4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2014, 2016, Intel Corporation.
26 * lustre/lfsck/lfsck_layout.c
28 * Author: Fan, Yong <fan.yong@intel.com>
32 # define EXPORT_SYMTAB
34 #define DEBUG_SUBSYSTEM S_LFSCK
36 #include <linux/bitops.h>
37 #include <linux/rbtree.h>
39 #include <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 down_write(&com->lc_sem);
1444 lo->ll_run_time_phase2 += ktime_get_seconds() -
1445 com->lc_time_last_checkpoint;
1446 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
1447 lo->ll_objs_checked_phase2 += com->lc_new_checked;
1450 if (lo->ll_flags & LF_INCOMPLETE) {
1451 lo->ll_status = LS_PARTIAL;
1453 if (lfsck->li_master) {
1454 struct lfsck_assistant_data *lad = com->lc_data;
1456 if (lad->lad_incomplete)
1457 lo->ll_status = LS_PARTIAL;
1459 lo->ll_status = LS_COMPLETED;
1461 lo->ll_status = LS_COMPLETED;
1464 lo->ll_flags &= ~LF_SCANNED_ONCE;
1465 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1466 lo->ll_flags &= ~LF_INCONSISTENT;
1467 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1468 lo->ll_success_count++;
1469 } else if (rc == 0) {
1470 if (lfsck->li_status != 0)
1471 lo->ll_status = lfsck->li_status;
1473 lo->ll_status = LS_STOPPED;
1475 lo->ll_status = LS_FAILED;
1478 rc = lfsck_layout_store(env, com);
1479 up_write(&com->lc_sem);
1484 static int lfsck_layout_trans_stop(const struct lu_env *env,
1485 struct dt_device *dev,
1486 struct thandle *handle, int result)
1490 /* XXX: If there is something worng or it needs to repair nothing,
1491 * then notify the lower to stop the modification. Currently,
1492 * we use th_result for such purpose, that may be replaced by
1493 * some rollback mechanism in the future. */
1494 handle->th_result = result;
1495 rc = dt_trans_stop(env, dev, handle);
1497 return result > 0 ? 0 : result;
1499 return rc == 0 ? 1 : rc;
1502 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1503 struct lfsck_component *com,
1504 const struct lu_fid *pfid,
1505 const struct lu_fid *cfid,
1506 __u32 comp_id, __u32 ea_off,
1509 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1510 struct lu_fid *rec = &lfsck_env_info(env)->lti_fid3;
1511 struct dt_device *dev;
1512 struct dt_object *obj;
1513 struct thandle *th = NULL;
1518 idx = lfsck_sub_trace_file_fid2idx(pfid);
1519 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1520 dev = lfsck_obj2dev(obj);
1522 fid_cpu_to_be(&key->lldk_fid, pfid);
1523 key->lldk_comp_id = cpu_to_be32(comp_id);
1524 key->lldk_ea_off = cpu_to_be32(ea_off);
1526 fid_cpu_to_be(rec, cfid);
1527 rec->f_ver = cpu_to_be32(ost_idx);
1529 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1531 th = dt_trans_create(env, dev);
1533 GOTO(unlock, rc = PTR_ERR(th));
1535 rc = dt_declare_insert(env, obj,
1536 (const struct dt_rec *)rec,
1537 (const struct dt_key *)key, th);
1541 rc = dt_trans_start_local(env, dev, th);
1545 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1546 (const struct dt_key *)key, th, 1);
1551 if (th && !IS_ERR(th))
1552 dt_trans_stop(env, dev, th);
1554 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1556 CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", comp_id = %u, "
1557 "ea_off = %u, ost_idx = %u, into the trace file for further "
1558 "dangling check: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1559 PFID(pfid), PFID(cfid), comp_id, ea_off, ost_idx, rc);
1564 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1565 struct lfsck_component *com,
1566 const struct lu_fid *fid,
1567 __u32 comp_id, __u32 ea_off)
1569 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1570 struct dt_device *dev;
1571 struct dt_object *obj;
1572 struct thandle *th = NULL;
1577 idx = lfsck_sub_trace_file_fid2idx(fid);
1578 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1579 dev = lfsck_obj2dev(obj);
1581 fid_cpu_to_be(&key->lldk_fid, fid);
1582 key->lldk_comp_id = cpu_to_be32(comp_id);
1583 key->lldk_ea_off = cpu_to_be32(ea_off);
1585 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1587 th = dt_trans_create(env, dev);
1589 GOTO(unlock, rc = PTR_ERR(th));
1591 rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1595 rc = dt_trans_start_local(env, dev, th);
1599 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1604 if (th && !IS_ERR(th))
1605 dt_trans_stop(env, dev, th);
1607 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1609 CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1610 ", comp_id = %u, ea_off = %u from the trace file: rc = %d\n",
1611 lfsck_lfsck2name(com->lc_lfsck), PFID(fid), comp_id, ea_off, rc);
1617 * Get the system default stripe size.
1619 * \param[in] env pointer to the thread context
1620 * \param[in] lfsck pointer to the lfsck instance
1621 * \param[out] size pointer to the default stripe size
1623 * \retval 0 for success
1624 * \retval negative error number on failure
1626 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1627 struct lfsck_instance *lfsck,
1630 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1631 struct dt_object *root;
1634 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1636 return PTR_ERR(root);
1638 /* Get the default stripe size via xattr_get on the backend root. */
1639 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1642 /* The lum->lmm_stripe_size is LE mode. The *size also
1643 * should be LE mode. So it is unnecessary to convert. */
1644 *size = lum->lmm_stripe_size;
1646 } else if (unlikely(rc == 0)) {
1650 lfsck_object_put(env, root);
1656 * \retval +1: repaired
1657 * \retval 0: did nothing
1658 * \retval -ve: on error
1660 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1661 struct lfsck_instance *lfsck,
1662 struct thandle *handle,
1663 struct dt_object *parent,
1664 const struct lu_fid *cfid,
1666 struct lov_mds_md_v1 *lmm,
1667 struct lov_ost_data_v1 *slot,
1668 int fl, __u32 ost_idx, int size)
1670 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1671 struct lu_buf ea_buf;
1678 magic = le32_to_cpu(lmm->lmm_magic);
1679 pattern = le32_to_cpu(lmm->lmm_pattern);
1680 count = le16_to_cpu(lmm->lmm_stripe_count);
1682 fid_to_ostid(cfid, oi);
1683 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1684 slot->l_ost_gen = cpu_to_le32(0);
1685 slot->l_ost_idx = cpu_to_le32(ost_idx);
1687 if (pattern & LOV_PATTERN_F_HOLE) {
1688 struct lov_ost_data_v1 *objs;
1691 if (magic == LOV_MAGIC_V1)
1692 objs = &lmm->lmm_objects[0];
1694 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1695 for (i = 0; i < count; i++, objs++) {
1696 if (lovea_slot_is_dummy(objs))
1700 /* If the @slot is the last dummy slot to be refilled,
1701 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1704 cpu_to_le32(pattern & ~LOV_PATTERN_F_HOLE);
1706 CDEBUG(D_LFSCK, "%s: remove layout HOLE for "DFID
1707 ": parent "DFID"\n", lfsck_lfsck2name(lfsck),
1708 PFID(cfid), PFID(lfsck_dto2fid(parent)));
1712 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
1713 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1720 static struct lov_ost_data_v1 *
1721 __lfsck_layout_new_v1_lovea(struct lov_mds_md_v1 *lmm,
1722 const struct lu_fid *pfid,
1723 __u32 stripe_size, __u32 ea_off,
1724 __u32 pattern, __u16 count)
1726 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1727 lmm->lmm_pattern = cpu_to_le32(pattern);
1728 fid_to_lmm_oi(pfid, &lmm->lmm_oi);
1729 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1730 lmm->lmm_stripe_size = cpu_to_le32(stripe_size);
1731 lmm->lmm_stripe_count = cpu_to_le16(count);
1732 lmm->lmm_layout_gen = cpu_to_le16(1);
1733 memset(&lmm->lmm_objects[0], 0,
1734 sizeof(struct lov_ost_data_v1) * count);
1736 return &lmm->lmm_objects[ea_off];
1739 static int lfsck_layout_new_v1_lovea(const struct lu_env *env,
1740 struct lfsck_instance *lfsck,
1741 struct ost_layout *ol,
1742 struct dt_object *parent,
1743 struct lu_buf *buf, __u32 ea_off,
1744 struct lov_mds_md_v1 **lmm,
1745 struct lov_ost_data_v1 **objs)
1748 __u32 stripe_size = ol->ol_stripe_size;
1749 __u32 pattern = LOV_PATTERN_RAID0;
1752 if (ol->ol_stripe_count != 0)
1753 count = ol->ol_stripe_count;
1757 size = lov_mds_md_size(count, LOV_MAGIC_V1);
1758 LASSERTF(buf->lb_len >= size,
1759 "buffer len %d is less than real size %d\n",
1760 (int)buf->lb_len, size);
1762 if (stripe_size == 0) {
1765 rc = lfsck_layout_get_def_stripesize(env, lfsck, &stripe_size);
1771 if (ol->ol_stripe_count > 1 ||
1772 (ol->ol_stripe_count == 0 && ea_off != 0)) {
1773 pattern |= LOV_PATTERN_F_HOLE;
1774 memset(&(*lmm)->lmm_objects[0], 0,
1775 count * sizeof(struct lov_ost_data_v1));
1778 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1779 stripe_size, ea_off, pattern, count);
1784 static int lfsck_layout_new_comp_lovea(const struct lu_env *env,
1785 struct ost_layout *ol,
1786 struct dt_object *parent,
1787 struct lu_buf *buf, __u32 ea_off,
1788 struct lov_mds_md_v1 **lmm,
1789 struct lov_ost_data_v1 **objs)
1791 struct lov_comp_md_v1 *lcm;
1792 struct lov_comp_md_entry_v1 *lcme;
1793 __u32 pattern = LOV_PATTERN_RAID0;
1794 __u32 offset = sizeof(*lcm) + sizeof(*lcme);
1795 int lcme_size = lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1796 int size = offset + lcme_size;
1798 LASSERTF(buf->lb_len >= size,
1799 "buffer len %d is less than real size %d\n",
1800 (int)buf->lb_len, size);
1803 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1804 lcm->lcm_size = cpu_to_le32(size);
1805 lcm->lcm_layout_gen = cpu_to_le32(1);
1807 lcm->lcm_entry_count = cpu_to_le16(1);
1809 lcme = &lcm->lcm_entries[0];
1810 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1811 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1812 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1813 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1814 lcme->lcme_offset = cpu_to_le32(offset);
1815 lcme->lcme_size = cpu_to_le32(lcme_size);
1816 if (ol->ol_stripe_count > 1)
1817 pattern |= LOV_PATTERN_F_HOLE;
1819 *lmm = buf->lb_buf + offset;
1820 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1821 ol->ol_stripe_size, ea_off,
1822 pattern, ol->ol_stripe_count);
1827 static int lfsck_layout_add_comp_comp(const struct lu_env *env,
1828 struct lfsck_instance *lfsck,
1829 struct thandle *handle,
1830 struct ost_layout *ol,
1831 struct dt_object *parent,
1832 const struct lu_fid *cfid,
1833 struct lu_buf *buf, __u32 ost_idx,
1834 __u32 ea_off, int pos)
1836 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1837 struct lov_comp_md_entry_v1 *lcme;
1838 struct lov_mds_md_v1 *lmm;
1839 struct lov_ost_data_v1 *objs;
1840 int added = sizeof(*lcme) +
1841 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1842 int size = le32_to_cpu(lcm->lcm_size) + added;
1846 __u32 pattern = LOV_PATTERN_RAID0;
1847 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1850 lu_buf_check_and_grow(buf, size);
1851 /* set the lcm again because lu_buf_check_and_grow() may
1852 * have reallocated the buf. */
1854 lcm->lcm_size = cpu_to_le32(size);
1855 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1856 lcm->lcm_entry_count = cpu_to_le16(count + 1);
1858 /* 1. Move the component bodies from [pos, count-1] to [pos+1, count]
1859 * with distance of 'added'. */
1862 for (i = pos; i < count; i++) {
1863 lcme = &lcm->lcm_entries[i];
1864 size += le32_to_cpu(lcme->lcme_size);
1867 offset = le32_to_cpu(lcm->lcm_entries[pos].lcme_offset);
1868 memmove(buf->lb_buf + offset + added,
1869 buf->lb_buf + offset, size);
1873 /* 2. Move the component header [0, pos-1] to [0, pos-1] with distance
1874 * of 'sizeof(struct lov_comp_md_entry_v1)' */
1876 for (i = 0; i < pos; i++) {
1877 lcme = &lcm->lcm_entries[i];
1878 size += le32_to_cpu(lcme->lcme_size);
1881 offset = le32_to_cpu(lcm->lcm_entries[0].lcme_offset);
1882 memmove(buf->lb_buf + offset + sizeof(*lcme),
1883 buf->lb_buf + offset, size);
1886 /* 3. Recalculate the enter offset for the component [pos, count-1] */
1887 for (i = count - 1; i >= pos; i--) {
1888 lcm->lcm_entries[i + 1] = lcm->lcm_entries[i];
1889 lcm->lcm_entries[i + 1].lcme_offset =
1890 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i + 1].
1891 lcme_offset) + added);
1894 /* 4. Recalculate the enter offset for the component [0, pos) */
1895 for (i = 0; i < pos; i++) {
1896 lcm->lcm_entries[i].lcme_offset =
1897 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i].
1898 lcme_offset) + sizeof(*lcme));
1901 offset = sizeof(*lcm) + sizeof(*lcme) * (count + 1) + size;
1902 /* 4. Insert the new component header (entry) at the slot 'pos'. */
1903 lcme = &lcm->lcm_entries[pos];
1904 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1905 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1906 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1907 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1908 lcme->lcme_offset = cpu_to_le32(offset);
1909 lcme->lcme_size = cpu_to_le32(lov_mds_md_size(ol->ol_stripe_count,
1912 if (ol->ol_stripe_count > 1)
1913 pattern |= LOV_PATTERN_F_HOLE;
1915 lmm = buf->lb_buf + offset;
1916 /* 5. Insert teh new component body at the 'offset'. */
1917 objs = __lfsck_layout_new_v1_lovea(lmm, lfsck_dto2fid(parent),
1918 ol->ol_stripe_size, ea_off,
1919 pattern, ol->ol_stripe_count);
1921 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1922 lmm, objs, LU_XATTR_REPLACE, ost_idx,
1923 le32_to_cpu(lcm->lcm_size));
1925 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant add new COMP for "
1926 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1927 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
1928 "comp_end %llu, %s LOV EA hole: rc = %d\n",
1929 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1930 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
1931 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
1932 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
1933 "with" : "without", rc);
1938 static int lfsck_layout_extend_v1v3_lovea(const struct lu_env *env,
1939 struct lfsck_instance *lfsck,
1940 struct thandle *handle,
1941 struct ost_layout *ol,
1942 struct dt_object *parent,
1943 const struct lu_fid *cfid,
1944 struct lu_buf *buf, __u32 ost_idx,
1947 struct lov_mds_md_v1 *lmm = buf->lb_buf;
1948 struct lov_ost_data_v1 *objs;
1949 __u16 count = le16_to_cpu(lmm->lmm_stripe_count);
1950 __u32 magic = le32_to_cpu(lmm->lmm_magic);
1956 /* The original LOVEA maybe re-generated via old filter_fid, at
1957 * that time, we do not know the stripe count and stripe size. */
1958 if (ol->ol_stripe_count > count)
1959 count = ol->ol_stripe_count;
1960 if (ol->ol_stripe_size != 0 &&
1961 ol->ol_stripe_size != le32_to_cpu(lmm->lmm_stripe_size))
1962 lmm->lmm_stripe_size = cpu_to_le32(ol->ol_stripe_size);
1964 if (magic == LOV_MAGIC_V1)
1965 objs = &lmm->lmm_objects[count];
1967 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[count];
1969 gap = ea_off - count;
1973 size = lov_mds_md_size(count, magic);
1974 LASSERTF(buf->lb_len >= size,
1975 "buffer len %d is less than real size %d\n",
1976 (int)buf->lb_len, size);
1979 memset(objs, 0, gap * sizeof(*objs));
1980 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1983 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1984 lmm->lmm_stripe_count = cpu_to_le16(count);
1987 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1988 lmm, objs, LU_XATTR_REPLACE, ost_idx, size);
1990 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1991 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1992 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
1993 "comp_end %llu, %s LOV EA hole: rc = %d\n",
1994 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1995 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
1996 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
1997 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
1998 "with" : "without", rc);
2004 * \retval +1: repaired
2005 * \retval 0: did nothing
2006 * \retval -ve: on error
2008 static int lfsck_layout_update_lovea(const struct lu_env *env,
2009 struct lfsck_instance *lfsck,
2010 struct thandle *handle,
2011 struct ost_layout *ol,
2012 struct dt_object *parent,
2013 const struct lu_fid *cfid,
2014 struct lu_buf *buf, int fl,
2015 __u32 ost_idx, __u32 ea_off)
2017 struct lov_mds_md_v1 *lmm = NULL;
2018 struct lov_ost_data_v1 *objs = NULL;
2022 if (ol->ol_comp_id != 0)
2023 rc = lfsck_layout_new_comp_lovea(env, ol, parent, buf, ea_off,
2026 rc = lfsck_layout_new_v1_lovea(env, lfsck, ol, parent, buf,
2027 ea_off, &lmm, &objs);
2030 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid,
2031 buf, lmm, objs, fl, ost_idx, rc);
2033 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant created layout EA for "
2034 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2035 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2036 "comp_end %llu, fl %d, %s LOV EA hole: rc = %d\n",
2037 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2038 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2039 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end, fl,
2040 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2041 "with" : "without", rc);
2046 static int __lfsck_layout_update_pfid(const struct lu_env *env,
2047 struct dt_object *child,
2048 const struct lu_fid *pfid,
2049 const struct ost_layout *ol, __u32 offset)
2051 struct dt_device *dev = lfsck_obj2dev(child);
2052 struct filter_fid *ff = &lfsck_env_info(env)->lti_ff;
2053 struct thandle *handle;
2054 struct lu_buf buf = { NULL };
2057 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2058 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2059 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2060 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2061 * parent MDT-object's layout EA. */
2062 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
2063 ost_layout_cpu_to_le(&ff->ff_layout, ol);
2064 lfsck_buf_init(&buf, ff, sizeof(*ff));
2066 handle = dt_trans_create(env, dev);
2068 RETURN(PTR_ERR(handle));
2070 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2074 rc = dt_trans_start_local(env, dev, handle);
2078 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2083 dt_trans_stop(env, dev, handle);
2089 * \retval +1: repaired
2090 * \retval 0: did nothing
2091 * \retval -ve: on error
2093 static int lfsck_layout_update_pfid(const struct lu_env *env,
2094 struct lfsck_component *com,
2095 struct dt_object *parent,
2096 struct lu_fid *cfid,
2097 struct dt_device *cdev,
2098 struct ost_layout *ol, __u32 ea_off)
2100 struct dt_object *child;
2104 child = lfsck_object_find_by_dev(env, cdev, cfid);
2106 RETURN(PTR_ERR(child));
2108 rc = __lfsck_layout_update_pfid(env, child,
2109 lu_object_fid(&parent->do_lu),
2111 lfsck_object_put(env, child);
2113 RETURN(rc == 0 ? 1 : rc);
2116 static int lfsck_lovea_size(struct ost_layout *ol, __u32 ea_off)
2118 if (ol->ol_comp_id != 0)
2119 return sizeof(struct lov_comp_md_v1) +
2120 sizeof(struct lov_comp_md_entry_v1) +
2121 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2123 if (ol->ol_stripe_count != 0)
2124 return lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2126 return lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2130 * This function will create the MDT-object with the given (partial) LOV EA.
2132 * Under some data corruption cases, the MDT-object of the file may be lost,
2133 * but its OST-objects, or some of them are there. The layout LFSCK needs to
2134 * re-create the MDT-object with the orphan OST-object(s) information.
2136 * On the other hand, the LFSCK may has created some OST-object for repairing
2137 * dangling LOV EA reference, but as the LFSCK processing, it may find that
2138 * the old OST-object is there and should replace the former new created OST
2139 * object. Unfortunately, some others have modified such newly created object.
2140 * To keep the data (both new and old), the LFSCK will create MDT-object with
2141 * new FID to reference the original OST-object.
2143 * \param[in] env pointer to the thread context
2144 * \param[in] com pointer to the lfsck component
2145 * \param[in] ltd pointer to target device descriptor
2146 * \param[in] rec pointer to the record for the orphan OST-object
2147 * \param[in] cfid pointer to FID for the orphan OST-object
2148 * \param[in] infix additional information, such as the FID for original
2149 * MDT-object and the stripe offset in the LOV EA
2150 * \param[in] type the type for describing why the orphan MDT-object is
2151 * created. The rules are as following:
2153 * type "C": Multiple OST-objects claim the same MDT-object and the
2154 * same slot in the layout EA. Then the LFSCK will create
2155 * new MDT-object(s) to hold the conflict OST-object(s).
2157 * type "N": The orphan OST-object does not know which one was the
2158 * real parent MDT-object, so the LFSCK uses new FID for
2159 * its parent MDT-object.
2161 * type "R": The orphan OST-object knows its parent MDT-object FID,
2162 * but does not know the position (the file name) in the
2165 * type "D": The MDT-object is a directory, it may knows its parent
2166 * but because there is no valid linkEA, the LFSCK cannot
2167 * know where to put it back to the namespace.
2168 * type "O": The MDT-object has no linkEA, and there is no name
2169 * entry that references the MDT-object.
2171 * type "P": The orphan object to be created was a parent directory
2172 * of some MDT-object which linkEA shows that the @orphan
2173 * object is missing.
2175 * The orphan name will be like:
2176 * ${FID}-${infix}-${type}-${conflict_version}
2178 * \param[in] ea_off the stripe offset in the LOV EA
2180 * \retval positive on repaired something
2181 * \retval 0 if needs to repair nothing
2182 * \retval negative error number on failure
2184 static int lfsck_layout_recreate_parent(const struct lu_env *env,
2185 struct lfsck_component *com,
2186 struct lfsck_tgt_desc *ltd,
2187 struct lu_orphan_rec_v2 *rec,
2188 struct lu_fid *cfid,
2193 struct lfsck_thread_info *info = lfsck_env_info(env);
2194 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
2195 char *name = info->lti_key;
2196 struct lu_attr *la = &info->lti_la2;
2197 struct dt_object_format *dof = &info->lti_dof;
2198 struct lfsck_instance *lfsck = com->lc_lfsck;
2199 struct ost_layout *ol = &rec->lor_layout;
2200 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
2201 struct lu_fid *tfid = &info->lti_fid3;
2202 struct dt_device *dev = lfsck->li_bottom;
2203 struct dt_object *lpf = lfsck->li_lpf_obj;
2204 struct dt_object *pobj = NULL;
2205 struct dt_object *cobj = NULL;
2206 struct thandle *th = NULL;
2207 struct lu_buf *ea_buf = &info->lti_big_buf;
2208 struct lu_buf lov_buf;
2209 struct lfsck_lock_handle *llh = &info->lti_llh;
2210 struct linkea_data ldata = { NULL };
2211 struct lu_buf linkea_buf;
2212 const struct lu_name *pname;
2218 if (unlikely(lpf == NULL))
2219 GOTO(log, rc = -ENXIO);
2221 /* We use two separated transactions to repair the inconsistency.
2223 * 1) create the MDT-object locally.
2224 * 2) update the OST-object's PFID EA if necessary.
2226 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
2227 * updated when the layout LFSCK run next time.
2229 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
2230 * when the layout LFSCK run next time. */
2232 if (fid_is_zero(pfid)) {
2233 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
2237 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2239 GOTO(log, rc = PTR_ERR(cobj));
2242 pobj = lfsck_object_find_by_dev(env, dev, pfid);
2244 GOTO(log, rc = PTR_ERR(pobj));
2246 LASSERT(infix != NULL);
2247 LASSERT(type != NULL);
2249 memset(la, 0, sizeof(*la));
2250 la->la_uid = rec->lor_rec.lor_uid;
2251 la->la_gid = rec->lor_rec.lor_gid;
2252 la->la_mode = S_IFREG | S_IRUSR;
2253 la->la_valid = LA_MODE | LA_UID | LA_GID;
2255 memset(dof, 0, sizeof(*dof));
2256 dof->dof_type = dt_mode_to_dft(S_IFREG);
2257 /* Because the dof->dof_reg.striped = 0, the LOD will not create
2258 * the stripe(s). The LFSCK will specify the LOV EA via
2259 * lfsck_layout_update_lovea(). */
2261 size = lfsck_lovea_size(ol, ea_off);
2262 if (ea_buf->lb_len < size) {
2263 lu_buf_realloc(ea_buf, size);
2264 if (ea_buf->lb_buf == NULL)
2265 GOTO(log, rc = -ENOMEM);
2270 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
2272 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2273 (const struct dt_key *)name);
2274 if (rc != 0 && rc != -ENOENT)
2278 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
2279 MDS_INODELOCK_UPDATE, LCK_PW);
2283 /* Re-check whether the name conflict with othrs after taken
2285 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2286 (const struct dt_key *)name);
2287 if (unlikely(rc == 0)) {
2295 pname = lfsck_name_get_const(env, name, strlen(name));
2296 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
2297 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
2301 /* The 1st transaction. */
2302 th = dt_trans_create(env, dev);
2304 GOTO(unlock, rc = PTR_ERR(th));
2306 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
2310 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
2311 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
2312 LU_XATTR_CREATE, th);
2316 dtrec->rec_fid = pfid;
2317 dtrec->rec_type = S_IFREG;
2318 rc = dt_declare_insert(env, lpf,
2319 (const struct dt_rec *)dtrec,
2320 (const struct dt_key *)name, th);
2324 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2325 ldata.ld_leh->leh_len);
2326 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
2327 XATTR_NAME_LINK, 0, th);
2331 rc = dt_trans_start_local(env, dev, th);
2335 dt_write_lock(env, pobj, 0);
2336 rc = dt_create(env, pobj, la, NULL, dof, th);
2338 rc = lfsck_layout_update_lovea(env, lfsck, th, ol, pobj, cfid,
2339 &lov_buf, LU_XATTR_CREATE, ltd->ltd_index, ea_off);
2340 dt_write_unlock(env, pobj);
2344 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
2345 (const struct dt_key *)name, th, 1);
2349 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
2350 if (rc == 0 && cobj != NULL) {
2351 dt_trans_stop(env, dev, th);
2354 /* The 2nd transaction. */
2355 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ol, ea_off);
2362 dt_trans_stop(env, dev, th);
2368 if (cobj != NULL && !IS_ERR(cobj))
2369 lfsck_object_put(env, cobj);
2370 if (pobj != NULL && !IS_ERR(pobj))
2371 lfsck_object_put(env, pobj);
2374 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2375 "recreate the lost MDT-object: parent "DFID
2376 ", child "DFID", OST-index %u, stripe-index %u, "
2377 "infix %s, type %s: rc = %d\n",
2378 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2379 ltd->ltd_index, ea_off, infix, type, rc);
2381 return rc >= 0 ? 1 : rc;
2384 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2385 struct lfsck_component *com,
2386 const struct lu_fid *fid,
2389 struct lfsck_thread_info *info = lfsck_env_info(env);
2390 struct lfsck_request *lr = &info->lti_lr;
2391 struct lfsck_instance *lfsck = com->lc_lfsck;
2392 struct lfsck_tgt_desc *ltd;
2393 struct ptlrpc_request *req;
2394 struct lfsck_request *tmp;
2395 struct obd_export *exp;
2399 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2400 if (unlikely(ltd == NULL))
2404 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2405 GOTO(put, rc = -EOPNOTSUPP);
2407 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2409 GOTO(put, rc = -ENOMEM);
2411 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2413 ptlrpc_request_free(req);
2418 memset(lr, 0, sizeof(*lr));
2419 lr->lr_event = LE_CONDITIONAL_DESTROY;
2420 lr->lr_active = LFSCK_TYPE_LAYOUT;
2423 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2425 ptlrpc_request_set_replen(req);
2427 rc = ptlrpc_queue_wait(req);
2428 ptlrpc_req_finished(req);
2438 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2439 struct lfsck_component *com,
2440 struct lfsck_request *lr)
2442 struct lfsck_thread_info *info = lfsck_env_info(env);
2443 struct lu_attr *la = &info->lti_la;
2444 union ldlm_policy_data *policy = &info->lti_policy;
2445 struct ldlm_res_id *resid = &info->lti_resid;
2446 struct lfsck_instance *lfsck = com->lc_lfsck;
2447 struct dt_device *dev = lfsck->li_bottom;
2448 struct lu_fid *fid = &lr->lr_fid;
2449 struct dt_object *obj;
2450 struct thandle *th = NULL;
2451 struct lustre_handle lh = { 0 };
2456 obj = lfsck_object_find_by_dev(env, dev, fid);
2458 RETURN(PTR_ERR(obj));
2460 dt_read_lock(env, obj, 0);
2461 if (dt_object_exists(obj) == 0 ||
2462 lfsck_is_dead_obj(obj)) {
2463 dt_read_unlock(env, obj);
2465 GOTO(put, rc = -ENOENT);
2468 /* Get obj's attr without lock firstly. */
2469 rc = dt_attr_get(env, obj, la);
2470 dt_read_unlock(env, obj);
2474 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2475 GOTO(put, rc = -ETXTBSY);
2477 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2478 LASSERT(lfsck->li_namespace != NULL);
2480 memset(policy, 0, sizeof(*policy));
2481 policy->l_extent.end = OBD_OBJECT_EOF;
2482 ost_fid_build_resid(fid, resid);
2483 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2484 policy, LCK_EX, &flags, ldlm_blocking_ast,
2485 ldlm_completion_ast, NULL, NULL, 0,
2486 LVB_T_NONE, NULL, &lh);
2488 GOTO(put, rc = -EIO);
2490 dt_write_lock(env, obj, 0);
2491 /* Get obj's attr within lock again. */
2492 rc = dt_attr_get(env, obj, la);
2496 if (la->la_ctime != 0)
2497 GOTO(unlock, rc = -ETXTBSY);
2499 th = dt_trans_create(env, dev);
2501 GOTO(unlock, rc = PTR_ERR(th));
2503 rc = dt_declare_ref_del(env, obj, th);
2507 rc = dt_declare_destroy(env, obj, th);
2511 rc = dt_trans_start_local(env, dev, th);
2515 rc = dt_ref_del(env, obj, th);
2519 rc = dt_destroy(env, obj, th);
2521 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2522 "OST-object "DFID" that was created for reparing "
2523 "dangling referenced case. But the original missing "
2524 "OST-object is found now.\n",
2525 lfsck_lfsck2name(lfsck), PFID(fid));
2530 dt_trans_stop(env, dev, th);
2533 dt_write_unlock(env, obj);
2534 ldlm_lock_decref(&lh, LCK_EX);
2537 lfsck_object_put(env, obj);
2543 * Some OST-object has occupied the specified layout EA slot.
2544 * Such OST-object may be generated by the LFSCK when repair
2545 * dangling referenced MDT-object, which can be indicated by
2546 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2547 * is true and such OST-object has not been modified yet, we
2548 * will replace it with the orphan OST-object; otherwise the
2549 * LFSCK will create new MDT-object to reference the orphan.
2551 * \retval +1: repaired
2552 * \retval 0: did nothing
2553 * \retval -ve: on error
2555 static int lfsck_layout_conflict_create(const struct lu_env *env,
2556 struct lfsck_component *com,
2557 struct lfsck_tgt_desc *ltd,
2558 struct lu_orphan_rec_v2 *rec,
2559 struct dt_object *parent,
2560 struct lu_fid *cfid,
2561 struct lu_buf *ea_buf,
2562 struct lov_mds_md_v1 *lmm,
2563 struct lov_ost_data_v1 *slot,
2564 __u32 ea_off, int lovea_size)
2566 struct lfsck_thread_info *info = lfsck_env_info(env);
2567 struct lu_fid *cfid2 = &info->lti_fid2;
2568 struct ost_id *oi = &info->lti_oi;
2569 struct dt_device *dev = lfsck_obj2dev(parent);
2570 struct thandle *th = NULL;
2571 struct lustre_handle lh = { 0 };
2572 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2576 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2577 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2581 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2582 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2586 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2587 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2592 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2594 /* If the conflict OST-obejct is not created for fixing dangling
2595 * referenced MDT-object in former LFSCK check/repair, or it has
2596 * been modified by others, then we cannot destroy it. Re-create
2597 * a new MDT-object for the orphan OST-object. */
2598 if (rc == -ETXTBSY) {
2599 /* No need the layout lock on the original parent. */
2600 lfsck_ibits_unlock(&lh, LCK_EX);
2602 fid_zero(&rec->lor_rec.lor_fid);
2603 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2604 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2606 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2607 info->lti_tmpbuf, "C", ea_off);
2612 if (rc != 0 && rc != -ENOENT)
2615 th = dt_trans_create(env, dev);
2617 GOTO(unlock, rc = PTR_ERR(th));
2619 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2620 LU_XATTR_REPLACE, th);
2624 rc = dt_trans_start_local(env, dev, th);
2628 dt_write_lock(env, parent, 0);
2629 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2630 rc = lfsck_layout_refill_lovea(env, com->lc_lfsck, th, parent, cfid,
2631 ea_buf, lmm, slot, LU_XATTR_REPLACE,
2632 ltd->ltd_index, lovea_size);
2633 dt_write_unlock(env, parent);
2638 dt_trans_stop(env, dev, th);
2641 lfsck_ibits_unlock(&lh, LCK_EX);
2644 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2645 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2646 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2647 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2648 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2651 return rc >= 0 ? 1 : rc;
2655 * \retval +1: repaired
2656 * \retval 0: did nothing
2657 * \retval -ve: on error
2659 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2660 struct lfsck_component *com,
2661 struct lfsck_tgt_desc *ltd,
2662 struct lu_orphan_rec_v2 *rec,
2663 struct dt_object *parent,
2664 struct lu_fid *cfid,
2665 __u32 ost_idx, __u32 ea_off)
2667 struct lfsck_thread_info *info = lfsck_env_info(env);
2668 struct lu_buf *buf = &info->lti_big_buf;
2669 struct lu_fid *fid = &info->lti_fid2;
2670 struct ost_id *oi = &info->lti_oi;
2671 struct lfsck_instance *lfsck = com->lc_lfsck;
2672 struct dt_device *dt = lfsck_obj2dev(parent);
2673 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2674 struct ost_layout *ol = &rec->lor_layout;
2675 struct lov_comp_md_v1 *lcm = NULL;
2676 struct lov_comp_md_entry_v1 *lcme = NULL;
2677 struct thandle *handle = NULL;
2679 struct lov_mds_md_v1 *lmm;
2680 struct lov_ost_data_v1 *objs;
2681 struct lustre_handle lh = { 0 };
2689 bool locked = false;
2692 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2693 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2696 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2697 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2698 "stripe-index %u, comp_id %u, comp_start %llu, "
2699 "comp_end %llu: rc = %d\n",
2700 lfsck_lfsck2name(lfsck), PFID(cfid),
2701 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off,
2702 ol->ol_comp_id, ol->ol_comp_start,
2703 ol->ol_comp_end, rc);
2710 dt_write_unlock(env, parent);
2714 if (handle != NULL) {
2715 dt_trans_stop(env, dt, handle);
2720 GOTO(unlock_layout, rc);
2723 if (buf->lb_len < lovea_size) {
2724 lu_buf_realloc(buf, lovea_size);
2725 if (buf->lb_buf == NULL)
2726 GOTO(unlock_layout, rc = -ENOMEM);
2729 if (!(bk->lb_param & LPF_DRYRUN)) {
2730 handle = dt_trans_create(env, dt);
2732 GOTO(unlock_layout, rc = PTR_ERR(handle));
2734 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2739 rc = dt_trans_start_local(env, dt, handle);
2744 dt_write_lock(env, parent, 0);
2746 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2747 if (rc == -ERANGE) {
2748 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2751 } else if (rc == -ENODATA || rc == 0) {
2752 lovea_size = lfsck_lovea_size(ol, ea_off);
2753 /* If the declared is not big enough, re-try. */
2754 if (buf->lb_len < lovea_size) {
2758 fl = LU_XATTR_CREATE;
2759 } else if (rc < 0) {
2760 GOTO(unlock_parent, rc);
2761 } else if (unlikely(buf->lb_len == 0)) {
2764 fl = LU_XATTR_REPLACE;
2768 if (fl == LU_XATTR_CREATE) {
2769 if (bk->lb_param & LPF_DRYRUN)
2770 GOTO(unlock_parent, rc = 1);
2772 LASSERT(buf->lb_len >= lovea_size);
2774 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2775 cfid, buf, fl, ost_idx, ea_off);
2777 GOTO(unlock_parent, rc);
2781 rc1 = lfsck_layout_verify_header(parent, lmm);
2783 /* If the LOV EA crashed, the rebuild it. */
2784 if (rc1 == -EINVAL) {
2785 if (bk->lb_param & LPF_DRYRUN)
2786 GOTO(unlock_parent, rc = 1);
2788 LASSERT(buf->lb_len >= lovea_size);
2790 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2791 cfid, buf, fl, ost_idx, ea_off);
2793 GOTO(unlock_parent, rc);
2796 /* For other unknown magic/pattern, keep the current LOV EA. */
2798 GOTO(unlock_parent, rc = rc1);
2800 magic = le32_to_cpu(lmm->lmm_magic);
2801 if (magic == LOV_MAGIC_COMP_V1) {
2806 count = le16_to_cpu(lcm->lcm_entry_count);
2807 for (i = 0; i < count; i++) {
2808 lcme = &lcm->lcm_entries[i];
2809 start = le64_to_cpu(lcme->lcme_extent.e_start);
2810 end = le64_to_cpu(lcme->lcme_extent.e_end);
2812 if (end <= ol->ol_comp_start)
2815 if (start >= ol->ol_comp_end)
2818 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
2819 magic = le32_to_cpu(lmm->lmm_magic);
2820 flags = le32_to_cpu(lcme->lcme_flags);
2824 rc = lfsck_layout_add_comp_comp(env, lfsck, handle, ol, parent,
2825 cfid, buf, ost_idx, ea_off, i);
2827 GOTO(unlock_parent, rc);
2831 count = le16_to_cpu(lmm->lmm_stripe_count);
2833 GOTO(unlock_parent, rc = -EINVAL);
2836 /* Exceed the current end of MDT-object layout EA. Then extend it. */
2837 if (count <= ea_off) {
2838 if (bk->lb_param & LPF_DRYRUN)
2839 GOTO(unlock_parent, rc = 1);
2841 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2842 /* If the declared is not big enough, re-try. */
2843 if (buf->lb_len < lovea_size) {
2848 if (lcme && !(flags & LCME_FL_INIT))
2849 lcme->lcme_flags = cpu_to_le32(flags | LCME_FL_INIT);
2851 rc = lfsck_layout_extend_v1v3_lovea(env, lfsck, handle, ol,
2852 parent, cfid, buf, ost_idx, ea_off);
2854 GOTO(unlock_parent, rc);
2857 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2859 if (magic == LOV_MAGIC_V1) {
2860 objs = &lmm->lmm_objects[0];
2862 LASSERT(magic == LOV_MAGIC_V3);
2863 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2866 for (i = 0; i < count; i++, objs++) {
2867 /* The MDT-object was created via lfsck_layout_recover_create()
2868 * by others before, and we fill the dummy layout EA. */
2869 if ((lcme && !(flags & LCME_FL_INIT)) ||
2870 lovea_slot_is_dummy(objs)) {
2874 if (bk->lb_param & LPF_DRYRUN)
2875 GOTO(unlock_parent, rc = 1);
2877 lmm->lmm_layout_gen =
2878 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2882 if (le32_to_cpu(lmm->lmm_stripe_size) !=
2883 ol->ol_stripe_size ||
2884 le16_to_cpu(lmm->lmm_stripe_count) !=
2885 ol->ol_stripe_count ||
2886 le64_to_cpu(lcme->lcme_extent.e_start) !=
2887 ol->ol_comp_start ||
2888 le64_to_cpu(lcme->lcme_extent.e_end) !=
2890 CDEBUG(D_LFSCK, "%s: found invalid "
2891 "component for "DFID ": parent "DFID
2892 ", stripe-index %u, stripe_size %u, "
2893 "stripe_count %u, comp_id %u, "
2894 "comp_start %llu, comp_end %llu, "
2895 "cur_stripe_size %u, "
2896 "cur_stripe_count %u, "
2897 "cur_comp_start %llu, "
2898 "cur_comp_end %llu\n",
2899 lfsck_lfsck2name(lfsck), PFID(cfid),
2900 PFID(lfsck_dto2fid(parent)), ea_off,
2902 ol->ol_stripe_count, ol->ol_comp_id,
2903 ol->ol_comp_start, ol->ol_comp_end,
2904 le32_to_cpu(lmm->lmm_stripe_size),
2905 le16_to_cpu(lmm->lmm_stripe_count),
2906 le64_to_cpu(lcme->lcme_extent.e_start),
2907 le64_to_cpu(lcme->lcme_extent.e_end));
2909 GOTO(unlock_parent, rc = -EINVAL);
2912 le32_add_cpu(&lcm->lcm_layout_gen, 1);
2913 lovea_size = le32_to_cpu(lcm->lcm_size);
2914 if (!(flags & LCME_FL_INIT))
2915 lcme->lcme_flags = cpu_to_le32(flags |
2919 LASSERTF(buf->lb_len >= lovea_size,
2920 "buffer len %d is less than real size %d\n",
2921 (int)buf->lb_len, (int)lovea_size);
2923 rc = lfsck_layout_refill_lovea(env, lfsck, handle,
2924 parent, cfid, buf, lmm, objs,
2925 fl, ost_idx, lovea_size);
2927 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2928 "dummy layout slot for "DFID": parent "DFID
2929 ", OST-index %u, stripe-index %u: rc = %d\n",
2930 lfsck_lfsck2name(lfsck), PFID(cfid),
2931 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2933 GOTO(unlock_parent, rc);
2936 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2937 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2939 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2940 "invalid layout EA at the slot %d, index %u\n",
2941 lfsck_lfsck2name(lfsck),
2942 PFID(lfsck_dto2fid(parent)), i,
2943 le32_to_cpu(objs->l_ost_idx));
2945 GOTO(unlock_parent, rc);
2948 /* It should be rare case, the slot is there, but the LFSCK
2949 * does not handle it during the first-phase cycle scanning. */
2950 if (unlikely(lu_fid_eq(fid, cfid))) {
2952 GOTO(unlock_parent, rc = 0);
2954 /* Rare case that the OST-object index
2955 * does not match the parent MDT-object
2956 * layout EA. We trust the later one. */
2957 if (bk->lb_param & LPF_DRYRUN)
2958 GOTO(unlock_parent, rc = 1);
2960 dt_write_unlock(env, parent);
2962 dt_trans_stop(env, dt, handle);
2963 lfsck_ibits_unlock(&lh, LCK_EX);
2964 rc = lfsck_layout_update_pfid(env, com, parent,
2968 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2969 "updated OST-object's pfid for "DFID
2970 ": parent "DFID", OST-index %u, "
2971 "stripe-index %u: rc = %d\n",
2972 lfsck_lfsck2name(lfsck), PFID(cfid),
2973 PFID(lfsck_dto2fid(parent)),
2974 ltd->ltd_index, i, rc);
2981 /* The MDT-object exists, but related layout EA slot is occupied
2983 if (bk->lb_param & LPF_DRYRUN)
2984 GOTO(unlock_parent, rc = 1);
2986 dt_write_unlock(env, parent);
2988 dt_trans_stop(env, dt, handle);
2989 lfsck_ibits_unlock(&lh, LCK_EX);
2990 if (magic == LOV_MAGIC_V1)
2991 objs = &lmm->lmm_objects[ea_off];
2993 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2994 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2995 buf, lmm, objs, ea_off, lovea_size);
3001 dt_write_unlock(env, parent);
3005 dt_trans_stop(env, dt, handle);
3008 lfsck_ibits_unlock(&lh, LCK_EX);
3013 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
3014 struct lfsck_component *com,
3015 struct lfsck_tgt_desc *ltd,
3016 struct lu_orphan_rec_v2 *rec,
3017 struct lu_fid *cfid)
3019 struct lfsck_layout *lo = com->lc_file_ram;
3020 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
3021 struct dt_object *parent = NULL;
3022 __u32 ea_off = pfid->f_stripe_idx;
3026 if (!fid_is_sane(cfid))
3027 GOTO(out, rc = -EINVAL);
3030 if (fid_is_zero(pfid)) {
3031 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3036 if (!fid_is_sane(pfid))
3037 GOTO(out, rc = -EINVAL);
3039 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
3041 GOTO(out, rc = PTR_ERR(parent));
3043 if (unlikely(dt_object_remote(parent) != 0))
3044 GOTO(put, rc = -EXDEV);
3046 if (dt_object_exists(parent) == 0) {
3047 lfsck_object_put(env, parent);
3048 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3053 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
3054 GOTO(put, rc = -EISDIR);
3056 /* The orphan OST-object claims to be the parent's stripe, then
3057 * related dangling record in the trace file is meaningless. */
3058 rc = lfsck_layout_del_dangling_rec(env, com, pfid,
3059 rec->lor_layout.ol_comp_id, ea_off);
3060 if (rc && rc != -ENOENT)
3063 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
3064 ltd->ltd_index, ea_off);
3070 lfsck_object_put(env, parent);
3072 /* The layout EA is changed, need to be reloaded next time. */
3073 dt_object_put_nocache(env, parent);
3076 down_write(&com->lc_sem);
3077 com->lc_new_scanned++;
3078 com->lc_new_checked++;
3080 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
3082 } else if (rc < 0) {
3083 lo->ll_objs_failed_phase2++;
3085 up_write(&com->lc_sem);
3090 static int lfsck_layout_scan_orphan(const struct lu_env *env,
3091 struct lfsck_component *com,
3092 struct lfsck_tgt_desc *ltd)
3094 struct lfsck_assistant_data *lad = com->lc_data;
3095 struct lfsck_instance *lfsck = com->lc_lfsck;
3096 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3097 struct lfsck_thread_info *info = lfsck_env_info(env);
3098 struct lu_fid *fid = &info->lti_fid;
3099 struct dt_object *obj;
3100 const struct dt_it_ops *iops;
3105 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
3106 "scanning for OST%04x\n",
3107 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3109 if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
3110 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
3111 "scanning for OST%04x\n",
3112 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3117 fid->f_seq = fid_idif_seq(0, ltd->ltd_index);
3118 fid->f_oid = fid->f_ver = 0;
3120 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
3121 if (unlikely(IS_ERR(obj)))
3122 GOTO(log, rc = PTR_ERR(obj));
3124 rc = obj->do_ops->do_index_try(env, obj,
3125 &dt_lfsck_layout_orphan_features);
3129 iops = &obj->do_index_ops->dio_it;
3130 di = iops->init(env, obj, 0);
3132 GOTO(put, rc = PTR_ERR(di));
3134 rc = iops->load(env, di, 0);
3136 /* -ESRCH means that the orphan OST-objects rbtree has been
3137 * cleanup because of the OSS server restart or other errors. */
3138 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
3143 rc = iops->next(env, di);
3155 struct lu_orphan_rec_v2 *rec = &info->lti_rec;
3157 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3158 unlikely(!thread_is_running(&lfsck->li_thread)))
3161 key = iops->key(env, di);
3162 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
3163 /* Remote target OST may be runnning old LFSCK */
3164 memset(rec, 0, sizeof(*rec));
3165 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
3167 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
3168 &com->lc_fid_latest_scanned_phase2);
3169 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3172 lfsck_control_speed_by_self(com);
3174 rc = iops->next(env, di);
3175 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
3182 iops->fini(env, di);
3184 lfsck_object_put(env, obj);
3187 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
3188 "scanning for OST%04x: rc = %d\n",
3189 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
3191 return rc > 0 ? 0 : rc;
3194 static int lfsck_lmm2layout(struct lov_mds_md_v1 *lmm, struct ost_layout *ol,
3197 __u32 magic = le32_to_cpu(lmm->lmm_magic);
3201 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3) {
3202 ol->ol_stripe_size = lmm->lmm_stripe_size;
3203 ol->ol_stripe_count = lmm->lmm_stripe_count;
3204 ol->ol_comp_start = 0;
3205 ol->ol_comp_end = 0;
3207 } else if (magic == LOV_MAGIC_COMP_V1) {
3208 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
3209 struct lov_comp_md_entry_v1 *lcme = NULL;
3210 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3213 for (i = 0; i < count; i++) {
3214 lcme = &lcm->lcm_entries[i];
3215 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3216 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3223 /* The comp has been removed, do nothing. */
3227 lmm = (void *)lmm + le32_to_cpu(lcme->lcme_offset);
3228 ol->ol_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
3229 ol->ol_stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
3230 ol->ol_comp_start = le64_to_cpu(lcme->lcme_extent.e_start);
3231 ol->ol_comp_end = le64_to_cpu(lcme->lcme_extent.e_end);
3232 ol->ol_comp_id = le32_to_cpu(lcme->lcme_id);
3234 GOTO(out, rc = -EINVAL);
3244 * Repair the MDT-object with dangling LOV EA reference.
3246 * we need to repair the inconsistency according to the users' requirement:
3248 * 1) Keep the inconsistency there and report the inconsistency case,
3249 * then give the chance to the application to find related issues,
3250 * and the users can make the decision about how to handle it with
3251 * more human knownledge. (by default)
3253 * 2) Re-create the missing OST-object with the FID/owner information.
3255 * \param[in] env pointer to the thread context
3256 * \param[in] com the layout LFSCK component
3257 * \param[in] parent the MDT-object with dangling LOV EA reference
3258 * \param[in] child the OST-object to be created
3259 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3260 * \param[in] ea_off the offset of the OST-object in the LOV EA
3261 * \param[in] ost_idx the index of OST on which the OST-object resides
3263 * \retval +1 for repair successfully
3264 * \retval 0 for did nothing
3265 * \retval negative error number on failure
3267 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
3268 struct lfsck_component *com,
3269 struct dt_object *parent,
3270 struct dt_object *child,
3271 __u32 comp_id, __u32 ea_off,
3272 __u32 ost_idx, bool log)
3274 struct lfsck_thread_info *info = lfsck_env_info(env);
3275 struct filter_fid *ff = &info->lti_ff;
3276 struct ost_layout *ol = &ff->ff_layout;
3277 struct dt_object_format *dof = &info->lti_dof;
3278 struct lu_attr *la = &info->lti_la;
3279 struct lfsck_instance *lfsck = com->lc_lfsck;
3280 struct dt_device *dev = lfsck_obj2dev(child);
3281 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3282 const struct lu_fid *cfid = lfsck_dto2fid(child);
3283 struct lu_buf *tbuf = &info->lti_big_buf;
3284 struct thandle *handle;
3286 struct lustre_handle lh = { 0 };
3290 if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
3293 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3294 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3299 rc = dt_attr_get(env, parent, la);
3303 la->la_mode = S_IFREG | 0666;
3304 la->la_atime = la->la_mtime = la->la_ctime = 0;
3305 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
3306 LA_ATIME | LA_MTIME | LA_CTIME;
3307 memset(dof, 0, sizeof(*dof));
3308 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
3309 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
3310 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3311 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3312 * parent MDT-object's layout EA. */
3313 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
3315 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3319 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, comp_id);
3323 buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
3324 handle = dt_trans_create(env, dev);
3326 GOTO(unlock1, rc = PTR_ERR(handle));
3328 rc = dt_declare_create(env, child, la, NULL, dof, handle);
3332 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
3333 LU_XATTR_CREATE, handle);
3337 rc = dt_trans_start_local(env, dev, handle);
3341 dt_read_lock(env, parent, 0);
3342 if (unlikely(lfsck_is_dead_obj(parent)))
3343 GOTO(unlock2, rc = 0);
3345 if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
3346 struct ost_id *oi = &info->lti_oi;
3347 struct lu_fid *tfid = &info->lti_fid2;
3348 struct lu_buf *lovea = &info->lti_big_buf;
3349 struct lov_mds_md_v1 *lmm;
3350 struct lov_ost_data_v1 *objs;
3355 rc = lfsck_layout_get_lovea(env, parent, lovea);
3359 lmm = lovea->lb_buf;
3360 magic = le32_to_cpu(lmm->lmm_magic);
3361 if (magic == LOV_MAGIC_COMP_V1) {
3362 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3363 struct lov_comp_md_entry_v1 *lcme;
3364 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3367 for (i = 0; i < count; i++) {
3368 lcme = &lcm->lcm_entries[i];
3369 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3370 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3373 lmm = lovea->lb_buf +
3374 le32_to_cpu(lcme->lcme_offset);
3375 magic = le32_to_cpu(lmm->lmm_magic);
3380 /* Someone removed the component, do nothing. */
3381 GOTO(unlock2, rc = 0);
3385 count = le16_to_cpu(lmm->lmm_stripe_count);
3386 /* Someone changed the LOV EA, do nothing. */
3387 if (count <= ea_off)
3388 GOTO(unlock2, rc = 0);
3390 if (magic == LOV_MAGIC_V1) {
3391 objs = &lmm->lmm_objects[ea_off];
3393 LASSERT(magic == LOV_MAGIC_V3);
3395 objs = &((struct lov_mds_md_v3 *)lmm)->\
3396 lmm_objects[ea_off];
3399 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3400 idx2 = le32_to_cpu(objs->l_ost_idx);
3401 rc = ostid_to_fid(tfid, oi, idx2);
3402 /* Someone changed the LOV EA, do nothing. */
3403 if (rc != 0 || !lu_fid_eq(tfid, cfid))
3407 rc = dt_create(env, child, la, NULL, dof, handle);
3411 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
3417 dt_read_unlock(env, parent);
3420 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3423 lfsck_ibits_unlock(&lh, LCK_EX);
3427 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3428 "dangling reference for: parent "DFID", child "
3429 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: "
3431 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3432 comp_id, ea_off, ost_idx,
3433 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3434 "Create the lost OST-object as required" :
3435 "Keep the MDT-object there by default", rc);
3441 * Repair the MDT-object with dangling LOV EA reference.
3443 * Prepare parameters and call __lfsck_layout_repair_dangling()
3444 * to repair the dangling LOV EA reference.
3446 * \param[in] env pointer to the thread context
3447 * \param[in] com the layout LFSCK component
3448 * \param[in] pfid the MDT-object's FID
3449 * \param[in] cfid the FID for the OST-object to be created
3450 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3451 * \param[in] ea_off the offset of the OST-object in the LOV EA
3452 * \param[in] ost_idx the index of OST on which the OST-object resides
3454 * \retval +1 for repair successfully
3455 * \retval 0 for did nothing
3456 * \retval negative error number on failure
3458 static int lfsck_layout_repair_dangling(const struct lu_env *env,
3459 struct lfsck_component *com,
3460 const struct lu_fid *pfid,
3461 const struct lu_fid *cfid,
3462 __u32 comp_id, __u32 ea_off,
3465 struct lfsck_instance *lfsck = com->lc_lfsck;
3466 struct dt_object *parent = NULL;
3467 struct dt_object *child = NULL;
3468 struct lfsck_tgt_desc *ltd;
3472 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3474 GOTO(log, rc = PTR_ERR(parent));
3476 /* The MDT-object has been removed. */
3477 if (dt_object_exists(parent) == 0)
3480 ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
3481 if (unlikely(ltd == NULL))
3482 GOTO(log, rc = -ENODEV);
3484 child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
3486 GOTO(log, rc = PTR_ERR(child));
3488 /* The OST-object has been created. */
3489 if (unlikely(dt_object_exists(child) != 0))
3492 rc = __lfsck_layout_repair_dangling(env, com, parent, child,
3493 comp_id, ea_off, ost_idx, false);
3498 if (child != NULL && !IS_ERR(child))
3499 lfsck_object_put(env, child);
3501 if (parent != NULL && !IS_ERR(parent))
3502 lfsck_object_put(env, parent);
3505 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3506 "dangling reference for: parent "DFID", child "
3507 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: rc = %d\n",
3508 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3509 comp_id, ea_off, ost_idx,
3510 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3511 "Create the lost OST-object as required" :
3512 "Keep the MDT-object there by default", rc);
3517 /* If the OST-object does not recognize the MDT-object as its parent, and
3518 * there is no other MDT-object claims as its parent, then just trust the
3519 * given MDT-object as its parent. So update the OST-object filter_fid. */
3520 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
3521 struct lfsck_component *com,
3522 struct dt_object *parent,
3523 struct lfsck_layout_req *llr,
3526 struct lfsck_thread_info *info = lfsck_env_info(env);
3527 struct filter_fid *ff = &info->lti_ff;
3528 struct ost_layout *ol = &ff->ff_layout;
3529 struct dt_object *child = llr->llr_child;
3530 struct dt_device *dev = lfsck_obj2dev(child);
3531 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
3532 struct lu_buf *tbuf = &info->lti_big_buf;
3533 struct thandle *handle;
3535 struct lustre_handle lh = { 0 };
3539 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3540 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3545 ff->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3546 ff->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3547 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3548 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3549 * parent MDT-object's layout EA. */
3550 ff->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3552 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3556 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, llr->llr_comp_id);
3560 buf = lfsck_buf_get(env, ff, sizeof(*ff));
3562 handle = dt_trans_create(env, dev);
3564 GOTO(unlock1, rc = PTR_ERR(handle));
3566 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3570 rc = dt_attr_get(env, parent, la);
3574 la->la_valid = LA_UID | LA_GID;
3575 rc = dt_declare_attr_set(env, child, la, handle);
3579 rc = dt_trans_start_local(env, dev, handle);
3583 dt_write_lock(env, parent, 0);
3584 if (unlikely(lfsck_is_dead_obj(parent)))
3585 GOTO(unlock2, rc = 1);
3587 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3591 /* Get the latest parent's owner. */
3592 rc = dt_attr_get(env, parent, la);
3596 la->la_valid = LA_UID | LA_GID;
3597 rc = dt_attr_set(env, child, la, handle);
3602 dt_write_unlock(env, parent);
3605 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3608 lfsck_ibits_unlock(&lh, LCK_EX);
3612 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3613 "unmatched MDT-OST pair for: parent "DFID
3614 ", child "DFID", comp_id %u, OST-index %u, "
3615 "stripe-index %u, owner %u/%u: rc = %d\n",
3616 lfsck_lfsck2name(com->lc_lfsck),
3617 PFID(lfsck_dto2fid(parent)),
3618 PFID(lfsck_dto2fid(child)),
3619 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3620 la->la_uid, la->la_gid, rc);
3625 /* If there are more than one MDT-objects claim as the OST-object's parent,
3626 * and the OST-object only recognizes one of them, then we need to generate
3627 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3628 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3629 struct lfsck_component *com,
3630 struct dt_object *parent,
3631 struct lfsck_layout_req *llr,
3634 struct lfsck_thread_info *info = lfsck_env_info(env);
3635 struct dt_allocation_hint *hint = &info->lti_hint;
3636 struct dt_object_format *dof = &info->lti_dof;
3637 struct ost_id *oi = &info->lti_oi;
3638 struct lu_buf *buf = &info->lti_big_buf;
3639 struct lfsck_instance *lfsck = com->lc_lfsck;
3640 struct dt_device *dev;
3641 struct lu_device *d =
3642 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3643 struct lu_object *o;
3644 struct lu_object *n;
3645 struct dt_object *child = NULL;
3646 struct thandle *handle = NULL;
3647 struct lov_mds_md_v1 *lmm;
3648 struct lov_ost_data_v1 *objs;
3649 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3651 struct lustre_handle lh = { 0 };
3657 /* We use two separated transactions to repair the inconsistency.
3659 * 1) create the child (OST-object).
3660 * 2) update the parent LOV EA according to the child's FID.
3662 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3663 * handled as orphan when the layout LFSCK run next time.
3665 * If 1) failed, but 2) succeed, then such OST-object will be re-created
3666 * as dangling referened case when the layout LFSCK run next time. */
3668 /* The 1st transaction. */
3669 o = lu_object_anon(env, d, NULL);
3671 GOTO(log, rc = PTR_ERR(o));
3673 n = lu_object_locate(o->lo_header, d->ld_type);
3674 if (unlikely(n == NULL)) {
3675 lu_object_put_nocache(env, o);
3677 GOTO(log, rc = -EINVAL);
3680 child = container_of(n, struct dt_object, do_lu);
3681 memset(hint, 0, sizeof(*hint));
3682 rc = dt_attr_get(env, parent, la);
3686 la->la_valid = LA_UID | LA_GID;
3687 memset(dof, 0, sizeof(*dof));
3689 dev = lfsck_obj2dev(child);
3690 handle = dt_trans_create(env, dev);
3692 GOTO(log, rc = PTR_ERR(handle));
3694 rc = dt_declare_create(env, child, la, hint, dof, handle);
3698 rc = dt_trans_start_local(env, dev, handle);
3702 rc = dt_create(env, child, la, hint, dof, handle);
3703 dt_trans_stop(env, dev, handle);
3708 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3709 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3714 /* The 2nd transaction. */
3716 /* XXX: Generally, we should use bottom device (OSD) to update parent
3717 * LOV EA. But because the LOD-object still references the wrong
3718 * OSP-object that should be detached after the parent's LOV EA
3719 * refreshed. Unfortunately, there is no suitable API for that.
3720 * So we have to make the LOD to re-load the OSP-object(s) via
3721 * replacing the LOV EA against the LOD-object.
3723 * Once the DNE2 patches have been landed, we can replace the
3724 * LOD device with the OSD device. LU-6230. */
3726 dev = lfsck->li_next;
3727 parent = lfsck_object_locate(dev, parent);
3729 GOTO(log, rc = PTR_ERR(parent));
3731 handle = dt_trans_create(env, dev);
3733 GOTO(log, rc = PTR_ERR(handle));
3735 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3736 LU_XATTR_REPLACE, handle);
3740 rc = dt_trans_start_local(env, dev, handle);
3744 dt_write_lock(env, parent, 0);
3745 if (unlikely(lfsck_is_dead_obj(parent)))
3746 GOTO(unlock, rc = 0);
3748 rc = lfsck_layout_get_lovea(env, parent, buf);
3749 if (unlikely(!rc || rc == -ENODATA))
3750 GOTO(unlock, rc = 0);
3753 magic = le32_to_cpu(lmm->lmm_magic);
3754 if (magic == LOV_MAGIC_COMP_V1) {
3755 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3756 struct lov_comp_md_entry_v1 *lcme;
3757 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3760 LASSERT(llr->llr_comp_id != 0);
3762 for (i = 0; i < count; i++) {
3763 lcme = &lcm->lcm_entries[i];
3764 if (le32_to_cpu(lcme->lcme_id) == llr->llr_comp_id) {
3765 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3768 le32_add_cpu(&lcm->lcm_layout_gen, 1);
3770 le32_to_cpu(lcme->lcme_offset);
3771 magic = le32_to_cpu(lmm->lmm_magic);
3776 GOTO(unlock, rc = 0);
3780 if (magic == LOV_MAGIC_V1) {
3781 objs = &lmm->lmm_objects[llr->llr_lov_idx];
3783 LASSERT(magic == LOV_MAGIC_V3);
3785 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
3788 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3789 index = le32_to_cpu(objs->l_ost_idx);
3790 rc = ostid_to_fid(&tfid, oi, index);
3791 /* Someone changed layout during the LFSCK, no need to repair then. */
3792 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
3793 GOTO(unlock, rc = 0);
3795 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3796 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3797 ostid_cpu_to_le(oi, &objs->l_ost_oi);
3798 objs->l_ost_gen = cpu_to_le32(0);
3799 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3800 rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3801 LU_XATTR_REPLACE, handle);
3803 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3806 dt_write_unlock(env, parent);
3810 dt_trans_stop(env, dev, handle);
3813 lfsck_ibits_unlock(&lh, LCK_EX);
3815 lfsck_object_put(env, child);
3818 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3819 "multiple references for: parent "DFID", comp_id %u, "
3820 "OST-index %u, stripe-index %u, owner %u/%u: rc = %d\n",
3821 lfsck_lfsck2name(lfsck), PFID(pfid),
3822 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3823 la->la_uid, la->la_gid, rc);
3828 /* If the MDT-object and the OST-object have different owner information,
3829 * then trust the MDT-object, because the normal chown/chgrp handle order
3830 * is from MDT to OST, and it is possible that some chown/chgrp operation
3831 * is partly done. */
3832 static int lfsck_layout_repair_owner(const struct lu_env *env,
3833 struct lfsck_component *com,
3834 struct dt_object *parent,
3835 struct lfsck_layout_req *llr,
3836 struct lu_attr *pla,
3837 const struct lu_attr *cla)
3839 struct lfsck_thread_info *info = lfsck_env_info(env);
3840 struct lu_attr *tla = &info->lti_la2;
3841 struct dt_object *child = llr->llr_child;
3842 struct dt_device *dev = lfsck_obj2dev(child);
3843 struct thandle *handle;
3847 tla->la_uid = pla->la_uid;
3848 tla->la_gid = pla->la_gid;
3849 tla->la_valid = LA_UID | LA_GID;
3850 handle = dt_trans_create(env, dev);
3852 GOTO(log, rc = PTR_ERR(handle));
3854 rc = dt_declare_attr_set(env, child, tla, handle);
3858 rc = dt_trans_start_local(env, dev, handle);
3862 /* Use the dt_object lock to serialize with destroy and attr_set. */
3863 dt_read_lock(env, parent, 0);
3864 if (unlikely(lfsck_is_dead_obj(parent)))
3865 GOTO(unlock, rc = 1);
3867 /* Get the latest parent's owner. */
3868 rc = dt_attr_get(env, parent, pla);
3872 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3873 if (unlikely(tla->la_uid != pla->la_uid ||
3874 tla->la_gid != pla->la_gid))
3877 rc = dt_attr_set(env, child, tla, handle);
3882 dt_read_unlock(env, parent);
3885 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3889 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3890 "inconsistent file owner for: parent "DFID", child "DFID
3891 ", OST-index %u, stripe-index %u, old owner %u/%u, "
3892 "new owner %u/%u: rc = %d\n",
3893 lfsck_lfsck2name(com->lc_lfsck),
3894 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3895 llr->llr_ost_idx, llr->llr_lov_idx,
3896 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3901 /* Check whether the OST-object correctly back points to the
3902 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3903 static int lfsck_layout_check_parent(const struct lu_env *env,
3904 struct lfsck_component *com,
3905 struct lfsck_assistant_object *lso,
3906 struct filter_fid *ff,
3907 const struct lu_fid *cfid,
3908 const struct lu_attr *cla,
3909 struct lfsck_layout_req *llr)
3911 struct lfsck_thread_info *info = lfsck_env_info(env);
3912 struct lu_buf *buf = &info->lti_big_buf;
3913 struct lu_fid *pfid = &info->lti_fid;
3914 struct dt_object *tobj;
3915 struct lov_mds_md_v1 *lmm;
3916 struct lov_ost_data_v1 *objs;
3917 struct lustre_handle lh = { 0 };
3925 *pfid = ff->ff_parent;
3926 idx = pfid->f_stripe_idx;
3929 if (unlikely(!fid_is_sane(pfid)))
3930 RETURN(LLIT_UNMATCHED_PAIR);
3932 if (lu_fid_eq(pfid, &lso->lso_fid)) {
3933 if (likely(llr->llr_lov_idx == idx))
3936 RETURN(LLIT_UNMATCHED_PAIR);
3939 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3941 RETURN(PTR_ERR(tobj));
3943 if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3944 !S_ISREG(lfsck_object_type(tobj)))
3945 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3947 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3948 * remote one on another MDT. Then check whether the given OST-object
3949 * is in such layout. If yes, it is multiple referenced, otherwise it
3950 * is unmatched referenced case. */
3951 rc = lfsck_layout_get_lovea(env, tobj, buf);
3952 if (rc == 0 || rc == -ENOENT)
3953 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3959 magic = le32_to_cpu(lmm->lmm_magic);
3960 if (magic == LOV_MAGIC_COMP_V1) {
3961 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3962 struct lov_comp_md_entry_v1 *lcme;
3964 if (ff->ff_layout.ol_comp_id == 0)
3965 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3967 count = le16_to_cpu(lcm->lcm_entry_count);
3968 for (i = 0; i < count; i++) {
3969 lcme = &lcm->lcm_entries[i];
3970 if (le32_to_cpu(lcme->lcme_id) ==
3971 ff->ff_layout.ol_comp_id) {
3973 le32_to_cpu(lcme->lcme_offset);
3974 magic = le32_to_cpu(lmm->lmm_magic);
3975 if (!(le32_to_cpu(lcme->lcme_flags) &
3977 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3983 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3987 if (magic == LOV_MAGIC_V1) {
3988 objs = &lmm->lmm_objects[0];
3990 LASSERT(magic == LOV_MAGIC_V3);
3991 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3994 count = le16_to_cpu(lmm->lmm_stripe_count);
3995 for (i = 0; i < count; i++, objs++) {
3996 struct lu_fid *tfid = &info->lti_fid2;
3997 struct ost_id *oi = &info->lti_oi;
4000 if (lovea_slot_is_dummy(objs))
4003 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4004 idx2 = le32_to_cpu(objs->l_ost_idx);
4005 rc = ostid_to_fid(tfid, oi, idx2);
4007 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
4008 "invalid layout EA at the slot %d, index %u\n",
4009 lfsck_lfsck2name(com->lc_lfsck),
4010 PFID(pfid), i, idx2);
4012 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4015 if (lu_fid_eq(cfid, tfid)) {
4016 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
4017 MDS_INODELOCK_UPDATE |
4018 MDS_INODELOCK_LAYOUT |
4019 MDS_INODELOCK_XATTR,
4024 dt_read_lock(env, tobj, 0);
4026 /* For local MDT-object, re-check existence
4027 * after taken the lock. */
4028 if (!dt_object_remote(tobj)) {
4029 if (dt_object_exists(tobj) == 0 ||
4030 lfsck_is_dead_obj(tobj))
4031 rc = LLIT_UNMATCHED_PAIR;
4033 rc = LLIT_MULTIPLE_REFERENCED;
4038 /* For migration case, the new MDT-object and old
4039 * MDT-object may reference the same OST-object at
4040 * some migration internal time.
4042 * For remote MDT-object, the local MDT may not know
4043 * whether it has been removed or not. Try checking
4044 * for a non-existent xattr to check if this object
4045 * has been been removed or not. */
4046 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
4048 if (unlikely(rc == -ENOENT || rc >= 0))
4049 rc = LLIT_UNMATCHED_PAIR;
4050 else if (rc == -ENODATA)
4051 rc = LLIT_MULTIPLE_REFERENCED;
4057 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4060 if (lustre_handle_is_used(&lh)) {
4061 dt_read_unlock(env, tobj);
4062 lfsck_ibits_unlock(&lh, LCK_EX);
4066 lfsck_object_put(env, tobj);
4071 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
4072 struct lfsck_component *com,
4073 struct lfsck_assistant_req *lar)
4075 struct lfsck_layout_req *llr =
4076 container_of0(lar, struct lfsck_layout_req, llr_lar);
4077 struct lfsck_assistant_object *lso = lar->lar_parent;
4078 struct lfsck_layout *lo = com->lc_file_ram;
4079 struct lfsck_thread_info *info = lfsck_env_info(env);
4080 struct filter_fid *ff = &info->lti_ff;
4081 struct lu_buf buf = { .lb_buf = ff,
4082 .lb_len = sizeof(*ff) };
4083 struct dt_object *parent = NULL;
4084 struct dt_object *child = llr->llr_child;
4085 struct lu_attr *pla = &lso->lso_attr;
4086 struct lu_attr *cla = &info->lti_la;
4087 struct lfsck_instance *lfsck = com->lc_lfsck;
4088 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4089 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
4096 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ASSISTANT_DIRECT, cfs_fail_val);
4098 rc = dt_attr_get(env, child, cla);
4099 if (rc == -ENOENT) {
4100 parent = lfsck_assistant_object_load(env, lfsck, lso);
4101 if (IS_ERR(parent)) {
4102 rc = PTR_ERR(parent);
4104 RETURN(rc == -ENOENT ? 0 : rc);
4107 type = LLIT_DANGLING;
4114 lfsck_buf_init(&buf, ff, sizeof(*ff));
4115 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
4116 if (unlikely(rc > 0 && rc < sizeof(struct lu_fid))) {
4117 type = LLIT_UNMATCHED_PAIR;
4121 if (rc < 0 && rc != -ENODATA)
4124 if (rc == 0 || rc == -ENODATA)
4125 GOTO(check_owner, rc = 0);
4127 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
4128 rc = lfsck_layout_check_parent(env, com, lso, ff,
4129 lu_object_fid(&child->do_lu), cla, llr);
4139 /* Someone may has changed the owner after the parent attr pre-loaded.
4140 * It can be handled later inside the lfsck_layout_repair_owner(). */
4141 if (unlikely(cla->la_uid != pla->la_uid ||
4142 cla->la_gid != pla->la_gid)) {
4143 type = LLIT_INCONSISTENT_OWNER;
4148 if (type == LLIT_NONE)
4151 if (bk->lb_param & LPF_DRYRUN)
4154 if (parent == NULL) {
4155 parent = lfsck_assistant_object_load(env, lfsck, lso);
4156 if (IS_ERR(parent)) {
4157 rc = PTR_ERR(parent);
4168 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
4169 rc = lfsck_layout_ins_dangling_rec(env, com,
4170 lfsck_dto2fid(parent), lfsck_dto2fid(child),
4171 llr->llr_comp_id, llr->llr_lov_idx,
4174 rc = __lfsck_layout_repair_dangling(env, com, parent,
4181 case LLIT_UNMATCHED_PAIR:
4182 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
4185 case LLIT_MULTIPLE_REFERENCED:
4186 rc = lfsck_layout_repair_multiple_references(env, com, parent,
4189 case LLIT_INCONSISTENT_OWNER:
4190 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
4200 down_write(&com->lc_sem);
4202 struct lfsck_assistant_data *lad = com->lc_data;
4204 if (unlikely(lad->lad_exit)) {
4206 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
4207 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
4208 rc == -EHOSTUNREACH) {
4209 /* If cannot touch the target server,
4210 * mark the LFSCK as INCOMPLETE. */
4211 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
4212 "talk with OST %x: rc = %d\n",
4213 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
4214 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
4215 lo->ll_objs_skipped++;
4218 lfsck_layout_record_failure(env, lfsck, lo);
4220 } else if (rc > 0 && (type != LLIT_DANGLING ||
4221 !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
4222 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
4223 "unknown type = %d\n", type);
4225 lo->ll_objs_repaired[type - 1]++;
4226 if (bk->lb_param & LPF_DRYRUN &&
4227 unlikely(lo->ll_pos_first_inconsistent == 0))
4228 lo->ll_pos_first_inconsistent =
4229 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
4232 up_write(&com->lc_sem);
4234 if (parent != NULL && !IS_ERR(parent))
4235 lfsck_object_put(env, parent);
4241 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
4242 struct lfsck_component *com,
4243 struct dt_object *obj, bool first)
4245 struct lfsck_instance *lfsck = com->lc_lfsck;
4246 struct ptlrpc_thread *thread = &lfsck->li_thread;
4247 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4248 struct lfsck_layout *lo = com->lc_file_ram;
4249 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
4252 struct lfsck_layout_dangling_key *parent =
4253 &lfsck_env_info(env)->lti_lldk;
4254 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
4259 di = iops->init(env, obj, 0);
4261 RETURN(PTR_ERR(di));
4264 lldk_cpu_to_be(parent, &lo->ll_lldk_latest_scanned_phase2);
4266 memset(parent, 0, sizeof(*parent));
4267 rc = iops->get(env, di, (const struct dt_key *)parent);
4272 /* The start one either has been processed or does not exist,
4274 rc = iops->next(env, di);
4280 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
4281 unlikely(!thread_is_running(thread)))
4284 key = iops->key(env, di);
4293 lldk_be_to_cpu(parent,
4294 (const struct lfsck_layout_dangling_key *)key);
4295 if (!fid_is_sane(&parent->lldk_fid)) {
4300 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
4302 fid_be_to_cpu(cfid, cfid);
4303 ost_idx = cfid->f_ver;
4305 if (!fid_is_sane(cfid)) {
4310 rc = lfsck_layout_repair_dangling(env, com,
4311 &parent->lldk_fid, cfid,
4312 parent->lldk_comp_id,
4313 parent->lldk_ea_off, ost_idx);
4317 down_write(&com->lc_sem);
4318 com->lc_new_checked++;
4319 com->lc_new_scanned++;
4321 lo->ll_lldk_latest_scanned_phase2 = *parent;
4324 lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
4326 lo->ll_objs_failed_phase2++;
4327 up_write(&com->lc_sem);
4329 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4332 if (unlikely(com->lc_time_next_checkpoint <=
4333 ktime_get_seconds()) &&
4334 com->lc_new_checked != 0) {
4335 down_write(&com->lc_sem);
4336 lo->ll_run_time_phase2 += ktime_get_seconds() -
4337 com->lc_time_last_checkpoint;
4338 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
4339 lo->ll_objs_checked_phase2 += com->lc_new_checked;
4340 com->lc_new_checked = 0;
4341 lfsck_layout_store(env, com);
4342 up_write(&com->lc_sem);
4344 com->lc_time_last_checkpoint = ktime_get_seconds();
4345 com->lc_time_next_checkpoint =
4346 com->lc_time_last_checkpoint +
4347 LFSCK_CHECKPOINT_INTERVAL;
4350 lfsck_control_speed_by_self(com);
4351 if (unlikely(!thread_is_running(thread)))
4354 rc = iops->next(env, di);
4363 iops->fini(env, di);
4368 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
4369 struct lfsck_component *com)
4371 struct lfsck_assistant_data *lad = com->lc_data;
4372 struct lfsck_instance *lfsck = com->lc_lfsck;
4373 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4374 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4375 struct lfsck_tgt_desc *ltd;
4379 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
4380 lfsck_lfsck2name(lfsck));
4382 spin_lock(<ds->ltd_lock);
4383 while (!list_empty(&lad->lad_ost_phase2_list)) {
4384 ltd = list_entry(lad->lad_ost_phase2_list.next,
4385 struct lfsck_tgt_desc,
4386 ltd_layout_phase_list);
4387 list_del_init(<d->ltd_layout_phase_list);
4388 if (bk->lb_param & LPF_OST_ORPHAN) {
4389 spin_unlock(<ds->ltd_lock);
4390 rc = lfsck_layout_scan_orphan(env, com, ltd);
4391 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
4394 if (unlikely(lad->lad_exit ||
4395 !thread_is_running(&lfsck->li_thread)))
4397 spin_lock(<ds->ltd_lock);
4401 if (list_empty(&lad->lad_ost_phase1_list))
4405 spin_unlock(<ds->ltd_lock);
4407 if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
4408 struct lfsck_layout *lo = com->lc_file_ram;
4411 com->lc_new_checked = 0;
4412 com->lc_new_scanned = 0;
4413 com->lc_time_last_checkpoint = ktime_get_seconds();
4414 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4415 LFSCK_CHECKPOINT_INTERVAL;
4417 i = lfsck_sub_trace_file_fid2idx(
4418 &lo->ll_lldk_latest_scanned_phase2.lldk_fid);
4419 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4420 com->lc_sub_trace_objs[i].lsto_obj, true);
4421 while (rc > 0 && ++i < LFSCK_STF_COUNT)
4422 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4423 com->lc_sub_trace_objs[i].lsto_obj, false);
4425 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
4426 "at the No. %d trace file: rc = %d\n",
4427 lfsck_lfsck2name(lfsck), i, rc);
4430 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
4431 lfsck_lfsck2name(lfsck), rc);
4437 lfsck_layout_slave_async_interpret(const struct lu_env *env,
4438 struct ptlrpc_request *req,
4441 struct lfsck_layout_slave_async_args *llsaa = args;
4442 struct obd_export *exp = llsaa->llsaa_exp;
4443 struct lfsck_component *com = llsaa->llsaa_com;
4444 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
4445 struct lfsck_layout_slave_data *llsd = com->lc_data;
4446 struct lfsck_reply *lr = NULL;
4450 /* It is probably caused by network trouble, or target crash,
4451 * it will try several times (depends on the obd_timeout, and
4452 * will not less than 3 times). But to make the LFSCK can go
4453 * ahead, we should not try for ever. After some try but still
4454 * hit failure, it will assume that the target exit the LFSCK
4455 * prcoessing and stop try. */
4456 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
4457 int max_try = max_t(int, obd_timeout / 30, 3);
4459 if (++(llst->llst_failures) > max_try)
4465 llst->llst_failures = 0;
4466 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
4467 if (lr->lr_status != LS_SCANNING_PHASE1 &&
4468 lr->lr_status != LS_SCANNING_PHASE2)
4473 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
4474 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
4475 llst->llst_index, lr != NULL ? lr->lr_status : rc,
4476 llst->llst_failures);
4478 lfsck_layout_llst_del(llsd, llst);
4481 lfsck_layout_llst_put(llst);
4482 lfsck_component_put(env, com);
4483 class_export_put(exp);
4488 static int lfsck_layout_async_query(const struct lu_env *env,
4489 struct lfsck_component *com,
4490 struct obd_export *exp,
4491 struct lfsck_layout_slave_target *llst,
4492 struct lfsck_request *lr,
4493 struct ptlrpc_request_set *set)
4495 struct lfsck_layout_slave_async_args *llsaa;
4496 struct ptlrpc_request *req;
4497 struct lfsck_request *tmp;
4501 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
4505 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
4507 ptlrpc_request_free(req);
4511 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4513 ptlrpc_request_set_replen(req);
4515 llsaa = ptlrpc_req_async_args(req);
4516 llsaa->llsaa_exp = exp;
4517 llsaa->llsaa_com = lfsck_component_get(com);
4518 llsaa->llsaa_llst = llst;
4519 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
4520 req->rq_allow_intr = 1;
4521 ptlrpc_set_add_req(set, req);
4526 static int lfsck_layout_async_notify(const struct lu_env *env,
4527 struct obd_export *exp,
4528 struct lfsck_request *lr,
4529 struct ptlrpc_request_set *set)
4531 struct ptlrpc_request *req;
4532 struct lfsck_request *tmp;
4536 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4540 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4542 ptlrpc_request_free(req);
4546 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4548 ptlrpc_request_set_replen(req);
4549 req->rq_allow_intr = 1;
4550 ptlrpc_set_add_req(set, req);
4556 lfsck_layout_slave_query_master(const struct lu_env *env,
4557 struct lfsck_component *com)
4559 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4560 struct lfsck_instance *lfsck = com->lc_lfsck;
4561 struct lfsck_layout_slave_data *llsd = com->lc_data;
4562 struct lfsck_layout_slave_target *llst;
4563 struct obd_export *exp;
4564 struct ptlrpc_request_set *set;
4569 set = ptlrpc_prep_set();
4571 GOTO(log, rc = -ENOMEM);
4573 memset(lr, 0, sizeof(*lr));
4574 lr->lr_event = LE_QUERY;
4575 lr->lr_active = LFSCK_TYPE_LAYOUT;
4577 llsd->llsd_touch_gen++;
4578 spin_lock(&llsd->llsd_lock);
4579 while (!list_empty(&llsd->llsd_master_list)) {
4580 llst = list_entry(llsd->llsd_master_list.next,
4581 struct lfsck_layout_slave_target,
4583 if (llst->llst_gen == llsd->llsd_touch_gen)
4586 llst->llst_gen = llsd->llsd_touch_gen;
4587 list_move_tail(&llst->llst_list,
4588 &llsd->llsd_master_list);
4589 atomic_inc(&llst->llst_ref);
4590 spin_unlock(&llsd->llsd_lock);
4592 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4595 lfsck_layout_llst_del(llsd, llst);
4596 lfsck_layout_llst_put(llst);
4597 spin_lock(&llsd->llsd_lock);
4601 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
4603 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4604 "query %s for layout: rc = %d\n",
4605 lfsck_lfsck2name(lfsck),
4606 exp->exp_obd->obd_name, rc);
4609 lfsck_layout_llst_put(llst);
4610 class_export_put(exp);
4612 spin_lock(&llsd->llsd_lock);
4614 spin_unlock(&llsd->llsd_lock);
4616 rc = ptlrpc_set_wait(set);
4617 ptlrpc_set_destroy(set);
4619 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4622 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4623 lfsck_lfsck2name(com->lc_lfsck), rc);
4629 lfsck_layout_slave_notify_master(const struct lu_env *env,
4630 struct lfsck_component *com,
4631 enum lfsck_events event, int result)
4633 struct lfsck_layout *lo = com->lc_file_ram;
4634 struct lfsck_instance *lfsck = com->lc_lfsck;
4635 struct lfsck_layout_slave_data *llsd = com->lc_data;
4636 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4637 struct lfsck_layout_slave_target *llst;
4638 struct obd_export *exp;
4639 struct ptlrpc_request_set *set;
4643 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4644 lfsck_lfsck2name(com->lc_lfsck));
4646 set = ptlrpc_prep_set();
4650 memset(lr, 0, sizeof(*lr));
4651 lr->lr_event = event;
4652 lr->lr_flags = LEF_FROM_OST;
4653 lr->lr_status = result;
4654 lr->lr_index = lfsck_dev_idx(lfsck);
4655 lr->lr_active = LFSCK_TYPE_LAYOUT;
4656 lr->lr_flags2 = lo->ll_flags;
4657 llsd->llsd_touch_gen++;
4658 spin_lock(&llsd->llsd_lock);
4659 while (!list_empty(&llsd->llsd_master_list)) {
4660 llst = list_entry(llsd->llsd_master_list.next,
4661 struct lfsck_layout_slave_target,
4663 if (llst->llst_gen == llsd->llsd_touch_gen)
4666 llst->llst_gen = llsd->llsd_touch_gen;
4667 list_move_tail(&llst->llst_list,
4668 &llsd->llsd_master_list);
4669 atomic_inc(&llst->llst_ref);
4670 spin_unlock(&llsd->llsd_lock);
4672 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4675 lfsck_layout_llst_del(llsd, llst);
4676 lfsck_layout_llst_put(llst);
4677 spin_lock(&llsd->llsd_lock);
4681 rc = lfsck_layout_async_notify(env, exp, lr, set);
4683 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4684 "notify %s for layout: rc = %d\n",
4685 lfsck_lfsck2name(lfsck),
4686 exp->exp_obd->obd_name, rc);
4688 lfsck_layout_llst_put(llst);
4689 class_export_put(exp);
4690 spin_lock(&llsd->llsd_lock);
4692 spin_unlock(&llsd->llsd_lock);
4694 ptlrpc_set_wait(set);
4695 ptlrpc_set_destroy(set);
4701 * \ret -ENODATA: unrecognized stripe
4702 * \ret = 0 : recognized stripe
4703 * \ret < 0 : other failures
4705 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4706 struct lfsck_component *com,
4707 struct lu_fid *cfid,
4708 struct lu_fid *pfid, __u32 comp_id)
4710 struct lfsck_thread_info *info = lfsck_env_info(env);
4711 struct lu_buf *buf = &info->lti_big_buf;
4712 struct ost_id *oi = &info->lti_oi;
4713 struct dt_object *obj;
4714 struct lov_mds_md_v1 *lmm;
4715 struct lov_ost_data_v1 *objs;
4716 __u32 idx = pfid->f_stripe_idx;
4724 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4726 RETURN(PTR_ERR(obj));
4728 dt_read_lock(env, obj, 0);
4729 if (unlikely(dt_object_exists(obj) == 0 ||
4730 lfsck_is_dead_obj(obj)))
4731 GOTO(unlock, rc = -ENOENT);
4733 if (!S_ISREG(lfsck_object_type(obj)))
4734 GOTO(unlock, rc = -ENODATA);
4736 rc = lfsck_layout_get_lovea(env, obj, buf);
4741 GOTO(unlock, rc = -ENODATA);
4744 magic = le32_to_cpu(lmm->lmm_magic);
4745 if (magic == LOV_MAGIC_COMP_V1) {
4746 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4747 struct lov_comp_md_entry_v1 *lcme;
4750 GOTO(unlock, rc = -ENODATA);
4752 count = le16_to_cpu(lcm->lcm_entry_count);
4753 for (i = 0; i < count; i++) {
4754 lcme = &lcm->lcm_entries[i];
4755 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
4757 le32_to_cpu(lcme->lcme_offset);
4758 magic = le32_to_cpu(lmm->lmm_magic);
4759 if (!(le32_to_cpu(lcme->lcme_flags) &
4761 GOTO(unlock, rc = -ENODATA);
4767 GOTO(unlock, rc = -ENODATA);
4771 if (magic == LOV_MAGIC_V1) {
4772 objs = &lmm->lmm_objects[0];
4774 LASSERT(magic == LOV_MAGIC_V3);
4775 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4778 fid_to_ostid(cfid, oi);
4779 count = le16_to_cpu(lmm->lmm_stripe_count);
4780 for (i = 0; i < count; i++, objs++) {
4783 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
4784 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
4785 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
4788 GOTO(unlock, rc = -ENODATA);
4791 dt_read_unlock(env, obj);
4792 lfsck_object_put(env, obj);
4798 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
4799 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
4800 * layout EA from MDT to OST. On one hand, the OST no need to understand
4801 * the layout EA structure; on the other hand, it may cause trouble when
4802 * transfer large layout EA from MDT to OST via normal OUT RPC.
4804 * \ret > 0: unrecognized stripe
4805 * \ret = 0: recognized stripe
4806 * \ret < 0: other failures
4808 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
4809 struct lfsck_component *com,
4810 struct lu_fid *cfid,
4811 struct lu_fid *pfid, __u32 comp_id)
4813 struct lfsck_instance *lfsck = com->lc_lfsck;
4814 struct obd_device *obd = lfsck->li_obd;
4815 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4816 struct obd_export *exp = NULL;
4817 struct ptlrpc_request *req = NULL;
4818 struct lfsck_request *lr;
4819 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
4823 if (unlikely(fid_is_idif(pfid)))
4826 fld_range_set_any(range);
4827 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
4829 RETURN(rc == -ENOENT ? 1 : rc);
4831 if (unlikely(!fld_range_is_mdt(range)))
4834 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
4835 if (unlikely(exp == NULL))
4838 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
4839 GOTO(out, rc = -EOPNOTSUPP);
4841 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4843 GOTO(out, rc = -ENOMEM);
4845 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4847 ptlrpc_request_free(req);
4852 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4853 memset(lr, 0, sizeof(*lr));
4854 lr->lr_event = LE_PAIRS_VERIFY;
4855 lr->lr_active = LFSCK_TYPE_LAYOUT;
4856 lr->lr_fid = *cfid; /* OST-object itself FID. */
4857 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
4858 lr->lr_comp_id = comp_id;
4860 ptlrpc_request_set_replen(req);
4861 rc = ptlrpc_queue_wait(req);
4862 ptlrpc_req_finished(req);
4864 if (rc == -ENOENT || rc == -ENODATA)
4871 class_export_put(exp);
4876 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
4877 struct lfsck_component *com,
4878 struct lfsck_req_local *lrl)
4880 struct dt_object *obj;
4884 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lrl->lrl_fid);
4886 GOTO(log, rc = PTR_ERR(obj));
4888 dt_write_lock(env, obj, 0);
4889 if (unlikely(dt_object_exists(obj) == 0 ||
4890 lfsck_is_dead_obj(obj)))
4891 GOTO(unlock, rc = 0);
4893 rc = __lfsck_layout_update_pfid(env, obj, &lrl->lrl_ff_client.ff_parent,
4894 &lrl->lrl_ff_client.ff_layout,
4895 lrl->lrl_ff_client.ff_parent.f_ver);
4900 dt_write_unlock(env, obj);
4901 lfsck_object_put(env, obj);
4904 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
4905 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
4906 PFID(&lrl->lrl_fid), PFID(&lrl->lrl_ff_client.ff_parent), rc);
4913 static void lfsck_layout_slave_quit(const struct lu_env *env,
4914 struct lfsck_component *com);
4916 static int lfsck_layout_reset(const struct lu_env *env,
4917 struct lfsck_component *com, bool init)
4919 struct lfsck_layout *lo = com->lc_file_ram;
4922 down_write(&com->lc_sem);
4924 memset(lo, 0, com->lc_file_size);
4926 __u32 count = lo->ll_success_count;
4927 time64_t last_time = lo->ll_time_last_complete;
4929 memset(lo, 0, com->lc_file_size);
4930 lo->ll_success_count = count;
4931 lo->ll_time_last_complete = last_time;
4934 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
4935 lo->ll_status = LS_INIT;
4937 if (com->lc_lfsck->li_master) {
4938 struct lfsck_assistant_data *lad = com->lc_data;
4940 lad->lad_incomplete = 0;
4941 CFS_RESET_BITMAP(lad->lad_bitmap);
4944 rc = lfsck_layout_store(env, com);
4945 if (rc == 0 && com->lc_lfsck->li_master)
4946 rc = lfsck_load_sub_trace_files(env, com,
4947 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
4948 up_write(&com->lc_sem);
4950 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
4951 lfsck_lfsck2name(com->lc_lfsck), rc);
4956 static void lfsck_layout_fail(const struct lu_env *env,
4957 struct lfsck_component *com, bool new_checked)
4959 struct lfsck_layout *lo = com->lc_file_ram;
4961 down_write(&com->lc_sem);
4963 com->lc_new_checked++;
4964 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
4965 up_write(&com->lc_sem);
4968 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
4969 struct lfsck_component *com, bool init)
4971 struct lfsck_instance *lfsck = com->lc_lfsck;
4972 struct lfsck_layout *lo = com->lc_file_ram;
4976 rc = lfsck_checkpoint_generic(env, com);
4978 return rc > 0 ? 0 : rc;
4981 down_write(&com->lc_sem);
4983 lo->ll_pos_latest_start =
4984 lfsck->li_pos_checkpoint.lp_oit_cookie;
4986 lo->ll_pos_last_checkpoint =
4987 lfsck->li_pos_checkpoint.lp_oit_cookie;
4988 lo->ll_run_time_phase1 += ktime_get_seconds() -
4989 lfsck->li_time_last_checkpoint;
4990 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
4991 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4992 com->lc_new_checked = 0;
4995 rc = lfsck_layout_store(env, com);
4996 up_write(&com->lc_sem);
4998 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
4999 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5000 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5005 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
5006 struct lfsck_component *com, bool init)
5008 struct lfsck_instance *lfsck = com->lc_lfsck;
5009 struct lfsck_layout *lo = com->lc_file_ram;
5012 if (com->lc_new_checked == 0 && !init)
5015 down_write(&com->lc_sem);
5017 lo->ll_pos_latest_start =
5018 lfsck->li_pos_checkpoint.lp_oit_cookie;
5020 lo->ll_pos_last_checkpoint =
5021 lfsck->li_pos_checkpoint.lp_oit_cookie;
5022 lo->ll_run_time_phase1 += ktime_get_seconds() -
5023 lfsck->li_time_last_checkpoint;
5024 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5025 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5026 com->lc_new_checked = 0;
5029 rc = lfsck_layout_store(env, com);
5030 up_write(&com->lc_sem);
5032 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
5033 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5034 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5039 static int lfsck_layout_prep(const struct lu_env *env,
5040 struct lfsck_component *com,
5041 struct lfsck_start *start)
5043 struct lfsck_instance *lfsck = com->lc_lfsck;
5044 struct lfsck_layout *lo = com->lc_file_ram;
5045 struct lfsck_position *pos = &com->lc_pos_start;
5047 fid_zero(&pos->lp_dir_parent);
5048 pos->lp_dir_cookie = 0;
5049 if (lo->ll_status == LS_COMPLETED ||
5050 lo->ll_status == LS_PARTIAL ||
5051 /* To handle orphan, must scan from the beginning. */
5052 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
5055 rc = lfsck_layout_reset(env, com, false);
5057 rc = lfsck_set_param(env, lfsck, start, true);
5060 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
5061 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
5067 down_write(&com->lc_sem);
5068 lo->ll_time_latest_start = ktime_get_real_seconds();
5069 spin_lock(&lfsck->li_lock);
5070 if (lo->ll_flags & LF_SCANNED_ONCE) {
5071 if (!lfsck->li_drop_dryrun ||
5072 lo->ll_pos_first_inconsistent == 0) {
5073 lo->ll_status = LS_SCANNING_PHASE2;
5074 list_move_tail(&com->lc_link,
5075 &lfsck->li_list_double_scan);
5076 pos->lp_oit_cookie = 0;
5080 lo->ll_status = LS_SCANNING_PHASE1;
5081 lo->ll_run_time_phase1 = 0;
5082 lo->ll_run_time_phase2 = 0;
5083 lo->ll_objs_checked_phase1 = 0;
5084 lo->ll_objs_checked_phase2 = 0;
5085 lo->ll_objs_failed_phase1 = 0;
5086 lo->ll_objs_failed_phase2 = 0;
5087 for (i = 0; i < LLIT_MAX; i++)
5088 lo->ll_objs_repaired[i] = 0;
5090 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5091 fid_zero(&com->lc_fid_latest_scanned_phase2);
5094 lo->ll_status = LS_SCANNING_PHASE1;
5095 if (!lfsck->li_drop_dryrun ||
5096 lo->ll_pos_first_inconsistent == 0)
5097 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
5099 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5101 spin_unlock(&lfsck->li_lock);
5102 up_write(&com->lc_sem);
5107 static int lfsck_layout_slave_prep(const struct lu_env *env,
5108 struct lfsck_component *com,
5109 struct lfsck_start_param *lsp)
5111 struct lfsck_layout_slave_data *llsd = com->lc_data;
5112 struct lfsck_instance *lfsck = com->lc_lfsck;
5113 struct lfsck_layout *lo = com->lc_file_ram;
5114 struct lfsck_start *start = lsp->lsp_start;
5117 rc = lfsck_layout_prep(env, com, start);
5121 if (lo->ll_flags & LF_CRASHED_LASTID &&
5122 list_empty(&llsd->llsd_master_list)) {
5123 LASSERT(lfsck->li_out_notify != NULL);
5125 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5126 LE_LASTID_REBUILDING);
5129 if (!lsp->lsp_index_valid)
5132 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5133 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
5134 LASSERT(!llsd->llsd_rbtree_valid);
5136 write_lock(&llsd->llsd_rb_lock);
5137 rc = lfsck_rbtree_setup(env, com);
5138 write_unlock(&llsd->llsd_rb_lock);
5141 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
5142 "%llu]\n", lfsck_lfsck2name(lfsck),
5143 com->lc_pos_start.lp_oit_cookie);
5148 static int lfsck_layout_master_prep(const struct lu_env *env,
5149 struct lfsck_component *com,
5150 struct lfsck_start_param *lsp)
5155 rc = lfsck_layout_load_bitmap(env, com);
5157 rc = lfsck_layout_reset(env, com, false);
5159 rc = lfsck_set_param(env, com->lc_lfsck,
5160 lsp->lsp_start, true);
5166 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
5170 rc = lfsck_start_assistant(env, com, lsp);
5175 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
5176 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
5177 com->lc_pos_start.lp_oit_cookie);
5182 /* Pre-fetch the attribute for each stripe in the given layout EA. */
5183 static int lfsck_layout_scan_stripes(const struct lu_env *env,
5184 struct lfsck_component *com,
5185 struct dt_object *parent,
5186 struct lov_mds_md_v1 *lmm, __u32 comp_id)
5188 struct lfsck_thread_info *info = lfsck_env_info(env);
5189 struct lfsck_instance *lfsck = com->lc_lfsck;
5190 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5191 struct lfsck_layout *lo = com->lc_file_ram;
5192 struct lfsck_assistant_data *lad = com->lc_data;
5193 struct lfsck_assistant_object *lso = NULL;
5194 struct lov_ost_data_v1 *objs;
5195 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
5196 struct ptlrpc_thread *mthread = &lfsck->li_thread;
5197 struct ptlrpc_thread *athread = &lad->lad_thread;
5198 struct l_wait_info lwi = { 0 };
5206 lfsck_buf_init(&buf, &info->lti_ff, sizeof(struct filter_fid));
5207 magic = le32_to_cpu(lmm->lmm_magic);
5208 if (magic == LOV_MAGIC_V1) {
5209 objs = &lmm->lmm_objects[0];
5211 LASSERT(magic == LOV_MAGIC_V3);
5212 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5215 count = le16_to_cpu(lmm->lmm_stripe_count);
5216 for (i = 0; i < count; i++, objs++) {
5217 struct lu_fid *fid = &info->lti_fid;
5218 struct ost_id *oi = &info->lti_oi;
5219 struct lfsck_layout_req *llr;
5220 struct lfsck_tgt_desc *tgt = NULL;
5221 struct dt_object *cobj = NULL;
5223 bool wakeup = false;
5225 if (unlikely(lovea_slot_is_dummy(objs)))
5228 l_wait_event(mthread->t_ctl_waitq,
5229 lad->lad_prefetched < bk->lb_async_windows ||
5230 !thread_is_running(mthread) ||
5231 thread_is_stopped(athread),
5234 if (unlikely(!thread_is_running(mthread)) ||
5235 thread_is_stopped(athread))
5238 if (unlikely(lfsck_is_dead_obj(parent)))
5241 ostid_le_to_cpu(&objs->l_ost_oi, oi);
5242 index = le32_to_cpu(objs->l_ost_idx);
5243 rc = ostid_to_fid(fid, oi, index);
5245 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
5246 ": "DOSTID", idx %u, comp_id %u\n",
5247 lfsck_lfsck2name(lfsck),
5248 PFID(lfsck_dto2fid(parent)), POSTID(oi),
5253 tgt = lfsck_tgt_get(ltds, index);
5254 if (unlikely(tgt == NULL)) {
5255 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
5256 "did not join the layout LFSCK, comp_id %u\n",
5257 lfsck_lfsck2name(lfsck), index, comp_id);
5258 lfsck_lad_set_bitmap(env, com, index);
5262 /* There is potential deadlock race condition between object
5263 * destroy and layout LFSCK. Consider the following scenario:
5265 * 1) The LFSCK thread obtained the parent object firstly, at
5266 * that time, the parent object has not been destroyed yet.
5268 * 2) One RPC service thread destroyed the parent and all its
5269 * children objects. Because the LFSCK is referencing the
5270 * parent object, then the parent object will be marked as
5271 * dying in RAM. On the other hand, the parent object is
5272 * referencing all its children objects, then all children
5273 * objects will be marked as dying in RAM also.
5275 * 3) The LFSCK thread tries to find some child object with
5276 * the parent object referenced. Then it will find that the
5277 * child object is dying. According to the object visibility
5278 * rules: the object with dying flag cannot be returned to
5279 * others. So the LFSCK thread has to wait until the dying
5280 * object has been purged from RAM, then it can allocate a
5281 * new object (with the same FID) in RAM. Unfortunately, the
5282 * LFSCK thread itself is referencing the parent object, and
5283 * cause the parent object cannot be purged, then cause the
5284 * child object cannot be purged also. So the LFSCK thread
5285 * will fall into deadlock.
5287 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
5289 if (lfsck_is_dead_obj(parent)) {
5299 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_ASSISTANT_DIRECT)) {
5300 rc = dt_declare_attr_get(env, cobj);
5304 rc = dt_declare_xattr_get(env, cobj, &buf,
5311 struct lu_attr *attr = &info->lti_la;
5313 rc = dt_attr_get(env, parent, attr);
5317 lso = lfsck_assistant_object_init(env,
5318 lfsck_dto2fid(parent), attr,
5319 lfsck->li_pos_current.lp_oit_cookie, false);
5328 llr = lfsck_layout_assistant_req_init(lso, cobj, comp_id,
5336 spin_lock(&lad->lad_lock);
5337 if (lad->lad_assistant_status < 0) {
5338 spin_unlock(&lad->lad_lock);
5339 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
5341 RETURN(lad->lad_assistant_status);
5344 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
5345 if (lad->lad_prefetched == 0)
5348 lad->lad_prefetched++;
5349 spin_unlock(&lad->lad_lock);
5351 wake_up_all(&athread->t_ctl_waitq);
5354 down_write(&com->lc_sem);
5355 com->lc_new_checked++;
5357 lfsck_layout_record_failure(env, lfsck, lo);
5358 up_write(&com->lc_sem);
5360 if (cobj != NULL && !IS_ERR(cobj))
5361 lfsck_object_put(env, cobj);
5363 if (likely(tgt != NULL))
5366 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5374 lfsck_assistant_object_put(env, lso);
5379 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
5380 * the OST-object's attribute and generate an structure lfsck_layout_req on the
5381 * list ::lad_req_list.
5383 * For each request on above list, the lfsck_layout_assistant thread compares
5384 * the OST side attribute with local attribute, if inconsistent, then repair it.
5386 * All above processing is async mode with pipeline. */
5387 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
5388 struct lfsck_component *com,
5389 struct dt_object *obj)
5391 struct lfsck_thread_info *info = lfsck_env_info(env);
5392 struct ost_id *oi = &info->lti_oi;
5393 struct lfsck_layout *lo = com->lc_file_ram;
5394 struct lfsck_assistant_data *lad = com->lc_data;
5395 struct lfsck_instance *lfsck = com->lc_lfsck;
5396 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5397 struct thandle *handle = NULL;
5398 struct lu_buf *buf = &info->lti_big_buf;
5399 struct lov_mds_md_v1 *lmm = NULL;
5400 struct dt_device *dev = lfsck_obj2dev(obj);
5401 struct lustre_handle lh = { 0 };
5402 struct lu_buf ea_buf = { NULL };
5403 struct lov_comp_md_v1 *lcm = NULL;
5404 struct lov_comp_md_entry_v1 *lcme = NULL;
5409 bool locked = false;
5410 bool stripe = false;
5411 bool bad_oi = false;
5414 if (!S_ISREG(lfsck_object_type(obj)))
5417 if (lad->lad_assistant_status < 0)
5418 GOTO(out, rc = -ESRCH);
5420 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
5421 lmm_oi_cpu_to_le(oi, oi);
5422 dt_read_lock(env, obj, 0);
5427 if (dt_object_exists(obj) == 0 ||
5428 lfsck_is_dead_obj(obj))
5431 rc = lfsck_layout_get_lovea(env, obj, buf);
5433 /* Skip bad lov EA during the 1st cycle scanning, and
5434 * try to recover it via orphan in the 2nd scanning. */
5435 GOTO(out, rc = (rc == -EINVAL ? 0 : rc));
5439 magic = le32_to_cpu(lmm->lmm_magic);
5440 if (magic == LOV_MAGIC_COMP_V1) {
5444 count = le16_to_cpu(lcm->lcm_entry_count);
5445 for (i = 0; i < count; i++) {
5446 lcme = &lcm->lcm_entries[i];
5447 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5448 if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) != 0)
5452 GOTO(out, stripe = true);
5453 } else if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0) {
5454 GOTO(out, stripe = true);
5458 /* Inconsistent lmm_oi, should be repaired. */
5461 if (bk->lb_param & LPF_DRYRUN) {
5462 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5464 GOTO(out, stripe = true);
5467 if (!lustre_handle_is_used(&lh)) {
5468 dt_read_unlock(env, obj);
5470 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5471 MDS_INODELOCK_LAYOUT |
5472 MDS_INODELOCK_XATTR, LCK_EX);
5476 handle = dt_trans_create(env, dev);
5478 GOTO(out, rc = PTR_ERR(handle));
5480 lfsck_buf_init(&ea_buf, lmm, size);
5481 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5482 LU_XATTR_REPLACE, handle);
5486 rc = dt_trans_start_local(env, dev, handle);
5490 dt_write_lock(env, obj, 0);
5496 if (magic == LOV_MAGIC_COMP_V1) {
5499 for (i = 0; i < count; i++) {
5500 lcme = &lcm->lcm_entries[i];
5501 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5508 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5509 LU_XATTR_REPLACE, handle);
5513 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5515 GOTO(out, stripe = true);
5519 if (lustre_handle_is_used(&lh))
5520 dt_write_unlock(env, obj);
5522 dt_read_unlock(env, obj);
5525 if (handle != NULL && !IS_ERR(handle))
5526 dt_trans_stop(env, dev, handle);
5528 lfsck_ibits_unlock(&lh, LCK_EX);
5531 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
5532 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
5533 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
5534 PFID(lfsck_dto2fid(obj)), rc);
5537 if (magic == LOV_MAGIC_COMP_V1) {
5540 for (i = 0; i < count; i++) {
5541 lcme = &lcm->lcm_entries[i];
5542 if (!(le32_to_cpu(lcme->lcme_flags) &
5546 rc = lfsck_layout_scan_stripes(env, com, obj,
5547 (struct lov_mds_md_v1 *)(buf->lb_buf +
5548 le32_to_cpu(lcme->lcme_offset)),
5549 le32_to_cpu(lcme->lcme_id));
5552 rc = lfsck_layout_scan_stripes(env, com, obj, lmm, 0);
5555 down_write(&com->lc_sem);
5556 com->lc_new_checked++;
5558 lfsck_layout_record_failure(env, lfsck, lo);
5559 up_write(&com->lc_sem);
5565 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
5566 struct lfsck_component *com,
5567 struct dt_object *obj)
5569 struct lfsck_instance *lfsck = com->lc_lfsck;
5570 struct lfsck_layout *lo = com->lc_file_ram;
5571 const struct lu_fid *fid = lfsck_dto2fid(obj);
5572 struct lfsck_layout_slave_data *llsd = com->lc_data;
5573 struct lfsck_layout_seq *lls;
5579 LASSERT(llsd != NULL);
5581 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
5582 cfs_fail_val == lfsck_dev_idx(lfsck)) {
5583 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(1),
5585 struct ptlrpc_thread *thread = &lfsck->li_thread;
5587 l_wait_event(thread->t_ctl_waitq,
5588 !thread_is_running(thread),
5592 lfsck_rbtree_update_bitmap(env, com, fid, false);
5594 down_write(&com->lc_sem);
5595 if (fid_is_idif(fid))
5597 else if (!fid_is_norm(fid) ||
5598 !fid_is_for_ostobj(env, lfsck, obj, fid))
5599 GOTO(unlock, rc = 0);
5602 com->lc_new_checked++;
5604 lls = lfsck_layout_seq_lookup(llsd, seq);
5607 if (unlikely(lls == NULL))
5608 GOTO(unlock, rc = -ENOMEM);
5610 INIT_LIST_HEAD(&lls->lls_list);
5612 rc = lfsck_layout_lastid_load(env, com, lls);
5614 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5615 "load LAST_ID for %#llx: rc = %d\n",
5616 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
5617 lo->ll_objs_failed_phase1++;
5622 lfsck_layout_seq_insert(llsd, lls);
5625 if (unlikely(fid_is_last_id(fid)))
5626 GOTO(unlock, rc = 0);
5628 if (fid_is_idif(fid))
5629 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
5633 if (oid > lls->lls_lastid_known)
5634 lls->lls_lastid_known = oid;
5636 if (oid > lls->lls_lastid) {
5637 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
5638 /* OFD may create new objects during LFSCK scanning. */
5639 rc = lfsck_layout_lastid_reload(env, com, lls);
5640 if (unlikely(rc != 0)) {
5641 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5642 "reload LAST_ID for %#llx: rc = %d\n",
5643 lfsck_lfsck2name(com->lc_lfsck),
5649 if (oid <= lls->lls_lastid ||
5650 lo->ll_flags & LF_CRASHED_LASTID)
5651 GOTO(unlock, rc = 0);
5653 LASSERT(lfsck->li_out_notify != NULL);
5655 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5656 LE_LASTID_REBUILDING);
5657 lo->ll_flags |= LF_CRASHED_LASTID;
5659 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
5660 "LAST_ID file (2) for the sequence %#llx"
5661 ", old value %llu, known value %llu\n",
5662 lfsck_lfsck2name(lfsck), lls->lls_seq,
5663 lls->lls_lastid, oid);
5666 lls->lls_lastid = oid;
5670 GOTO(unlock, rc = 0);
5673 up_write(&com->lc_sem);
5678 static int lfsck_layout_exec_dir(const struct lu_env *env,
5679 struct lfsck_component *com,
5680 struct lfsck_assistant_object *lso,
5681 struct lu_dirent *ent, __u16 type)
5686 static int lfsck_layout_master_post(const struct lu_env *env,
5687 struct lfsck_component *com,
5688 int result, bool init)
5690 struct lfsck_instance *lfsck = com->lc_lfsck;
5691 struct lfsck_layout *lo = com->lc_file_ram;
5695 lfsck_post_generic(env, com, &result);
5697 down_write(&com->lc_sem);
5698 spin_lock(&lfsck->li_lock);
5700 lo->ll_pos_last_checkpoint =
5701 lfsck->li_pos_checkpoint.lp_oit_cookie;
5704 if (lo->ll_flags & LF_INCOMPLETE)
5705 lo->ll_status = LS_PARTIAL;
5707 lo->ll_status = LS_SCANNING_PHASE2;
5708 lo->ll_flags |= LF_SCANNED_ONCE;
5709 lo->ll_flags &= ~LF_UPGRADE;
5710 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5711 } else if (result == 0) {
5712 if (lfsck->li_status != 0)
5713 lo->ll_status = lfsck->li_status;
5715 lo->ll_status = LS_STOPPED;
5716 if (lo->ll_status != LS_PAUSED)
5717 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5719 lo->ll_status = LS_FAILED;
5720 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5722 spin_unlock(&lfsck->li_lock);
5725 lo->ll_run_time_phase1 += ktime_get_seconds() -
5726 lfsck->li_time_last_checkpoint;
5727 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5728 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5729 com->lc_new_checked = 0;
5732 rc = lfsck_layout_store(env, com);
5733 up_write(&com->lc_sem);
5735 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
5736 lfsck_lfsck2name(lfsck), rc);
5741 static int lfsck_layout_slave_post(const struct lu_env *env,
5742 struct lfsck_component *com,
5743 int result, bool init)
5745 struct lfsck_instance *lfsck = com->lc_lfsck;
5746 struct lfsck_layout *lo = com->lc_file_ram;
5750 down_write(&com->lc_sem);
5751 rc = lfsck_layout_lastid_store(env, com);
5755 LASSERT(lfsck->li_out_notify != NULL);
5757 spin_lock(&lfsck->li_lock);
5759 lo->ll_pos_last_checkpoint =
5760 lfsck->li_pos_checkpoint.lp_oit_cookie;
5763 lo->ll_status = LS_SCANNING_PHASE2;
5764 lo->ll_flags |= LF_SCANNED_ONCE;
5765 if (lo->ll_flags & LF_CRASHED_LASTID) {
5767 lo->ll_flags &= ~LF_CRASHED_LASTID;
5769 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
5770 "crashed LAST_ID files successfully\n",
5771 lfsck_lfsck2name(lfsck));
5773 lo->ll_flags &= ~LF_UPGRADE;
5774 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5775 } else if (result == 0) {
5776 if (lfsck->li_status != 0)
5777 lo->ll_status = lfsck->li_status;
5779 lo->ll_status = LS_STOPPED;
5780 if (lo->ll_status != LS_PAUSED)
5781 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5783 lo->ll_status = LS_FAILED;
5784 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5786 spin_unlock(&lfsck->li_lock);
5789 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5793 lo->ll_run_time_phase1 += ktime_get_seconds() -
5794 lfsck->li_time_last_checkpoint;
5795 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5796 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5797 com->lc_new_checked = 0;
5800 rc = lfsck_layout_store(env, com);
5801 up_write(&com->lc_sem);
5803 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
5805 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
5806 lfsck_lfsck2name(lfsck), rc);
5811 static void lfsck_layout_dump(const struct lu_env *env,
5812 struct lfsck_component *com, struct seq_file *m)
5814 struct lfsck_instance *lfsck = com->lc_lfsck;
5815 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5816 struct lfsck_layout *lo = com->lc_file_ram;
5819 down_read(&com->lc_sem);
5820 if (bk->lb_param & LPF_DRYRUN)
5821 prefix = "inconsistent";
5823 prefix = "repaired";
5825 seq_printf(m, "name: lfsck_layout\n"
5831 lfsck_status2name(lo->ll_status));
5833 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
5835 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
5837 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
5839 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
5841 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
5843 seq_printf(m, "latest_start_position: %llu\n"
5844 "last_checkpoint_position: %llu\n"
5845 "first_failure_position: %llu\n",
5846 lo->ll_pos_latest_start,
5847 lo->ll_pos_last_checkpoint,
5848 lo->ll_pos_first_inconsistent);
5850 seq_printf(m, "success_count: %u\n"
5851 "%s_dangling: %llu\n"
5852 "%s_unmatched_pair: %llu\n"
5853 "%s_multiple_referenced: %llu\n"
5855 "%s_inconsistent_owner: %llu\n"
5858 "failed_phase1: %llu\n"
5859 "failed_phase2: %llu\n",
5860 lo->ll_success_count,
5861 prefix, lo->ll_objs_repaired[LLIT_DANGLING - 1],
5862 prefix, lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
5863 prefix, lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
5864 prefix, lo->ll_objs_repaired[LLIT_ORPHAN - 1],
5865 prefix, lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
5866 prefix, lo->ll_objs_repaired[LLIT_OTHERS - 1],
5867 lo->ll_objs_skipped,
5868 lo->ll_objs_failed_phase1,
5869 lo->ll_objs_failed_phase2);
5871 if (lo->ll_status == LS_SCANNING_PHASE1) {
5872 time64_t duration = ktime_get_seconds() -
5873 lfsck->li_time_last_checkpoint;
5874 u64 checked = lo->ll_objs_checked_phase1 +
5875 com->lc_new_checked;
5876 u64 speed = checked;
5877 u64 new_checked = com->lc_new_checked;
5878 time64_t rtime = lo->ll_run_time_phase1 + duration;
5882 div_u64(new_checked, duration);
5884 div_u64(speed, rtime);
5885 seq_printf(m, "checked_phase1: %llu\n"
5886 "checked_phase2: %llu\n"
5887 "run_time_phase1: %lld seconds\n"
5888 "run_time_phase2: %lld seconds\n"
5889 "average_speed_phase1: %llu items/sec\n"
5890 "average_speed_phase2: N/A\n"
5891 "real-time_speed_phase1: %llu items/sec\n"
5892 "real-time_speed_phase2: N/A\n",
5894 lo->ll_objs_checked_phase2,
5896 lo->ll_run_time_phase2,
5900 if (likely(lfsck->li_di_oit)) {
5901 const struct dt_it_ops *iops =
5902 &lfsck->li_obj_oit->do_index_ops->dio_it;
5904 /* The low layer otable-based iteration position may NOT
5905 * exactly match the layout-based directory traversal
5906 * cookie. Generally, it is not a serious issue. But the
5907 * caller should NOT make assumption on that. */
5908 pos = iops->store(env, lfsck->li_di_oit);
5909 if (!lfsck->li_current_oit_processed)
5912 pos = lo->ll_pos_last_checkpoint;
5915 seq_printf(m, "current_position: %llu\n", pos);
5916 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
5917 time64_t duration = ktime_get_seconds() -
5918 com->lc_time_last_checkpoint;
5919 u64 checked = lo->ll_objs_checked_phase2 +
5920 com->lc_new_checked;
5921 u64 speed1 = lo->ll_objs_checked_phase1;
5922 u64 speed2 = checked;
5923 u64 new_checked = com->lc_new_checked;
5924 time64_t rtime = lo->ll_run_time_phase2 + duration;
5927 div_u64(new_checked, duration);
5928 if (lo->ll_run_time_phase1 != 0)
5929 div_u64(speed1, lo->ll_run_time_phase1);
5931 div_u64(speed2, rtime);
5932 seq_printf(m, "checked_phase1: %llu\n"
5933 "checked_phase2: %llu\n"
5934 "run_time_phase1: %lld seconds\n"
5935 "run_time_phase2: %lld seconds\n"
5936 "average_speed_phase1: %llu items/sec\n"
5937 "average_speed_phase2: %llu items/sec\n"
5938 "real-time_speed_phase1: N/A\n"
5939 "real-time_speed_phase2: %llu items/sec\n"
5940 "current_position: "DFID"\n",
5941 lo->ll_objs_checked_phase1,
5943 lo->ll_run_time_phase1,
5948 PFID(&com->lc_fid_latest_scanned_phase2));
5950 __u64 speed1 = lo->ll_objs_checked_phase1;
5951 __u64 speed2 = lo->ll_objs_checked_phase2;
5953 if (lo->ll_run_time_phase1 != 0)
5954 div_u64(speed1, lo->ll_run_time_phase1);
5955 if (lo->ll_run_time_phase2 != 0)
5956 div_u64(speed2, lo->ll_run_time_phase2);
5957 seq_printf(m, "checked_phase1: %llu\n"
5958 "checked_phase2: %llu\n"
5959 "run_time_phase1: %lld seconds\n"
5960 "run_time_phase2: %lld seconds\n"
5961 "average_speed_phase1: %llu items/sec\n"
5962 "average_speed_phase2: %llu objs/sec\n"
5963 "real-time_speed_phase1: N/A\n"
5964 "real-time_speed_phase2: N/A\n"
5965 "current_position: N/A\n",
5966 lo->ll_objs_checked_phase1,
5967 lo->ll_objs_checked_phase2,
5968 lo->ll_run_time_phase1,
5969 lo->ll_run_time_phase2,
5974 up_read(&com->lc_sem);
5977 static int lfsck_layout_master_double_scan(const struct lu_env *env,
5978 struct lfsck_component *com)
5980 struct lfsck_layout *lo = com->lc_file_ram;
5981 struct lfsck_assistant_data *lad = com->lc_data;
5982 struct lfsck_instance *lfsck = com->lc_lfsck;
5983 struct lfsck_tgt_descs *ltds;
5984 struct lfsck_tgt_desc *ltd;
5985 struct lfsck_tgt_desc *next;
5988 rc = lfsck_double_scan_generic(env, com, lo->ll_status);
5990 if (thread_is_stopped(&lad->lad_thread)) {
5991 LASSERT(list_empty(&lad->lad_req_list));
5992 LASSERT(list_empty(&lad->lad_ost_phase1_list));
5993 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
5995 ltds = &lfsck->li_ost_descs;
5996 spin_lock(<ds->ltd_lock);
5997 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5998 ltd_layout_phase_list) {
5999 list_del_init(<d->ltd_layout_phase_list);
6001 spin_unlock(<ds->ltd_lock);
6003 ltds = &lfsck->li_mdt_descs;
6004 spin_lock(<ds->ltd_lock);
6005 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6006 ltd_layout_phase_list) {
6007 list_del_init(<d->ltd_layout_phase_list);
6009 spin_unlock(<ds->ltd_lock);
6015 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
6016 struct lfsck_component *com)
6018 struct lfsck_instance *lfsck = com->lc_lfsck;
6019 struct lfsck_layout_slave_data *llsd = com->lc_data;
6020 struct lfsck_layout *lo = com->lc_file_ram;
6021 struct ptlrpc_thread *thread = &lfsck->li_thread;
6025 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
6026 lfsck_lfsck2name(lfsck));
6028 atomic_inc(&lfsck->li_double_scan_count);
6030 if (lo->ll_flags & LF_INCOMPLETE)
6033 com->lc_new_checked = 0;
6034 com->lc_new_scanned = 0;
6035 com->lc_time_last_checkpoint = ktime_get_seconds();
6036 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6037 LFSCK_CHECKPOINT_INTERVAL;
6040 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
6043 rc = lfsck_layout_slave_query_master(env, com);
6044 if (list_empty(&llsd->llsd_master_list)) {
6045 if (unlikely(!thread_is_running(thread)))
6056 rc = l_wait_event(thread->t_ctl_waitq,
6057 !thread_is_running(thread) ||
6058 lo->ll_flags & LF_INCOMPLETE ||
6059 list_empty(&llsd->llsd_master_list),
6061 if (unlikely(!thread_is_running(thread)))
6064 if (lo->ll_flags & LF_INCOMPLETE)
6067 if (rc == -ETIMEDOUT)
6070 GOTO(done, rc = (rc < 0 ? rc : 1));
6074 rc = lfsck_layout_double_scan_result(env, com, rc);
6075 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
6076 (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
6077 lfsck_layout_slave_quit(env, com);
6078 if (atomic_dec_and_test(&lfsck->li_double_scan_count))
6079 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6081 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
6082 "status %d: rc = %d\n",
6083 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
6088 static void lfsck_layout_master_data_release(const struct lu_env *env,
6089 struct lfsck_component *com)
6091 struct lfsck_assistant_data *lad = com->lc_data;
6092 struct lfsck_instance *lfsck = com->lc_lfsck;
6093 struct lfsck_tgt_descs *ltds;
6094 struct lfsck_tgt_desc *ltd;
6095 struct lfsck_tgt_desc *next;
6097 LASSERT(lad != NULL);
6098 LASSERT(thread_is_init(&lad->lad_thread) ||
6099 thread_is_stopped(&lad->lad_thread));
6100 LASSERT(list_empty(&lad->lad_req_list));
6102 com->lc_data = NULL;
6104 ltds = &lfsck->li_ost_descs;
6105 spin_lock(<ds->ltd_lock);
6106 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6107 ltd_layout_phase_list) {
6108 list_del_init(<d->ltd_layout_phase_list);
6110 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6111 ltd_layout_phase_list) {
6112 list_del_init(<d->ltd_layout_phase_list);
6114 list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
6116 list_del_init(<d->ltd_layout_list);
6118 spin_unlock(<ds->ltd_lock);
6120 ltds = &lfsck->li_mdt_descs;
6121 spin_lock(<ds->ltd_lock);
6122 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6123 ltd_layout_phase_list) {
6124 list_del_init(<d->ltd_layout_phase_list);
6126 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6127 ltd_layout_phase_list) {
6128 list_del_init(<d->ltd_layout_phase_list);
6130 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
6132 list_del_init(<d->ltd_layout_list);
6134 spin_unlock(<ds->ltd_lock);
6136 if (likely(lad->lad_bitmap != NULL))
6137 CFS_FREE_BITMAP(lad->lad_bitmap);
6142 static void lfsck_layout_slave_data_release(const struct lu_env *env,
6143 struct lfsck_component *com)
6145 struct lfsck_layout_slave_data *llsd = com->lc_data;
6147 lfsck_layout_slave_quit(env, com);
6148 com->lc_data = NULL;
6152 static void lfsck_layout_master_quit(const struct lu_env *env,
6153 struct lfsck_component *com)
6155 struct lfsck_assistant_data *lad = com->lc_data;
6156 struct lfsck_instance *lfsck = com->lc_lfsck;
6157 struct lfsck_tgt_descs *ltds;
6158 struct lfsck_tgt_desc *ltd;
6159 struct lfsck_tgt_desc *next;
6161 LASSERT(lad != NULL);
6163 lfsck_quit_generic(env, com);
6165 LASSERT(thread_is_init(&lad->lad_thread) ||
6166 thread_is_stopped(&lad->lad_thread));
6167 LASSERT(list_empty(&lad->lad_req_list));
6169 ltds = &lfsck->li_ost_descs;
6170 spin_lock(<ds->ltd_lock);
6171 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6172 ltd_layout_phase_list) {
6173 list_del_init(<d->ltd_layout_phase_list);
6175 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6176 ltd_layout_phase_list) {
6177 list_del_init(<d->ltd_layout_phase_list);
6179 spin_unlock(<ds->ltd_lock);
6181 ltds = &lfsck->li_mdt_descs;
6182 spin_lock(<ds->ltd_lock);
6183 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6184 ltd_layout_phase_list) {
6185 list_del_init(<d->ltd_layout_phase_list);
6187 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6188 ltd_layout_phase_list) {
6189 list_del_init(<d->ltd_layout_phase_list);
6191 spin_unlock(<ds->ltd_lock);
6194 static void lfsck_layout_slave_quit(const struct lu_env *env,
6195 struct lfsck_component *com)
6197 struct lfsck_layout_slave_data *llsd = com->lc_data;
6198 struct lfsck_layout_seq *lls;
6199 struct lfsck_layout_seq *next;
6200 struct lfsck_layout_slave_target *llst;
6202 LASSERT(llsd != NULL);
6204 down_write(&com->lc_sem);
6205 list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
6207 list_del_init(&lls->lls_list);
6208 lfsck_object_put(env, lls->lls_lastid_obj);
6211 up_write(&com->lc_sem);
6213 spin_lock(&llsd->llsd_lock);
6214 while (!list_empty(&llsd->llsd_master_list)) {
6215 llst = list_entry(llsd->llsd_master_list.next,
6216 struct lfsck_layout_slave_target, llst_list);
6217 list_del_init(&llst->llst_list);
6218 spin_unlock(&llsd->llsd_lock);
6219 lfsck_layout_llst_put(llst);
6220 spin_lock(&llsd->llsd_lock);
6222 spin_unlock(&llsd->llsd_lock);
6224 lfsck_rbtree_cleanup(env, com);
6227 static int lfsck_layout_master_in_notify(const struct lu_env *env,
6228 struct lfsck_component *com,
6229 struct lfsck_request *lr)
6231 struct lfsck_instance *lfsck = com->lc_lfsck;
6232 struct lfsck_layout *lo = com->lc_file_ram;
6233 struct lfsck_assistant_data *lad = com->lc_data;
6234 struct lfsck_tgt_descs *ltds;
6235 struct lfsck_tgt_desc *ltd;
6239 if (lr->lr_event == LE_PAIRS_VERIFY) {
6242 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
6249 CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
6250 "from %s %x, status %d, flags %x, flags2 %x\n",
6251 lfsck_lfsck2name(lfsck), lr->lr_event,
6252 (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
6253 lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
6255 if (lr->lr_event != LE_PHASE1_DONE &&
6256 lr->lr_event != LE_PHASE2_DONE &&
6257 lr->lr_event != LE_PEER_EXIT)
6260 if (lr->lr_flags & LEF_FROM_OST)
6261 ltds = &lfsck->li_ost_descs;
6263 ltds = &lfsck->li_mdt_descs;
6264 spin_lock(<ds->ltd_lock);
6265 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
6267 spin_unlock(<ds->ltd_lock);
6272 list_del_init(<d->ltd_layout_phase_list);
6273 switch (lr->lr_event) {
6274 case LE_PHASE1_DONE:
6275 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
6276 if (lr->lr_flags2 & LF_INCOMPLETE) {
6277 if (lr->lr_flags & LEF_FROM_OST)
6278 lfsck_lad_set_bitmap(env, com,
6281 lo->ll_flags |= LF_INCOMPLETE;
6283 ltd->ltd_layout_done = 1;
6284 list_del_init(<d->ltd_layout_list);
6289 if (lr->lr_flags & LEF_FROM_OST) {
6290 if (list_empty(<d->ltd_layout_list))
6291 list_add_tail(<d->ltd_layout_list,
6292 &lad->lad_ost_list);
6293 list_add_tail(<d->ltd_layout_phase_list,
6294 &lad->lad_ost_phase2_list);
6296 if (list_empty(<d->ltd_layout_list))
6297 list_add_tail(<d->ltd_layout_list,
6298 &lad->lad_mdt_list);
6299 list_add_tail(<d->ltd_layout_phase_list,
6300 &lad->lad_mdt_phase2_list);
6303 case LE_PHASE2_DONE:
6304 ltd->ltd_layout_done = 1;
6305 if (!list_empty(<d->ltd_layout_list)) {
6306 list_del_init(<d->ltd_layout_list);
6307 if (lr->lr_flags2 & LF_INCOMPLETE) {
6308 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
6316 ltd->ltd_layout_done = 1;
6317 list_del_init(<d->ltd_layout_list);
6318 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
6319 !(lr->lr_flags & LEF_FROM_OST))
6320 lo->ll_flags |= LF_INCOMPLETE;
6325 spin_unlock(<ds->ltd_lock);
6327 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
6328 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6330 memset(stop, 0, sizeof(*stop));
6331 stop->ls_status = lr->lr_status;
6332 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6333 lfsck_stop(env, lfsck->li_bottom, stop);
6334 } else if (lfsck_phase2_next_ready(lad)) {
6335 wake_up_all(&lad->lad_thread.t_ctl_waitq);
6341 static int lfsck_layout_slave_in_notify_local(const struct lu_env *env,
6342 struct lfsck_component *com,
6343 struct lfsck_req_local *lrl,
6348 switch (lrl->lrl_event) {
6349 case LEL_FID_ACCESSED:
6350 lfsck_rbtree_update_bitmap(env, com, &lrl->lrl_fid, true);
6352 case LEL_PAIRS_VERIFY_LOCAL: {
6355 lrl->lrl_status = LPVS_INIT;
6356 /* Firstly, if the MDT-object which is claimed via OST-object
6357 * local stored PFID xattr recognizes the OST-object, then it
6358 * must be that the client given PFID is wrong. */
6359 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6360 &lrl->lrl_ff_local.ff_parent,
6361 lrl->lrl_ff_local.ff_layout.ol_comp_id);
6365 lrl->lrl_status = LPVS_INCONSISTENT;
6366 /* The OST-object local stored PFID xattr is stale. We need to
6367 * check whether the MDT-object that is claimed via the client
6368 * given PFID information recognizes the OST-object or not. If
6369 * matches, then need to update the OST-object's PFID xattr. */
6370 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6371 &lrl->lrl_ff_client.ff_parent,
6372 lrl->lrl_ff_client.ff_layout.ol_comp_id);
6374 * We are not sure whether the client given PFID information
6375 * is correct or not, do nothing to avoid improper fixing.
6378 * The client given PFID information is also invalid, we can
6379 * NOT fix the OST-object inconsistency.
6382 lrl->lrl_status = LPVS_INCONSISTENT_TOFIX;
6383 rc = lfsck_layout_slave_repair_pfid(env, com, lrl);
6392 RETURN(-EOPNOTSUPP);
6395 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
6396 struct lfsck_component *com,
6397 struct lfsck_request *lr)
6399 struct lfsck_instance *lfsck = com->lc_lfsck;
6400 struct lfsck_layout_slave_data *llsd = com->lc_data;
6401 struct lfsck_layout_slave_target *llst;
6405 switch (lr->lr_event) {
6406 case LE_CONDITIONAL_DESTROY:
6407 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
6409 case LE_PHASE1_DONE: {
6410 if (lr->lr_flags2 & LF_INCOMPLETE) {
6411 struct lfsck_layout *lo = com->lc_file_ram;
6413 lo->ll_flags |= LF_INCOMPLETE;
6414 llst = lfsck_layout_llst_find_and_del(llsd,
6418 lfsck_layout_llst_put(llst);
6419 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6425 case LE_PHASE2_DONE:
6427 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
6428 "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
6429 lr->lr_event, lr->lr_index, lr->lr_status);
6435 llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
6439 lfsck_layout_llst_put(llst);
6440 if (list_empty(&llsd->llsd_master_list))
6441 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6443 if (lr->lr_event == LE_PEER_EXIT &&
6444 (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
6445 (list_empty(&llsd->llsd_master_list) &&
6446 (lr->lr_status == LS_STOPPED ||
6447 lr->lr_status == LS_CO_STOPPED)))) {
6448 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6450 memset(stop, 0, sizeof(*stop));
6451 stop->ls_status = lr->lr_status;
6452 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6453 lfsck_stop(env, lfsck->li_bottom, stop);
6459 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
6463 for (i = 0; i < LLIT_MAX; i++)
6464 *count += lo->ll_objs_repaired[i];
6467 static int lfsck_layout_query_all(const struct lu_env *env,
6468 struct lfsck_component *com,
6469 __u32 *mdts_count, __u32 *osts_count,
6472 struct lfsck_layout *lo = com->lc_file_ram;
6473 struct lfsck_tgt_descs *ltds;
6474 struct lfsck_tgt_desc *ltd;
6479 rc = lfsck_query_all(env, com);
6483 ltds = &com->lc_lfsck->li_mdt_descs;
6484 down_read(<ds->ltd_rw_sem);
6485 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
6486 ltd = lfsck_ltd2tgt(ltds, idx);
6487 LASSERT(ltd != NULL);
6489 mdts_count[ltd->ltd_layout_status]++;
6490 *repaired += ltd->ltd_layout_repaired;
6492 up_read(<ds->ltd_rw_sem);
6494 ltds = &com->lc_lfsck->li_ost_descs;
6495 down_read(<ds->ltd_rw_sem);
6496 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
6497 ltd = lfsck_ltd2tgt(ltds, idx);
6498 LASSERT(ltd != NULL);
6500 osts_count[ltd->ltd_layout_status]++;
6501 *repaired += ltd->ltd_layout_repaired;
6503 up_read(<ds->ltd_rw_sem);
6505 down_read(&com->lc_sem);
6506 mdts_count[lo->ll_status]++;
6507 lfsck_layout_repaired(lo, repaired);
6508 up_read(&com->lc_sem);
6513 static int lfsck_layout_query(const struct lu_env *env,
6514 struct lfsck_component *com,
6515 struct lfsck_request *req,
6516 struct lfsck_reply *rep,
6517 struct lfsck_query *que, int idx)
6519 struct lfsck_layout *lo = com->lc_file_ram;
6523 LASSERT(com->lc_lfsck->li_master);
6525 rc = lfsck_layout_query_all(env, com,
6526 que->lu_mdts_count[idx],
6527 que->lu_osts_count[idx],
6528 &que->lu_repaired[idx]);
6530 down_read(&com->lc_sem);
6531 rep->lr_status = lo->ll_status;
6532 if (req->lr_flags & LEF_QUERY_ALL)
6533 lfsck_layout_repaired(lo, &rep->lr_repaired);
6534 up_read(&com->lc_sem);
6540 /* with lfsck::li_lock held */
6541 static int lfsck_layout_slave_join(const struct lu_env *env,
6542 struct lfsck_component *com,
6543 struct lfsck_start_param *lsp)
6545 struct lfsck_instance *lfsck = com->lc_lfsck;
6546 struct lfsck_layout_slave_data *llsd = com->lc_data;
6547 struct lfsck_layout_slave_target *llst;
6548 struct lfsck_start *start = lsp->lsp_start;
6552 if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
6555 if (!lsp->lsp_index_valid)
6558 /* If someone is running the LFSCK without orphan handling,
6559 * it will not maintain the object accessing rbtree. So we
6560 * cannot join it for orphan handling. */
6561 if (!llsd->llsd_rbtree_valid)
6564 spin_unlock(&lfsck->li_lock);
6565 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
6566 spin_lock(&lfsck->li_lock);
6567 if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
6568 spin_unlock(&lfsck->li_lock);
6569 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
6572 lfsck_layout_llst_put(llst);
6573 spin_lock(&lfsck->li_lock);
6580 static struct lfsck_operations lfsck_layout_master_ops = {
6581 .lfsck_reset = lfsck_layout_reset,
6582 .lfsck_fail = lfsck_layout_fail,
6583 .lfsck_checkpoint = lfsck_layout_master_checkpoint,
6584 .lfsck_prep = lfsck_layout_master_prep,
6585 .lfsck_exec_oit = lfsck_layout_master_exec_oit,
6586 .lfsck_exec_dir = lfsck_layout_exec_dir,
6587 .lfsck_post = lfsck_layout_master_post,
6588 .lfsck_dump = lfsck_layout_dump,
6589 .lfsck_double_scan = lfsck_layout_master_double_scan,
6590 .lfsck_data_release = lfsck_layout_master_data_release,
6591 .lfsck_quit = lfsck_layout_master_quit,
6592 .lfsck_in_notify = lfsck_layout_master_in_notify,
6593 .lfsck_query = lfsck_layout_query,
6596 static struct lfsck_operations lfsck_layout_slave_ops = {
6597 .lfsck_reset = lfsck_layout_reset,
6598 .lfsck_fail = lfsck_layout_fail,
6599 .lfsck_checkpoint = lfsck_layout_slave_checkpoint,
6600 .lfsck_prep = lfsck_layout_slave_prep,
6601 .lfsck_exec_oit = lfsck_layout_slave_exec_oit,
6602 .lfsck_exec_dir = lfsck_layout_exec_dir,
6603 .lfsck_post = lfsck_layout_slave_post,
6604 .lfsck_dump = lfsck_layout_dump,
6605 .lfsck_double_scan = lfsck_layout_slave_double_scan,
6606 .lfsck_data_release = lfsck_layout_slave_data_release,
6607 .lfsck_quit = lfsck_layout_slave_quit,
6608 .lfsck_in_notify_local = lfsck_layout_slave_in_notify_local,
6609 .lfsck_in_notify = lfsck_layout_slave_in_notify,
6610 .lfsck_query = lfsck_layout_query,
6611 .lfsck_join = lfsck_layout_slave_join,
6614 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
6615 struct lfsck_component *com,
6616 struct lfsck_position *pos)
6618 struct lfsck_assistant_data *lad = com->lc_data;
6619 struct lfsck_layout_req *llr;
6621 if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
6625 if (list_empty(&lad->lad_req_list))
6628 llr = list_entry(lad->lad_req_list.next,
6629 struct lfsck_layout_req,
6631 pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
6634 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
6635 .la_handler_p1 = lfsck_layout_assistant_handler_p1,
6636 .la_handler_p2 = lfsck_layout_assistant_handler_p2,
6637 .la_fill_pos = lfsck_layout_assistant_fill_pos,
6638 .la_double_scan_result = lfsck_layout_double_scan_result,
6639 .la_req_fini = lfsck_layout_assistant_req_fini,
6640 .la_sync_failures = lfsck_layout_assistant_sync_failures,
6643 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
6645 struct lfsck_component *com;
6646 struct lfsck_layout *lo;
6647 struct dt_object *root = NULL;
6648 struct dt_object *obj;
6657 INIT_LIST_HEAD(&com->lc_link);
6658 INIT_LIST_HEAD(&com->lc_link_dir);
6659 init_rwsem(&com->lc_sem);
6660 atomic_set(&com->lc_ref, 1);
6661 com->lc_lfsck = lfsck;
6662 com->lc_type = LFSCK_TYPE_LAYOUT;
6663 if (lfsck->li_master) {
6664 com->lc_ops = &lfsck_layout_master_ops;
6665 com->lc_data = lfsck_assistant_data_init(
6666 &lfsck_layout_assistant_ops,
6668 if (com->lc_data == NULL)
6669 GOTO(out, rc = -ENOMEM);
6671 for (i = 0; i < LFSCK_STF_COUNT; i++)
6672 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6674 struct lfsck_layout_slave_data *llsd;
6676 com->lc_ops = &lfsck_layout_slave_ops;
6677 OBD_ALLOC_PTR(llsd);
6679 GOTO(out, rc = -ENOMEM);
6681 INIT_LIST_HEAD(&llsd->llsd_seq_list);
6682 INIT_LIST_HEAD(&llsd->llsd_master_list);
6683 spin_lock_init(&llsd->llsd_lock);
6684 llsd->llsd_rb_root = RB_ROOT;
6685 rwlock_init(&llsd->llsd_rb_lock);
6686 com->lc_data = llsd;
6688 com->lc_file_size = sizeof(*lo);
6689 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6690 if (com->lc_file_ram == NULL)
6691 GOTO(out, rc = -ENOMEM);
6693 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6694 if (com->lc_file_disk == NULL)
6695 GOTO(out, rc = -ENOMEM);
6697 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6699 GOTO(out, rc = PTR_ERR(root));
6701 if (unlikely(!dt_try_as_dir(env, root)))
6702 GOTO(out, rc = -ENOTDIR);
6704 obj = local_file_find_or_create(env, lfsck->li_los, root,
6706 S_IFREG | S_IRUGO | S_IWUSR);
6708 GOTO(out, rc = PTR_ERR(obj));
6711 rc = lfsck_layout_load(env, com);
6713 rc = lfsck_layout_reset(env, com, true);
6714 else if (rc == -ENOENT)
6715 rc = lfsck_layout_init(env, com);
6716 else if (lfsck->li_master)
6717 rc = lfsck_load_sub_trace_files(env, com,
6718 &dt_lfsck_layout_dangling_features,
6719 LFSCK_LAYOUT, false);
6724 lo = com->lc_file_ram;
6725 switch (lo->ll_status) {
6731 spin_lock(&lfsck->li_lock);
6732 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6733 spin_unlock(&lfsck->li_lock);
6736 CERROR("%s: unknown lfsck_layout status %d\n",
6737 lfsck_lfsck2name(lfsck), lo->ll_status);
6739 case LS_SCANNING_PHASE1:
6740 case LS_SCANNING_PHASE2:
6741 /* No need to store the status to disk right now.
6742 * If the system crashed before the status stored,
6743 * it will be loaded back when next time. */
6744 lo->ll_status = LS_CRASHED;
6745 if (!lfsck->li_master)
6746 lo->ll_flags |= LF_INCOMPLETE;
6753 spin_lock(&lfsck->li_lock);
6754 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6755 spin_unlock(&lfsck->li_lock);
6759 if (lo->ll_flags & LF_CRASHED_LASTID) {
6760 LASSERT(lfsck->li_out_notify != NULL);
6762 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6763 LE_LASTID_REBUILDING);
6769 if (root != NULL && !IS_ERR(root))
6770 lfsck_object_put(env, root);
6773 lfsck_component_cleanup(env, com);
6774 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
6775 lfsck_lfsck2name(lfsck), rc);
6781 struct lfsck_orphan_it {
6782 struct lfsck_component *loi_com;
6783 struct lfsck_rbtree_node *loi_lrn;
6784 struct lfsck_layout_slave_target *loi_llst;
6785 struct lu_fid loi_key;
6786 struct lu_orphan_rec_v2 loi_rec;
6788 unsigned int loi_over:1;
6791 static int lfsck_fid_match_idx(const struct lu_env *env,
6792 struct lfsck_instance *lfsck,
6793 const struct lu_fid *fid, int idx)
6795 struct seq_server_site *ss;
6796 struct lu_server_fld *sf;
6797 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
6800 /* All abnormal cases will be returned to MDT0. */
6801 if (!fid_is_norm(fid)) {
6808 ss = lfsck_dev_site(lfsck);
6809 if (unlikely(ss == NULL))
6812 sf = ss->ss_server_fld;
6813 LASSERT(sf != NULL);
6815 fld_range_set_any(range);
6816 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
6820 if (!fld_range_is_mdt(range))
6823 if (range->lsr_index == idx)
6829 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
6830 struct dt_object *obj)
6832 struct dt_device *dev = lfsck_obj2dev(obj);
6833 struct thandle *handle;
6837 handle = dt_trans_create(env, dev);
6841 rc = dt_declare_ref_del(env, obj, handle);
6845 rc = dt_declare_destroy(env, obj, handle);
6849 rc = dt_trans_start_local(env, dev, handle);
6853 dt_write_lock(env, obj, 0);
6854 rc = dt_ref_del(env, obj, handle);
6856 rc = dt_destroy(env, obj, handle);
6857 dt_write_unlock(env, obj);
6862 dt_trans_stop(env, dev, handle);
6864 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
6865 PFID(lfsck_dto2fid(obj)), rc);
6870 static int lfsck_orphan_index_lookup(const struct lu_env *env,
6871 struct dt_object *dt,
6873 const struct dt_key *key)
6878 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
6879 struct dt_object *dt,
6880 const struct dt_rec *rec,
6881 const struct dt_key *key,
6882 struct thandle *handle)
6887 static int lfsck_orphan_index_insert(const struct lu_env *env,
6888 struct dt_object *dt,
6889 const struct dt_rec *rec,
6890 const struct dt_key *key,
6891 struct thandle *handle,
6897 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
6898 struct dt_object *dt,
6899 const struct dt_key *key,
6900 struct thandle *handle)
6905 static int lfsck_orphan_index_delete(const struct lu_env *env,
6906 struct dt_object *dt,
6907 const struct dt_key *key,
6908 struct thandle *handle)
6913 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
6914 struct dt_object *dt,
6917 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
6918 struct lfsck_instance *lfsck;
6919 struct lfsck_component *com = NULL;
6920 struct lfsck_layout_slave_data *llsd;
6921 struct lfsck_orphan_it *it = NULL;
6922 struct lfsck_layout *lo;
6926 lfsck = lfsck_instance_find(dev, true, false);
6927 if (unlikely(lfsck == NULL))
6928 RETURN(ERR_PTR(-ENXIO));
6930 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
6931 if (unlikely(com == NULL))
6932 GOTO(out, rc = -ENOENT);
6934 lo = com->lc_file_ram;
6935 if (lo->ll_flags & LF_INCOMPLETE)
6936 GOTO(out, rc = -ESRCH);
6938 llsd = com->lc_data;
6939 if (!llsd->llsd_rbtree_valid)
6940 GOTO(out, rc = -ESRCH);
6944 GOTO(out, rc = -ENOMEM);
6946 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
6947 if (it->loi_llst == NULL)
6948 GOTO(out, rc = -ENXIO);
6950 if (dev->dd_record_fid_accessed) {
6951 /* The first iteration against the rbtree, scan the whole rbtree
6952 * to remove the nodes which do NOT need to be handled. */
6953 write_lock(&llsd->llsd_rb_lock);
6954 if (dev->dd_record_fid_accessed) {
6955 struct rb_node *node;
6956 struct rb_node *next;
6957 struct lfsck_rbtree_node *lrn;
6959 /* No need to record the fid accessing anymore. */
6960 dev->dd_record_fid_accessed = 0;
6962 node = rb_first(&llsd->llsd_rb_root);
6963 while (node != NULL) {
6964 next = rb_next(node);
6965 lrn = rb_entry(node, struct lfsck_rbtree_node,
6967 if (atomic_read(&lrn->lrn_known_count) <=
6968 atomic_read(&lrn->lrn_accessed_count)) {
6969 rb_erase(node, &llsd->llsd_rb_root);
6970 lfsck_rbtree_free(lrn);
6975 write_unlock(&llsd->llsd_rb_lock);
6978 /* read lock the rbtree when init, and unlock when fini */
6979 read_lock(&llsd->llsd_rb_lock);
6987 lfsck_component_put(env, com);
6989 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
6990 lfsck_lfsck2name(lfsck), rc);
6992 lfsck_instance_put(env, lfsck);
6997 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
7000 return (struct dt_it *)it;
7003 static void lfsck_orphan_it_fini(const struct lu_env *env,
7006 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7007 struct lfsck_component *com = it->loi_com;
7008 struct lfsck_layout_slave_data *llsd;
7009 struct lfsck_layout_slave_target *llst;
7012 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
7013 lfsck_lfsck2name(com->lc_lfsck));
7015 llsd = com->lc_data;
7016 read_unlock(&llsd->llsd_rb_lock);
7017 llst = it->loi_llst;
7018 LASSERT(llst != NULL);
7020 /* Save the key and hash for iterate next. */
7021 llst->llst_fid = it->loi_key;
7022 llst->llst_hash = it->loi_hash;
7023 lfsck_layout_llst_put(llst);
7024 lfsck_component_put(env, com);
7030 * \retval +1: the iteration finished
7031 * \retval 0: on success, not finished
7032 * \retval -ve: on error
7034 static int lfsck_orphan_it_next(const struct lu_env *env,
7037 struct lfsck_thread_info *info = lfsck_env_info(env);
7038 struct filter_fid *ff = &info->lti_ff;
7039 struct lu_attr *la = &info->lti_la;
7040 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7041 struct lu_fid *key = &it->loi_key;
7042 struct lu_orphan_rec_v2 *rec = &it->loi_rec;
7043 struct ost_layout *ol = &rec->lor_layout;
7044 struct lfsck_component *com = it->loi_com;
7045 struct lfsck_instance *lfsck = com->lc_lfsck;
7046 struct lfsck_layout_slave_data *llsd = com->lc_data;
7047 struct dt_object *obj;
7048 struct lfsck_rbtree_node *lrn;
7052 __u32 idx = it->loi_llst->llst_index;
7062 lrn = lfsck_rbtree_search(llsd, key, &exact);
7070 key->f_seq = lrn->lrn_seq;
7071 key->f_oid = lrn->lrn_first_oid;
7076 if (unlikely(key->f_oid == 0)) {
7083 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
7089 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
7090 atomic_read(&lrn->lrn_accessed_count))) {
7091 struct rb_node *next = rb_next(&lrn->lrn_node);
7093 while (next != NULL) {
7094 lrn = rb_entry(next, struct lfsck_rbtree_node,
7096 if (atomic_read(&lrn->lrn_known_count) >
7097 atomic_read(&lrn->lrn_accessed_count))
7099 next = rb_next(next);
7108 key->f_seq = lrn->lrn_seq;
7109 key->f_oid = lrn->lrn_first_oid;
7113 pos = key->f_oid - lrn->lrn_first_oid;
7116 pos = find_next_bit(lrn->lrn_known_bitmap,
7117 LFSCK_RBTREE_BITMAP_WIDTH, pos);
7118 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
7119 key->f_oid = lrn->lrn_first_oid + pos;
7120 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
7128 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
7133 key->f_oid = lrn->lrn_first_oid + pos;
7134 obj = lfsck_object_find_bottom(env, lfsck, key);
7137 if (rc == -ENOENT) {
7144 dt_read_lock(env, obj, 0);
7145 if (dt_object_exists(obj) == 0 ||
7146 lfsck_is_dead_obj(obj)) {
7147 dt_read_unlock(env, obj);
7148 lfsck_object_put(env, obj);
7153 rc = dt_attr_get(env, obj, la);
7157 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, ff, sizeof(*ff)),
7159 if (rc == -ENODATA) {
7160 /* For the pre-created OST-object, update the bitmap to avoid
7161 * others LFSCK (second phase) iteration to touch it again. */
7162 if (la->la_ctime == 0) {
7163 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
7164 atomic_inc(&lrn->lrn_accessed_count);
7166 /* For the race between repairing dangling referenced
7167 * MDT-object and unlink the file, it may left orphan
7168 * OST-object there. Destroy it now! */
7169 if (unlikely(!(la->la_mode & S_ISUID))) {
7170 dt_read_unlock(env, obj);
7171 lfsck_layout_destroy_orphan(env, obj);
7172 lfsck_object_put(env, obj);
7176 } else if (idx == 0) {
7177 /* If the orphan OST-object has no parent information,
7178 * regard it as referenced by the MDT-object on MDT0. */
7179 fid_zero(&rec->lor_rec.lor_fid);
7180 rec->lor_rec.lor_uid = la->la_uid;
7181 rec->lor_rec.lor_gid = la->la_gid;
7182 memset(ol, 0, sizeof(*ol));
7187 dt_read_unlock(env, obj);
7188 lfsck_object_put(env, obj);
7193 if (rc < sizeof(struct lu_fid))
7194 GOTO(out, rc = (rc < 0 ? rc : -EINVAL));
7196 fid_le_to_cpu(&rec->lor_rec.lor_fid, &ff->ff_parent);
7197 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
7198 * MDT-object's FID::f_ver, instead it is the OST-object index in its
7199 * parent MDT-object's layout EA. */
7200 save = rec->lor_rec.lor_fid.f_stripe_idx;
7201 rec->lor_rec.lor_fid.f_ver = 0;
7202 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_rec.lor_fid, idx);
7203 /* If the orphan OST-object does not claim the MDT, then next.
7205 * If we do not know whether it matches or not, then return it
7206 * to the MDT for further check. */
7208 dt_read_unlock(env, obj);
7209 lfsck_object_put(env, obj);
7214 rec->lor_rec.lor_fid.f_stripe_idx = save;
7215 rec->lor_rec.lor_uid = la->la_uid;
7216 rec->lor_rec.lor_gid = la->la_gid;
7217 ost_layout_le_to_cpu(ol, &ff->ff_layout);
7219 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u, "
7220 "stripe size %u, stripe count %u, COMP id %u, COMP start %llu, "
7221 "COMP end %llu\n", lfsck_lfsck2name(com->lc_lfsck), PFID(key),
7222 PFID(&rec->lor_rec.lor_fid), rec->lor_rec.lor_uid,
7223 rec->lor_rec.lor_gid, ol->ol_stripe_size, ol->ol_stripe_count,
7224 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end);
7229 dt_read_unlock(env, obj);
7230 lfsck_object_put(env, obj);
7238 * \retval +1: locate to the exactly position
7239 * \retval 0: cannot locate to the exactly position,
7240 * call next() to move to a valid position.
7241 * \retval -ve: on error
7243 static int lfsck_orphan_it_get(const struct lu_env *env,
7245 const struct dt_key *key)
7247 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7250 it->loi_key = *(struct lu_fid *)key;
7251 rc = lfsck_orphan_it_next(env, di);
7261 static void lfsck_orphan_it_put(const struct lu_env *env,
7266 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
7267 const struct dt_it *di)
7269 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7271 return (struct dt_key *)&it->loi_key;
7274 static int lfsck_orphan_it_key_size(const struct lu_env *env,
7275 const struct dt_it *di)
7277 return sizeof(struct lu_fid);
7280 static int lfsck_orphan_it_rec(const struct lu_env *env,
7281 const struct dt_it *di,
7285 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7287 *(struct lu_orphan_rec_v2 *)rec = it->loi_rec;
7292 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
7293 const struct dt_it *di)
7295 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7297 return it->loi_hash;
7301 * \retval +1: locate to the exactly position
7302 * \retval 0: cannot locate to the exactly position,
7303 * call next() to move to a valid position.
7304 * \retval -ve: on error
7306 static int lfsck_orphan_it_load(const struct lu_env *env,
7307 const struct dt_it *di,
7310 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7311 struct lfsck_layout_slave_target *llst = it->loi_llst;
7314 LASSERT(llst != NULL);
7316 if (hash != llst->llst_hash) {
7317 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
7318 "iteration does not match the one when fini "
7319 "%llu, to be reset.\n",
7320 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
7322 fid_zero(&llst->llst_fid);
7323 llst->llst_hash = 0;
7326 it->loi_key = llst->llst_fid;
7327 it->loi_hash = llst->llst_hash;
7328 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
7338 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
7339 const struct dt_it *di,
7345 const struct dt_index_operations lfsck_orphan_index_ops = {
7346 .dio_lookup = lfsck_orphan_index_lookup,
7347 .dio_declare_insert = lfsck_orphan_index_declare_insert,
7348 .dio_insert = lfsck_orphan_index_insert,
7349 .dio_declare_delete = lfsck_orphan_index_declare_delete,
7350 .dio_delete = lfsck_orphan_index_delete,
7352 .init = lfsck_orphan_it_init,
7353 .fini = lfsck_orphan_it_fini,
7354 .get = lfsck_orphan_it_get,
7355 .put = lfsck_orphan_it_put,
7356 .next = lfsck_orphan_it_next,
7357 .key = lfsck_orphan_it_key,
7358 .key_size = lfsck_orphan_it_key_size,
7359 .rec = lfsck_orphan_it_rec,
7360 .store = lfsck_orphan_it_store,
7361 .load = lfsck_orphan_it_load,
7362 .key_rec = lfsck_orphan_it_key_rec,