_is_zero(pfid)) {
3037 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3042 if (!fid_is_sane(pfid))
3043 GOTO(out, rc = -EINVAL);
3045 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
3047 GOTO(out, rc = PTR_ERR(parent));
3049 if (unlikely(dt_object_remote(parent) != 0))
3050 GOTO(put, rc = -EXDEV);
3052 if (dt_object_exists(parent) == 0) {
3053 lfsck_object_put(env, parent);
3054 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3059 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
3060 GOTO(put, rc = -EISDIR);
3062 /* The orphan OST-object claims to be the parent's stripe, then
3063 * related dangling record in the trace file is meaningless. */
3064 rc = lfsck_layout_del_dangling_rec(env, com, pfid,
3065 rec->lor_layout.ol_comp_id, ea_off);
3066 if (rc && rc != -ENOENT)
3069 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
3070 ltd->ltd_index, ea_off);
3076 lfsck_object_put(env, parent);
3078 /* The layout EA is changed, need to be reloaded next time. */
3079 dt_object_put_nocache(env, parent);
3082 down_write(&com->lc_sem);
3083 com->lc_new_scanned++;
3084 com->lc_new_checked++;
3086 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
3088 } else if (rc < 0) {
3089 lo->ll_objs_failed_phase2++;
3091 up_write(&com->lc_sem);
3096 static int lfsck_layout_scan_orphan(const struct lu_env *env,
3097 struct lfsck_component *com,
3098 struct lfsck_tgt_desc *ltd)
3100 struct lfsck_assistant_data *lad = com->lc_data;
3101 struct lfsck_instance *lfsck = com->lc_lfsck;
3102 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3103 struct lfsck_thread_info *info = lfsck_env_info(env);
3104 struct lu_fid *fid = &info->lti_fid;
3105 struct dt_object *obj;
3106 const struct dt_it_ops *iops;
3111 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
3112 "scanning for OST%04x\n",
3113 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3115 if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
3116 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
3117 "scanning for OST%04x\n",
3118 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3123 fid->f_seq = fid_idif_seq(0, ltd->ltd_index);
3124 fid->f_oid = fid->f_ver = 0;
3126 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
3127 if (unlikely(IS_ERR(obj)))
3128 GOTO(log, rc = PTR_ERR(obj));
3130 rc = obj->do_ops->do_index_try(env, obj,
3131 &dt_lfsck_layout_orphan_features);
3135 iops = &obj->do_index_ops->dio_it;
3136 di = iops->init(env, obj, 0);
3138 GOTO(put, rc = PTR_ERR(di));
3140 rc = iops->load(env, di, 0);
3142 /* -ESRCH means that the orphan OST-objects rbtree has been
3143 * cleanup because of the OSS server restart or other errors. */
3144 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
3149 rc = iops->next(env, di);
3161 struct lu_orphan_rec_v2 *rec = &info->lti_rec;
3163 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3164 unlikely(!thread_is_running(&lfsck->li_thread)))
3167 key = iops->key(env, di);
3168 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
3169 /* Remote target OST may be runnning old LFSCK */
3170 memset(rec, 0, sizeof(*rec));
3171 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
3173 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
3174 &com->lc_fid_latest_scanned_phase2);
3175 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3178 lfsck_control_speed_by_self(com);
3180 rc = iops->next(env, di);
3181 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
3188 iops->fini(env, di);
3190 lfsck_object_put(env, obj);
3193 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
3194 "scanning for OST%04x: rc = %d\n",
3195 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
3197 return rc > 0 ? 0 : rc;
3200 static int lfsck_lmm2layout(struct lov_mds_md_v1 *lmm, struct ost_layout *ol,
3203 __u32 magic = le32_to_cpu(lmm->lmm_magic);
3207 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3) {
3208 ol->ol_stripe_size = lmm->lmm_stripe_size;
3209 ol->ol_stripe_count = lmm->lmm_stripe_count;
3210 ol->ol_comp_start = 0;
3211 ol->ol_comp_end = 0;
3213 } else if (magic == LOV_MAGIC_COMP_V1) {
3214 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
3215 struct lov_comp_md_entry_v1 *lcme = NULL;
3216 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3219 for (i = 0; i < count; i++) {
3220 lcme = &lcm->lcm_entries[i];
3221 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3222 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3229 /* The comp has been removed, do nothing. */
3233 lmm = (void *)lmm + le32_to_cpu(lcme->lcme_offset);
3234 ol->ol_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
3235 ol->ol_stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
3236 ol->ol_comp_start = le64_to_cpu(lcme->lcme_extent.e_start);
3237 ol->ol_comp_end = le64_to_cpu(lcme->lcme_extent.e_end);
3238 ol->ol_comp_id = le32_to_cpu(lcme->lcme_id);
3240 GOTO(out, rc = -EINVAL);
3250 * Repair the MDT-object with dangling LOV EA reference.
3252 * we need to repair the inconsistency according to the users' requirement:
3254 * 1) Keep the inconsistency there and report the inconsistency case,
3255 * then give the chance to the application to find related issues,
3256 * and the users can make the decision about how to handle it with
3257 * more human knownledge. (by default)
3259 * 2) Re-create the missing OST-object with the FID/owner information.
3261 * \param[in] env pointer to the thread context
3262 * \param[in] com the layout LFSCK component
3263 * \param[in] parent the MDT-object with dangling LOV EA reference
3264 * \param[in] child the OST-object to be created
3265 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3266 * \param[in] ea_off the offset of the OST-object in the LOV EA
3267 * \param[in] ost_idx the index of OST on which the OST-object resides
3269 * \retval +1 for repair successfully
3270 * \retval 0 for did nothing
3271 * \retval negative error number on failure
3273 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
3274 struct lfsck_component *com,
3275 struct dt_object *parent,
3276 struct dt_object *child,
3277 __u32 comp_id, __u32 ea_off,
3278 __u32 ost_idx, bool log)
3280 struct lfsck_thread_info *info = lfsck_env_info(env);
3281 struct filter_fid *ff = &info->lti_ff;
3282 struct ost_layout *ol = &ff->ff_layout;
3283 struct dt_object_format *dof = &info->lti_dof;
3284 struct lu_attr *la = &info->lti_la;
3285 struct lfsck_instance *lfsck = com->lc_lfsck;
3286 struct dt_device *dev = lfsck_obj2dev(child);
3287 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3288 const struct lu_fid *cfid = lfsck_dto2fid(child);
3289 struct lu_buf *tbuf = &info->lti_big_buf;
3290 struct thandle *handle;
3292 struct lustre_handle lh = { 0 };
3296 if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
3299 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3300 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3305 rc = dt_attr_get(env, parent, la);
3309 la->la_mode = S_IFREG | 0666;
3310 la->la_atime = la->la_mtime = la->la_ctime = 0;
3311 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
3312 LA_ATIME | LA_MTIME | LA_CTIME;
3313 memset(dof, 0, sizeof(*dof));
3314 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
3315 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
3316 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3317 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3318 * parent MDT-object's layout EA. */
3319 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
3321 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3325 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, comp_id);
3329 buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
3330 handle = dt_trans_create(env, dev);
3332 GOTO(unlock1, rc = PTR_ERR(handle));
3334 rc = dt_declare_create(env, child, la, NULL, dof, handle);
3338 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
3339 LU_XATTR_CREATE, handle);
3343 rc = dt_trans_start_local(env, dev, handle);
3347 dt_read_lock(env, parent, 0);
3348 if (unlikely(lfsck_is_dead_obj(parent)))
3349 GOTO(unlock2, rc = 0);
3351 if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
3352 struct ost_id *oi = &info->lti_oi;
3353 struct lu_fid *tfid = &info->lti_fid2;
3354 struct lu_buf *lovea = &info->lti_big_buf;
3355 struct lov_mds_md_v1 *lmm;
3356 struct lov_ost_data_v1 *objs;
3361 rc = lfsck_layout_get_lovea(env, parent, lovea);
3365 lmm = lovea->lb_buf;
3366 magic = le32_to_cpu(lmm->lmm_magic);
3367 if (magic == LOV_MAGIC_COMP_V1) {
3368 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3369 struct lov_comp_md_entry_v1 *lcme;
3370 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3373 for (i = 0; i < count; i++) {
3374 lcme = &lcm->lcm_entries[i];
3375 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3376 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3379 lmm = lovea->lb_buf +
3380 le32_to_cpu(lcme->lcme_offset);
3381 magic = le32_to_cpu(lmm->lmm_magic);
3386 /* Someone removed the component, do nothing. */
3387 GOTO(unlock2, rc = 0);
3391 count = le16_to_cpu(lmm->lmm_stripe_count);
3392 /* Someone changed the LOV EA, do nothing. */
3393 if (count <= ea_off)
3394 GOTO(unlock2, rc = 0);
3396 if (magic == LOV_MAGIC_V1) {
3397 objs = &lmm->lmm_objects[ea_off];
3399 LASSERT(magic == LOV_MAGIC_V3);
3401 objs = &((struct lov_mds_md_v3 *)lmm)->\
3402 lmm_objects[ea_off];
3405 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3406 idx2 = le32_to_cpu(objs->l_ost_idx);
3407 rc = ostid_to_fid(tfid, oi, idx2);
3408 /* Someone changed the LOV EA, do nothing. */
3409 if (rc != 0 || !lu_fid_eq(tfid, cfid))
3413 rc = dt_create(env, child, la, NULL, dof, handle);
3417 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
3423 dt_read_unlock(env, parent);
3426 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3429 lfsck_ibits_unlock(&lh, LCK_EX);
3433 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3434 "dangling reference for: parent "DFID", child "
3435 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: "
3437 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3438 comp_id, ea_off, ost_idx,
3439 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3440 "Create the lost OST-object as required" :
3441 "Keep the MDT-object there by default", rc);
3447 * Repair the MDT-object with dangling LOV EA reference.
3449 * Prepare parameters and call __lfsck_layout_repair_dangling()
3450 * to repair the dangling LOV EA reference.
3452 * \param[in] env pointer to the thread context
3453 * \param[in] com the layout LFSCK component
3454 * \param[in] pfid the MDT-object's FID
3455 * \param[in] cfid the FID for the OST-object to be created
3456 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3457 * \param[in] ea_off the offset of the OST-object in the LOV EA
3458 * \param[in] ost_idx the index of OST on which the OST-object resides
3460 * \retval +1 for repair successfully
3461 * \retval 0 for did nothing
3462 * \retval negative error number on failure
3464 static int lfsck_layout_repair_dangling(const struct lu_env *env,
3465 struct lfsck_component *com,
3466 const struct lu_fid *pfid,
3467 const struct lu_fid *cfid,
3468 __u32 comp_id, __u32 ea_off,
3471 struct lfsck_instance *lfsck = com->lc_lfsck;
3472 struct dt_object *parent = NULL;
3473 struct dt_object *child = NULL;
3474 struct lfsck_tgt_desc *ltd;
3478 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3480 GOTO(log, rc = PTR_ERR(parent));
3482 /* The MDT-object has been removed. */
3483 if (dt_object_exists(parent) == 0)
3486 ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
3487 if (unlikely(ltd == NULL))
3488 GOTO(log, rc = -ENODEV);
3490 child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
3492 GOTO(log, rc = PTR_ERR(child));
3494 /* The OST-object has been created. */
3495 if (unlikely(dt_object_exists(child) != 0))
3498 rc = __lfsck_layout_repair_dangling(env, com, parent, child,
3499 comp_id, ea_off, ost_idx, false);
3504 if (child != NULL && !IS_ERR(child))
3505 lfsck_object_put(env, child);
3507 if (parent != NULL && !IS_ERR(parent))
3508 lfsck_object_put(env, parent);
3511 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3512 "dangling reference for: parent "DFID", child "
3513 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: rc = %d\n",
3514 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3515 comp_id, ea_off, ost_idx,
3516 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3517 "Create the lost OST-object as required" :
3518 "Keep the MDT-object there by default", rc);
3523 /* If the OST-object does not recognize the MDT-object as its parent, and
3524 * there is no other MDT-object claims as its parent, then just trust the
3525 * given MDT-object as its parent. So update the OST-object filter_fid. */
3526 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
3527 struct lfsck_component *com,
3528 struct dt_object *parent,
3529 struct lfsck_layout_req *llr,
3532 struct lfsck_thread_info *info = lfsck_env_info(env);
3533 struct filter_fid *ff = &info->lti_ff;
3534 struct ost_layout *ol = &ff->ff_layout;
3535 struct dt_object *child = llr->llr_child;
3536 struct dt_device *dev = lfsck_obj2dev(child);
3537 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
3538 struct lu_buf *tbuf = &info->lti_big_buf;
3539 struct thandle *handle;
3541 struct lustre_handle lh = { 0 };
3545 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3546 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3551 ff->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3552 ff->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3553 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3554 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3555 * parent MDT-object's layout EA. */
3556 ff->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3558 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3562 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, llr->llr_comp_id);
3566 buf = lfsck_buf_get(env, ff, sizeof(*ff));
3568 handle = dt_trans_create(env, dev);
3570 GOTO(unlock1, rc = PTR_ERR(handle));
3572 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3576 rc = dt_attr_get(env, parent, la);
3580 la->la_valid = LA_UID | LA_GID;
3581 rc = dt_declare_attr_set(env, child, la, handle);
3585 rc = dt_trans_start_local(env, dev, handle);
3589 dt_write_lock(env, parent, 0);
3590 if (unlikely(lfsck_is_dead_obj(parent)))
3591 GOTO(unlock2, rc = 1);
3593 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3597 /* Get the latest parent's owner. */
3598 rc = dt_attr_get(env, parent, la);
3602 la->la_valid = LA_UID | LA_GID;
3603 rc = dt_attr_set(env, child, la, handle);
3608 dt_write_unlock(env, parent);
3611 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3614 lfsck_ibits_unlock(&lh, LCK_EX);
3618 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3619 "unmatched MDT-OST pair for: parent "DFID
3620 ", child "DFID", comp_id %u, OST-index %u, "
3621 "stripe-index %u, owner %u/%u: rc = %d\n",
3622 lfsck_lfsck2name(com->lc_lfsck),
3623 PFID(lfsck_dto2fid(parent)),
3624 PFID(lfsck_dto2fid(child)),
3625 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3626 la->la_uid, la->la_gid, rc);
3631 /* If there are more than one MDT-objects claim as the OST-object's parent,
3632 * and the OST-object only recognizes one of them, then we need to generate
3633 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3634 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3635 struct lfsck_component *com,
3636 struct dt_object *parent,
3637 struct lfsck_layout_req *llr,
3640 struct lfsck_thread_info *info = lfsck_env_info(env);
3641 struct dt_allocation_hint *hint = &info->lti_hint;
3642 struct dt_object_format *dof = &info->lti_dof;
3643 struct ost_id *oi = &info->lti_oi;
3644 struct lu_buf *buf = &info->lti_big_buf;
3645 struct lfsck_instance *lfsck = com->lc_lfsck;
3646 struct dt_device *dev;
3647 struct lu_device *d =
3648 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3649 struct lu_object *o;
3650 struct lu_object *n;
3651 struct dt_object *child = NULL;
3652 struct thandle *handle = NULL;
3653 struct lov_mds_md_v1 *lmm;
3654 struct lov_ost_data_v1 *objs;
3655 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3657 struct lustre_handle lh = { 0 };
3663 /* We use two separated transactions to repair the inconsistency.
3665 * 1) create the child (OST-object).
3666 * 2) update the parent LOV EA according to the child's FID.
3668 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3669 * handled as orphan when the layout LFSCK run next time.
3671 * If 1) failed, but 2) succeed, then such OST-object will be re-created
3672 * as dangling referened case when the layout LFSCK run next time. */
3674 /* The 1st transaction. */
3675 o = lu_object_anon(env, d, NULL);
3677 GOTO(log, rc = PTR_ERR(o));
3679 n = lu_object_locate(o->lo_header, d->ld_type);
3680 if (unlikely(n == NULL)) {
3681 lu_object_put_nocache(env, o);
3683 GOTO(log, rc = -EINVAL);
3686 child = container_of(n, struct dt_object, do_lu);
3687 memset(hint, 0, sizeof(*hint));
3688 rc = dt_attr_get(env, parent, la);
3692 la->la_valid = LA_UID | LA_GID;
3693 memset(dof, 0, sizeof(*dof));
3695 dev = lfsck_obj2dev(child);
3696 handle = dt_trans_create(env, dev);
3698 GOTO(log, rc = PTR_ERR(handle));
3700 rc = dt_declare_create(env, child, la, hint, dof, handle);
3704 rc = dt_trans_start_local(env, dev, handle);
3708 rc = dt_create(env, child, la, hint, dof, handle);
3709 dt_trans_stop(env, dev, handle);
3714 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3715 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3720 /* The 2nd transaction. */
3722 /* XXX: Generally, we should use bottom device (OSD) to update parent
3723 * LOV EA. But because the LOD-object still references the wrong
3724 * OSP-object that should be detached after the parent's LOV EA
3725 * refreshed. Unfortunately, there is no suitable API for that.
3726 * So we have to make the LOD to re-load the OSP-object(s) via
3727 * replacing the LOV EA against the LOD-object.
3729 * Once the DNE2 patches have been landed, we can replace the
3730 * LOD device with the OSD device. LU-6230. */
3732 dev = lfsck->li_next;
3733 parent = lfsck_object_locate(dev, parent);
3735 GOTO(log, rc = PTR_ERR(parent));
3737 handle = dt_trans_create(env, dev);
3739 GOTO(log, rc = PTR_ERR(handle));
3741 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3742 LU_XATTR_REPLACE, handle);
3746 rc = dt_trans_start_local(env, dev, handle);
3750 dt_write_lock(env, parent, 0);
3751 if (unlikely(lfsck_is_dead_obj(parent)))
3752 GOTO(unlock, rc = 0);
3754 rc = lfsck_layout_get_lovea(env, parent, buf);
3755 if (unlikely(!rc || rc == -ENODATA))
3756 GOTO(unlock, rc = 0);
3759 magic = le32_to_cpu(lmm->lmm_magic);
3760 if (magic == LOV_MAGIC_COMP_V1) {
3761 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3762 struct lov_comp_md_entry_v1 *lcme;
3763 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3766 LASSERT(llr->llr_comp_id != 0);
3768 for (i = 0; i < count; i++) {
3769 lcme = &lcm->lcm_entries[i];
3770 if (le32_to_cpu(lcme->lcme_id) == llr->llr_comp_id) {
3771 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3774 le32_add_cpu(&lcm->lcm_layout_gen, 1);
3776 le32_to_cpu(lcme->lcme_offset);
3777 magic = le32_to_cpu(lmm->lmm_magic);
3782 GOTO(unlock, rc = 0);
3786 if (magic == LOV_MAGIC_V1) {
3787 objs = &lmm->lmm_objects[llr->llr_lov_idx];
3789 LASSERT(magic == LOV_MAGIC_V3);
3791 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
3794 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3795 index = le32_to_cpu(objs->l_ost_idx);
3796 rc = ostid_to_fid(&tfid, oi, index);
3797 /* Someone changed layout during the LFSCK, no need to repair then. */
3798 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
3799 GOTO(unlock, rc = 0);
3801 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3802 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3803 ostid_cpu_to_le(oi, &objs->l_ost_oi);
3804 objs->l_ost_gen = cpu_to_le32(0);
3805 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3806 rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3807 LU_XATTR_REPLACE, handle);
3809 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3812 dt_write_unlock(env, parent);
3816 dt_trans_stop(env, dev, handle);
3819 lfsck_ibits_unlock(&lh, LCK_EX);
3821 lfsck_object_put(env, child);
3824 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3825 "multiple references for: parent "DFID", comp_id %u, "
3826 "OST-index %u, stripe-index %u, owner %u/%u: rc = %d\n",
3827 lfsck_lfsck2name(lfsck), PFID(pfid),
3828 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3829 la->la_uid, la->la_gid, rc);
3834 /* If the MDT-object and the OST-object have different owner information,
3835 * then trust the MDT-object, because the normal chown/chgrp handle order
3836 * is from MDT to OST, and it is possible that some chown/chgrp operation
3837 * is partly done. */
3838 static int lfsck_layout_repair_owner(const struct lu_env *env,
3839 struct lfsck_component *com,
3840 struct dt_object *parent,
3841 struct lfsck_layout_req *llr,
3842 struct lu_attr *pla,
3843 const struct lu_attr *cla)
3845 struct lfsck_thread_info *info = lfsck_env_info(env);
3846 struct lu_attr *tla = &info->lti_la2;
3847 struct dt_object *child = llr->llr_child;
3848 struct dt_device *dev = lfsck_obj2dev(child);
3849 struct thandle *handle;
3853 tla->la_uid = pla->la_uid;
3854 tla->la_gid = pla->la_gid;
3855 tla->la_valid = LA_UID | LA_GID;
3856 handle = dt_trans_create(env, dev);
3858 GOTO(log, rc = PTR_ERR(handle));
3860 rc = dt_declare_attr_set(env, child, tla, handle);
3864 rc = dt_trans_start_local(env, dev, handle);
3868 /* Use the dt_object lock to serialize with destroy and attr_set. */
3869 dt_read_lock(env, parent, 0);
3870 if (unlikely(lfsck_is_dead_obj(parent)))
3871 GOTO(unlock, rc = 1);
3873 /* Get the latest parent's owner. */
3874 rc = dt_attr_get(env, parent, pla);
3878 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3879 if (unlikely(tla->la_uid != pla->la_uid ||
3880 tla->la_gid != pla->la_gid))
3883 rc = dt_attr_set(env, child, tla, handle);
3888 dt_read_unlock(env, parent);
3891 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3895 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3896 "inconsistent file owner for: parent "DFID", child "DFID
3897 ", OST-index %u, stripe-index %u, old owner %u/%u, "
3898 "new owner %u/%u: rc = %d\n",
3899 lfsck_lfsck2name(com->lc_lfsck),
3900 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3901 llr->llr_ost_idx, llr->llr_lov_idx,
3902 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3907 /* Check whether the OST-object correctly back points to the
3908 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3909 static int lfsck_layout_check_parent(const struct lu_env *env,
3910 struct lfsck_component *com,
3911 struct lfsck_assistant_object *lso,
3912 struct filter_fid *ff,
3913 const struct lu_fid *cfid,
3914 const struct lu_attr *cla,
3915 struct lfsck_layout_req *llr)
3917 struct lfsck_thread_info *info = lfsck_env_info(env);
3918 struct lu_buf *buf = &info->lti_big_buf;
3919 struct lu_fid *pfid = &info->lti_fid;
3920 struct dt_object *tobj;
3921 struct lov_mds_md_v1 *lmm;
3922 struct lov_ost_data_v1 *objs;
3923 struct lustre_handle lh = { 0 };
3931 *pfid = ff->ff_parent;
3932 idx = pfid->f_stripe_idx;
3935 if (unlikely(!fid_is_sane(pfid)))
3936 RETURN(LLIT_UNMATCHED_PAIR);
3938 if (lu_fid_eq(pfid, &lso->lso_fid)) {
3939 if (likely(llr->llr_lov_idx == idx))
3942 RETURN(LLIT_UNMATCHED_PAIR);
3945 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3947 RETURN(PTR_ERR(tobj));
3949 if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3950 !S_ISREG(lfsck_object_type(tobj)))
3951 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3953 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3954 * remote one on another MDT. Then check whether the given OST-object
3955 * is in such layout. If yes, it is multiple referenced, otherwise it
3956 * is unmatched referenced case. */
3957 rc = lfsck_layout_get_lovea(env, tobj, buf);
3958 if (rc == 0 || rc == -ENOENT)
3959 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3965 magic = le32_to_cpu(lmm->lmm_magic);
3966 if (magic == LOV_MAGIC_COMP_V1) {
3967 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3968 struct lov_comp_md_entry_v1 *lcme;
3970 if (ff->ff_layout.ol_comp_id == 0)
3971 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3973 count = le16_to_cpu(lcm->lcm_entry_count);
3974 for (i = 0; i < count; i++) {
3975 lcme = &lcm->lcm_entries[i];
3976 if (le32_to_cpu(lcme->lcme_id) ==
3977 ff->ff_layout.ol_comp_id) {
3979 le32_to_cpu(lcme->lcme_offset);
3980 magic = le32_to_cpu(lmm->lmm_magic);
3981 if (!(le32_to_cpu(lcme->lcme_flags) &
3983 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3989 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3993 if (magic == LOV_MAGIC_V1) {
3994 objs = &lmm->lmm_objects[0];
3996 LASSERT(magic == LOV_MAGIC_V3);
3997 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4000 count = le16_to_cpu(lmm->lmm_stripe_count);
4001 for (i = 0; i < count; i++, objs++) {
4002 struct lu_fid *tfid = &info->lti_fid2;
4003 struct ost_id *oi = &info->lti_oi;
4006 if (lovea_slot_is_dummy(objs))
4009 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4010 idx2 = le32_to_cpu(objs->l_ost_idx);
4011 rc = ostid_to_fid(tfid, oi, idx2);
4013 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
4014 "invalid layout EA at the slot %d, index %u\n",
4015 lfsck_lfsck2name(com->lc_lfsck),
4016 PFID(pfid), i, idx2);
4018 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4021 if (lu_fid_eq(cfid, tfid)) {
4022 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
4023 MDS_INODELOCK_UPDATE |
4024 MDS_INODELOCK_LAYOUT |
4025 MDS_INODELOCK_XATTR,
4030 dt_read_lock(env, tobj, 0);
4032 /* For local MDT-object, re-check existence
4033 * after taken the lock. */
4034 if (!dt_object_remote(tobj)) {
4035 if (dt_object_exists(tobj) == 0 ||
4036 lfsck_is_dead_obj(tobj))
4037 rc = LLIT_UNMATCHED_PAIR;
4039 rc = LLIT_MULTIPLE_REFERENCED;
4044 /* For migration case, the new MDT-object and old
4045 * MDT-object may reference the same OST-object at
4046 * some migration internal time.
4048 * For remote MDT-object, the local MDT may not know
4049 * whether it has been removed or not. Try checking
4050 * for a non-existent xattr to check if this object
4051 * has been been removed or not. */
4052 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
4054 if (unlikely(rc == -ENOENT || rc >= 0))
4055 rc = LLIT_UNMATCHED_PAIR;
4056 else if (rc == -ENODATA)
4057 rc = LLIT_MULTIPLE_REFERENCED;
4063 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4066 if (lustre_handle_is_used(&lh)) {
4067 dt_read_unlock(env, tobj);
4068 lfsck_ibits_unlock(&lh, LCK_EX);
4072 lfsck_object_put(env, tobj);
4077 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
4078 struct lfsck_component *com,
4079 struct lfsck_assistant_req *lar)
4081 struct lfsck_layout_req *llr =
4082 container_of0(lar, struct lfsck_layout_req, llr_lar);
4083 struct lfsck_assistant_object *lso = lar->lar_parent;
4084 struct lfsck_layout *lo = com->lc_file_ram;
4085 struct lfsck_thread_info *info = lfsck_env_info(env);
4086 struct filter_fid *ff = &info->lti_ff;
4087 struct lu_buf buf = { .lb_buf = ff,
4088 .lb_len = sizeof(*ff) };
4089 struct dt_object *parent = NULL;
4090 struct dt_object *child = llr->llr_child;
4091 struct lu_attr *pla = &lso->lso_attr;
4092 struct lu_attr *cla = &info->lti_la;
4093 struct lfsck_instance *lfsck = com->lc_lfsck;
4094 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4095 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
4102 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ENGINE_DELAY, cfs_fail_val);
4104 rc = dt_attr_get(env, child, cla);
4105 if (rc == -ENOENT) {
4106 parent = lfsck_assistant_object_load(env, lfsck, lso);
4107 if (IS_ERR(parent)) {
4108 rc = PTR_ERR(parent);
4110 RETURN(rc == -ENOENT ? 0 : rc);
4113 type = LLIT_DANGLING;
4120 lfsck_buf_init(&buf, ff, sizeof(*ff));
4121 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
4122 if (unlikely(rc > 0 && rc < sizeof(struct lu_fid))) {
4123 type = LLIT_UNMATCHED_PAIR;
4127 if (rc < 0 && rc != -ENODATA)
4130 if (rc == 0 || rc == -ENODATA)
4131 GOTO(check_owner, rc = 0);
4133 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
4134 rc = lfsck_layout_check_parent(env, com, lso, ff,
4135 lu_object_fid(&child->do_lu), cla, llr);
4145 /* Someone may has changed the owner after the parent attr pre-loaded.
4146 * It can be handled later inside the lfsck_layout_repair_owner(). */
4147 if (unlikely(cla->la_uid != pla->la_uid ||
4148 cla->la_gid != pla->la_gid)) {
4149 type = LLIT_INCONSISTENT_OWNER;
4154 if (type == LLIT_NONE)
4157 if (bk->lb_param & LPF_DRYRUN)
4160 if (parent == NULL) {
4161 parent = lfsck_assistant_object_load(env, lfsck, lso);
4162 if (IS_ERR(parent)) {
4163 rc = PTR_ERR(parent);
4174 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
4175 rc = lfsck_layout_ins_dangling_rec(env, com,
4176 lfsck_dto2fid(parent), lfsck_dto2fid(child),
4177 llr->llr_comp_id, llr->llr_lov_idx,
4180 rc = __lfsck_layout_repair_dangling(env, com, parent,
4187 case LLIT_UNMATCHED_PAIR:
4188 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
4191 case LLIT_MULTIPLE_REFERENCED:
4192 rc = lfsck_layout_repair_multiple_references(env, com, parent,
4195 case LLIT_INCONSISTENT_OWNER:
4196 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
4206 down_write(&com->lc_sem);
4208 struct lfsck_assistant_data *lad = com->lc_data;
4210 if (unlikely(lad->lad_exit)) {
4212 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
4213 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
4214 rc == -EHOSTUNREACH) {
4215 /* If cannot touch the target server,
4216 * mark the LFSCK as INCOMPLETE. */
4217 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
4218 "talk with OST %x: rc = %d\n",
4219 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
4220 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
4221 lo->ll_objs_skipped++;
4224 lfsck_layout_record_failure(env, lfsck, lo);
4226 } else if (rc > 0 && (type != LLIT_DANGLING ||
4227 !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
4228 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
4229 "unknown type = %d\n", type);
4231 lo->ll_objs_repaired[type - 1]++;
4232 if (bk->lb_param & LPF_DRYRUN &&
4233 unlikely(lo->ll_pos_first_inconsistent == 0))
4234 lo->ll_pos_first_inconsistent =
4235 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
4238 up_write(&com->lc_sem);
4240 if (parent != NULL && !IS_ERR(parent))
4241 lfsck_object_put(env, parent);
4247 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
4248 struct lfsck_component *com,
4249 struct dt_object *obj, bool first)
4251 struct lfsck_instance *lfsck = com->lc_lfsck;
4252 struct ptlrpc_thread *thread = &lfsck->li_thread;
4253 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4254 struct lfsck_layout *lo = com->lc_file_ram;
4255 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
4258 struct lfsck_layout_dangling_key *parent =
4259 &lfsck_env_info(env)->lti_lldk;
4260 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
4265 di = iops->init(env, obj, 0);
4267 RETURN(PTR_ERR(di));
4270 lldk_cpu_to_be(parent, &lo->ll_lldk_latest_scanned_phase2);
4272 memset(parent, 0, sizeof(*parent));
4273 rc = iops->get(env, di, (const struct dt_key *)parent);
4278 /* The start one either has been processed or does not exist,
4280 rc = iops->next(env, di);
4286 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
4287 unlikely(!thread_is_running(thread)))
4290 key = iops->key(env, di);
4299 lldk_be_to_cpu(parent,
4300 (const struct lfsck_layout_dangling_key *)key);
4301 if (!fid_is_sane(&parent->lldk_fid)) {
4306 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
4308 fid_be_to_cpu(cfid, cfid);
4309 ost_idx = cfid->f_ver;
4311 if (!fid_is_sane(cfid)) {
4316 rc = lfsck_layout_repair_dangling(env, com,
4317 &parent->lldk_fid, cfid,
4318 parent->lldk_comp_id,
4319 parent->lldk_ea_off, ost_idx);
4323 down_write(&com->lc_sem);
4324 com->lc_new_checked++;
4325 com->lc_new_scanned++;
4327 lo->ll_lldk_latest_scanned_phase2 = *parent;
4330 lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
4332 lo->ll_objs_failed_phase2++;
4333 up_write(&com->lc_sem);
4335 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4338 if (unlikely(com->lc_time_next_checkpoint <=
4339 ktime_get_seconds()) &&
4340 com->lc_new_checked != 0) {
4341 down_write(&com->lc_sem);
4342 lo->ll_run_time_phase2 += ktime_get_seconds() -
4343 com->lc_time_last_checkpoint;
4344 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
4345 lo->ll_objs_checked_phase2 += com->lc_new_checked;
4346 com->lc_new_checked = 0;
4347 lfsck_layout_store(env, com);
4348 up_write(&com->lc_sem);
4350 com->lc_time_last_checkpoint = ktime_get_seconds();
4351 com->lc_time_next_checkpoint =
4352 com->lc_time_last_checkpoint +
4353 LFSCK_CHECKPOINT_INTERVAL;
4356 lfsck_control_speed_by_self(com);
4357 if (unlikely(!thread_is_running(thread)))
4360 rc = iops->next(env, di);
4369 iops->fini(env, di);
4374 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
4375 struct lfsck_component *com)
4377 struct lfsck_assistant_data *lad = com->lc_data;
4378 struct lfsck_instance *lfsck = com->lc_lfsck;
4379 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4380 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4381 struct lfsck_tgt_desc *ltd;
4385 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
4386 lfsck_lfsck2name(lfsck));
4388 spin_lock(<ds->ltd_lock);
4389 while (!list_empty(&lad->lad_ost_phase2_list)) {
4390 ltd = list_entry(lad->lad_ost_phase2_list.next,
4391 struct lfsck_tgt_desc,
4392 ltd_layout_phase_list);
4393 list_del_init(<d->ltd_layout_phase_list);
4394 if (bk->lb_param & LPF_OST_ORPHAN) {
4395 spin_unlock(<ds->ltd_lock);
4396 rc = lfsck_layout_scan_orphan(env, com, ltd);
4397 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
4400 if (unlikely(lad->lad_exit ||
4401 !thread_is_running(&lfsck->li_thread)))
4403 spin_lock(<ds->ltd_lock);
4407 if (list_empty(&lad->lad_ost_phase1_list))
4411 spin_unlock(<ds->ltd_lock);
4413 if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
4414 struct lfsck_layout *lo = com->lc_file_ram;
4417 com->lc_new_checked = 0;
4418 com->lc_new_scanned = 0;
4419 com->lc_time_last_checkpoint = ktime_get_seconds();
4420 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4421 LFSCK_CHECKPOINT_INTERVAL;
4423 i = lfsck_sub_trace_file_fid2idx(
4424 &lo->ll_lldk_latest_scanned_phase2.lldk_fid);
4425 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4426 com->lc_sub_trace_objs[i].lsto_obj, true);
4427 while (rc > 0 && ++i < LFSCK_STF_COUNT)
4428 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4429 com->lc_sub_trace_objs[i].lsto_obj, false);
4431 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
4432 "at the No. %d trace file: rc = %d\n",
4433 lfsck_lfsck2name(lfsck), i, rc);
4436 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
4437 lfsck_lfsck2name(lfsck), rc);
4443 lfsck_layout_slave_async_interpret(const struct lu_env *env,
4444 struct ptlrpc_request *req,
4447 struct lfsck_layout_slave_async_args *llsaa = args;
4448 struct obd_export *exp = llsaa->llsaa_exp;
4449 struct lfsck_component *com = llsaa->llsaa_com;
4450 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
4451 struct lfsck_layout_slave_data *llsd = com->lc_data;
4452 struct lfsck_reply *lr = NULL;
4456 /* It is probably caused by network trouble, or target crash,
4457 * it will try several times (depends on the obd_timeout, and
4458 * will not less than 3 times). But to make the LFSCK can go
4459 * ahead, we should not try for ever. After some try but still
4460 * hit failure, it will assume that the target exit the LFSCK
4461 * prcoessing and stop try. */
4462 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
4463 int max_try = max_t(int, obd_timeout / 30, 3);
4465 if (++(llst->llst_failures) > max_try)
4471 llst->llst_failures = 0;
4472 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
4473 if (lr->lr_status != LS_SCANNING_PHASE1 &&
4474 lr->lr_status != LS_SCANNING_PHASE2)
4479 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
4480 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
4481 llst->llst_index, lr != NULL ? lr->lr_status : rc,
4482 llst->llst_failures);
4484 lfsck_layout_llst_del(llsd, llst);
4487 lfsck_layout_llst_put(llst);
4488 lfsck_component_put(env, com);
4489 class_export_put(exp);
4494 static int lfsck_layout_async_query(const struct lu_env *env,
4495 struct lfsck_component *com,
4496 struct obd_export *exp,
4497 struct lfsck_layout_slave_target *llst,
4498 struct lfsck_request *lr,
4499 struct ptlrpc_request_set *set)
4501 struct lfsck_layout_slave_async_args *llsaa;
4502 struct ptlrpc_request *req;
4503 struct lfsck_request *tmp;
4507 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
4511 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
4513 ptlrpc_request_free(req);
4517 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4519 ptlrpc_request_set_replen(req);
4521 llsaa = ptlrpc_req_async_args(req);
4522 llsaa->llsaa_exp = exp;
4523 llsaa->llsaa_com = lfsck_component_get(com);
4524 llsaa->llsaa_llst = llst;
4525 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
4526 req->rq_allow_intr = 1;
4527 req->rq_no_delay = 1;
4528 ptlrpc_set_add_req(set, req);
4533 static int lfsck_layout_async_notify(const struct lu_env *env,
4534 struct obd_export *exp,
4535 struct lfsck_request *lr,
4536 struct ptlrpc_request_set *set)
4538 struct ptlrpc_request *req;
4539 struct lfsck_request *tmp;
4543 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4547 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4549 ptlrpc_request_free(req);
4553 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4555 ptlrpc_request_set_replen(req);
4556 req->rq_allow_intr = 1;
4557 req->rq_no_delay = 1;
4558 ptlrpc_set_add_req(set, req);
4564 lfsck_layout_slave_query_master(const struct lu_env *env,
4565 struct lfsck_component *com)
4567 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4568 struct lfsck_instance *lfsck = com->lc_lfsck;
4569 struct lfsck_layout_slave_data *llsd = com->lc_data;
4570 struct lfsck_layout_slave_target *llst;
4571 struct obd_export *exp;
4572 struct ptlrpc_request_set *set;
4577 set = ptlrpc_prep_set();
4579 GOTO(log, rc = -ENOMEM);
4581 memset(lr, 0, sizeof(*lr));
4582 lr->lr_event = LE_QUERY;
4583 lr->lr_active = LFSCK_TYPE_LAYOUT;
4585 llsd->llsd_touch_gen++;
4586 spin_lock(&llsd->llsd_lock);
4587 while (!list_empty(&llsd->llsd_master_list)) {
4588 llst = list_entry(llsd->llsd_master_list.next,
4589 struct lfsck_layout_slave_target,
4591 if (llst->llst_gen == llsd->llsd_touch_gen)
4594 llst->llst_gen = llsd->llsd_touch_gen;
4595 list_move_tail(&llst->llst_list,
4596 &llsd->llsd_master_list);
4597 atomic_inc(&llst->llst_ref);
4598 spin_unlock(&llsd->llsd_lock);
4600 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4603 lfsck_layout_llst_del(llsd, llst);
4604 lfsck_layout_llst_put(llst);
4605 spin_lock(&llsd->llsd_lock);
4609 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
4611 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4612 "query %s for layout: rc = %d\n",
4613 lfsck_lfsck2name(lfsck),
4614 exp->exp_obd->obd_name, rc);
4617 lfsck_layout_llst_put(llst);
4618 class_export_put(exp);
4620 spin_lock(&llsd->llsd_lock);
4622 spin_unlock(&llsd->llsd_lock);
4624 rc = ptlrpc_set_wait(set);
4625 ptlrpc_set_destroy(set);
4627 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4630 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4631 lfsck_lfsck2name(com->lc_lfsck), rc);
4637 lfsck_layout_slave_notify_master(const struct lu_env *env,
4638 struct lfsck_component *com,
4639 enum lfsck_events event, int result)
4641 struct lfsck_layout *lo = com->lc_file_ram;
4642 struct lfsck_instance *lfsck = com->lc_lfsck;
4643 struct lfsck_layout_slave_data *llsd = com->lc_data;
4644 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4645 struct lfsck_layout_slave_target *llst;
4646 struct obd_export *exp;
4647 struct ptlrpc_request_set *set;
4651 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4652 lfsck_lfsck2name(com->lc_lfsck));
4654 set = ptlrpc_prep_set();
4658 memset(lr, 0, sizeof(*lr));
4659 lr->lr_event = event;
4660 lr->lr_flags = LEF_FROM_OST;
4661 lr->lr_status = result;
4662 lr->lr_index = lfsck_dev_idx(lfsck);
4663 lr->lr_active = LFSCK_TYPE_LAYOUT;
4664 lr->lr_flags2 = lo->ll_flags;
4665 llsd->llsd_touch_gen++;
4666 spin_lock(&llsd->llsd_lock);
4667 while (!list_empty(&llsd->llsd_master_list)) {
4668 llst = list_entry(llsd->llsd_master_list.next,
4669 struct lfsck_layout_slave_target,
4671 if (llst->llst_gen == llsd->llsd_touch_gen)
4674 llst->llst_gen = llsd->llsd_touch_gen;
4675 list_move_tail(&llst->llst_list,
4676 &llsd->llsd_master_list);
4677 atomic_inc(&llst->llst_ref);
4678 spin_unlock(&llsd->llsd_lock);
4680 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4683 lfsck_layout_llst_del(llsd, llst);
4684 lfsck_layout_llst_put(llst);
4685 spin_lock(&llsd->llsd_lock);
4689 rc = lfsck_layout_async_notify(env, exp, lr, set);
4691 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4692 "notify %s for layout: rc = %d\n",
4693 lfsck_lfsck2name(lfsck),
4694 exp->exp_obd->obd_name, rc);
4696 lfsck_layout_llst_put(llst);
4697 class_export_put(exp);
4698 spin_lock(&llsd->llsd_lock);
4700 spin_unlock(&llsd->llsd_lock);
4702 ptlrpc_set_wait(set);
4703 ptlrpc_set_destroy(set);
4709 * \ret -ENODATA: unrecognized stripe
4710 * \ret = 0 : recognized stripe
4711 * \ret < 0 : other failures
4713 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4714 struct lfsck_component *com,
4715 struct lu_fid *cfid,
4716 struct lu_fid *pfid, __u32 comp_id)
4718 struct lfsck_thread_info *info = lfsck_env_info(env);
4719 struct lu_buf *buf = &info->lti_big_buf;
4720 struct ost_id *oi = &info->lti_oi;
4721 struct dt_object *obj;
4722 struct lov_mds_md_v1 *lmm;
4723 struct lov_ost_data_v1 *objs;
4724 __u32 idx = pfid->f_stripe_idx;
4732 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4734 RETURN(PTR_ERR(obj));
4736 dt_read_lock(env, obj, 0);
4737 if (unlikely(dt_object_exists(obj) == 0 ||
4738 lfsck_is_dead_obj(obj)))
4739 GOTO(unlock, rc = -ENOENT);
4741 if (!S_ISREG(lfsck_object_type(obj)))
4742 GOTO(unlock, rc = -ENODATA);
4744 rc = lfsck_layout_get_lovea(env, obj, buf);
4749 GOTO(unlock, rc = -ENODATA);
4752 magic = le32_to_cpu(lmm->lmm_magic);
4753 if (magic == LOV_MAGIC_COMP_V1) {
4754 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4755 struct lov_comp_md_entry_v1 *lcme;
4758 GOTO(unlock, rc = -ENODATA);
4760 count = le16_to_cpu(lcm->lcm_entry_count);
4761 for (i = 0; i < count; i++) {
4762 lcme = &lcm->lcm_entries[i];
4763 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
4765 le32_to_cpu(lcme->lcme_offset);
4766 magic = le32_to_cpu(lmm->lmm_magic);
4767 if (!(le32_to_cpu(lcme->lcme_flags) &
4769 GOTO(unlock, rc = -ENODATA);
4775 GOTO(unlock, rc = -ENODATA);
4779 if (magic == LOV_MAGIC_V1) {
4780 objs = &lmm->lmm_objects[0];
4782 LASSERT(magic == LOV_MAGIC_V3);
4783 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4786 fid_to_ostid(cfid, oi);
4787 count = le16_to_cpu(lmm->lmm_stripe_count);
4788 for (i = 0; i < count; i++, objs++) {
4791 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
4792 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
4793 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
4796 GOTO(unlock, rc = -ENODATA);
4799 dt_read_unlock(env, obj);
4800 lfsck_object_put(env, obj);
4806 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
4807 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
4808 * layout EA from MDT to OST. On one hand, the OST no need to understand
4809 * the layout EA structure; on the other hand, it may cause trouble when
4810 * transfer large layout EA from MDT to OST via normal OUT RPC.
4812 * \ret > 0: unrecognized stripe
4813 * \ret = 0: recognized stripe
4814 * \ret < 0: other failures
4816 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
4817 struct lfsck_component *com,
4818 struct lu_fid *cfid,
4819 struct lu_fid *pfid, __u32 comp_id)
4821 struct lfsck_instance *lfsck = com->lc_lfsck;
4822 struct obd_device *obd = lfsck->li_obd;
4823 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4824 struct obd_export *exp = NULL;
4825 struct ptlrpc_request *req = NULL;
4826 struct lfsck_request *lr;
4827 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
4831 if (unlikely(fid_is_idif(pfid)))
4834 fld_range_set_any(range);
4835 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
4837 RETURN(rc == -ENOENT ? 1 : rc);
4839 if (unlikely(!fld_range_is_mdt(range)))
4842 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
4843 if (unlikely(exp == NULL))
4846 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
4847 GOTO(out, rc = -EOPNOTSUPP);
4849 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4851 GOTO(out, rc = -ENOMEM);
4853 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4855 ptlrpc_request_free(req);
4860 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4861 memset(lr, 0, sizeof(*lr));
4862 lr->lr_event = LE_PAIRS_VERIFY;
4863 lr->lr_active = LFSCK_TYPE_LAYOUT;
4864 lr->lr_fid = *cfid; /* OST-object itself FID. */
4865 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
4866 lr->lr_comp_id = comp_id;
4868 ptlrpc_request_set_replen(req);
4869 rc = ptlrpc_queue_wait(req);
4870 ptlrpc_req_finished(req);
4872 if (rc == -ENOENT || rc == -ENODATA)
4879 class_export_put(exp);
4884 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
4885 struct lfsck_component *com,
4886 struct lfsck_req_local *lrl)
4888 struct dt_object *obj;
4892 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lrl->lrl_fid);
4894 GOTO(log, rc = PTR_ERR(obj));
4896 dt_write_lock(env, obj, 0);
4897 if (unlikely(dt_object_exists(obj) == 0 ||
4898 lfsck_is_dead_obj(obj)))
4899 GOTO(unlock, rc = 0);
4901 rc = __lfsck_layout_update_pfid(env, obj, &lrl->lrl_ff_client.ff_parent,
4902 &lrl->lrl_ff_client.ff_layout,
4903 lrl->lrl_ff_client.ff_parent.f_ver);
4908 dt_write_unlock(env, obj);
4909 lfsck_object_put(env, obj);
4912 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
4913 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
4914 PFID(&lrl->lrl_fid), PFID(&lrl->lrl_ff_client.ff_parent), rc);
4921 static void lfsck_layout_slave_quit(const struct lu_env *env,
4922 struct lfsck_component *com);
4924 static int lfsck_layout_reset(const struct lu_env *env,
4925 struct lfsck_component *com, bool init)
4927 struct lfsck_layout *lo = com->lc_file_ram;
4930 down_write(&com->lc_sem);
4932 memset(lo, 0, com->lc_file_size);
4934 __u32 count = lo->ll_success_count;
4935 time64_t last_time = lo->ll_time_last_complete;
4937 memset(lo, 0, com->lc_file_size);
4938 lo->ll_success_count = count;
4939 lo->ll_time_last_complete = last_time;
4942 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
4943 lo->ll_status = LS_INIT;
4945 if (com->lc_lfsck->li_master) {
4946 struct lfsck_assistant_data *lad = com->lc_data;
4948 lad->lad_incomplete = 0;
4949 CFS_RESET_BITMAP(lad->lad_bitmap);
4952 rc = lfsck_layout_store(env, com);
4953 if (rc == 0 && com->lc_lfsck->li_master)
4954 rc = lfsck_load_sub_trace_files(env, com,
4955 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
4956 up_write(&com->lc_sem);
4958 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
4959 lfsck_lfsck2name(com->lc_lfsck), rc);
4964 static void lfsck_layout_fail(const struct lu_env *env,
4965 struct lfsck_component *com, bool new_checked)
4967 struct lfsck_layout *lo = com->lc_file_ram;
4969 down_write(&com->lc_sem);
4971 com->lc_new_checked++;
4972 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
4973 up_write(&com->lc_sem);
4976 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
4977 struct lfsck_component *com, bool init)
4979 struct lfsck_instance *lfsck = com->lc_lfsck;
4980 struct lfsck_layout *lo = com->lc_file_ram;
4984 rc = lfsck_checkpoint_generic(env, com);
4986 return rc > 0 ? 0 : rc;
4989 down_write(&com->lc_sem);
4991 lo->ll_pos_latest_start =
4992 lfsck->li_pos_checkpoint.lp_oit_cookie;
4994 lo->ll_pos_last_checkpoint =
4995 lfsck->li_pos_checkpoint.lp_oit_cookie;
4996 lo->ll_run_time_phase1 += ktime_get_seconds() -
4997 lfsck->li_time_last_checkpoint;
4998 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
4999 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5000 com->lc_new_checked = 0;
5003 rc = lfsck_layout_store(env, com);
5004 up_write(&com->lc_sem);
5006 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
5007 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5008 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5013 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
5014 struct lfsck_component *com, bool init)
5016 struct lfsck_instance *lfsck = com->lc_lfsck;
5017 struct lfsck_layout *lo = com->lc_file_ram;
5020 if (com->lc_new_checked == 0 && !init)
5023 down_write(&com->lc_sem);
5025 lo->ll_pos_latest_start =
5026 lfsck->li_pos_checkpoint.lp_oit_cookie;
5028 lo->ll_pos_last_checkpoint =
5029 lfsck->li_pos_checkpoint.lp_oit_cookie;
5030 lo->ll_run_time_phase1 += ktime_get_seconds() -
5031 lfsck->li_time_last_checkpoint;
5032 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5033 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5034 com->lc_new_checked = 0;
5037 rc = lfsck_layout_store(env, com);
5038 up_write(&com->lc_sem);
5040 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
5041 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5042 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5047 static int lfsck_layout_prep(const struct lu_env *env,
5048 struct lfsck_component *com,
5049 struct lfsck_start *start)
5051 struct lfsck_instance *lfsck = com->lc_lfsck;
5052 struct lfsck_layout *lo = com->lc_file_ram;
5053 struct lfsck_position *pos = &com->lc_pos_start;
5055 fid_zero(&pos->lp_dir_parent);
5056 pos->lp_dir_cookie = 0;
5057 if (lo->ll_status == LS_COMPLETED ||
5058 lo->ll_status == LS_PARTIAL ||
5059 /* To handle orphan, must scan from the beginning. */
5060 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
5063 rc = lfsck_layout_reset(env, com, false);
5065 rc = lfsck_set_param(env, lfsck, start, true);
5068 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
5069 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
5075 down_write(&com->lc_sem);
5076 lo->ll_time_latest_start = ktime_get_real_seconds();
5077 spin_lock(&lfsck->li_lock);
5078 if (lo->ll_flags & LF_SCANNED_ONCE) {
5079 if (!lfsck->li_drop_dryrun ||
5080 lo->ll_pos_first_inconsistent == 0) {
5081 lo->ll_status = LS_SCANNING_PHASE2;
5082 list_move_tail(&com->lc_link,
5083 &lfsck->li_list_double_scan);
5084 pos->lp_oit_cookie = 0;
5088 lo->ll_status = LS_SCANNING_PHASE1;
5089 lo->ll_run_time_phase1 = 0;
5090 lo->ll_run_time_phase2 = 0;
5091 lo->ll_objs_checked_phase1 = 0;
5092 lo->ll_objs_checked_phase2 = 0;
5093 lo->ll_objs_failed_phase1 = 0;
5094 lo->ll_objs_failed_phase2 = 0;
5095 for (i = 0; i < LLIT_MAX; i++)
5096 lo->ll_objs_repaired[i] = 0;
5098 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5099 fid_zero(&com->lc_fid_latest_scanned_phase2);
5102 lo->ll_status = LS_SCANNING_PHASE1;
5103 if (!lfsck->li_drop_dryrun ||
5104 lo->ll_pos_first_inconsistent == 0)
5105 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
5107 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5109 spin_unlock(&lfsck->li_lock);
5110 up_write(&com->lc_sem);
5115 static int lfsck_layout_slave_prep(const struct lu_env *env,
5116 struct lfsck_component *com,
5117 struct lfsck_start_param *lsp)
5119 struct lfsck_layout_slave_data *llsd = com->lc_data;
5120 struct lfsck_instance *lfsck = com->lc_lfsck;
5121 struct lfsck_layout *lo = com->lc_file_ram;
5122 struct lfsck_start *start = lsp->lsp_start;
5125 rc = lfsck_layout_prep(env, com, start);
5129 if (lo->ll_flags & LF_CRASHED_LASTID &&
5130 list_empty(&llsd->llsd_master_list)) {
5131 LASSERT(lfsck->li_out_notify != NULL);
5133 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5134 LE_LASTID_REBUILDING);
5137 if (!lsp->lsp_index_valid)
5140 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5141 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
5142 LASSERT(!llsd->llsd_rbtree_valid);
5144 write_lock(&llsd->llsd_rb_lock);
5145 rc = lfsck_rbtree_setup(env, com);
5146 write_unlock(&llsd->llsd_rb_lock);
5149 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
5150 "%llu]\n", lfsck_lfsck2name(lfsck),
5151 com->lc_pos_start.lp_oit_cookie);
5156 static int lfsck_layout_master_prep(const struct lu_env *env,
5157 struct lfsck_component *com,
5158 struct lfsck_start_param *lsp)
5163 rc = lfsck_layout_load_bitmap(env, com);
5165 rc = lfsck_layout_reset(env, com, false);
5167 rc = lfsck_set_param(env, com->lc_lfsck,
5168 lsp->lsp_start, true);
5174 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
5178 rc = lfsck_start_assistant(env, com, lsp);
5183 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
5184 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
5185 com->lc_pos_start.lp_oit_cookie);
5190 /* Pre-fetch the attribute for each stripe in the given layout EA. */
5191 static int lfsck_layout_scan_stripes(const struct lu_env *env,
5192 struct lfsck_component *com,
5193 struct dt_object *parent,
5194 struct lov_mds_md_v1 *lmm, __u32 comp_id)
5196 struct lfsck_thread_info *info = lfsck_env_info(env);
5197 struct lfsck_instance *lfsck = com->lc_lfsck;
5198 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5199 struct lfsck_layout *lo = com->lc_file_ram;
5200 struct lfsck_assistant_data *lad = com->lc_data;
5201 struct lfsck_assistant_object *lso = NULL;
5202 struct lov_ost_data_v1 *objs;
5203 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
5204 struct ptlrpc_thread *mthread = &lfsck->li_thread;
5205 struct ptlrpc_thread *athread = &lad->lad_thread;
5206 struct l_wait_info lwi = { 0 };
5214 lfsck_buf_init(&buf, &info->lti_ff, sizeof(struct filter_fid));
5215 magic = le32_to_cpu(lmm->lmm_magic);
5216 if (magic == LOV_MAGIC_V1) {
5217 objs = &lmm->lmm_objects[0];
5219 LASSERT(magic == LOV_MAGIC_V3);
5220 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5223 count = le16_to_cpu(lmm->lmm_stripe_count);
5224 for (i = 0; i < count; i++, objs++) {
5225 struct lu_fid *fid = &info->lti_fid;
5226 struct ost_id *oi = &info->lti_oi;
5227 struct lfsck_layout_req *llr;
5228 struct lfsck_tgt_desc *tgt = NULL;
5229 struct dt_object *cobj = NULL;
5231 bool wakeup = false;
5233 if (unlikely(lovea_slot_is_dummy(objs)))
5236 l_wait_event(mthread->t_ctl_waitq,
5237 lad->lad_prefetched < bk->lb_async_windows ||
5238 !thread_is_running(mthread) ||
5239 thread_is_stopped(athread),
5242 if (unlikely(!thread_is_running(mthread)) ||
5243 thread_is_stopped(athread))
5246 if (unlikely(lfsck_is_dead_obj(parent)))
5249 ostid_le_to_cpu(&objs->l_ost_oi, oi);
5250 index = le32_to_cpu(objs->l_ost_idx);
5251 rc = ostid_to_fid(fid, oi, index);
5253 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
5254 ": "DOSTID", idx %u, comp_id %u\n",
5255 lfsck_lfsck2name(lfsck),
5256 PFID(lfsck_dto2fid(parent)), POSTID(oi),
5261 tgt = lfsck_tgt_get(ltds, index);
5262 if (unlikely(tgt == NULL)) {
5263 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
5264 "did not join the layout LFSCK, comp_id %u\n",
5265 lfsck_lfsck2name(lfsck), index, comp_id);
5266 lfsck_lad_set_bitmap(env, com, index);
5270 /* There is potential deadlock race condition between object
5271 * destroy and layout LFSCK. Consider the following scenario:
5273 * 1) The LFSCK thread obtained the parent object firstly, at
5274 * that time, the parent object has not been destroyed yet.
5276 * 2) One RPC service thread destroyed the parent and all its
5277 * children objects. Because the LFSCK is referencing the
5278 * parent object, then the parent object will be marked as
5279 * dying in RAM. On the other hand, the parent object is
5280 * referencing all its children objects, then all children
5281 * objects will be marked as dying in RAM also.
5283 * 3) The LFSCK thread tries to find some child object with
5284 * the parent object referenced. Then it will find that the
5285 * child object is dying. According to the object visibility
5286 * rules: the object with dying flag cannot be returned to
5287 * others. So the LFSCK thread has to wait until the dying
5288 * object has been purged from RAM, then it can allocate a
5289 * new object (with the same FID) in RAM. Unfortunately, the
5290 * LFSCK thread itself is referencing the parent object, and
5291 * cause the parent object cannot be purged, then cause the
5292 * child object cannot be purged also. So the LFSCK thread
5293 * will fall into deadlock.
5295 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
5297 if (lfsck_is_dead_obj(parent)) {
5307 rc = dt_declare_attr_get(env, cobj);
5311 rc = dt_declare_xattr_get(env, cobj, &buf, XATTR_NAME_FID);
5316 struct lu_attr *attr = &info->lti_la;
5318 rc = dt_attr_get(env, parent, attr);
5322 lso = lfsck_assistant_object_init(env,
5323 lfsck_dto2fid(parent), attr,
5324 lfsck->li_pos_current.lp_oit_cookie, false);
5333 llr = lfsck_layout_assistant_req_init(lso, cobj, comp_id,
5341 spin_lock(&lad->lad_lock);
5342 if (lad->lad_assistant_status < 0) {
5343 spin_unlock(&lad->lad_lock);
5344 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
5346 RETURN(lad->lad_assistant_status);
5349 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
5350 if (lad->lad_prefetched == 0)
5353 lad->lad_prefetched++;
5354 spin_unlock(&lad->lad_lock);
5356 wake_up_all(&athread->t_ctl_waitq);
5359 down_write(&com->lc_sem);
5360 com->lc_new_checked++;
5362 lfsck_layout_record_failure(env, lfsck, lo);
5363 up_write(&com->lc_sem);
5365 if (cobj != NULL && !IS_ERR(cobj))
5366 lfsck_object_put(env, cobj);
5368 if (likely(tgt != NULL))
5371 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5379 lfsck_assistant_object_put(env, lso);
5384 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
5385 * the OST-object's attribute and generate an structure lfsck_layout_req on the
5386 * list ::lad_req_list.
5388 * For each request on above list, the lfsck_layout_assistant thread compares
5389 * the OST side attribute with local attribute, if inconsistent, then repair it.
5391 * All above processing is async mode with pipeline. */
5392 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
5393 struct lfsck_component *com,
5394 struct dt_object *obj)
5396 struct lfsck_thread_info *info = lfsck_env_info(env);
5397 struct ost_id *oi = &info->lti_oi;
5398 struct lfsck_layout *lo = com->lc_file_ram;
5399 struct lfsck_assistant_data *lad = com->lc_data;
5400 struct lfsck_instance *lfsck = com->lc_lfsck;
5401 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5402 struct thandle *handle = NULL;
5403 struct lu_buf *buf = &info->lti_big_buf;
5404 struct lov_mds_md_v1 *lmm = NULL;
5405 struct dt_device *dev = lfsck_obj2dev(obj);
5406 struct lustre_handle lh = { 0 };
5407 struct lu_buf ea_buf = { NULL };
5408 struct lov_comp_md_v1 *lcm = NULL;
5409 struct lov_comp_md_entry_v1 *lcme = NULL;
5414 bool locked = false;
5415 bool stripe = false;
5416 bool bad_oi = false;
5419 if (!S_ISREG(lfsck_object_type(obj)))
5422 if (lad->lad_assistant_status < 0)
5423 GOTO(out, rc = -ESRCH);
5425 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
5426 lmm_oi_cpu_to_le(oi, oi);
5427 dt_read_lock(env, obj, 0);
5432 if (dt_object_exists(obj) == 0 ||
5433 lfsck_is_dead_obj(obj))
5436 rc = lfsck_layout_get_lovea(env, obj, buf);
5438 /* Skip bad lov EA during the 1st cycle scanning, and
5439 * try to recover it via orphan in the 2nd scanning. */
5440 GOTO(out, rc = (rc == -EINVAL ? 0 : rc));
5444 magic = le32_to_cpu(lmm->lmm_magic);
5445 if (magic == LOV_MAGIC_COMP_V1) {
5449 count = le16_to_cpu(lcm->lcm_entry_count);
5450 for (i = 0; i < count; i++) {
5451 lcme = &lcm->lcm_entries[i];
5452 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5453 if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) != 0)
5457 GOTO(out, stripe = true);
5458 } else if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0) {
5459 GOTO(out, stripe = true);
5463 /* Inconsistent lmm_oi, should be repaired. */
5466 if (bk->lb_param & LPF_DRYRUN) {
5467 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5469 GOTO(out, stripe = true);
5472 if (!lustre_handle_is_used(&lh)) {
5473 dt_read_unlock(env, obj);
5475 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5476 MDS_INODELOCK_LAYOUT |
5477 MDS_INODELOCK_XATTR, LCK_EX);
5481 handle = dt_trans_create(env, dev);
5483 GOTO(out, rc = PTR_ERR(handle));
5485 lfsck_buf_init(&ea_buf, lmm, size);
5486 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5487 LU_XATTR_REPLACE, handle);
5491 rc = dt_trans_start_local(env, dev, handle);
5495 dt_write_lock(env, obj, 0);
5501 if (magic == LOV_MAGIC_COMP_V1) {
5504 for (i = 0; i < count; i++) {
5505 lcme = &lcm->lcm_entries[i];
5506 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5513 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5514 LU_XATTR_REPLACE, handle);
5518 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5520 GOTO(out, stripe = true);
5524 if (lustre_handle_is_used(&lh))
5525 dt_write_unlock(env, obj);
5527 dt_read_unlock(env, obj);
5530 if (handle != NULL && !IS_ERR(handle))
5531 dt_trans_stop(env, dev, handle);
5533 lfsck_ibits_unlock(&lh, LCK_EX);
5536 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
5537 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
5538 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
5539 PFID(lfsck_dto2fid(obj)), rc);
5542 if (magic == LOV_MAGIC_COMP_V1) {
5545 for (i = 0; i < count; i++) {
5546 lcme = &lcm->lcm_entries[i];
5547 if (!(le32_to_cpu(lcme->lcme_flags) &
5551 rc = lfsck_layout_scan_stripes(env, com, obj,
5552 (struct lov_mds_md_v1 *)(buf->lb_buf +
5553 le32_to_cpu(lcme->lcme_offset)),
5554 le32_to_cpu(lcme->lcme_id));
5557 rc = lfsck_layout_scan_stripes(env, com, obj, lmm, 0);
5560 down_write(&com->lc_sem);
5561 com->lc_new_checked++;
5563 lfsck_layout_record_failure(env, lfsck, lo);
5564 up_write(&com->lc_sem);
5570 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
5571 struct lfsck_component *com,
5572 struct dt_object *obj)
5574 struct lfsck_instance *lfsck = com->lc_lfsck;
5575 struct lfsck_layout *lo = com->lc_file_ram;
5576 const struct lu_fid *fid = lfsck_dto2fid(obj);
5577 struct lfsck_layout_slave_data *llsd = com->lc_data;
5578 struct lfsck_layout_seq *lls;
5584 LASSERT(llsd != NULL);
5586 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
5587 cfs_fail_val == lfsck_dev_idx(lfsck)) {
5588 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(1),
5590 struct ptlrpc_thread *thread = &lfsck->li_thread;
5592 l_wait_event(thread->t_ctl_waitq,
5593 !thread_is_running(thread),
5597 lfsck_rbtree_update_bitmap(env, com, fid, false);
5599 down_write(&com->lc_sem);
5600 if (fid_is_idif(fid))
5602 else if (!fid_is_norm(fid) ||
5603 !fid_is_for_ostobj(env, lfsck, obj, fid))
5604 GOTO(unlock, rc = 0);
5607 com->lc_new_checked++;
5609 lls = lfsck_layout_seq_lookup(llsd, seq);
5612 if (unlikely(lls == NULL))
5613 GOTO(unlock, rc = -ENOMEM);
5615 INIT_LIST_HEAD(&lls->lls_list);
5617 rc = lfsck_layout_lastid_load(env, com, lls);
5619 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5620 "load LAST_ID for %#llx: rc = %d\n",
5621 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
5622 lo->ll_objs_failed_phase1++;
5627 lfsck_layout_seq_insert(llsd, lls);
5630 if (unlikely(fid_is_last_id(fid)))
5631 GOTO(unlock, rc = 0);
5633 if (fid_is_idif(fid))
5634 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
5638 if (oid > lls->lls_lastid_known)
5639 lls->lls_lastid_known = oid;
5641 if (oid > lls->lls_lastid) {
5642 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
5643 /* OFD may create new objects during LFSCK scanning. */
5644 rc = lfsck_layout_lastid_reload(env, com, lls);
5645 if (unlikely(rc != 0)) {
5646 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5647 "reload LAST_ID for %#llx: rc = %d\n",
5648 lfsck_lfsck2name(com->lc_lfsck),
5654 if (oid <= lls->lls_lastid ||
5655 lo->ll_flags & LF_CRASHED_LASTID)
5656 GOTO(unlock, rc = 0);
5658 LASSERT(lfsck->li_out_notify != NULL);
5660 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5661 LE_LASTID_REBUILDING);
5662 lo->ll_flags |= LF_CRASHED_LASTID;
5664 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
5665 "LAST_ID file (2) for the sequence %#llx"
5666 ", old value %llu, known value %llu\n",
5667 lfsck_lfsck2name(lfsck), lls->lls_seq,
5668 lls->lls_lastid, oid);
5671 lls->lls_lastid = oid;
5675 GOTO(unlock, rc = 0);
5678 up_write(&com->lc_sem);
5683 static int lfsck_layout_exec_dir(const struct lu_env *env,
5684 struct lfsck_component *com,
5685 struct lfsck_assistant_object *lso,
5686 struct lu_dirent *ent, __u16 type)
5691 static int lfsck_layout_master_post(const struct lu_env *env,
5692 struct lfsck_component *com,
5693 int result, bool init)
5695 struct lfsck_instance *lfsck = com->lc_lfsck;
5696 struct lfsck_layout *lo = com->lc_file_ram;
5700 lfsck_post_generic(env, com, &result);
5702 down_write(&com->lc_sem);
5703 spin_lock(&lfsck->li_lock);
5705 lo->ll_pos_last_checkpoint =
5706 lfsck->li_pos_checkpoint.lp_oit_cookie;
5709 if (lo->ll_flags & LF_INCOMPLETE)
5710 lo->ll_status = LS_PARTIAL;
5712 lo->ll_status = LS_SCANNING_PHASE2;
5713 lo->ll_flags |= LF_SCANNED_ONCE;
5714 lo->ll_flags &= ~LF_UPGRADE;
5715 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5716 } else if (result == 0) {
5717 if (lfsck->li_status != 0)
5718 lo->ll_status = lfsck->li_status;
5720 lo->ll_status = LS_STOPPED;
5721 if (lo->ll_status != LS_PAUSED)
5722 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5724 lo->ll_status = LS_FAILED;
5725 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5727 spin_unlock(&lfsck->li_lock);
5730 lo->ll_run_time_phase1 += ktime_get_seconds() -
5731 lfsck->li_time_last_checkpoint;
5732 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5733 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5734 com->lc_new_checked = 0;
5737 rc = lfsck_layout_store(env, com);
5738 up_write(&com->lc_sem);
5740 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
5741 lfsck_lfsck2name(lfsck), rc);
5746 static int lfsck_layout_slave_post(const struct lu_env *env,
5747 struct lfsck_component *com,
5748 int result, bool init)
5750 struct lfsck_instance *lfsck = com->lc_lfsck;
5751 struct lfsck_layout *lo = com->lc_file_ram;
5755 down_write(&com->lc_sem);
5756 rc = lfsck_layout_lastid_store(env, com);
5760 LASSERT(lfsck->li_out_notify != NULL);
5762 spin_lock(&lfsck->li_lock);
5764 lo->ll_pos_last_checkpoint =
5765 lfsck->li_pos_checkpoint.lp_oit_cookie;
5768 lo->ll_status = LS_SCANNING_PHASE2;
5769 lo->ll_flags |= LF_SCANNED_ONCE;
5770 if (lo->ll_flags & LF_CRASHED_LASTID) {
5772 lo->ll_flags &= ~LF_CRASHED_LASTID;
5774 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
5775 "crashed LAST_ID files successfully\n",
5776 lfsck_lfsck2name(lfsck));
5778 lo->ll_flags &= ~LF_UPGRADE;
5779 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5780 } else if (result == 0) {
5781 if (lfsck->li_status != 0)
5782 lo->ll_status = lfsck->li_status;
5784 lo->ll_status = LS_STOPPED;
5785 if (lo->ll_status != LS_PAUSED)
5786 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5788 lo->ll_status = LS_FAILED;
5789 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5791 spin_unlock(&lfsck->li_lock);
5794 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5798 lo->ll_run_time_phase1 += ktime_get_seconds() -
5799 lfsck->li_time_last_checkpoint;
5800 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5801 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5802 com->lc_new_checked = 0;
5805 rc = lfsck_layout_store(env, com);
5806 up_write(&com->lc_sem);
5808 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
5810 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
5811 lfsck_lfsck2name(lfsck), rc);
5816 static void lfsck_layout_dump(const struct lu_env *env,
5817 struct lfsck_component *com, struct seq_file *m)
5819 struct lfsck_instance *lfsck = com->lc_lfsck;
5820 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5821 struct lfsck_layout *lo = com->lc_file_ram;
5824 down_read(&com->lc_sem);
5825 if (bk->lb_param & LPF_DRYRUN)
5826 prefix = "inconsistent";
5828 prefix = "repaired";
5830 seq_printf(m, "name: lfsck_layout\n"
5836 lfsck_status2name(lo->ll_status));
5838 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
5840 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
5842 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
5844 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
5846 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
5848 seq_printf(m, "latest_start_position: %llu\n"
5849 "last_checkpoint_position: %llu\n"
5850 "first_failure_position: %llu\n",
5851 lo->ll_pos_latest_start,
5852 lo->ll_pos_last_checkpoint,
5853 lo->ll_pos_first_inconsistent);
5855 seq_printf(m, "success_count: %u\n"
5856 "%s_dangling: %llu\n"
5857 "%s_unmatched_pair: %llu\n"
5858 "%s_multiple_referenced: %llu\n"
5860 "%s_inconsistent_owner: %llu\n"
5863 "failed_phase1: %llu\n"
5864 "failed_phase2: %llu\n",
5865 lo->ll_success_count,
5866 prefix, lo->ll_objs_repaired[LLIT_DANGLING - 1],
5867 prefix, lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
5868 prefix, lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
5869 prefix, lo->ll_objs_repaired[LLIT_ORPHAN - 1],
5870 prefix, lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
5871 prefix, lo->ll_objs_repaired[LLIT_OTHERS - 1],
5872 lo->ll_objs_skipped,
5873 lo->ll_objs_failed_phase1,
5874 lo->ll_objs_failed_phase2);
5876 if (lo->ll_status == LS_SCANNING_PHASE1) {
5877 time64_t duration = ktime_get_seconds() -
5878 lfsck->li_time_last_checkpoint;
5879 u64 checked = lo->ll_objs_checked_phase1 +
5880 com->lc_new_checked;
5881 u64 speed = checked;
5882 u64 new_checked = com->lc_new_checked;
5883 time64_t rtime = lo->ll_run_time_phase1 + duration;
5887 new_checked = div64_s64(new_checked, duration);
5889 speed = div64_s64(speed, rtime);
5890 seq_printf(m, "checked_phase1: %llu\n"
5891 "checked_phase2: %llu\n"
5892 "run_time_phase1: %lld seconds\n"
5893 "run_time_phase2: %lld seconds\n"
5894 "average_speed_phase1: %llu items/sec\n"
5895 "average_speed_phase2: N/A\n"
5896 "real-time_speed_phase1: %llu items/sec\n"
5897 "real-time_speed_phase2: N/A\n",
5899 lo->ll_objs_checked_phase2,
5901 lo->ll_run_time_phase2,
5905 if (likely(lfsck->li_di_oit)) {
5906 const struct dt_it_ops *iops =
5907 &lfsck->li_obj_oit->do_index_ops->dio_it;
5909 /* The low layer otable-based iteration position may NOT
5910 * exactly match the layout-based directory traversal
5911 * cookie. Generally, it is not a serious issue. But the
5912 * caller should NOT make assumption on that. */
5913 pos = iops->store(env, lfsck->li_di_oit);
5914 if (!lfsck->li_current_oit_processed)
5917 pos = lo->ll_pos_last_checkpoint;
5920 seq_printf(m, "current_position: %llu\n", pos);
5921 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
5922 time64_t duration = ktime_get_seconds() -
5923 com->lc_time_last_checkpoint;
5924 u64 checked = lo->ll_objs_checked_phase2 +
5925 com->lc_new_checked;
5926 u64 speed1 = lo->ll_objs_checked_phase1;
5927 u64 speed2 = checked;
5928 u64 new_checked = com->lc_new_checked;
5929 time64_t rtime = lo->ll_run_time_phase2 + duration;
5932 new_checked = div64_s64(new_checked, duration);
5933 if (lo->ll_run_time_phase1 != 0)
5934 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
5936 speed2 = div64_s64(speed2, rtime);
5937 seq_printf(m, "checked_phase1: %llu\n"
5938 "checked_phase2: %llu\n"
5939 "run_time_phase1: %lld seconds\n"
5940 "run_time_phase2: %lld 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 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
5960 if (lo->ll_run_time_phase2 != 0)
5961 speed2 = div64_s64(speed2, lo->ll_run_time_phase2);
5962 seq_printf(m, "checked_phase1: %llu\n"
5963 "checked_phase2: %llu\n"
5964 "run_time_phase1: %lld seconds\n"
5965 "run_time_phase2: %lld 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 = ktime_get_seconds();
6041 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6042 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);
6313 if (lr->lr_flags2 & LF_INCOMPLETE) {
6314 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);
6726 rc = lfsck_layout_reset(env, com, true);
6732 lo = com->lc_file_ram;
6733 switch (lo->ll_status) {
6739 spin_lock(&lfsck->li_lock);
6740 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6741 spin_unlock(&lfsck->li_lock);
6744 CERROR("%s: unknown lfsck_layout status %d\n",
6745 lfsck_lfsck2name(lfsck), lo->ll_status);
6747 case LS_SCANNING_PHASE1:
6748 case LS_SCANNING_PHASE2:
6749 /* No need to store the status to disk right now.
6750 * If the system crashed before the status stored,
6751 * it will be loaded back when next time. */
6752 lo->ll_status = LS_CRASHED;
6753 if (!lfsck->li_master)
6754 lo->ll_flags |= LF_INCOMPLETE;
6761 spin_lock(&lfsck->li_lock);
6762 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6763 spin_unlock(&lfsck->li_lock);
6767 if (lo->ll_flags & LF_CRASHED_LASTID) {
6768 LASSERT(lfsck->li_out_notify != NULL);
6770 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6771 LE_LASTID_REBUILDING);
6777 if (root != NULL && !IS_ERR(root))
6778 lfsck_object_put(env, root);
6781 lfsck_component_cleanup(env, com);
6782 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
6783 lfsck_lfsck2name(lfsck), rc);
6789 struct lfsck_orphan_it {
6790 struct lfsck_component *loi_com;
6791 struct lfsck_rbtree_node *loi_lrn;
6792 struct lfsck_layout_slave_target *loi_llst;
6793 struct lu_fid loi_key;
6794 struct lu_orphan_rec_v2 loi_rec;
6796 unsigned int loi_over:1;
6799 static int lfsck_fid_match_idx(const struct lu_env *env,
6800 struct lfsck_instance *lfsck,
6801 const struct lu_fid *fid, int idx)
6803 struct seq_server_site *ss;
6804 struct lu_server_fld *sf;
6805 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
6808 /* All abnormal cases will be returned to MDT0. */
6809 if (!fid_is_norm(fid)) {
6816 ss = lfsck_dev_site(lfsck);
6817 if (unlikely(ss == NULL))
6820 sf = ss->ss_server_fld;
6821 LASSERT(sf != NULL);
6823 fld_range_set_any(range);
6824 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
6828 if (!fld_range_is_mdt(range))
6831 if (range->lsr_index == idx)
6837 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
6838 struct dt_object *obj)
6840 struct dt_device *dev = lfsck_obj2dev(obj);
6841 struct thandle *handle;
6845 handle = dt_trans_create(env, dev);
6849 rc = dt_declare_ref_del(env, obj, handle);
6853 rc = dt_declare_destroy(env, obj, handle);
6857 rc = dt_trans_start_local(env, dev, handle);
6861 dt_write_lock(env, obj, 0);
6862 rc = dt_ref_del(env, obj, handle);
6864 rc = dt_destroy(env, obj, handle);
6865 dt_write_unlock(env, obj);
6870 dt_trans_stop(env, dev, handle);
6872 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
6873 PFID(lfsck_dto2fid(obj)), rc);
6878 static int lfsck_orphan_index_lookup(const struct lu_env *env,
6879 struct dt_object *dt,
6881 const struct dt_key *key)
6886 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
6887 struct dt_object *dt,
6888 const struct dt_rec *rec,
6889 const struct dt_key *key,
6890 struct thandle *handle)
6895 static int lfsck_orphan_index_insert(const struct lu_env *env,
6896 struct dt_object *dt,
6897 const struct dt_rec *rec,
6898 const struct dt_key *key,
6899 struct thandle *handle,
6905 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
6906 struct dt_object *dt,
6907 const struct dt_key *key,
6908 struct thandle *handle)
6913 static int lfsck_orphan_index_delete(const struct lu_env *env,
6914 struct dt_object *dt,
6915 const struct dt_key *key,
6916 struct thandle *handle)
6921 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
6922 struct dt_object *dt,
6925 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
6926 struct lfsck_instance *lfsck;
6927 struct lfsck_component *com = NULL;
6928 struct lfsck_layout_slave_data *llsd;
6929 struct lfsck_orphan_it *it = NULL;
6930 struct lfsck_layout *lo;
6934 lfsck = lfsck_instance_find(dev, true, false);
6935 if (unlikely(lfsck == NULL))
6936 RETURN(ERR_PTR(-ENXIO));
6938 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
6939 if (unlikely(com == NULL))
6940 GOTO(out, rc = -ENOENT);
6942 lo = com->lc_file_ram;
6943 if (lo->ll_flags & LF_INCOMPLETE)
6944 GOTO(out, rc = -ESRCH);
6946 llsd = com->lc_data;
6947 if (!llsd->llsd_rbtree_valid)
6948 GOTO(out, rc = -ESRCH);
6952 GOTO(out, rc = -ENOMEM);
6954 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
6955 if (it->loi_llst == NULL)
6956 GOTO(out, rc = -ENXIO);
6958 if (dev->dd_record_fid_accessed) {
6959 /* The first iteration against the rbtree, scan the whole rbtree
6960 * to remove the nodes which do NOT need to be handled. */
6961 write_lock(&llsd->llsd_rb_lock);
6962 if (dev->dd_record_fid_accessed) {
6963 struct rb_node *node;
6964 struct rb_node *next;
6965 struct lfsck_rbtree_node *lrn;
6967 /* No need to record the fid accessing anymore. */
6968 dev->dd_record_fid_accessed = 0;
6970 node = rb_first(&llsd->llsd_rb_root);
6971 while (node != NULL) {
6972 next = rb_next(node);
6973 lrn = rb_entry(node, struct lfsck_rbtree_node,
6975 if (atomic_read(&lrn->lrn_known_count) <=
6976 atomic_read(&lrn->lrn_accessed_count)) {
6977 rb_erase(node, &llsd->llsd_rb_root);
6978 lfsck_rbtree_free(lrn);
6983 write_unlock(&llsd->llsd_rb_lock);
6986 /* read lock the rbtree when init, and unlock when fini */
6987 read_lock(&llsd->llsd_rb_lock);
6995 lfsck_component_put(env, com);
6997 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
6998 lfsck_lfsck2name(lfsck), rc);
7000 lfsck_instance_put(env, lfsck);
7005 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
7008 return (struct dt_it *)it;
7011 static void lfsck_orphan_it_fini(const struct lu_env *env,
7014 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7015 struct lfsck_component *com = it->loi_com;
7016 struct lfsck_layout_slave_data *llsd;
7017 struct lfsck_layout_slave_target *llst;
7020 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
7021 lfsck_lfsck2name(com->lc_lfsck));
7023 llsd = com->lc_data;
7024 read_unlock(&llsd->llsd_rb_lock);
7025 llst = it->loi_llst;
7026 LASSERT(llst != NULL);
7028 /* Save the key and hash for iterate next. */
7029 llst->llst_fid = it->loi_key;
7030 llst->llst_hash = it->loi_hash;
7031 lfsck_layout_llst_put(llst);
7032 lfsck_component_put(env, com);
7038 * \retval +1: the iteration finished
7039 * \retval 0: on success, not finished
7040 * \retval -ve: on error
7042 static int lfsck_orphan_it_next(const struct lu_env *env,
7045 struct lfsck_thread_info *info = lfsck_env_info(env);
7046 struct filter_fid *ff = &info->lti_ff;
7047 struct lu_attr *la = &info->lti_la;
7048 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7049 struct lu_fid *key = &it->loi_key;
7050 struct lu_orphan_rec_v2 *rec = &it->loi_rec;
7051 struct ost_layout *ol = &rec->lor_layout;
7052 struct lfsck_component *com = it->loi_com;
7053 struct lfsck_instance *lfsck = com->lc_lfsck;
7054 struct lfsck_layout_slave_data *llsd = com->lc_data;
7055 struct dt_object *obj;
7056 struct lfsck_rbtree_node *lrn;
7060 __u32 idx = it->loi_llst->llst_index;
7070 lrn = lfsck_rbtree_search(llsd, key, &exact);
7078 key->f_seq = lrn->lrn_seq;
7079 key->f_oid = lrn->lrn_first_oid;
7084 if (unlikely(key->f_oid == 0)) {
7091 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
7097 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
7098 atomic_read(&lrn->lrn_accessed_count))) {
7099 struct rb_node *next = rb_next(&lrn->lrn_node);
7101 while (next != NULL) {
7102 lrn = rb_entry(next, struct lfsck_rbtree_node,
7104 if (atomic_read(&lrn->lrn_known_count) >
7105 atomic_read(&lrn->lrn_accessed_count))
7107 next = rb_next(next);
7116 key->f_seq = lrn->lrn_seq;
7117 key->f_oid = lrn->lrn_first_oid;
7121 pos = key->f_oid - lrn->lrn_first_oid;
7124 pos = find_next_bit(lrn->lrn_known_bitmap,
7125 LFSCK_RBTREE_BITMAP_WIDTH, pos);
7126 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
7127 key->f_oid = lrn->lrn_first_oid + pos;
7128 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
7136 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
7141 key->f_oid = lrn->lrn_first_oid + pos;
7142 obj = lfsck_object_find_bottom(env, lfsck, key);
7145 if (rc == -ENOENT) {
7152 dt_read_lock(env, obj, 0);
7153 if (dt_object_exists(obj) == 0 ||
7154 lfsck_is_dead_obj(obj)) {
7155 dt_read_unlock(env, obj);
7156 lfsck_object_put(env, obj);
7161 rc = dt_attr_get(env, obj, la);
7165 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, ff, sizeof(*ff)),
7167 if (rc == -ENODATA) {
7168 /* For the pre-created OST-object, update the bitmap to avoid
7169 * others LFSCK (second phase) iteration to touch it again. */
7170 if (la->la_ctime == 0) {
7171 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
7172 atomic_inc(&lrn->lrn_accessed_count);
7174 /* For the race between repairing dangling referenced
7175 * MDT-object and unlink the file, it may left orphan
7176 * OST-object there. Destroy it now! */
7177 if (unlikely(!(la->la_mode & S_ISUID))) {
7178 dt_read_unlock(env, obj);
7179 lfsck_layout_destroy_orphan(env, obj);
7180 lfsck_object_put(env, obj);
7184 } else if (idx == 0) {
7185 /* If the orphan OST-object has no parent information,
7186 * regard it as referenced by the MDT-object on MDT0. */
7187 fid_zero(&rec->lor_rec.lor_fid);
7188 rec->lor_rec.lor_uid = la->la_uid;
7189 rec->lor_rec.lor_gid = la->la_gid;
7190 memset(ol, 0, sizeof(*ol));
7195 dt_read_unlock(env, obj);
7196 lfsck_object_put(env, obj);
7201 if (rc < sizeof(struct lu_fid))
7202 GOTO(out, rc = (rc < 0 ? rc : -EINVAL));
7204 fid_le_to_cpu(&rec->lor_rec.lor_fid, &ff->ff_parent);
7205 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
7206 * MDT-object's FID::f_ver, instead it is the OST-object index in its
7207 * parent MDT-object's layout EA. */
7208 save = rec->lor_rec.lor_fid.f_stripe_idx;
7209 rec->lor_rec.lor_fid.f_ver = 0;
7210 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_rec.lor_fid, idx);
7211 /* If the orphan OST-object does not claim the MDT, then next.
7213 * If we do not know whether it matches or not, then return it
7214 * to the MDT for further check. */
7216 dt_read_unlock(env, obj);
7217 lfsck_object_put(env, obj);
7222 rec->lor_rec.lor_fid.f_stripe_idx = save;
7223 rec->lor_rec.lor_uid = la->la_uid;
7224 rec->lor_rec.lor_gid = la->la_gid;
7225 ost_layout_le_to_cpu(ol, &ff->ff_layout);
7227 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u, "
7228 "stripe size %u, stripe count %u, COMP id %u, COMP start %llu, "
7229 "COMP end %llu\n", lfsck_lfsck2name(com->lc_lfsck), PFID(key),
7230 PFID(&rec->lor_rec.lor_fid), rec->lor_rec.lor_uid,
7231 rec->lor_rec.lor_gid, ol->ol_stripe_size, ol->ol_stripe_count,
7232 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end);
7237 dt_read_unlock(env, obj);
7238 lfsck_object_put(env, obj);
7246 * \retval +1: locate to the exactly position
7247 * \retval 0: cannot locate to the exactly position,
7248 * call next() to move to a valid position.
7249 * \retval -ve: on error
7251 static int lfsck_orphan_it_get(const struct lu_env *env,
7253 const struct dt_key *key)
7255 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7258 it->loi_key = *(struct lu_fid *)key;
7259 rc = lfsck_orphan_it_next(env, di);
7269 static void lfsck_orphan_it_put(const struct lu_env *env,
7274 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
7275 const struct dt_it *di)
7277 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7279 return (struct dt_key *)&it->loi_key;
7282 static int lfsck_orphan_it_key_size(const struct lu_env *env,
7283 const struct dt_it *di)
7285 return sizeof(struct lu_fid);
7288 static int lfsck_orphan_it_rec(const struct lu_env *env,
7289 const struct dt_it *di,
7293 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7295 *(struct lu_orphan_rec_v2 *)rec = it->loi_rec;
7300 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
7301 const struct dt_it *di)
7303 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7305 return it->loi_hash;
7309 * \retval +1: locate to the exactly position
7310 * \retval 0: cannot locate to the exactly position,
7311 * call next() to move to a valid position.
7312 * \retval -ve: on error
7314 static int lfsck_orphan_it_load(const struct lu_env *env,
7315 const struct dt_it *di,
7318 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7319 struct lfsck_layout_slave_target *llst = it->loi_llst;
7322 LASSERT(llst != NULL);
7324 if (hash != llst->llst_hash) {
7325 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
7326 "iteration does not match the one when fini "
7327 "%llu, to be reset.\n",
7328 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
7330 fid_zero(&llst->llst_fid);
7331 llst->llst_hash = 0;
7334 it->loi_key = llst->llst_fid;
7335 it->loi_hash = llst->llst_hash;
7336 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
7346 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
7347 const struct dt_it *di,
7353 const struct dt_index_operations lfsck_orphan_index_ops = {
7354 .dio_lookup = lfsck_orphan_index_lookup,
7355 .dio_declare_insert = lfsck_orphan_index_declare_insert,
7356 .dio_insert = lfsck_orphan_index_insert,
7357 .dio_declare_delete = lfsck_orphan_index_declare_delete,
7358 .dio_delete = lfsck_orphan_index_delete,
7360 .init = lfsck_orphan_it_init,
7361 .fini = lfsck_orphan_it_fini,
7362 .get = lfsck_orphan_it_get,
7363 .put = lfsck_orphan_it_put,
7364 .next = lfsck_orphan_it_next,
7365 .key = lfsck_orphan_it_key,
7366 .key_size = lfsck_orphan_it_key_size,
7367 .rec = lfsck_orphan_it_rec,
7368 .store = lfsck_orphan_it_store,
7369 .load = lfsck_orphan_it_load,
7370 .key_rec = lfsck_orphan_it_key_rec,