4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2014, 2016, Intel Corporation.
26 * lustre/lfsck/lfsck_layout.c
28 * Author: Fan, Yong <fan.yong@intel.com>
32 # define EXPORT_SYMTAB
34 #define DEBUG_SUBSYSTEM S_LFSCK
36 #include <linux/bitops.h>
37 #include <linux/rbtree.h>
39 #include <lustre/lustre_idl.h>
40 #include <lu_object.h>
41 #include <dt_object.h>
42 #include <lustre_fid.h>
43 #include <lustre_lib.h>
44 #include <lustre_net.h>
45 #include <lustre/lustre_user.h>
46 #include <md_object.h>
47 #include <obd_class.h>
49 #include "lfsck_internal.h"
51 #define LFSCK_LAYOUT_MAGIC_V1 0xB173AE14
52 #define LFSCK_LAYOUT_MAGIC_V2 0xB1734D76
53 #define LFSCK_LAYOUT_MAGIC_V3 0xB17371B9
55 #define LFSCK_LAYOUT_MAGIC LFSCK_LAYOUT_MAGIC_V3
57 struct lfsck_layout_seq {
58 struct list_head lls_list;
61 __u64 lls_lastid_known;
62 struct dt_object *lls_lastid_obj;
63 unsigned int lls_dirty:1;
66 struct lfsck_layout_slave_target {
67 /* link into lfsck_layout_slave_data::llsd_master_list. */
68 struct list_head llst_list;
69 /* The position for next record in the rbtree for iteration. */
70 struct lu_fid llst_fid;
71 /* Dummy hash for iteration against the rbtree. */
76 /* How many times we have failed to get the master status. */
80 struct lfsck_layout_slave_data {
81 /* list for lfsck_layout_seq */
82 struct list_head llsd_seq_list;
84 /* list for the masters involve layout verification. */
85 struct list_head llsd_master_list;
88 struct dt_object *llsd_rb_obj;
89 struct rb_root llsd_rb_root;
90 rwlock_t llsd_rb_lock;
91 unsigned int llsd_rbtree_valid:1;
94 struct lfsck_layout_slave_async_args {
95 struct obd_export *llsaa_exp;
96 struct lfsck_component *llsaa_com;
97 struct lfsck_layout_slave_target *llsaa_llst;
100 static inline bool lfsck_comp_extent_aligned(__u64 size)
102 return (size & (LOV_MIN_STRIPE_SIZE - 1)) == 0;
106 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
108 if (atomic_dec_and_test(&llst->llst_ref)) {
109 LASSERT(list_empty(&llst->llst_list));
116 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
118 struct lfsck_layout_slave_target *llst;
119 struct lfsck_layout_slave_target *tmp;
126 INIT_LIST_HEAD(&llst->llst_list);
128 llst->llst_index = index;
129 atomic_set(&llst->llst_ref, 1);
131 spin_lock(&llsd->llsd_lock);
132 list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
133 if (tmp->llst_index == index) {
139 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
140 spin_unlock(&llsd->llsd_lock);
149 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
150 struct lfsck_layout_slave_target *llst)
154 spin_lock(&llsd->llsd_lock);
155 if (!list_empty(&llst->llst_list)) {
156 list_del_init(&llst->llst_list);
159 spin_unlock(&llsd->llsd_lock);
162 lfsck_layout_llst_put(llst);
165 static inline struct lfsck_layout_slave_target *
166 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
167 __u32 index, bool unlink)
169 struct lfsck_layout_slave_target *llst;
171 spin_lock(&llsd->llsd_lock);
172 list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
173 if (llst->llst_index == index) {
175 list_del_init(&llst->llst_list);
177 atomic_inc(&llst->llst_ref);
178 spin_unlock(&llsd->llsd_lock);
183 spin_unlock(&llsd->llsd_lock);
188 static struct lfsck_layout_req *
189 lfsck_layout_assistant_req_init(struct lfsck_assistant_object *lso,
190 struct dt_object *child, __u32 comp_id,
191 __u32 ost_idx, __u32 lov_idx)
193 struct lfsck_layout_req *llr;
197 return ERR_PTR(-ENOMEM);
199 INIT_LIST_HEAD(&llr->llr_lar.lar_list);
200 llr->llr_lar.lar_parent = lfsck_assistant_object_get(lso);
201 llr->llr_child = child;
202 llr->llr_comp_id = comp_id;
203 llr->llr_ost_idx = ost_idx;
204 llr->llr_lov_idx = lov_idx;
209 static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
210 struct lfsck_assistant_req *lar)
212 struct lfsck_layout_req *llr =
213 container_of0(lar, struct lfsck_layout_req, llr_lar);
215 lfsck_object_put(env, llr->llr_child);
216 lfsck_assistant_object_put(env, lar->lar_parent);
221 lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
222 struct ptlrpc_request *req,
226 struct lfsck_async_interpret_args *laia = args;
227 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
229 ltd->ltd_synced_failures = 1;
230 atomic_dec(laia->laia_count);
237 * Notify remote LFSCK instances about former failures.
239 * The local LFSCK instance has recorded which OSTs have ever failed to respond
240 * some LFSCK verification requests (maybe because of network issues or the OST
241 * itself trouble). During the respond gap, the OST may missed some OST-objects
242 * verification, then the OST cannot know whether related OST-objects have been
243 * referenced by related MDT-objects or not, then in the second-stage scanning,
244 * these OST-objects will be regarded as orphan, if the OST-object contains bad
245 * parent FID for back reference, then it will misguide the LFSCK to make wrong
246 * fixing for the fake orphan.
248 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
249 * it will scan the bitmap for the ever failed OSTs, and notify them that they
250 * have ever missed some OST-object verification and should skip the handling
251 * for orphan OST-objects on all MDTs that are in the layout LFSCK.
253 * \param[in] env pointer to the thread context
254 * \param[in] com pointer to the lfsck component
255 * \param[in] lr pointer to the lfsck request
257 static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
258 struct lfsck_component *com,
259 struct lfsck_request *lr)
261 struct lfsck_async_interpret_args *laia =
262 &lfsck_env_info(env)->lti_laia2;
263 struct lfsck_assistant_data *lad = com->lc_data;
264 struct lfsck_layout *lo = com->lc_file_ram;
265 struct lfsck_instance *lfsck = com->lc_lfsck;
266 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
267 struct lfsck_tgt_desc *ltd;
268 struct ptlrpc_request_set *set;
274 if (!lad->lad_incomplete)
277 /* If the MDT has ever failed to verfiy some OST-objects,
278 * then sync failures with them firstly. */
279 lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
281 atomic_set(&count, 0);
282 memset(laia, 0, sizeof(*laia));
283 laia->laia_count = &count;
284 set = ptlrpc_prep_set();
286 GOTO(out, rc = -ENOMEM);
288 down_read(<ds->ltd_rw_sem);
289 cfs_foreach_bit(lad->lad_bitmap, idx) {
290 ltd = lfsck_ltd2tgt(ltds, idx);
294 laia->laia_ltd = ltd;
295 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
296 lfsck_layout_assistant_sync_failures_interpret,
299 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
300 "notify target %x for %s phase1 done: "
301 "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
302 ltd->ltd_index, lad->lad_name, rc);
309 up_read(<ds->ltd_rw_sem);
311 if (rc == 0 && atomic_read(&count) > 0)
312 rc = ptlrpc_set_wait(set);
314 ptlrpc_set_destroy(set);
316 if (rc == 0 && atomic_read(&count) > 0)
323 /* If failed to sync failures with the OSTs, then have to
324 * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
325 * subsequent orphan OST-object handling. */
326 lo->ll_flags |= LF_INCOMPLETE;
328 lr->lr_flags2 = lo->ll_flags;
331 static int lfsck_layout_verify_header_v1v3(struct dt_object *obj,
332 struct lov_mds_md_v1 *lmm)
337 magic = le32_to_cpu(lmm->lmm_magic);
338 /* If magic crashed, keep it there. Sometime later, during OST-object
339 * orphan handling, if some OST-object(s) back-point to it, it can be
340 * verified and repaired. */
341 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
344 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
349 CDEBUG(D_LFSCK, "%s LOV EA magic %u for the file "DFID"\n",
350 rc == -EINVAL ? "Unknown" : "Unsupported",
351 magic, PFID(lfsck_dto2fid(obj)));
356 pattern = le32_to_cpu(lmm->lmm_pattern);
357 /* XXX: currently, we only support LOV_PATTERN_RAID0. */
358 if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
359 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u for the file "
360 DFID"\n", pattern, PFID(lfsck_dto2fid(obj)));
368 static int lfsck_layout_verify_header(struct dt_object *obj,
369 struct lov_mds_md_v1 *lmm)
373 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1) {
374 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
376 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
378 if (unlikely(count == 0)) {
379 CDEBUG(D_LFSCK, "the PFL file "DFID" contains invalid "
380 "components count 0\n",
381 PFID(lfsck_dto2fid(obj)));
386 for (i = 0; i < count; i++) {
387 struct lov_comp_md_entry_v1 *lcme =
388 &lcm->lcm_entries[i];
389 __u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
390 __u64 end = le64_to_cpu(lcme->lcme_extent.e_end);
391 __u32 comp_id = le32_to_cpu(lcme->lcme_id);
393 if (unlikely(comp_id == LCME_ID_INVAL ||
394 comp_id > LCME_ID_MAX)) {
395 CDEBUG(D_LFSCK, "found invalid FPL ID %u "
396 "for the file "DFID" at idx %d\n",
397 comp_id, PFID(lfsck_dto2fid(obj)), i);
402 if (unlikely(start >= end ||
403 !lfsck_comp_extent_aligned(start) ||
404 (!lfsck_comp_extent_aligned(end) &&
405 end != LUSTRE_EOF))) {
406 CDEBUG(D_LFSCK, "found invalid FPL extent "
407 "range [%llu - %llu) for the file "
409 start, end, PFID(lfsck_dto2fid(obj)), i);
414 rc = lfsck_layout_verify_header_v1v3(obj,
415 (struct lov_mds_md_v1 *)((char *)lmm +
416 le32_to_cpu(lcme->lcme_offset)));
421 rc = lfsck_layout_verify_header_v1v3(obj, lmm);
427 static int lfsck_layout_get_lovea(const struct lu_env *env,
428 struct dt_object *obj, struct lu_buf *buf)
434 rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
436 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
440 lu_buf_realloc(buf, rc);
441 if (buf->lb_buf == NULL)
453 if (unlikely(buf->lb_buf == NULL)) {
454 lu_buf_alloc(buf, rc);
455 if (buf->lb_buf == NULL)
461 rc1 = lfsck_layout_verify_header(obj, buf->lb_buf);
463 return rc1 ? rc1 : rc;
466 #define LFSCK_RBTREE_BITMAP_SIZE PAGE_SIZE
467 #define LFSCK_RBTREE_BITMAP_WIDTH (LFSCK_RBTREE_BITMAP_SIZE << 3)
468 #define LFSCK_RBTREE_BITMAP_MASK (LFSCK_RBTREE_BITMAP_WIDTH - 1)
470 struct lfsck_rbtree_node {
471 struct rb_node lrn_node;
474 atomic_t lrn_known_count;
475 atomic_t lrn_accessed_count;
476 void *lrn_known_bitmap;
477 void *lrn_accessed_bitmap;
480 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
481 __u64 seq, __u32 oid)
483 if (seq < lrn->lrn_seq)
486 if (seq > lrn->lrn_seq)
489 if (oid < lrn->lrn_first_oid)
492 if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
498 /* The caller should hold llsd->llsd_rb_lock. */
499 static struct lfsck_rbtree_node *
500 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
501 const struct lu_fid *fid, bool *exact)
503 struct rb_node *node = llsd->llsd_rb_root.rb_node;
504 struct rb_node *prev = NULL;
505 struct lfsck_rbtree_node *lrn = NULL;
511 while (node != NULL) {
513 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
514 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
516 node = node->rb_left;
518 node = node->rb_right;
526 /* If there is no exactly matched one, then to the next valid one. */
529 /* The rbtree is empty. */
536 node = rb_next(prev);
538 /* The end of the rbtree. */
542 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
547 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
548 const struct lu_fid *fid)
550 struct lfsck_rbtree_node *lrn;
554 return ERR_PTR(-ENOMEM);
556 OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
557 if (lrn->lrn_known_bitmap == NULL) {
560 return ERR_PTR(-ENOMEM);
563 OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
564 if (lrn->lrn_accessed_bitmap == NULL) {
565 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
568 return ERR_PTR(-ENOMEM);
571 RB_CLEAR_NODE(&lrn->lrn_node);
572 lrn->lrn_seq = fid_seq(fid);
573 lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
574 atomic_set(&lrn->lrn_known_count, 0);
575 atomic_set(&lrn->lrn_accessed_count, 0);
580 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
582 OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
583 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
587 /* The caller should hold lock. */
588 static struct lfsck_rbtree_node *
589 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
590 struct lfsck_rbtree_node *lrn)
592 struct rb_node **pos = &llsd->llsd_rb_root.rb_node;
593 struct rb_node *parent = NULL;
594 struct lfsck_rbtree_node *tmp;
597 while (*pos != NULL) {
599 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
600 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
602 pos = &(*pos)->rb_left;
604 pos = &(*pos)->rb_right;
609 rb_link_node(&lrn->lrn_node, parent, pos);
610 rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
615 extern const struct dt_index_operations lfsck_orphan_index_ops;
617 static int lfsck_rbtree_setup(const struct lu_env *env,
618 struct lfsck_component *com)
620 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
621 struct lfsck_instance *lfsck = com->lc_lfsck;
622 struct dt_device *dev = lfsck->li_bottom;
623 struct lfsck_layout_slave_data *llsd = com->lc_data;
624 struct dt_object *obj;
626 fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
627 fid->f_oid = lfsck_dev_idx(lfsck);
629 obj = dt_locate(env, dev, fid);
631 RETURN(PTR_ERR(obj));
633 /* Generate an in-RAM object to stand for the layout rbtree.
634 * Scanning the layout rbtree will be via the iteration over
635 * the object. In the future, the rbtree may be written onto
636 * disk with the object.
638 * Mark the object to be as exist. */
639 obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
640 obj->do_index_ops = &lfsck_orphan_index_ops;
641 llsd->llsd_rb_obj = obj;
642 llsd->llsd_rbtree_valid = 1;
643 dev->dd_record_fid_accessed = 1;
645 CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
646 lfsck_lfsck2name(lfsck));
651 static void lfsck_rbtree_cleanup(const struct lu_env *env,
652 struct lfsck_component *com)
654 struct lfsck_instance *lfsck = com->lc_lfsck;
655 struct lfsck_layout_slave_data *llsd = com->lc_data;
656 struct rb_node *node = rb_first(&llsd->llsd_rb_root);
657 struct rb_node *next;
658 struct lfsck_rbtree_node *lrn;
660 lfsck->li_bottom->dd_record_fid_accessed = 0;
661 /* Invalid the rbtree, then no others will use it. */
662 write_lock(&llsd->llsd_rb_lock);
663 llsd->llsd_rbtree_valid = 0;
664 write_unlock(&llsd->llsd_rb_lock);
666 while (node != NULL) {
667 next = rb_next(node);
668 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
669 rb_erase(node, &llsd->llsd_rb_root);
670 lfsck_rbtree_free(lrn);
674 if (llsd->llsd_rb_obj != NULL) {
675 lfsck_object_put(env, llsd->llsd_rb_obj);
676 llsd->llsd_rb_obj = NULL;
679 CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
680 lfsck_lfsck2name(lfsck));
683 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
684 struct lfsck_component *com,
685 const struct lu_fid *fid,
688 struct lfsck_layout_slave_data *llsd = com->lc_data;
689 struct lfsck_rbtree_node *lrn;
695 if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
698 if (!fid_is_idif(fid) && !fid_is_norm(fid))
701 read_lock(&llsd->llsd_rb_lock);
702 if (!llsd->llsd_rbtree_valid)
703 GOTO(unlock, rc = 0);
705 lrn = lfsck_rbtree_search(llsd, fid, NULL);
707 struct lfsck_rbtree_node *tmp;
711 read_unlock(&llsd->llsd_rb_lock);
712 tmp = lfsck_rbtree_new(env, fid);
714 GOTO(out, rc = PTR_ERR(tmp));
717 write_lock(&llsd->llsd_rb_lock);
718 if (!llsd->llsd_rbtree_valid) {
719 lfsck_rbtree_free(tmp);
720 GOTO(unlock, rc = 0);
723 lrn = lfsck_rbtree_insert(llsd, tmp);
725 lfsck_rbtree_free(tmp);
728 idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
729 /* Any accessed object must be a known object. */
730 if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
731 atomic_inc(&lrn->lrn_known_count);
732 if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
733 atomic_inc(&lrn->lrn_accessed_count);
735 GOTO(unlock, rc = 0);
739 write_unlock(&llsd->llsd_rb_lock);
741 read_unlock(&llsd->llsd_rb_lock);
743 if (rc != 0 && accessed) {
744 struct lfsck_layout *lo = com->lc_file_ram;
746 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
747 "bitmap, and will cause incorrect LFSCK OST-object "
748 "handling, so disable it to cancel orphan handling "
749 "for related device. rc = %d\n",
750 lfsck_lfsck2name(com->lc_lfsck), rc);
752 lo->ll_flags |= LF_INCOMPLETE;
753 lfsck_rbtree_cleanup(env, com);
757 static inline void lldk_le_to_cpu(struct lfsck_layout_dangling_key *des,
758 const struct lfsck_layout_dangling_key *src)
760 fid_le_to_cpu(&des->lldk_fid, &src->lldk_fid);
761 des->lldk_comp_id = le32_to_cpu(src->lldk_comp_id);
762 des->lldk_ea_off = le32_to_cpu(src->lldk_ea_off);
765 static inline void lldk_cpu_to_le(struct lfsck_layout_dangling_key *des,
766 const struct lfsck_layout_dangling_key *src)
768 fid_cpu_to_le(&des->lldk_fid, &src->lldk_fid);
769 des->lldk_comp_id = cpu_to_le32(src->lldk_comp_id);
770 des->lldk_ea_off = cpu_to_le32(src->lldk_ea_off);
773 static inline void lldk_be_to_cpu(struct lfsck_layout_dangling_key *des,
774 const struct lfsck_layout_dangling_key *src)
776 fid_be_to_cpu(&des->lldk_fid, &src->lldk_fid);
777 des->lldk_comp_id = be32_to_cpu(src->lldk_comp_id);
778 des->lldk_ea_off = be32_to_cpu(src->lldk_ea_off);
781 static inline void lldk_cpu_to_be(struct lfsck_layout_dangling_key *des,
782 const struct lfsck_layout_dangling_key *src)
784 fid_cpu_to_be(&des->lldk_fid, &src->lldk_fid);
785 des->lldk_comp_id = cpu_to_be32(src->lldk_comp_id);
786 des->lldk_ea_off = cpu_to_be32(src->lldk_ea_off);
789 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
790 const struct lfsck_layout *src)
794 des->ll_magic = le32_to_cpu(src->ll_magic);
795 des->ll_status = le32_to_cpu(src->ll_status);
796 des->ll_flags = le32_to_cpu(src->ll_flags);
797 des->ll_success_count = le32_to_cpu(src->ll_success_count);
798 des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
799 des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
800 des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
801 des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
802 des->ll_time_last_checkpoint =
803 le64_to_cpu(src->ll_time_last_checkpoint);
804 des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
805 des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
806 des->ll_pos_first_inconsistent =
807 le64_to_cpu(src->ll_pos_first_inconsistent);
808 des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
809 des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
810 des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
811 des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
812 for (i = 0; i < LLIT_MAX; i++)
813 des->ll_objs_repaired[i] =
814 le64_to_cpu(src->ll_objs_repaired[i]);
815 des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
816 des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
817 lldk_le_to_cpu(&des->ll_lldk_latest_scanned_phase2,
818 &src->ll_lldk_latest_scanned_phase2);
821 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
822 const struct lfsck_layout *src)
826 des->ll_magic = cpu_to_le32(src->ll_magic);
827 des->ll_status = cpu_to_le32(src->ll_status);
828 des->ll_flags = cpu_to_le32(src->ll_flags);
829 des->ll_success_count = cpu_to_le32(src->ll_success_count);
830 des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
831 des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
832 des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
833 des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
834 des->ll_time_last_checkpoint =
835 cpu_to_le64(src->ll_time_last_checkpoint);
836 des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
837 des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
838 des->ll_pos_first_inconsistent =
839 cpu_to_le64(src->ll_pos_first_inconsistent);
840 des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
841 des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
842 des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
843 des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
844 for (i = 0; i < LLIT_MAX; i++)
845 des->ll_objs_repaired[i] =
846 cpu_to_le64(src->ll_objs_repaired[i]);
847 des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
848 des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
849 lldk_cpu_to_le(&des->ll_lldk_latest_scanned_phase2,
850 &src->ll_lldk_latest_scanned_phase2);
854 * Load the OST bitmap from the lfsck_layout trace file.
856 * \param[in] env pointer to the thread context
857 * \param[in] com pointer to the lfsck component
859 * \retval 0 for success
860 * \retval negative error number on failure or data corruption
862 static int lfsck_layout_load_bitmap(const struct lu_env *env,
863 struct lfsck_component *com)
865 struct dt_object *obj = com->lc_obj;
866 struct lfsck_assistant_data *lad = com->lc_data;
867 struct lfsck_layout *lo = com->lc_file_ram;
868 struct cfs_bitmap *bitmap = lad->lad_bitmap;
869 loff_t pos = com->lc_file_size;
875 if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
877 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
879 nbits = lo->ll_bitmap_size;
881 if (unlikely(nbits < BITS_PER_LONG))
882 nbits = BITS_PER_LONG;
884 if (nbits > bitmap->size) {
885 __u32 new_bits = bitmap->size;
886 struct cfs_bitmap *new_bitmap;
888 while (new_bits < nbits)
891 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
892 if (new_bitmap == NULL)
895 lad->lad_bitmap = new_bitmap;
896 CFS_FREE_BITMAP(bitmap);
900 if (lo->ll_bitmap_size == 0) {
901 lad->lad_incomplete = 0;
902 CFS_RESET_BITMAP(bitmap);
907 size = (lo->ll_bitmap_size + 7) >> 3;
908 rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
910 RETURN(rc >= 0 ? -EINVAL : rc);
912 if (cfs_bitmap_check_empty(bitmap))
913 lad->lad_incomplete = 0;
915 lad->lad_incomplete = 1;
921 * Load the layout LFSCK trace file from disk.
923 * The layout LFSCK trace file records the layout LFSCK status information
924 * and other statistics, such as how many objects have been scanned, and how
925 * many objects have been repaired, and etc. It also contains the bitmap for
926 * failed OSTs during the layout LFSCK. All these information will be loaded
927 * from disk to RAM when the layout LFSCK component setup.
929 * \param[in] env pointer to the thread context
930 * \param[in] com pointer to the lfsck component
932 * \retval positive number for file data corruption, the caller
933 * should reset the layout LFSCK trace file
934 * \retval 0 for success
935 * \retval negative error number on failure
937 static int lfsck_layout_load(const struct lu_env *env,
938 struct lfsck_component *com)
940 struct lfsck_layout *lo = com->lc_file_ram;
941 ssize_t size = com->lc_file_size;
945 rc = dt_read(env, com->lc_obj,
946 lfsck_buf_get(env, com->lc_file_disk, size), &pos);
950 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
951 lfsck_lfsck2name(com->lc_lfsck), rc);
953 } else if (rc != size) {
954 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
955 lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
959 lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
960 if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
961 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
962 "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
963 lo->ll_magic, LFSCK_LAYOUT_MAGIC);
971 * Store the layout LFSCK trace file on disk.
973 * The layout LFSCK trace file records the layout LFSCK status information
974 * and other statistics, such as how many objects have been scanned, and how
975 * many objects have been repaired, and etc. It also contains the bitmap for
976 * failed OSTs during the layout LFSCK. All these information will be synced
977 * from RAM to disk periodically.
979 * \param[in] env pointer to the thread context
980 * \param[in] com pointer to the lfsck component
982 * \retval 0 for success
983 * \retval negative error number on failure
985 static int lfsck_layout_store(const struct lu_env *env,
986 struct lfsck_component *com)
988 struct dt_object *obj = com->lc_obj;
989 struct lfsck_instance *lfsck = com->lc_lfsck;
990 struct lfsck_layout *lo_ram = com->lc_file_ram;
991 struct lfsck_layout *lo = com->lc_file_disk;
993 struct dt_device *dev = lfsck_obj2dev(obj);
994 struct cfs_bitmap *bitmap = NULL;
996 ssize_t size = com->lc_file_size;
1001 if (lfsck->li_master) {
1002 struct lfsck_assistant_data *lad = com->lc_data;
1004 bitmap = lad->lad_bitmap;
1005 nbits = bitmap->size;
1008 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
1011 lo_ram->ll_bitmap_size = nbits;
1012 lfsck_layout_cpu_to_le(lo, lo_ram);
1013 th = dt_trans_create(env, dev);
1015 GOTO(log, rc = PTR_ERR(th));
1017 rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
1022 if (bitmap != NULL) {
1023 rc = dt_declare_record_write(env, obj,
1024 lfsck_buf_get(env, bitmap->data, nbits >> 3),
1030 rc = dt_trans_start_local(env, dev, th);
1035 rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
1039 if (bitmap != NULL) {
1041 rc = dt_record_write(env, obj,
1042 lfsck_buf_get(env, bitmap->data, nbits >> 3),
1049 dt_trans_stop(env, dev, th);
1053 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
1054 lfsck_lfsck2name(lfsck), rc);
1059 static int lfsck_layout_init(const struct lu_env *env,
1060 struct lfsck_component *com)
1062 struct lfsck_layout *lo = com->lc_file_ram;
1065 memset(lo, 0, com->lc_file_size);
1066 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
1067 lo->ll_status = LS_INIT;
1068 down_write(&com->lc_sem);
1069 rc = lfsck_layout_store(env, com);
1070 if (rc == 0 && com->lc_lfsck->li_master)
1071 rc = lfsck_load_sub_trace_files(env, com,
1072 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
1073 up_write(&com->lc_sem);
1078 static int fid_is_for_ostobj(const struct lu_env *env,
1079 struct lfsck_instance *lfsck,
1080 struct dt_object *obj, const struct lu_fid *fid)
1082 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1083 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
1084 struct lustre_ost_attrs *loa;
1087 fld_range_set_any(range);
1088 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1090 if (fld_range_is_ost(range))
1096 loa = &lfsck_env_info(env)->lti_loa;
1097 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, loa, sizeof(*loa)),
1099 if (rc >= sizeof(struct lustre_mdt_attrs)) {
1100 lustre_lma_swab(&loa->loa_lma);
1102 return loa->loa_lma.lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1105 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1110 static struct lfsck_layout_seq *
1111 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1113 struct lfsck_layout_seq *lls;
1115 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1116 if (lls->lls_seq == seq)
1119 if (lls->lls_seq > seq)
1127 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1128 struct lfsck_layout_seq *lls)
1130 struct lfsck_layout_seq *tmp;
1131 struct list_head *pos = &llsd->llsd_seq_list;
1133 list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1134 if (lls->lls_seq < tmp->lls_seq) {
1135 pos = &tmp->lls_list;
1139 list_add_tail(&lls->lls_list, pos);
1143 lfsck_layout_lastid_create(const struct lu_env *env,
1144 struct lfsck_instance *lfsck,
1145 struct dt_object *obj)
1147 struct lfsck_thread_info *info = lfsck_env_info(env);
1148 struct lu_attr *la = &info->lti_la;
1149 struct dt_object_format *dof = &info->lti_dof;
1150 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1151 struct dt_device *dt = lfsck_obj2dev(obj);
1158 if (bk->lb_param & LPF_DRYRUN)
1161 memset(la, 0, sizeof(*la));
1162 la->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1163 la->la_valid = LA_MODE | LA_UID | LA_GID;
1164 memset(dof, 0, sizeof(*dof));
1165 dof->dof_type = dt_mode_to_dft(S_IFREG);
1167 th = dt_trans_create(env, dt);
1169 GOTO(log, rc = PTR_ERR(th));
1171 rc = dt_declare_create(env, obj, la, NULL, dof, th);
1175 rc = dt_declare_record_write(env, obj,
1176 lfsck_buf_get(env, &lastid,
1182 rc = dt_trans_start_local(env, dt, th);
1186 dt_write_lock(env, obj, 0);
1187 if (likely(dt_object_exists(obj) == 0)) {
1188 rc = dt_create(env, obj, la, NULL, dof, th);
1190 rc = dt_record_write(env, obj,
1191 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1194 dt_write_unlock(env, obj);
1199 dt_trans_stop(env, dt, th);
1202 CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1204 lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1210 lfsck_layout_lastid_reload(const struct lu_env *env,
1211 struct lfsck_component *com,
1212 struct lfsck_layout_seq *lls)
1218 dt_read_lock(env, lls->lls_lastid_obj, 0);
1219 rc = dt_record_read(env, lls->lls_lastid_obj,
1220 lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1221 dt_read_unlock(env, lls->lls_lastid_obj);
1222 if (unlikely(rc != 0))
1225 lastid = le64_to_cpu(lastid);
1226 if (lastid < lls->lls_lastid_known) {
1227 struct lfsck_instance *lfsck = com->lc_lfsck;
1228 struct lfsck_layout *lo = com->lc_file_ram;
1230 lls->lls_lastid = lls->lls_lastid_known;
1232 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1233 LASSERT(lfsck->li_out_notify != NULL);
1235 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1236 LE_LASTID_REBUILDING);
1237 lo->ll_flags |= LF_CRASHED_LASTID;
1239 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1240 "LAST_ID file (1) for the sequence %#llx"
1241 ", old value %llu, known value %llu\n",
1242 lfsck_lfsck2name(lfsck), lls->lls_seq,
1243 lastid, lls->lls_lastid);
1245 } else if (lastid >= lls->lls_lastid) {
1246 lls->lls_lastid = lastid;
1254 lfsck_layout_lastid_store(const struct lu_env *env,
1255 struct lfsck_component *com)
1257 struct lfsck_instance *lfsck = com->lc_lfsck;
1258 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1259 struct dt_device *dt = lfsck->li_bottom;
1260 struct lfsck_layout_slave_data *llsd = com->lc_data;
1261 struct lfsck_layout_seq *lls;
1267 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1270 if (!lls->lls_dirty)
1273 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1274 "<seq> %#llx as <oid> %llu\n",
1275 lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1277 if (bk->lb_param & LPF_DRYRUN) {
1282 th = dt_trans_create(env, dt);
1285 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1286 "the LAST_ID for <seq> %#llx(1): rc = %d\n",
1287 lfsck_lfsck2name(com->lc_lfsck),
1292 lastid = cpu_to_le64(lls->lls_lastid);
1293 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1294 lfsck_buf_get(env, &lastid,
1300 rc = dt_trans_start_local(env, dt, th);
1304 dt_write_lock(env, lls->lls_lastid_obj, 0);
1305 rc = dt_record_write(env, lls->lls_lastid_obj,
1306 lfsck_buf_get(env, &lastid,
1307 sizeof(lastid)), &pos, th);
1308 dt_write_unlock(env, lls->lls_lastid_obj);
1313 dt_trans_stop(env, dt, th);
1316 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1317 "the LAST_ID for <seq> %#llx(2): rc = %d\n",
1318 lfsck_lfsck2name(com->lc_lfsck),
1327 lfsck_layout_lastid_load(const struct lu_env *env,
1328 struct lfsck_component *com,
1329 struct lfsck_layout_seq *lls)
1331 struct lfsck_instance *lfsck = com->lc_lfsck;
1332 struct lfsck_layout *lo = com->lc_file_ram;
1333 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
1334 struct dt_object *obj;
1339 lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1340 obj = dt_locate(env, lfsck->li_bottom, fid);
1342 RETURN(PTR_ERR(obj));
1344 /* LAST_ID crashed, to be rebuilt */
1345 if (dt_object_exists(obj) == 0) {
1346 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1347 LASSERT(lfsck->li_out_notify != NULL);
1349 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1350 LE_LASTID_REBUILDING);
1351 lo->ll_flags |= LF_CRASHED_LASTID;
1353 CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1354 "LAST_ID file for sequence %#llx\n",
1355 lfsck_lfsck2name(lfsck), lls->lls_seq);
1357 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1359 struct l_wait_info lwi = LWI_TIMEOUT(
1360 cfs_time_seconds(cfs_fail_val),
1363 /* Some others may changed the cfs_fail_val
1364 * as zero after above check, re-check it for
1365 * sure to avoid falling into wait for ever. */
1366 if (likely(lwi.lwi_timeout > 0)) {
1367 struct ptlrpc_thread *thread =
1370 up_write(&com->lc_sem);
1371 l_wait_event(thread->t_ctl_waitq,
1372 !thread_is_running(thread),
1374 down_write(&com->lc_sem);
1379 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1381 dt_read_lock(env, obj, 0);
1382 rc = dt_read(env, obj,
1383 lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1385 dt_read_unlock(env, obj);
1386 if (rc != 0 && rc != sizeof(__u64))
1387 GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1389 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1390 LASSERT(lfsck->li_out_notify != NULL);
1392 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1393 LE_LASTID_REBUILDING);
1394 lo->ll_flags |= LF_CRASHED_LASTID;
1396 CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1397 "LAST_ID file for the sequence %#llx"
1399 lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1402 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1410 lfsck_object_put(env, obj);
1412 lls->lls_lastid_obj = obj;
1417 static void lfsck_layout_record_failure(const struct lu_env *env,
1418 struct lfsck_instance *lfsck,
1419 struct lfsck_layout *lo)
1423 lo->ll_objs_failed_phase1++;
1424 cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1426 if (lo->ll_pos_first_inconsistent == 0 ||
1427 lo->ll_pos_first_inconsistent < cookie) {
1428 lo->ll_pos_first_inconsistent = cookie;
1430 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1431 "inconsistency at the pos [%llu]\n",
1432 lfsck_lfsck2name(lfsck),
1433 lo->ll_pos_first_inconsistent);
1437 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1438 struct lfsck_component *com,
1441 struct lfsck_instance *lfsck = com->lc_lfsck;
1442 struct lfsck_layout *lo = com->lc_file_ram;
1444 down_write(&com->lc_sem);
1445 lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1446 HALF_SEC - com->lc_time_last_checkpoint);
1447 lo->ll_time_last_checkpoint = cfs_time_current_sec();
1448 lo->ll_objs_checked_phase2 += com->lc_new_checked;
1451 if (lo->ll_flags & LF_INCOMPLETE) {
1452 lo->ll_status = LS_PARTIAL;
1454 if (lfsck->li_master) {
1455 struct lfsck_assistant_data *lad = com->lc_data;
1457 if (lad->lad_incomplete)
1458 lo->ll_status = LS_PARTIAL;
1460 lo->ll_status = LS_COMPLETED;
1462 lo->ll_status = LS_COMPLETED;
1465 lo->ll_flags &= ~LF_SCANNED_ONCE;
1466 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1467 lo->ll_flags &= ~LF_INCONSISTENT;
1468 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1469 lo->ll_success_count++;
1470 } else if (rc == 0) {
1471 if (lfsck->li_status != 0)
1472 lo->ll_status = lfsck->li_status;
1474 lo->ll_status = LS_STOPPED;
1476 lo->ll_status = LS_FAILED;
1479 rc = lfsck_layout_store(env, com);
1480 up_write(&com->lc_sem);
1485 static int lfsck_layout_trans_stop(const struct lu_env *env,
1486 struct dt_device *dev,
1487 struct thandle *handle, int result)
1491 /* XXX: If there is something worng or it needs to repair nothing,
1492 * then notify the lower to stop the modification. Currently,
1493 * we use th_result for such purpose, that may be replaced by
1494 * some rollback mechanism in the future. */
1495 handle->th_result = result;
1496 rc = dt_trans_stop(env, dev, handle);
1498 return result > 0 ? 0 : result;
1500 return rc == 0 ? 1 : rc;
1503 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1504 struct lfsck_component *com,
1505 const struct lu_fid *pfid,
1506 const struct lu_fid *cfid,
1507 __u32 comp_id, __u32 ea_off,
1510 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1511 struct lu_fid *rec = &lfsck_env_info(env)->lti_fid3;
1512 struct dt_device *dev;
1513 struct dt_object *obj;
1514 struct thandle *th = NULL;
1519 idx = lfsck_sub_trace_file_fid2idx(pfid);
1520 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1521 dev = lfsck_obj2dev(obj);
1523 fid_cpu_to_be(&key->lldk_fid, pfid);
1524 key->lldk_comp_id = cpu_to_be32(comp_id);
1525 key->lldk_ea_off = cpu_to_be32(ea_off);
1527 fid_cpu_to_be(rec, cfid);
1528 rec->f_ver = cpu_to_be32(ost_idx);
1530 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1532 th = dt_trans_create(env, dev);
1534 GOTO(unlock, rc = PTR_ERR(th));
1536 rc = dt_declare_insert(env, obj,
1537 (const struct dt_rec *)rec,
1538 (const struct dt_key *)key, th);
1542 rc = dt_trans_start_local(env, dev, th);
1546 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1547 (const struct dt_key *)key, th, 1);
1552 if (th && !IS_ERR(th))
1553 dt_trans_stop(env, dev, th);
1555 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1557 CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", comp_id = %u, "
1558 "ea_off = %u, ost_idx = %u, into the trace file for further "
1559 "dangling check: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1560 PFID(pfid), PFID(cfid), comp_id, ea_off, ost_idx, rc);
1565 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1566 struct lfsck_component *com,
1567 const struct lu_fid *fid,
1568 __u32 comp_id, __u32 ea_off)
1570 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1571 struct dt_device *dev;
1572 struct dt_object *obj;
1573 struct thandle *th = NULL;
1578 idx = lfsck_sub_trace_file_fid2idx(fid);
1579 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1580 dev = lfsck_obj2dev(obj);
1582 fid_cpu_to_be(&key->lldk_fid, fid);
1583 key->lldk_comp_id = cpu_to_be32(comp_id);
1584 key->lldk_ea_off = cpu_to_be32(ea_off);
1586 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1588 th = dt_trans_create(env, dev);
1590 GOTO(unlock, rc = PTR_ERR(th));
1592 rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1596 rc = dt_trans_start_local(env, dev, th);
1600 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1605 if (th && !IS_ERR(th))
1606 dt_trans_stop(env, dev, th);
1608 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1610 CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1611 ", comp_id = %u, ea_off = %u from the trace file: rc = %d\n",
1612 lfsck_lfsck2name(com->lc_lfsck), PFID(fid), comp_id, ea_off, rc);
1618 * Get the system default stripe size.
1620 * \param[in] env pointer to the thread context
1621 * \param[in] lfsck pointer to the lfsck instance
1622 * \param[out] size pointer to the default stripe size
1624 * \retval 0 for success
1625 * \retval negative error number on failure
1627 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1628 struct lfsck_instance *lfsck,
1631 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1632 struct dt_object *root;
1635 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1637 return PTR_ERR(root);
1639 /* Get the default stripe size via xattr_get on the backend root. */
1640 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1643 /* The lum->lmm_stripe_size is LE mode. The *size also
1644 * should be LE mode. So it is unnecessary to convert. */
1645 *size = lum->lmm_stripe_size;
1647 } else if (unlikely(rc == 0)) {
1651 lfsck_object_put(env, root);
1657 * \retval +1: repaired
1658 * \retval 0: did nothing
1659 * \retval -ve: on error
1661 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1662 struct lfsck_instance *lfsck,
1663 struct thandle *handle,
1664 struct dt_object *parent,
1665 const struct lu_fid *cfid,
1667 struct lov_mds_md_v1 *lmm,
1668 struct lov_ost_data_v1 *slot,
1669 int fl, __u32 ost_idx, int size)
1671 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1672 struct lu_buf ea_buf;
1679 magic = le32_to_cpu(lmm->lmm_magic);
1680 pattern = le32_to_cpu(lmm->lmm_pattern);
1681 count = le16_to_cpu(lmm->lmm_stripe_count);
1683 fid_to_ostid(cfid, oi);
1684 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1685 slot->l_ost_gen = cpu_to_le32(0);
1686 slot->l_ost_idx = cpu_to_le32(ost_idx);
1688 if (pattern & LOV_PATTERN_F_HOLE) {
1689 struct lov_ost_data_v1 *objs;
1692 if (magic == LOV_MAGIC_V1)
1693 objs = &lmm->lmm_objects[0];
1695 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1696 for (i = 0; i < count; i++, objs++) {
1697 if (lovea_slot_is_dummy(objs))
1701 /* If the @slot is the last dummy slot to be refilled,
1702 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1705 cpu_to_le32(pattern & ~LOV_PATTERN_F_HOLE);
1707 CDEBUG(D_LFSCK, "%s: remove layout HOLE for "DFID
1708 ": parent "DFID"\n", lfsck_lfsck2name(lfsck),
1709 PFID(cfid), PFID(lfsck_dto2fid(parent)));
1713 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
1714 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1721 static struct lov_ost_data_v1 *
1722 __lfsck_layout_new_v1_lovea(struct lov_mds_md_v1 *lmm,
1723 const struct lu_fid *pfid,
1724 __u32 stripe_size, __u32 ea_off,
1725 __u32 pattern, __u16 count)
1727 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1728 lmm->lmm_pattern = cpu_to_le32(pattern);
1729 fid_to_lmm_oi(pfid, &lmm->lmm_oi);
1730 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1731 lmm->lmm_stripe_size = cpu_to_le32(stripe_size);
1732 lmm->lmm_stripe_count = cpu_to_le16(count);
1733 lmm->lmm_layout_gen = cpu_to_le16(1);
1734 memset(&lmm->lmm_objects[0], 0,
1735 sizeof(struct lov_ost_data_v1) * count);
1737 return &lmm->lmm_objects[ea_off];
1740 static int lfsck_layout_new_v1_lovea(const struct lu_env *env,
1741 struct lfsck_instance *lfsck,
1742 struct ost_layout *ol,
1743 struct dt_object *parent,
1744 struct lu_buf *buf, __u32 ea_off,
1745 struct lov_mds_md_v1 **lmm,
1746 struct lov_ost_data_v1 **objs)
1749 __u32 stripe_size = ol->ol_stripe_size;
1750 __u32 pattern = LOV_PATTERN_RAID0;
1753 if (ol->ol_stripe_count != 0)
1754 count = ol->ol_stripe_count;
1758 size = lov_mds_md_size(count, LOV_MAGIC_V1);
1759 LASSERTF(buf->lb_len >= size,
1760 "buffer len %d is less than real size %d\n",
1761 (int)buf->lb_len, size);
1763 if (stripe_size == 0) {
1766 rc = lfsck_layout_get_def_stripesize(env, lfsck, &stripe_size);
1772 if (ol->ol_stripe_count > 1 ||
1773 (ol->ol_stripe_count == 0 && ea_off != 0)) {
1774 pattern |= LOV_PATTERN_F_HOLE;
1775 memset(&(*lmm)->lmm_objects[0], 0,
1776 count * sizeof(struct lov_ost_data_v1));
1779 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1780 stripe_size, ea_off, pattern, count);
1785 static int lfsck_layout_new_comp_lovea(const struct lu_env *env,
1786 struct ost_layout *ol,
1787 struct dt_object *parent,
1788 struct lu_buf *buf, __u32 ea_off,
1789 struct lov_mds_md_v1 **lmm,
1790 struct lov_ost_data_v1 **objs)
1792 struct lov_comp_md_v1 *lcm;
1793 struct lov_comp_md_entry_v1 *lcme;
1794 __u32 pattern = LOV_PATTERN_RAID0;
1795 __u32 offset = sizeof(*lcm) + sizeof(*lcme);
1796 int lcme_size = lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1797 int size = offset + lcme_size;
1799 LASSERTF(buf->lb_len >= size,
1800 "buffer len %d is less than real size %d\n",
1801 (int)buf->lb_len, size);
1804 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1805 lcm->lcm_size = cpu_to_le32(size);
1806 lcm->lcm_layout_gen = cpu_to_le32(1);
1808 lcm->lcm_entry_count = cpu_to_le16(1);
1810 lcme = &lcm->lcm_entries[0];
1811 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1812 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1813 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1814 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1815 lcme->lcme_offset = cpu_to_le32(offset);
1816 lcme->lcme_size = cpu_to_le32(lcme_size);
1817 if (ol->ol_stripe_count > 1)
1818 pattern |= LOV_PATTERN_F_HOLE;
1820 *lmm = buf->lb_buf + offset;
1821 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1822 ol->ol_stripe_size, ea_off,
1823 pattern, ol->ol_stripe_count);
1828 static int lfsck_layout_add_comp_comp(const struct lu_env *env,
1829 struct lfsck_instance *lfsck,
1830 struct thandle *handle,
1831 struct ost_layout *ol,
1832 struct dt_object *parent,
1833 const struct lu_fid *cfid,
1834 struct lu_buf *buf, __u32 ost_idx,
1835 __u32 ea_off, int pos)
1837 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1838 struct lov_comp_md_entry_v1 *lcme;
1839 struct lov_mds_md_v1 *lmm;
1840 struct lov_ost_data_v1 *objs;
1841 int added = sizeof(*lcme) +
1842 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1843 int size = le32_to_cpu(lcm->lcm_size) + added;
1847 __u32 pattern = LOV_PATTERN_RAID0;
1848 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1851 lu_buf_check_and_grow(buf, size);
1852 /* set the lcm again because lu_buf_check_and_grow() may
1853 * have reallocated the buf. */
1855 lcm->lcm_size = cpu_to_le32(size);
1856 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1857 lcm->lcm_entry_count = cpu_to_le16(count + 1);
1859 /* 1. Move the component bodies from [pos, count-1] to [pos+1, count]
1860 * with distance of 'added'. */
1863 for (i = pos; i < count; i++) {
1864 lcme = &lcm->lcm_entries[i];
1865 size += le32_to_cpu(lcme->lcme_size);
1868 offset = le32_to_cpu(lcm->lcm_entries[pos].lcme_offset);
1869 memmove(buf->lb_buf + offset + added,
1870 buf->lb_buf + offset, size);
1874 /* 2. Move the component header [0, pos-1] to [0, pos-1] with distance
1875 * of 'sizeof(struct lov_comp_md_entry_v1)' */
1877 for (i = 0; i < pos; i++) {
1878 lcme = &lcm->lcm_entries[i];
1879 size += le32_to_cpu(lcme->lcme_size);
1882 offset = le32_to_cpu(lcm->lcm_entries[0].lcme_offset);
1883 memmove(buf->lb_buf + offset + sizeof(*lcme),
1884 buf->lb_buf + offset, size);
1887 /* 3. Recalculate the enter offset for the component [pos, count-1] */
1888 for (i = count - 1; i >= pos; i--) {
1889 lcm->lcm_entries[i + 1] = lcm->lcm_entries[i];
1890 lcm->lcm_entries[i + 1].lcme_offset =
1891 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i + 1].
1892 lcme_offset) + added);
1895 /* 4. Recalculate the enter offset for the component [0, pos) */
1896 for (i = 0; i < pos; i++) {
1897 lcm->lcm_entries[i].lcme_offset =
1898 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i].
1899 lcme_offset) + sizeof(*lcme));
1902 offset = sizeof(*lcm) + sizeof(*lcme) * (count + 1) + size;
1903 /* 4. Insert the new component header (entry) at the slot 'pos'. */
1904 lcme = &lcm->lcm_entries[pos];
1905 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1906 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1907 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1908 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1909 lcme->lcme_offset = cpu_to_le32(offset);
1910 lcme->lcme_size = cpu_to_le32(lov_mds_md_size(ol->ol_stripe_count,
1913 if (ol->ol_stripe_count > 1)
1914 pattern |= LOV_PATTERN_F_HOLE;
1916 lmm = buf->lb_buf + offset;
1917 /* 5. Insert teh new component body at the 'offset'. */
1918 objs = __lfsck_layout_new_v1_lovea(lmm, lfsck_dto2fid(parent),
1919 ol->ol_stripe_size, ea_off,
1920 pattern, ol->ol_stripe_count);
1922 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1923 lmm, objs, LU_XATTR_REPLACE, ost_idx,
1924 le32_to_cpu(lcm->lcm_size));
1926 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant add new COMP for "
1927 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1928 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
1929 "comp_end %llu, %s LOV EA hole: rc = %d\n",
1930 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1931 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
1932 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
1933 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
1934 "with" : "without", rc);
1939 static int lfsck_layout_extend_v1v3_lovea(const struct lu_env *env,
1940 struct lfsck_instance *lfsck,
1941 struct thandle *handle,
1942 struct ost_layout *ol,
1943 struct dt_object *parent,
1944 const struct lu_fid *cfid,
1945 struct lu_buf *buf, __u32 ost_idx,
1948 struct lov_mds_md_v1 *lmm = buf->lb_buf;
1949 struct lov_ost_data_v1 *objs;
1950 __u16 count = le16_to_cpu(lmm->lmm_stripe_count);
1951 __u32 magic = le32_to_cpu(lmm->lmm_magic);
1957 /* The original LOVEA maybe re-generated via old filter_fid, at
1958 * that time, we do not know the stripe count and stripe size. */
1959 if (ol->ol_stripe_count > count)
1960 count = ol->ol_stripe_count;
1961 if (ol->ol_stripe_size != 0 &&
1962 ol->ol_stripe_size != le32_to_cpu(lmm->lmm_stripe_size))
1963 lmm->lmm_stripe_size = cpu_to_le32(ol->ol_stripe_size);
1965 if (magic == LOV_MAGIC_V1)
1966 objs = &lmm->lmm_objects[count];
1968 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[count];
1970 gap = ea_off - count;
1974 size = lov_mds_md_size(count, magic);
1975 LASSERTF(buf->lb_len >= size,
1976 "buffer len %d is less than real size %d\n",
1977 (int)buf->lb_len, size);
1980 memset(objs, 0, gap * sizeof(*objs));
1981 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1984 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1985 lmm->lmm_stripe_count = cpu_to_le16(count);
1988 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1989 lmm, objs, LU_XATTR_REPLACE, ost_idx, size);
1991 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1992 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1993 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
1994 "comp_end %llu, %s LOV EA hole: rc = %d\n",
1995 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1996 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
1997 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
1998 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
1999 "with" : "without", rc);
2005 * \retval +1: repaired
2006 * \retval 0: did nothing
2007 * \retval -ve: on error
2009 static int lfsck_layout_update_lovea(const struct lu_env *env,
2010 struct lfsck_instance *lfsck,
2011 struct thandle *handle,
2012 struct ost_layout *ol,
2013 struct dt_object *parent,
2014 const struct lu_fid *cfid,
2015 struct lu_buf *buf, int fl,
2016 __u32 ost_idx, __u32 ea_off)
2018 struct lov_mds_md_v1 *lmm = NULL;
2019 struct lov_ost_data_v1 *objs = NULL;
2023 if (ol->ol_comp_id != 0)
2024 rc = lfsck_layout_new_comp_lovea(env, ol, parent, buf, ea_off,
2027 rc = lfsck_layout_new_v1_lovea(env, lfsck, ol, parent, buf,
2028 ea_off, &lmm, &objs);
2031 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid,
2032 buf, lmm, objs, fl, ost_idx, rc);
2034 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant created layout EA for "
2035 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2036 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2037 "comp_end %llu, fl %d, %s LOV EA hole: rc = %d\n",
2038 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2039 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2040 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end, fl,
2041 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2042 "with" : "without", rc);
2047 static int __lfsck_layout_update_pfid(const struct lu_env *env,
2048 struct dt_object *child,
2049 const struct lu_fid *pfid,
2050 const struct ost_layout *ol, __u32 offset)
2052 struct dt_device *dev = lfsck_obj2dev(child);
2053 struct filter_fid *ff = &lfsck_env_info(env)->lti_ff;
2054 struct thandle *handle;
2055 struct lu_buf buf = { NULL };
2058 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2059 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2060 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2061 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2062 * parent MDT-object's layout EA. */
2063 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
2064 ost_layout_cpu_to_le(&ff->ff_layout, ol);
2065 lfsck_buf_init(&buf, ff, sizeof(*ff));
2067 handle = dt_trans_create(env, dev);
2069 RETURN(PTR_ERR(handle));
2071 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2075 rc = dt_trans_start_local(env, dev, handle);
2079 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2084 dt_trans_stop(env, dev, handle);
2090 * \retval +1: repaired
2091 * \retval 0: did nothing
2092 * \retval -ve: on error
2094 static int lfsck_layout_update_pfid(const struct lu_env *env,
2095 struct lfsck_component *com,
2096 struct dt_object *parent,
2097 struct lu_fid *cfid,
2098 struct dt_device *cdev,
2099 struct ost_layout *ol, __u32 ea_off)
2101 struct dt_object *child;
2105 child = lfsck_object_find_by_dev(env, cdev, cfid);
2107 RETURN(PTR_ERR(child));
2109 rc = __lfsck_layout_update_pfid(env, child,
2110 lu_object_fid(&parent->do_lu),
2112 lfsck_object_put(env, child);
2114 RETURN(rc == 0 ? 1 : rc);
2117 static int lfsck_lovea_size(struct ost_layout *ol, __u32 ea_off)
2119 if (ol->ol_comp_id != 0)
2120 return sizeof(struct lov_comp_md_v1) +
2121 sizeof(struct lov_comp_md_entry_v1) +
2122 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2124 if (ol->ol_stripe_count != 0)
2125 return lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2127 return lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2131 * This function will create the MDT-object with the given (partial) LOV EA.
2133 * Under some data corruption cases, the MDT-object of the file may be lost,
2134 * but its OST-objects, or some of them are there. The layout LFSCK needs to
2135 * re-create the MDT-object with the orphan OST-object(s) information.
2137 * On the other hand, the LFSCK may has created some OST-object for repairing
2138 * dangling LOV EA reference, but as the LFSCK processing, it may find that
2139 * the old OST-object is there and should replace the former new created OST
2140 * object. Unfortunately, some others have modified such newly created object.
2141 * To keep the data (both new and old), the LFSCK will create MDT-object with
2142 * new FID to reference the original OST-object.
2144 * \param[in] env pointer to the thread context
2145 * \param[in] com pointer to the lfsck component
2146 * \param[in] ltd pointer to target device descriptor
2147 * \param[in] rec pointer to the record for the orphan OST-object
2148 * \param[in] cfid pointer to FID for the orphan OST-object
2149 * \param[in] infix additional information, such as the FID for original
2150 * MDT-object and the stripe offset in the LOV EA
2151 * \param[in] type the type for describing why the orphan MDT-object is
2152 * created. The rules are as following:
2154 * type "C": Multiple OST-objects claim the same MDT-object and the
2155 * same slot in the layout EA. Then the LFSCK will create
2156 * new MDT-object(s) to hold the conflict OST-object(s).
2158 * type "N": The orphan OST-object does not know which one was the
2159 * real parent MDT-object, so the LFSCK uses new FID for
2160 * its parent MDT-object.
2162 * type "R": The orphan OST-object knows its parent MDT-object FID,
2163 * but does not know the position (the file name) in the
2166 * type "D": The MDT-object is a directory, it may knows its parent
2167 * but because there is no valid linkEA, the LFSCK cannot
2168 * know where to put it back to the namespace.
2169 * type "O": The MDT-object has no linkEA, and there is no name
2170 * entry that references the MDT-object.
2172 * type "P": The orphan object to be created was a parent directory
2173 * of some MDT-object which linkEA shows that the @orphan
2174 * object is missing.
2176 * The orphan name will be like:
2177 * ${FID}-${infix}-${type}-${conflict_version}
2179 * \param[in] ea_off the stripe offset in the LOV EA
2181 * \retval positive on repaired something
2182 * \retval 0 if needs to repair nothing
2183 * \retval negative error number on failure
2185 static int lfsck_layout_recreate_parent(const struct lu_env *env,
2186 struct lfsck_component *com,
2187 struct lfsck_tgt_desc *ltd,
2188 struct lu_orphan_rec_v2 *rec,
2189 struct lu_fid *cfid,
2194 struct lfsck_thread_info *info = lfsck_env_info(env);
2195 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
2196 char *name = info->lti_key;
2197 struct lu_attr *la = &info->lti_la2;
2198 struct dt_object_format *dof = &info->lti_dof;
2199 struct lfsck_instance *lfsck = com->lc_lfsck;
2200 struct ost_layout *ol = &rec->lor_layout;
2201 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
2202 struct lu_fid *tfid = &info->lti_fid3;
2203 struct dt_device *dev = lfsck->li_bottom;
2204 struct dt_object *lpf = lfsck->li_lpf_obj;
2205 struct dt_object *pobj = NULL;
2206 struct dt_object *cobj = NULL;
2207 struct thandle *th = NULL;
2208 struct lu_buf *ea_buf = &info->lti_big_buf;
2209 struct lu_buf lov_buf;
2210 struct lfsck_lock_handle *llh = &info->lti_llh;
2211 struct linkea_data ldata = { NULL };
2212 struct lu_buf linkea_buf;
2213 const struct lu_name *pname;
2219 if (unlikely(lpf == NULL))
2220 GOTO(log, rc = -ENXIO);
2222 /* We use two separated transactions to repair the inconsistency.
2224 * 1) create the MDT-object locally.
2225 * 2) update the OST-object's PFID EA if necessary.
2227 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
2228 * updated when the layout LFSCK run next time.
2230 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
2231 * when the layout LFSCK run next time. */
2233 if (fid_is_zero(pfid)) {
2234 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
2238 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2240 GOTO(log, rc = PTR_ERR(cobj));
2243 pobj = lfsck_object_find_by_dev(env, dev, pfid);
2245 GOTO(log, rc = PTR_ERR(pobj));
2247 LASSERT(infix != NULL);
2248 LASSERT(type != NULL);
2250 memset(la, 0, sizeof(*la));
2251 la->la_uid = rec->lor_rec.lor_uid;
2252 la->la_gid = rec->lor_rec.lor_gid;
2253 la->la_mode = S_IFREG | S_IRUSR;
2254 la->la_valid = LA_MODE | LA_UID | LA_GID;
2256 memset(dof, 0, sizeof(*dof));
2257 dof->dof_type = dt_mode_to_dft(S_IFREG);
2258 /* Because the dof->dof_reg.striped = 0, the LOD will not create
2259 * the stripe(s). The LFSCK will specify the LOV EA via
2260 * lfsck_layout_update_lovea(). */
2262 size = lfsck_lovea_size(ol, ea_off);
2263 if (ea_buf->lb_len < size) {
2264 lu_buf_realloc(ea_buf, size);
2265 if (ea_buf->lb_buf == NULL)
2266 GOTO(log, rc = -ENOMEM);
2271 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
2273 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2274 (const struct dt_key *)name);
2275 if (rc != 0 && rc != -ENOENT)
2279 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
2280 MDS_INODELOCK_UPDATE, LCK_PW);
2284 /* Re-check whether the name conflict with othrs after taken
2286 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2287 (const struct dt_key *)name);
2288 if (unlikely(rc == 0)) {
2296 pname = lfsck_name_get_const(env, name, strlen(name));
2297 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
2298 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
2302 /* The 1st transaction. */
2303 th = dt_trans_create(env, dev);
2305 GOTO(unlock, rc = PTR_ERR(th));
2307 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
2311 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
2312 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
2313 LU_XATTR_CREATE, th);
2317 dtrec->rec_fid = pfid;
2318 dtrec->rec_type = S_IFREG;
2319 rc = dt_declare_insert(env, lpf,
2320 (const struct dt_rec *)dtrec,
2321 (const struct dt_key *)name, th);
2325 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2326 ldata.ld_leh->leh_len);
2327 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
2328 XATTR_NAME_LINK, 0, th);
2332 rc = dt_trans_start_local(env, dev, th);
2336 dt_write_lock(env, pobj, 0);
2337 rc = dt_create(env, pobj, la, NULL, dof, th);
2339 rc = lfsck_layout_update_lovea(env, lfsck, th, ol, pobj, cfid,
2340 &lov_buf, LU_XATTR_CREATE, ltd->ltd_index, ea_off);
2341 dt_write_unlock(env, pobj);
2345 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
2346 (const struct dt_key *)name, th, 1);
2350 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
2351 if (rc == 0 && cobj != NULL) {
2352 dt_trans_stop(env, dev, th);
2355 /* The 2nd transaction. */
2356 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ol, ea_off);
2363 dt_trans_stop(env, dev, th);
2369 if (cobj != NULL && !IS_ERR(cobj))
2370 lfsck_object_put(env, cobj);
2371 if (pobj != NULL && !IS_ERR(pobj))
2372 lfsck_object_put(env, pobj);
2375 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2376 "recreate the lost MDT-object: parent "DFID
2377 ", child "DFID", OST-index %u, stripe-index %u, "
2378 "infix %s, type %s: rc = %d\n",
2379 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2380 ltd->ltd_index, ea_off, infix, type, rc);
2382 return rc >= 0 ? 1 : rc;
2385 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2386 struct lfsck_component *com,
2387 const struct lu_fid *fid,
2390 struct lfsck_thread_info *info = lfsck_env_info(env);
2391 struct lfsck_request *lr = &info->lti_lr;
2392 struct lfsck_instance *lfsck = com->lc_lfsck;
2393 struct lfsck_tgt_desc *ltd;
2394 struct ptlrpc_request *req;
2395 struct lfsck_request *tmp;
2396 struct obd_export *exp;
2400 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2401 if (unlikely(ltd == NULL))
2405 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2406 GOTO(put, rc = -EOPNOTSUPP);
2408 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2410 GOTO(put, rc = -ENOMEM);
2412 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2414 ptlrpc_request_free(req);
2419 memset(lr, 0, sizeof(*lr));
2420 lr->lr_event = LE_CONDITIONAL_DESTROY;
2421 lr->lr_active = LFSCK_TYPE_LAYOUT;
2424 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2426 ptlrpc_request_set_replen(req);
2428 rc = ptlrpc_queue_wait(req);
2429 ptlrpc_req_finished(req);
2439 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2440 struct lfsck_component *com,
2441 struct lfsck_request *lr)
2443 struct lfsck_thread_info *info = lfsck_env_info(env);
2444 struct lu_attr *la = &info->lti_la;
2445 union ldlm_policy_data *policy = &info->lti_policy;
2446 struct ldlm_res_id *resid = &info->lti_resid;
2447 struct lfsck_instance *lfsck = com->lc_lfsck;
2448 struct dt_device *dev = lfsck->li_bottom;
2449 struct lu_fid *fid = &lr->lr_fid;
2450 struct dt_object *obj;
2451 struct thandle *th = NULL;
2452 struct lustre_handle lh = { 0 };
2457 obj = lfsck_object_find_by_dev(env, dev, fid);
2459 RETURN(PTR_ERR(obj));
2461 dt_read_lock(env, obj, 0);
2462 if (dt_object_exists(obj) == 0 ||
2463 lfsck_is_dead_obj(obj)) {
2464 dt_read_unlock(env, obj);
2466 GOTO(put, rc = -ENOENT);
2469 /* Get obj's attr without lock firstly. */
2470 rc = dt_attr_get(env, obj, la);
2471 dt_read_unlock(env, obj);
2475 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2476 GOTO(put, rc = -ETXTBSY);
2478 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2479 LASSERT(lfsck->li_namespace != NULL);
2481 memset(policy, 0, sizeof(*policy));
2482 policy->l_extent.end = OBD_OBJECT_EOF;
2483 ost_fid_build_resid(fid, resid);
2484 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2485 policy, LCK_EX, &flags, ldlm_blocking_ast,
2486 ldlm_completion_ast, NULL, NULL, 0,
2487 LVB_T_NONE, NULL, &lh);
2489 GOTO(put, rc = -EIO);
2491 dt_write_lock(env, obj, 0);
2492 /* Get obj's attr within lock again. */
2493 rc = dt_attr_get(env, obj, la);
2497 if (la->la_ctime != 0)
2498 GOTO(unlock, rc = -ETXTBSY);
2500 th = dt_trans_create(env, dev);
2502 GOTO(unlock, rc = PTR_ERR(th));
2504 rc = dt_declare_ref_del(env, obj, th);
2508 rc = dt_declare_destroy(env, obj, th);
2512 rc = dt_trans_start_local(env, dev, th);
2516 rc = dt_ref_del(env, obj, th);
2520 rc = dt_destroy(env, obj, th);
2522 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2523 "OST-object "DFID" that was created for reparing "
2524 "dangling referenced case. But the original missing "
2525 "OST-object is found now.\n",
2526 lfsck_lfsck2name(lfsck), PFID(fid));
2531 dt_trans_stop(env, dev, th);
2534 dt_write_unlock(env, obj);
2535 ldlm_lock_decref(&lh, LCK_EX);
2538 lfsck_object_put(env, obj);
2544 * Some OST-object has occupied the specified layout EA slot.
2545 * Such OST-object may be generated by the LFSCK when repair
2546 * dangling referenced MDT-object, which can be indicated by
2547 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2548 * is true and such OST-object has not been modified yet, we
2549 * will replace it with the orphan OST-object; otherwise the
2550 * LFSCK will create new MDT-object to reference the orphan.
2552 * \retval +1: repaired
2553 * \retval 0: did nothing
2554 * \retval -ve: on error
2556 static int lfsck_layout_conflict_create(const struct lu_env *env,
2557 struct lfsck_component *com,
2558 struct lfsck_tgt_desc *ltd,
2559 struct lu_orphan_rec_v2 *rec,
2560 struct dt_object *parent,
2561 struct lu_fid *cfid,
2562 struct lu_buf *ea_buf,
2563 struct lov_mds_md_v1 *lmm,
2564 struct lov_ost_data_v1 *slot,
2565 __u32 ea_off, int lovea_size)
2567 struct lfsck_thread_info *info = lfsck_env_info(env);
2568 struct lu_fid *cfid2 = &info->lti_fid2;
2569 struct ost_id *oi = &info->lti_oi;
2570 struct dt_device *dev = lfsck_obj2dev(parent);
2571 struct thandle *th = NULL;
2572 struct lustre_handle lh = { 0 };
2573 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2577 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2578 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2582 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2583 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2587 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2588 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2593 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2595 /* If the conflict OST-obejct is not created for fixing dangling
2596 * referenced MDT-object in former LFSCK check/repair, or it has
2597 * been modified by others, then we cannot destroy it. Re-create
2598 * a new MDT-object for the orphan OST-object. */
2599 if (rc == -ETXTBSY) {
2600 /* No need the layout lock on the original parent. */
2601 lfsck_ibits_unlock(&lh, LCK_EX);
2603 fid_zero(&rec->lor_rec.lor_fid);
2604 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2605 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2607 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2608 info->lti_tmpbuf, "C", ea_off);
2613 if (rc != 0 && rc != -ENOENT)
2616 th = dt_trans_create(env, dev);
2618 GOTO(unlock, rc = PTR_ERR(th));
2620 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2621 LU_XATTR_REPLACE, th);
2625 rc = dt_trans_start_local(env, dev, th);
2629 dt_write_lock(env, parent, 0);
2630 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2631 rc = lfsck_layout_refill_lovea(env, com->lc_lfsck, th, parent, cfid,
2632 ea_buf, lmm, slot, LU_XATTR_REPLACE,
2633 ltd->ltd_index, lovea_size);
2634 dt_write_unlock(env, parent);
2639 dt_trans_stop(env, dev, th);
2642 lfsck_ibits_unlock(&lh, LCK_EX);
2645 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2646 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2647 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2648 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2649 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2652 return rc >= 0 ? 1 : rc;
2656 * \retval +1: repaired
2657 * \retval 0: did nothing
2658 * \retval -ve: on error
2660 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2661 struct lfsck_component *com,
2662 struct lfsck_tgt_desc *ltd,
2663 struct lu_orphan_rec_v2 *rec,
2664 struct dt_object *parent,
2665 struct lu_fid *cfid,
2666 __u32 ost_idx, __u32 ea_off)
2668 struct lfsck_thread_info *info = lfsck_env_info(env);
2669 struct lu_buf *buf = &info->lti_big_buf;
2670 struct lu_fid *fid = &info->lti_fid2;
2671 struct ost_id *oi = &info->lti_oi;
2672 struct lfsck_instance *lfsck = com->lc_lfsck;
2673 struct dt_device *dt = lfsck_obj2dev(parent);
2674 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2675 struct ost_layout *ol = &rec->lor_layout;
2676 struct lov_comp_md_v1 *lcm = NULL;
2677 struct lov_comp_md_entry_v1 *lcme = NULL;
2678 struct thandle *handle = NULL;
2680 struct lov_mds_md_v1 *lmm;
2681 struct lov_ost_data_v1 *objs;
2682 struct lustre_handle lh = { 0 };
2690 bool locked = false;
2693 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2694 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2697 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2698 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2699 "stripe-index %u, comp_id %u, comp_start %llu, "
2700 "comp_end %llu: rc = %d\n",
2701 lfsck_lfsck2name(lfsck), PFID(cfid),
2702 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off,
2703 ol->ol_comp_id, ol->ol_comp_start,
2704 ol->ol_comp_end, rc);
2711 dt_write_unlock(env, parent);
2715 if (handle != NULL) {
2716 dt_trans_stop(env, dt, handle);
2721 GOTO(unlock_layout, rc);
2724 if (buf->lb_len < lovea_size) {
2725 lu_buf_realloc(buf, lovea_size);
2726 if (buf->lb_buf == NULL)
2727 GOTO(unlock_layout, rc = -ENOMEM);
2730 if (!(bk->lb_param & LPF_DRYRUN)) {
2731 handle = dt_trans_create(env, dt);
2733 GOTO(unlock_layout, rc = PTR_ERR(handle));
2735 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2740 rc = dt_trans_start_local(env, dt, handle);
2745 dt_write_lock(env, parent, 0);
2747 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2748 if (rc == -ERANGE) {
2749 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2752 } else if (rc == -ENODATA || rc == 0) {
2753 lovea_size = lfsck_lovea_size(ol, ea_off);
2754 /* If the declared is not big enough, re-try. */
2755 if (buf->lb_len < lovea_size) {
2759 fl = LU_XATTR_CREATE;
2760 } else if (rc < 0) {
2761 GOTO(unlock_parent, rc);
2762 } else if (unlikely(buf->lb_len == 0)) {
2765 fl = LU_XATTR_REPLACE;
2769 if (fl == LU_XATTR_CREATE) {
2770 if (bk->lb_param & LPF_DRYRUN)
2771 GOTO(unlock_parent, rc = 1);
2773 LASSERT(buf->lb_len >= lovea_size);
2775 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2776 cfid, buf, fl, ost_idx, ea_off);
2778 GOTO(unlock_parent, rc);
2782 rc1 = lfsck_layout_verify_header(parent, lmm);
2784 /* If the LOV EA crashed, the rebuild it. */
2785 if (rc1 == -EINVAL) {
2786 if (bk->lb_param & LPF_DRYRUN)
2787 GOTO(unlock_parent, rc = 1);
2789 LASSERT(buf->lb_len >= lovea_size);
2791 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2792 cfid, buf, fl, ost_idx, ea_off);
2794 GOTO(unlock_parent, rc);
2797 /* For other unknown magic/pattern, keep the current LOV EA. */
2799 GOTO(unlock_parent, rc = rc1);
2801 magic = le32_to_cpu(lmm->lmm_magic);
2802 if (magic == LOV_MAGIC_COMP_V1) {
2807 count = le16_to_cpu(lcm->lcm_entry_count);
2808 for (i = 0; i < count; i++) {
2809 lcme = &lcm->lcm_entries[i];
2810 start = le64_to_cpu(lcme->lcme_extent.e_start);
2811 end = le64_to_cpu(lcme->lcme_extent.e_end);
2813 if (end <= ol->ol_comp_start)
2816 if (start >= ol->ol_comp_end)
2819 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
2820 magic = le32_to_cpu(lmm->lmm_magic);
2821 flags = le32_to_cpu(lcme->lcme_flags);
2825 rc = lfsck_layout_add_comp_comp(env, lfsck, handle, ol, parent,
2826 cfid, buf, ost_idx, ea_off, i);
2828 GOTO(unlock_parent, rc);
2832 count = le16_to_cpu(lmm->lmm_stripe_count);
2834 GOTO(unlock_parent, rc = -EINVAL);
2837 /* Exceed the current end of MDT-object layout EA. Then extend it. */
2838 if (count <= ea_off) {
2839 if (bk->lb_param & LPF_DRYRUN)
2840 GOTO(unlock_parent, rc = 1);
2842 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2843 /* If the declared is not big enough, re-try. */
2844 if (buf->lb_len < lovea_size) {
2849 if (lcme && !(flags & LCME_FL_INIT))
2850 lcme->lcme_flags = cpu_to_le32(flags | LCME_FL_INIT);
2852 rc = lfsck_layout_extend_v1v3_lovea(env, lfsck, handle, ol,
2853 parent, cfid, buf, ost_idx, ea_off);
2855 GOTO(unlock_parent, rc);
2858 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2860 if (magic == LOV_MAGIC_V1) {
2861 objs = &lmm->lmm_objects[0];
2863 LASSERT(magic == LOV_MAGIC_V3);
2864 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2867 for (i = 0; i < count; i++, objs++) {
2868 /* The MDT-object was created via lfsck_layout_recover_create()
2869 * by others before, and we fill the dummy layout EA. */
2870 if ((lcme && !(flags & LCME_FL_INIT)) ||
2871 lovea_slot_is_dummy(objs)) {
2875 if (bk->lb_param & LPF_DRYRUN)
2876 GOTO(unlock_parent, rc = 1);
2878 lmm->lmm_layout_gen =
2879 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2883 if (le32_to_cpu(lmm->lmm_stripe_size) !=
2884 ol->ol_stripe_size ||
2885 le16_to_cpu(lmm->lmm_stripe_count) !=
2886 ol->ol_stripe_count ||
2887 le64_to_cpu(lcme->lcme_extent.e_start) !=
2888 ol->ol_comp_start ||
2889 le64_to_cpu(lcme->lcme_extent.e_end) !=
2891 CDEBUG(D_LFSCK, "%s: found invalid "
2892 "component for "DFID ": parent "DFID
2893 ", stripe-index %u, stripe_size %u, "
2894 "stripe_count %u, comp_id %u, "
2895 "comp_start %llu, comp_end %llu, "
2896 "cur_stripe_size %u, "
2897 "cur_stripe_count %u, "
2898 "cur_comp_start %llu, "
2899 "cur_comp_end %llu\n",
2900 lfsck_lfsck2name(lfsck), PFID(cfid),
2901 PFID(lfsck_dto2fid(parent)), ea_off,
2903 ol->ol_stripe_count, ol->ol_comp_id,
2904 ol->ol_comp_start, ol->ol_comp_end,
2905 le32_to_cpu(lmm->lmm_stripe_size),
2906 le16_to_cpu(lmm->lmm_stripe_count),
2907 le64_to_cpu(lcme->lcme_extent.e_start),
2908 le64_to_cpu(lcme->lcme_extent.e_end));
2910 GOTO(unlock_parent, rc = -EINVAL);
2913 le32_add_cpu(&lcm->lcm_layout_gen, 1);
2914 lovea_size = le32_to_cpu(lcm->lcm_size);
2915 if (!(flags & LCME_FL_INIT))
2916 lcme->lcme_flags = cpu_to_le32(flags |
2920 LASSERTF(buf->lb_len >= lovea_size,
2921 "buffer len %d is less than real size %d\n",
2922 (int)buf->lb_len, (int)lovea_size);
2924 rc = lfsck_layout_refill_lovea(env, lfsck, handle,
2925 parent, cfid, buf, lmm, objs,
2926 fl, ost_idx, lovea_size);
2928 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2929 "dummy layout slot for "DFID": parent "DFID
2930 ", OST-index %u, stripe-index %u: rc = %d\n",
2931 lfsck_lfsck2name(lfsck), PFID(cfid),
2932 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2934 GOTO(unlock_parent, rc);
2937 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2938 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2940 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2941 "invalid layout EA at the slot %d, index %u\n",
2942 lfsck_lfsck2name(lfsck),
2943 PFID(lfsck_dto2fid(parent)), i,
2944 le32_to_cpu(objs->l_ost_idx));
2946 GOTO(unlock_parent, rc);
2949 /* It should be rare case, the slot is there, but the LFSCK
2950 * does not handle it during the first-phase cycle scanning. */
2951 if (unlikely(lu_fid_eq(fid, cfid))) {
2953 GOTO(unlock_parent, rc = 0);
2955 /* Rare case that the OST-object index
2956 * does not match the parent MDT-object
2957 * layout EA. We trust the later one. */
2958 if (bk->lb_param & LPF_DRYRUN)
2959 GOTO(unlock_parent, rc = 1);
2961 dt_write_unlock(env, parent);
2963 dt_trans_stop(env, dt, handle);
2964 lfsck_ibits_unlock(&lh, LCK_EX);
2965 rc = lfsck_layout_update_pfid(env, com, parent,
2969 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2970 "updated OST-object's pfid for "DFID
2971 ": parent "DFID", OST-index %u, "
2972 "stripe-index %u: rc = %d\n",
2973 lfsck_lfsck2name(lfsck), PFID(cfid),
2974 PFID(lfsck_dto2fid(parent)),
2975 ltd->ltd_index, i, rc);
2982 /* The MDT-object exists, but related layout EA slot is occupied
2984 if (bk->lb_param & LPF_DRYRUN)
2985 GOTO(unlock_parent, rc = 1);
2987 dt_write_unlock(env, parent);
2989 dt_trans_stop(env, dt, handle);
2990 lfsck_ibits_unlock(&lh, LCK_EX);
2991 if (magic == LOV_MAGIC_V1)
2992 objs = &lmm->lmm_objects[ea_off];
2994 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2995 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2996 buf, lmm, objs, ea_off, lovea_size);
3002 dt_write_unlock(env, parent);
3006 dt_trans_stop(env, dt, handle);
3009 lfsck_ibits_unlock(&lh, LCK_EX);
3014 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
3015 struct lfsck_component *com,
3016 struct lfsck_tgt_desc *ltd,
3017 struct lu_orphan_rec_v2 *rec,
3018 struct lu_fid *cfid)
3020 struct lfsck_layout *lo = com->lc_file_ram;
3021 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
3022 struct dt_object *parent = NULL;
3023 __u32 ea_off = pfid->f_stripe_idx;
3027 if (!fid_is_sane(cfid))
3028 GOTO(out, rc = -EINVAL);
3031 if (fid_is_zero(pfid)) {
3032 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3037 if (!fid_is_sane(pfid))
3038 GOTO(out, rc = -EINVAL);
3040 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
3042 GOTO(out, rc = PTR_ERR(parent));
3044 if (unlikely(dt_object_remote(parent) != 0))
3045 GOTO(put, rc = -EXDEV);
3047 if (dt_object_exists(parent) == 0) {
3048 lfsck_object_put(env, parent);
3049 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3054 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
3055 GOTO(put, rc = -EISDIR);
3057 /* The orphan OST-object claims to be the parent's stripe, then
3058 * related dangling record in the trace file is meaningless. */
3059 rc = lfsck_layout_del_dangling_rec(env, com, pfid,
3060 rec->lor_layout.ol_comp_id, ea_off);
3061 if (rc && rc != -ENOENT)
3064 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
3065 ltd->ltd_index, ea_off);
3071 lfsck_object_put(env, parent);
3073 /* The layout EA is changed, need to be reloaded next time. */
3074 dt_object_put_nocache(env, parent);
3077 down_write(&com->lc_sem);
3078 com->lc_new_scanned++;
3079 com->lc_new_checked++;
3081 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
3083 } else if (rc < 0) {
3084 lo->ll_objs_failed_phase2++;
3086 up_write(&com->lc_sem);
3091 static int lfsck_layout_scan_orphan(const struct lu_env *env,
3092 struct lfsck_component *com,
3093 struct lfsck_tgt_desc *ltd)
3095 struct lfsck_assistant_data *lad = com->lc_data;
3096 struct lfsck_instance *lfsck = com->lc_lfsck;
3097 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3098 struct lfsck_thread_info *info = lfsck_env_info(env);
3099 struct lu_fid *fid = &info->lti_fid;
3100 struct dt_object *obj;
3101 const struct dt_it_ops *iops;
3106 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
3107 "scanning for OST%04x\n",
3108 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3110 if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
3111 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
3112 "scanning for OST%04x\n",
3113 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3118 fid->f_seq = fid_idif_seq(0, ltd->ltd_index);
3119 fid->f_oid = fid->f_ver = 0;
3121 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
3122 if (unlikely(IS_ERR(obj)))
3123 GOTO(log, rc = PTR_ERR(obj));
3125 rc = obj->do_ops->do_index_try(env, obj,
3126 &dt_lfsck_layout_orphan_features);
3130 iops = &obj->do_index_ops->dio_it;
3131 di = iops->init(env, obj, 0);
3133 GOTO(put, rc = PTR_ERR(di));
3135 rc = iops->load(env, di, 0);
3137 /* -ESRCH means that the orphan OST-objects rbtree has been
3138 * cleanup because of the OSS server restart or other errors. */
3139 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
3144 rc = iops->next(env, di);
3156 struct lu_orphan_rec_v2 *rec = &info->lti_rec;
3158 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3159 unlikely(!thread_is_running(&lfsck->li_thread)))
3162 key = iops->key(env, di);
3163 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
3164 /* Remote target OST may be runnning old LFSCK */
3165 memset(rec, 0, sizeof(*rec));
3166 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
3168 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
3169 &com->lc_fid_latest_scanned_phase2);
3170 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3173 lfsck_control_speed_by_self(com);
3175 rc = iops->next(env, di);
3176 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
3183 iops->fini(env, di);
3185 lfsck_object_put(env, obj);
3188 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
3189 "scanning for OST%04x: rc = %d\n",
3190 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
3192 return rc > 0 ? 0 : rc;
3195 static int lfsck_lmm2layout(struct lov_mds_md_v1 *lmm, struct ost_layout *ol,
3198 __u32 magic = le32_to_cpu(lmm->lmm_magic);
3202 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3) {
3203 ol->ol_stripe_size = lmm->lmm_stripe_size;
3204 ol->ol_stripe_count = lmm->lmm_stripe_count;
3205 ol->ol_comp_start = 0;
3206 ol->ol_comp_end = 0;
3208 } else if (magic == LOV_MAGIC_COMP_V1) {
3209 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
3210 struct lov_comp_md_entry_v1 *lcme = NULL;
3211 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3214 for (i = 0; i < count; i++) {
3215 lcme = &lcm->lcm_entries[i];
3216 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3217 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3224 /* The comp has been removed, do nothing. */
3228 lmm = (void *)lmm + le32_to_cpu(lcme->lcme_offset);
3229 ol->ol_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
3230 ol->ol_stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
3231 ol->ol_comp_start = le64_to_cpu(lcme->lcme_extent.e_start);
3232 ol->ol_comp_end = le64_to_cpu(lcme->lcme_extent.e_end);
3233 ol->ol_comp_id = le32_to_cpu(lcme->lcme_id);
3235 GOTO(out, rc = -EINVAL);
3245 * Repair the MDT-object with dangling LOV EA reference.
3247 * we need to repair the inconsistency according to the users' requirement:
3249 * 1) Keep the inconsistency there and report the inconsistency case,
3250 * then give the chance to the application to find related issues,
3251 * and the users can make the decision about how to handle it with
3252 * more human knownledge. (by default)
3254 * 2) Re-create the missing OST-object with the FID/owner information.
3256 * \param[in] env pointer to the thread context
3257 * \param[in] com the layout LFSCK component
3258 * \param[in] parent the MDT-object with dangling LOV EA reference
3259 * \param[in] child the OST-object to be created
3260 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3261 * \param[in] ea_off the offset of the OST-object in the LOV EA
3262 * \param[in] ost_idx the index of OST on which the OST-object resides
3264 * \retval +1 for repair successfully
3265 * \retval 0 for did nothing
3266 * \retval negative error number on failure
3268 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
3269 struct lfsck_component *com,
3270 struct dt_object *parent,
3271 struct dt_object *child,
3272 __u32 comp_id, __u32 ea_off,
3273 __u32 ost_idx, bool log)
3275 struct lfsck_thread_info *info = lfsck_env_info(env);
3276 struct filter_fid *ff = &info->lti_ff;
3277 struct ost_layout *ol = &ff->ff_layout;
3278 struct dt_object_format *dof = &info->lti_dof;
3279 struct lu_attr *la = &info->lti_la;
3280 struct lfsck_instance *lfsck = com->lc_lfsck;
3281 struct dt_device *dev = lfsck_obj2dev(child);
3282 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3283 const struct lu_fid *cfid = lfsck_dto2fid(child);
3284 struct lu_buf *tbuf = &info->lti_big_buf;
3285 struct thandle *handle;
3287 struct lustre_handle lh = { 0 };
3291 if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
3294 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3295 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3300 rc = dt_attr_get(env, parent, la);
3304 la->la_mode = S_IFREG | 0666;
3305 la->la_atime = la->la_mtime = la->la_ctime = 0;
3306 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
3307 LA_ATIME | LA_MTIME | LA_CTIME;
3308 memset(dof, 0, sizeof(*dof));
3309 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
3310 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
3311 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3312 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3313 * parent MDT-object's layout EA. */
3314 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
3316 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3320 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, comp_id);
3324 buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
3325 handle = dt_trans_create(env, dev);
3327 GOTO(unlock1, rc = PTR_ERR(handle));
3329 rc = dt_declare_create(env, child, la, NULL, dof, handle);
3333 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
3334 LU_XATTR_CREATE, handle);
3338 rc = dt_trans_start_local(env, dev, handle);
3342 dt_read_lock(env, parent, 0);
3343 if (unlikely(lfsck_is_dead_obj(parent)))
3344 GOTO(unlock2, rc = 0);
3346 if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
3347 struct ost_id *oi = &info->lti_oi;
3348 struct lu_fid *tfid = &info->lti_fid2;
3349 struct lu_buf *lovea = &info->lti_big_buf;
3350 struct lov_mds_md_v1 *lmm;
3351 struct lov_ost_data_v1 *objs;
3356 rc = lfsck_layout_get_lovea(env, parent, lovea);
3360 lmm = lovea->lb_buf;
3361 magic = le32_to_cpu(lmm->lmm_magic);
3362 if (magic == LOV_MAGIC_COMP_V1) {
3363 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3364 struct lov_comp_md_entry_v1 *lcme;
3365 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3368 for (i = 0; i < count; i++) {
3369 lcme = &lcm->lcm_entries[i];
3370 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3371 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3374 lmm = lovea->lb_buf +
3375 le32_to_cpu(lcme->lcme_offset);
3376 magic = le32_to_cpu(lmm->lmm_magic);
3381 /* Someone removed the component, do nothing. */
3382 GOTO(unlock2, rc = 0);
3386 count = le16_to_cpu(lmm->lmm_stripe_count);
3387 /* Someone changed the LOV EA, do nothing. */
3388 if (count <= ea_off)
3389 GOTO(unlock2, rc = 0);
3391 if (magic == LOV_MAGIC_V1) {
3392 objs = &lmm->lmm_objects[ea_off];
3394 LASSERT(magic == LOV_MAGIC_V3);
3396 objs = &((struct lov_mds_md_v3 *)lmm)->\
3397 lmm_objects[ea_off];
3400 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3401 idx2 = le32_to_cpu(objs->l_ost_idx);
3402 rc = ostid_to_fid(tfid, oi, idx2);
3403 /* Someone changed the LOV EA, do nothing. */
3404 if (rc != 0 || !lu_fid_eq(tfid, cfid))
3408 rc = dt_create(env, child, la, NULL, dof, handle);
3412 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
3418 dt_read_unlock(env, parent);
3421 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3424 lfsck_ibits_unlock(&lh, LCK_EX);
3428 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3429 "dangling reference for: parent "DFID", child "
3430 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: "
3432 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3433 comp_id, ea_off, ost_idx,
3434 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3435 "Create the lost OST-object as required" :
3436 "Keep the MDT-object there by default", rc);
3442 * Repair the MDT-object with dangling LOV EA reference.
3444 * Prepare parameters and call __lfsck_layout_repair_dangling()
3445 * to repair the dangling LOV EA reference.
3447 * \param[in] env pointer to the thread context
3448 * \param[in] com the layout LFSCK component
3449 * \param[in] pfid the MDT-object's FID
3450 * \param[in] cfid the FID for the OST-object to be created
3451 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3452 * \param[in] ea_off the offset of the OST-object in the LOV EA
3453 * \param[in] ost_idx the index of OST on which the OST-object resides
3455 * \retval +1 for repair successfully
3456 * \retval 0 for did nothing
3457 * \retval negative error number on failure
3459 static int lfsck_layout_repair_dangling(const struct lu_env *env,
3460 struct lfsck_component *com,
3461 const struct lu_fid *pfid,
3462 const struct lu_fid *cfid,
3463 __u32 comp_id, __u32 ea_off,
3466 struct lfsck_instance *lfsck = com->lc_lfsck;
3467 struct dt_object *parent = NULL;
3468 struct dt_object *child = NULL;
3469 struct lfsck_tgt_desc *ltd;
3473 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3475 GOTO(log, rc = PTR_ERR(parent));
3477 /* The MDT-object has been removed. */
3478 if (dt_object_exists(parent) == 0)
3481 ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
3482 if (unlikely(ltd == NULL))
3483 GOTO(log, rc = -ENODEV);
3485 child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
3487 GOTO(log, rc = PTR_ERR(child));
3489 /* The OST-object has been created. */
3490 if (unlikely(dt_object_exists(child) != 0))
3493 rc = __lfsck_layout_repair_dangling(env, com, parent, child,
3494 comp_id, ea_off, ost_idx, false);
3499 if (child != NULL && !IS_ERR(child))
3500 lfsck_object_put(env, child);
3502 if (parent != NULL && !IS_ERR(parent))
3503 lfsck_object_put(env, parent);
3506 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3507 "dangling reference for: parent "DFID", child "
3508 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: rc = %d\n",
3509 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3510 comp_id, ea_off, ost_idx,
3511 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3512 "Create the lost OST-object as required" :
3513 "Keep the MDT-object there by default", rc);
3518 /* If the OST-object does not recognize the MDT-object as its parent, and
3519 * there is no other MDT-object claims as its parent, then just trust the
3520 * given MDT-object as its parent. So update the OST-object filter_fid. */
3521 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
3522 struct lfsck_component *com,
3523 struct dt_object *parent,
3524 struct lfsck_layout_req *llr,
3527 struct lfsck_thread_info *info = lfsck_env_info(env);
3528 struct filter_fid *ff = &info->lti_ff;
3529 struct ost_layout *ol = &ff->ff_layout;
3530 struct dt_object *child = llr->llr_child;
3531 struct dt_device *dev = lfsck_obj2dev(child);
3532 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
3533 struct lu_buf *tbuf = &info->lti_big_buf;
3534 struct thandle *handle;
3536 struct lustre_handle lh = { 0 };
3540 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3541 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3546 ff->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3547 ff->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3548 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3549 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3550 * parent MDT-object's layout EA. */
3551 ff->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3553 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3557 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, llr->llr_comp_id);
3561 buf = lfsck_buf_get(env, ff, sizeof(*ff));
3563 handle = dt_trans_create(env, dev);
3565 GOTO(unlock1, rc = PTR_ERR(handle));
3567 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3571 rc = dt_attr_get(env, parent, la);
3575 la->la_valid = LA_UID | LA_GID;
3576 rc = dt_declare_attr_set(env, child, la, handle);
3580 rc = dt_trans_start_local(env, dev, handle);
3584 dt_write_lock(env, parent, 0);
3585 if (unlikely(lfsck_is_dead_obj(parent)))
3586 GOTO(unlock2, rc = 1);
3588 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3592 /* Get the latest parent's owner. */
3593 rc = dt_attr_get(env, parent, la);
3597 la->la_valid = LA_UID | LA_GID;
3598 rc = dt_attr_set(env, child, la, handle);
3603 dt_write_unlock(env, parent);
3606 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3609 lfsck_ibits_unlock(&lh, LCK_EX);
3613 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3614 "unmatched MDT-OST pair for: parent "DFID
3615 ", child "DFID", comp_id %u, OST-index %u, "
3616 "stripe-index %u, owner %u/%u: rc = %d\n",
3617 lfsck_lfsck2name(com->lc_lfsck),
3618 PFID(lfsck_dto2fid(parent)),
3619 PFID(lfsck_dto2fid(child)),
3620 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3621 la->la_uid, la->la_gid, rc);
3626 /* If there are more than one MDT-objects claim as the OST-object's parent,
3627 * and the OST-object only recognizes one of them, then we need to generate
3628 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3629 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3630 struct lfsck_component *com,
3631 struct dt_object *parent,
3632 struct lfsck_layout_req *llr,
3635 struct lfsck_thread_info *info = lfsck_env_info(env);
3636 struct dt_allocation_hint *hint = &info->lti_hint;
3637 struct dt_object_format *dof = &info->lti_dof;
3638 struct ost_id *oi = &info->lti_oi;
3639 struct lu_buf *buf = &info->lti_big_buf;
3640 struct lfsck_instance *lfsck = com->lc_lfsck;
3641 struct dt_device *dev;
3642 struct lu_device *d =
3643 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3644 struct lu_object *o;
3645 struct lu_object *n;
3646 struct dt_object *child = NULL;
3647 struct thandle *handle = NULL;
3648 struct lov_mds_md_v1 *lmm;
3649 struct lov_ost_data_v1 *objs;
3650 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3652 struct lustre_handle lh = { 0 };
3658 /* We use two separated transactions to repair the inconsistency.
3660 * 1) create the child (OST-object).
3661 * 2) update the parent LOV EA according to the child's FID.
3663 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3664 * handled as orphan when the layout LFSCK run next time.
3666 * If 1) failed, but 2) succeed, then such OST-object will be re-created
3667 * as dangling referened case when the layout LFSCK run next time. */
3669 /* The 1st transaction. */
3670 o = lu_object_anon(env, d, NULL);
3672 GOTO(log, rc = PTR_ERR(o));
3674 n = lu_object_locate(o->lo_header, d->ld_type);
3675 if (unlikely(n == NULL)) {
3676 lu_object_put_nocache(env, o);
3678 GOTO(log, rc = -EINVAL);
3681 child = container_of(n, struct dt_object, do_lu);
3682 memset(hint, 0, sizeof(*hint));
3683 rc = dt_attr_get(env, parent, la);
3687 la->la_valid = LA_UID | LA_GID;
3688 memset(dof, 0, sizeof(*dof));
3690 dev = lfsck_obj2dev(child);
3691 handle = dt_trans_create(env, dev);
3693 GOTO(log, rc = PTR_ERR(handle));
3695 rc = dt_declare_create(env, child, la, hint, dof, handle);
3699 rc = dt_trans_start_local(env, dev, handle);
3703 rc = dt_create(env, child, la, hint, dof, handle);
3704 dt_trans_stop(env, dev, handle);
3709 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3710 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3715 /* The 2nd transaction. */
3717 /* XXX: Generally, we should use bottom device (OSD) to update parent
3718 * LOV EA. But because the LOD-object still references the wrong
3719 * OSP-object that should be detached after the parent's LOV EA
3720 * refreshed. Unfortunately, there is no suitable API for that.
3721 * So we have to make the LOD to re-load the OSP-object(s) via
3722 * replacing the LOV EA against the LOD-object.
3724 * Once the DNE2 patches have been landed, we can replace the
3725 * LOD device with the OSD device. LU-6230. */
3727 dev = lfsck->li_next;
3728 parent = lfsck_object_locate(dev, parent);
3730 GOTO(log, rc = PTR_ERR(parent));
3732 handle = dt_trans_create(env, dev);
3734 GOTO(log, rc = PTR_ERR(handle));
3736 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3737 LU_XATTR_REPLACE, handle);
3741 rc = dt_trans_start_local(env, dev, handle);
3745 dt_write_lock(env, parent, 0);
3746 if (unlikely(lfsck_is_dead_obj(parent)))
3747 GOTO(unlock, rc = 0);
3749 rc = lfsck_layout_get_lovea(env, parent, buf);
3750 if (unlikely(!rc || rc == -ENODATA))
3751 GOTO(unlock, rc = 0);
3754 magic = le32_to_cpu(lmm->lmm_magic);
3755 if (magic == LOV_MAGIC_COMP_V1) {
3756 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3757 struct lov_comp_md_entry_v1 *lcme;
3758 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3761 LASSERT(llr->llr_comp_id != 0);
3763 for (i = 0; i < count; i++) {
3764 lcme = &lcm->lcm_entries[i];
3765 if (le32_to_cpu(lcme->lcme_id) == llr->llr_comp_id) {
3766 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3769 le32_add_cpu(&lcm->lcm_layout_gen, 1);
3771 le32_to_cpu(lcme->lcme_offset);
3772 magic = le32_to_cpu(lmm->lmm_magic);
3777 GOTO(unlock, rc = 0);
3781 if (magic == LOV_MAGIC_V1) {
3782 objs = &lmm->lmm_objects[llr->llr_lov_idx];
3784 LASSERT(magic == LOV_MAGIC_V3);
3786 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
3789 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3790 index = le32_to_cpu(objs->l_ost_idx);
3791 rc = ostid_to_fid(&tfid, oi, index);
3792 /* Someone changed layout during the LFSCK, no need to repair then. */
3793 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
3794 GOTO(unlock, rc = 0);
3796 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3797 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3798 ostid_cpu_to_le(oi, &objs->l_ost_oi);
3799 objs->l_ost_gen = cpu_to_le32(0);
3800 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3801 rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3802 LU_XATTR_REPLACE, handle);
3804 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3807 dt_write_unlock(env, parent);
3811 dt_trans_stop(env, dev, handle);
3814 lfsck_ibits_unlock(&lh, LCK_EX);
3816 lfsck_object_put(env, child);
3819 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3820 "multiple references for: parent "DFID", comp_id %u, "
3821 "OST-index %u, stripe-index %u, owner %u/%u: rc = %d\n",
3822 lfsck_lfsck2name(lfsck), PFID(pfid),
3823 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3824 la->la_uid, la->la_gid, rc);
3829 /* If the MDT-object and the OST-object have different owner information,
3830 * then trust the MDT-object, because the normal chown/chgrp handle order
3831 * is from MDT to OST, and it is possible that some chown/chgrp operation
3832 * is partly done. */
3833 static int lfsck_layout_repair_owner(const struct lu_env *env,
3834 struct lfsck_component *com,
3835 struct dt_object *parent,
3836 struct lfsck_layout_req *llr,
3837 struct lu_attr *pla,
3838 const struct lu_attr *cla)
3840 struct lfsck_thread_info *info = lfsck_env_info(env);
3841 struct lu_attr *tla = &info->lti_la2;
3842 struct dt_object *child = llr->llr_child;
3843 struct dt_device *dev = lfsck_obj2dev(child);
3844 struct thandle *handle;
3848 tla->la_uid = pla->la_uid;
3849 tla->la_gid = pla->la_gid;
3850 tla->la_valid = LA_UID | LA_GID;
3851 handle = dt_trans_create(env, dev);
3853 GOTO(log, rc = PTR_ERR(handle));
3855 rc = dt_declare_attr_set(env, child, tla, handle);
3859 rc = dt_trans_start_local(env, dev, handle);
3863 /* Use the dt_object lock to serialize with destroy and attr_set. */
3864 dt_read_lock(env, parent, 0);
3865 if (unlikely(lfsck_is_dead_obj(parent)))
3866 GOTO(unlock, rc = 1);
3868 /* Get the latest parent's owner. */
3869 rc = dt_attr_get(env, parent, pla);
3873 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3874 if (unlikely(tla->la_uid != pla->la_uid ||
3875 tla->la_gid != pla->la_gid))
3878 rc = dt_attr_set(env, child, tla, handle);
3883 dt_read_unlock(env, parent);
3886 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3890 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3891 "inconsistent file owner for: parent "DFID", child "DFID
3892 ", OST-index %u, stripe-index %u, old owner %u/%u, "
3893 "new owner %u/%u: rc = %d\n",
3894 lfsck_lfsck2name(com->lc_lfsck),
3895 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3896 llr->llr_ost_idx, llr->llr_lov_idx,
3897 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3902 /* Check whether the OST-object correctly back points to the
3903 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3904 static int lfsck_layout_check_parent(const struct lu_env *env,
3905 struct lfsck_component *com,
3906 struct lfsck_assistant_object *lso,
3907 struct filter_fid *ff,
3908 const struct lu_fid *cfid,
3909 const struct lu_attr *cla,
3910 struct lfsck_layout_req *llr)
3912 struct lfsck_thread_info *info = lfsck_env_info(env);
3913 struct lu_buf *buf = &info->lti_big_buf;
3914 struct lu_fid *pfid = &info->lti_fid;
3915 struct dt_object *tobj;
3916 struct lov_mds_md_v1 *lmm;
3917 struct lov_ost_data_v1 *objs;
3918 struct lustre_handle lh = { 0 };
3926 *pfid = ff->ff_parent;
3927 idx = pfid->f_stripe_idx;
3930 if (unlikely(!fid_is_sane(pfid)))
3931 RETURN(LLIT_UNMATCHED_PAIR);
3933 if (lu_fid_eq(pfid, &lso->lso_fid)) {
3934 if (likely(llr->llr_lov_idx == idx))
3937 RETURN(LLIT_UNMATCHED_PAIR);
3940 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3942 RETURN(PTR_ERR(tobj));
3944 if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3945 !S_ISREG(lfsck_object_type(tobj)))
3946 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3948 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3949 * remote one on another MDT. Then check whether the given OST-object
3950 * is in such layout. If yes, it is multiple referenced, otherwise it
3951 * is unmatched referenced case. */
3952 rc = lfsck_layout_get_lovea(env, tobj, buf);
3953 if (rc == 0 || rc == -ENOENT)
3954 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3960 magic = le32_to_cpu(lmm->lmm_magic);
3961 if (magic == LOV_MAGIC_COMP_V1) {
3962 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3963 struct lov_comp_md_entry_v1 *lcme;
3965 if (ff->ff_layout.ol_comp_id == 0)
3966 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3968 count = le16_to_cpu(lcm->lcm_entry_count);
3969 for (i = 0; i < count; i++) {
3970 lcme = &lcm->lcm_entries[i];
3971 if (le32_to_cpu(lcme->lcme_id) ==
3972 ff->ff_layout.ol_comp_id) {
3974 le32_to_cpu(lcme->lcme_offset);
3975 magic = le32_to_cpu(lmm->lmm_magic);
3976 if (!(le32_to_cpu(lcme->lcme_flags) &
3978 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3984 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3988 if (magic == LOV_MAGIC_V1) {
3989 objs = &lmm->lmm_objects[0];
3991 LASSERT(magic == LOV_MAGIC_V3);
3992 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3995 count = le16_to_cpu(lmm->lmm_stripe_count);
3996 for (i = 0; i < count; i++, objs++) {
3997 struct lu_fid *tfid = &info->lti_fid2;
3998 struct ost_id *oi = &info->lti_oi;
4001 if (lovea_slot_is_dummy(objs))
4004 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4005 idx2 = le32_to_cpu(objs->l_ost_idx);
4006 rc = ostid_to_fid(tfid, oi, idx2);
4008 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
4009 "invalid layout EA at the slot %d, index %u\n",
4010 lfsck_lfsck2name(com->lc_lfsck),
4011 PFID(pfid), i, idx2);
4013 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4016 if (lu_fid_eq(cfid, tfid)) {
4017 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
4018 MDS_INODELOCK_UPDATE |
4019 MDS_INODELOCK_LAYOUT |
4020 MDS_INODELOCK_XATTR,
4025 dt_read_lock(env, tobj, 0);
4027 /* For local MDT-object, re-check existence
4028 * after taken the lock. */
4029 if (!dt_object_remote(tobj)) {
4030 if (dt_object_exists(tobj) == 0 ||
4031 lfsck_is_dead_obj(tobj))
4032 rc = LLIT_UNMATCHED_PAIR;
4034 rc = LLIT_MULTIPLE_REFERENCED;
4039 /* For migration case, the new MDT-object and old
4040 * MDT-object may reference the same OST-object at
4041 * some migration internal time.
4043 * For remote MDT-object, the local MDT may not know
4044 * whether it has been removed or not. Try checking
4045 * for a non-existent xattr to check if this object
4046 * has been been removed or not. */
4047 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
4049 if (unlikely(rc == -ENOENT || rc >= 0))
4050 rc = LLIT_UNMATCHED_PAIR;
4051 else if (rc == -ENODATA)
4052 rc = LLIT_MULTIPLE_REFERENCED;
4058 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4061 if (lustre_handle_is_used(&lh)) {
4062 dt_read_unlock(env, tobj);
4063 lfsck_ibits_unlock(&lh, LCK_EX);
4067 lfsck_object_put(env, tobj);
4072 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
4073 struct lfsck_component *com,
4074 struct lfsck_assistant_req *lar)
4076 struct lfsck_layout_req *llr =
4077 container_of0(lar, struct lfsck_layout_req, llr_lar);
4078 struct lfsck_assistant_object *lso = lar->lar_parent;
4079 struct lfsck_layout *lo = com->lc_file_ram;
4080 struct lfsck_thread_info *info = lfsck_env_info(env);
4081 struct filter_fid *ff = &info->lti_ff;
4082 struct lu_buf buf = { .lb_buf = ff,
4083 .lb_len = sizeof(*ff) };
4084 struct dt_object *parent = NULL;
4085 struct dt_object *child = llr->llr_child;
4086 struct lu_attr *pla = &lso->lso_attr;
4087 struct lu_attr *cla = &info->lti_la;
4088 struct lfsck_instance *lfsck = com->lc_lfsck;
4089 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4090 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
4097 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ASSISTANT_DIRECT, cfs_fail_val);
4099 rc = dt_attr_get(env, child, cla);
4100 if (rc == -ENOENT) {
4101 parent = lfsck_assistant_object_load(env, lfsck, lso);
4102 if (IS_ERR(parent)) {
4103 rc = PTR_ERR(parent);
4105 RETURN(rc == -ENOENT ? 0 : rc);
4108 type = LLIT_DANGLING;
4115 lfsck_buf_init(&buf, ff, sizeof(*ff));
4116 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
4117 if (unlikely(rc > 0 && rc < sizeof(struct lu_fid))) {
4118 type = LLIT_UNMATCHED_PAIR;
4122 if (rc < 0 && rc != -ENODATA)
4125 if (rc == 0 || rc == -ENODATA)
4126 GOTO(check_owner, rc = 0);
4128 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
4129 rc = lfsck_layout_check_parent(env, com, lso, ff,
4130 lu_object_fid(&child->do_lu), cla, llr);
4140 /* Someone may has changed the owner after the parent attr pre-loaded.
4141 * It can be handled later inside the lfsck_layout_repair_owner(). */
4142 if (unlikely(cla->la_uid != pla->la_uid ||
4143 cla->la_gid != pla->la_gid)) {
4144 type = LLIT_INCONSISTENT_OWNER;
4149 if (type == LLIT_NONE)
4152 if (bk->lb_param & LPF_DRYRUN)
4155 if (parent == NULL) {
4156 parent = lfsck_assistant_object_load(env, lfsck, lso);
4157 if (IS_ERR(parent)) {
4158 rc = PTR_ERR(parent);
4169 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
4170 rc = lfsck_layout_ins_dangling_rec(env, com,
4171 lfsck_dto2fid(parent), lfsck_dto2fid(child),
4172 llr->llr_comp_id, llr->llr_lov_idx,
4175 rc = __lfsck_layout_repair_dangling(env, com, parent,
4182 case LLIT_UNMATCHED_PAIR:
4183 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
4186 case LLIT_MULTIPLE_REFERENCED:
4187 rc = lfsck_layout_repair_multiple_references(env, com, parent,
4190 case LLIT_INCONSISTENT_OWNER:
4191 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
4201 down_write(&com->lc_sem);
4203 struct lfsck_assistant_data *lad = com->lc_data;
4205 if (unlikely(lad->lad_exit)) {
4207 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
4208 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
4209 rc == -EHOSTUNREACH) {
4210 /* If cannot touch the target server,
4211 * mark the LFSCK as INCOMPLETE. */
4212 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
4213 "talk with OST %x: rc = %d\n",
4214 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
4215 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
4216 lo->ll_objs_skipped++;
4219 lfsck_layout_record_failure(env, lfsck, lo);
4221 } else if (rc > 0 && (type != LLIT_DANGLING ||
4222 !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
4223 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
4224 "unknown type = %d\n", type);
4226 lo->ll_objs_repaired[type - 1]++;
4227 if (bk->lb_param & LPF_DRYRUN &&
4228 unlikely(lo->ll_pos_first_inconsistent == 0))
4229 lo->ll_pos_first_inconsistent =
4230 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
4233 up_write(&com->lc_sem);
4235 if (parent != NULL && !IS_ERR(parent))
4236 lfsck_object_put(env, parent);
4242 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
4243 struct lfsck_component *com,
4244 struct dt_object *obj, bool first)
4246 struct lfsck_instance *lfsck = com->lc_lfsck;
4247 struct ptlrpc_thread *thread = &lfsck->li_thread;
4248 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4249 struct lfsck_layout *lo = com->lc_file_ram;
4250 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
4253 struct lfsck_layout_dangling_key *parent =
4254 &lfsck_env_info(env)->lti_lldk;
4255 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
4260 di = iops->init(env, obj, 0);
4262 RETURN(PTR_ERR(di));
4265 lldk_cpu_to_be(parent, &lo->ll_lldk_latest_scanned_phase2);
4267 memset(parent, 0, sizeof(*parent));
4268 rc = iops->get(env, di, (const struct dt_key *)parent);
4273 /* The start one either has been processed or does not exist,
4275 rc = iops->next(env, di);
4281 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
4282 unlikely(!thread_is_running(thread)))
4285 key = iops->key(env, di);
4294 lldk_be_to_cpu(parent,
4295 (const struct lfsck_layout_dangling_key *)key);
4296 if (!fid_is_sane(&parent->lldk_fid)) {
4301 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
4303 fid_be_to_cpu(cfid, cfid);
4304 ost_idx = cfid->f_ver;
4306 if (!fid_is_sane(cfid)) {
4311 rc = lfsck_layout_repair_dangling(env, com,
4312 &parent->lldk_fid, cfid,
4313 parent->lldk_comp_id,
4314 parent->lldk_ea_off, ost_idx);
4318 down_write(&com->lc_sem);
4319 com->lc_new_checked++;
4320 com->lc_new_scanned++;
4322 lo->ll_lldk_latest_scanned_phase2 = *parent;
4325 lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
4327 lo->ll_objs_failed_phase2++;
4328 up_write(&com->lc_sem);
4330 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4333 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
4334 cfs_time_current())) &&
4335 com->lc_new_checked != 0) {
4336 down_write(&com->lc_sem);
4337 lo->ll_run_time_phase2 +=
4338 cfs_duration_sec(cfs_time_current() +
4339 HALF_SEC - com->lc_time_last_checkpoint);
4340 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4341 lo->ll_objs_checked_phase2 += com->lc_new_checked;
4342 com->lc_new_checked = 0;
4343 lfsck_layout_store(env, com);
4344 up_write(&com->lc_sem);
4346 com->lc_time_last_checkpoint = cfs_time_current();
4347 com->lc_time_next_checkpoint =
4348 com->lc_time_last_checkpoint +
4349 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4352 lfsck_control_speed_by_self(com);
4353 if (unlikely(!thread_is_running(thread)))
4356 rc = iops->next(env, di);
4365 iops->fini(env, di);
4370 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
4371 struct lfsck_component *com)
4373 struct lfsck_assistant_data *lad = com->lc_data;
4374 struct lfsck_instance *lfsck = com->lc_lfsck;
4375 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4376 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4377 struct lfsck_tgt_desc *ltd;
4381 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
4382 lfsck_lfsck2name(lfsck));
4384 spin_lock(<ds->ltd_lock);
4385 while (!list_empty(&lad->lad_ost_phase2_list)) {
4386 ltd = list_entry(lad->lad_ost_phase2_list.next,
4387 struct lfsck_tgt_desc,
4388 ltd_layout_phase_list);
4389 list_del_init(<d->ltd_layout_phase_list);
4390 if (bk->lb_param & LPF_OST_ORPHAN) {
4391 spin_unlock(<ds->ltd_lock);
4392 rc = lfsck_layout_scan_orphan(env, com, ltd);
4393 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
4396 if (unlikely(lad->lad_exit ||
4397 !thread_is_running(&lfsck->li_thread)))
4399 spin_lock(<ds->ltd_lock);
4403 if (list_empty(&lad->lad_ost_phase1_list))
4407 spin_unlock(<ds->ltd_lock);
4409 if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
4410 struct lfsck_layout *lo = com->lc_file_ram;
4413 com->lc_new_checked = 0;
4414 com->lc_new_scanned = 0;
4415 com->lc_time_last_checkpoint = cfs_time_current();
4416 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4417 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4419 i = lfsck_sub_trace_file_fid2idx(
4420 &lo->ll_lldk_latest_scanned_phase2.lldk_fid);
4421 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4422 com->lc_sub_trace_objs[i].lsto_obj, true);
4423 while (rc > 0 && ++i < LFSCK_STF_COUNT)
4424 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4425 com->lc_sub_trace_objs[i].lsto_obj, false);
4427 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
4428 "at the No. %d trace file: rc = %d\n",
4429 lfsck_lfsck2name(lfsck), i, rc);
4432 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
4433 lfsck_lfsck2name(lfsck), rc);
4439 lfsck_layout_slave_async_interpret(const struct lu_env *env,
4440 struct ptlrpc_request *req,
4443 struct lfsck_layout_slave_async_args *llsaa = args;
4444 struct obd_export *exp = llsaa->llsaa_exp;
4445 struct lfsck_component *com = llsaa->llsaa_com;
4446 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
4447 struct lfsck_layout_slave_data *llsd = com->lc_data;
4448 struct lfsck_reply *lr = NULL;
4452 /* It is probably caused by network trouble, or target crash,
4453 * it will try several times (depends on the obd_timeout, and
4454 * will not less than 3 times). But to make the LFSCK can go
4455 * ahead, we should not try for ever. After some try but still
4456 * hit failure, it will assume that the target exit the LFSCK
4457 * prcoessing and stop try. */
4458 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
4459 int max_try = max_t(int, obd_timeout / 30, 3);
4461 if (++(llst->llst_failures) > max_try)
4467 llst->llst_failures = 0;
4468 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
4469 if (lr->lr_status != LS_SCANNING_PHASE1 &&
4470 lr->lr_status != LS_SCANNING_PHASE2)
4475 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
4476 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
4477 llst->llst_index, lr != NULL ? lr->lr_status : rc,
4478 llst->llst_failures);
4480 lfsck_layout_llst_del(llsd, llst);
4483 lfsck_layout_llst_put(llst);
4484 lfsck_component_put(env, com);
4485 class_export_put(exp);
4490 static int lfsck_layout_async_query(const struct lu_env *env,
4491 struct lfsck_component *com,
4492 struct obd_export *exp,
4493 struct lfsck_layout_slave_target *llst,
4494 struct lfsck_request *lr,
4495 struct ptlrpc_request_set *set)
4497 struct lfsck_layout_slave_async_args *llsaa;
4498 struct ptlrpc_request *req;
4499 struct lfsck_request *tmp;
4503 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
4507 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
4509 ptlrpc_request_free(req);
4513 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4515 ptlrpc_request_set_replen(req);
4517 llsaa = ptlrpc_req_async_args(req);
4518 llsaa->llsaa_exp = exp;
4519 llsaa->llsaa_com = lfsck_component_get(com);
4520 llsaa->llsaa_llst = llst;
4521 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
4522 req->rq_allow_intr = 1;
4523 ptlrpc_set_add_req(set, req);
4528 static int lfsck_layout_async_notify(const struct lu_env *env,
4529 struct obd_export *exp,
4530 struct lfsck_request *lr,
4531 struct ptlrpc_request_set *set)
4533 struct ptlrpc_request *req;
4534 struct lfsck_request *tmp;
4538 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4542 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4544 ptlrpc_request_free(req);
4548 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4550 ptlrpc_request_set_replen(req);
4551 req->rq_allow_intr = 1;
4552 ptlrpc_set_add_req(set, req);
4558 lfsck_layout_slave_query_master(const struct lu_env *env,
4559 struct lfsck_component *com)
4561 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4562 struct lfsck_instance *lfsck = com->lc_lfsck;
4563 struct lfsck_layout_slave_data *llsd = com->lc_data;
4564 struct lfsck_layout_slave_target *llst;
4565 struct obd_export *exp;
4566 struct ptlrpc_request_set *set;
4571 set = ptlrpc_prep_set();
4573 GOTO(log, rc = -ENOMEM);
4575 memset(lr, 0, sizeof(*lr));
4576 lr->lr_event = LE_QUERY;
4577 lr->lr_active = LFSCK_TYPE_LAYOUT;
4579 llsd->llsd_touch_gen++;
4580 spin_lock(&llsd->llsd_lock);
4581 while (!list_empty(&llsd->llsd_master_list)) {
4582 llst = list_entry(llsd->llsd_master_list.next,
4583 struct lfsck_layout_slave_target,
4585 if (llst->llst_gen == llsd->llsd_touch_gen)
4588 llst->llst_gen = llsd->llsd_touch_gen;
4589 list_move_tail(&llst->llst_list,
4590 &llsd->llsd_master_list);
4591 atomic_inc(&llst->llst_ref);
4592 spin_unlock(&llsd->llsd_lock);
4594 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4597 lfsck_layout_llst_del(llsd, llst);
4598 lfsck_layout_llst_put(llst);
4599 spin_lock(&llsd->llsd_lock);
4603 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
4605 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4606 "query %s for layout: rc = %d\n",
4607 lfsck_lfsck2name(lfsck),
4608 exp->exp_obd->obd_name, rc);
4611 lfsck_layout_llst_put(llst);
4612 class_export_put(exp);
4614 spin_lock(&llsd->llsd_lock);
4616 spin_unlock(&llsd->llsd_lock);
4618 rc = ptlrpc_set_wait(set);
4619 ptlrpc_set_destroy(set);
4621 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4624 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4625 lfsck_lfsck2name(com->lc_lfsck), rc);
4631 lfsck_layout_slave_notify_master(const struct lu_env *env,
4632 struct lfsck_component *com,
4633 enum lfsck_events event, int result)
4635 struct lfsck_layout *lo = com->lc_file_ram;
4636 struct lfsck_instance *lfsck = com->lc_lfsck;
4637 struct lfsck_layout_slave_data *llsd = com->lc_data;
4638 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4639 struct lfsck_layout_slave_target *llst;
4640 struct obd_export *exp;
4641 struct ptlrpc_request_set *set;
4645 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4646 lfsck_lfsck2name(com->lc_lfsck));
4648 set = ptlrpc_prep_set();
4652 memset(lr, 0, sizeof(*lr));
4653 lr->lr_event = event;
4654 lr->lr_flags = LEF_FROM_OST;
4655 lr->lr_status = result;
4656 lr->lr_index = lfsck_dev_idx(lfsck);
4657 lr->lr_active = LFSCK_TYPE_LAYOUT;
4658 lr->lr_flags2 = lo->ll_flags;
4659 llsd->llsd_touch_gen++;
4660 spin_lock(&llsd->llsd_lock);
4661 while (!list_empty(&llsd->llsd_master_list)) {
4662 llst = list_entry(llsd->llsd_master_list.next,
4663 struct lfsck_layout_slave_target,
4665 if (llst->llst_gen == llsd->llsd_touch_gen)
4668 llst->llst_gen = llsd->llsd_touch_gen;
4669 list_move_tail(&llst->llst_list,
4670 &llsd->llsd_master_list);
4671 atomic_inc(&llst->llst_ref);
4672 spin_unlock(&llsd->llsd_lock);
4674 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4677 lfsck_layout_llst_del(llsd, llst);
4678 lfsck_layout_llst_put(llst);
4679 spin_lock(&llsd->llsd_lock);
4683 rc = lfsck_layout_async_notify(env, exp, lr, set);
4685 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4686 "notify %s for layout: rc = %d\n",
4687 lfsck_lfsck2name(lfsck),
4688 exp->exp_obd->obd_name, rc);
4690 lfsck_layout_llst_put(llst);
4691 class_export_put(exp);
4692 spin_lock(&llsd->llsd_lock);
4694 spin_unlock(&llsd->llsd_lock);
4696 ptlrpc_set_wait(set);
4697 ptlrpc_set_destroy(set);
4703 * \ret -ENODATA: unrecognized stripe
4704 * \ret = 0 : recognized stripe
4705 * \ret < 0 : other failures
4707 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4708 struct lfsck_component *com,
4709 struct lu_fid *cfid,
4710 struct lu_fid *pfid, __u32 comp_id)
4712 struct lfsck_thread_info *info = lfsck_env_info(env);
4713 struct lu_buf *buf = &info->lti_big_buf;
4714 struct ost_id *oi = &info->lti_oi;
4715 struct dt_object *obj;
4716 struct lov_mds_md_v1 *lmm;
4717 struct lov_ost_data_v1 *objs;
4718 __u32 idx = pfid->f_stripe_idx;
4726 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4728 RETURN(PTR_ERR(obj));
4730 dt_read_lock(env, obj, 0);
4731 if (unlikely(dt_object_exists(obj) == 0 ||
4732 lfsck_is_dead_obj(obj)))
4733 GOTO(unlock, rc = -ENOENT);
4735 if (!S_ISREG(lfsck_object_type(obj)))
4736 GOTO(unlock, rc = -ENODATA);
4738 rc = lfsck_layout_get_lovea(env, obj, buf);
4743 GOTO(unlock, rc = -ENODATA);
4746 magic = le32_to_cpu(lmm->lmm_magic);
4747 if (magic == LOV_MAGIC_COMP_V1) {
4748 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4749 struct lov_comp_md_entry_v1 *lcme;
4752 GOTO(unlock, rc = -ENODATA);
4754 count = le16_to_cpu(lcm->lcm_entry_count);
4755 for (i = 0; i < count; i++) {
4756 lcme = &lcm->lcm_entries[i];
4757 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
4759 le32_to_cpu(lcme->lcme_offset);
4760 magic = le32_to_cpu(lmm->lmm_magic);
4761 if (!(le32_to_cpu(lcme->lcme_flags) &
4763 GOTO(unlock, rc = -ENODATA);
4769 GOTO(unlock, rc = -ENODATA);
4773 if (magic == LOV_MAGIC_V1) {
4774 objs = &lmm->lmm_objects[0];
4776 LASSERT(magic == LOV_MAGIC_V3);
4777 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4780 fid_to_ostid(cfid, oi);
4781 count = le16_to_cpu(lmm->lmm_stripe_count);
4782 for (i = 0; i < count; i++, objs++) {
4785 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
4786 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
4787 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
4790 GOTO(unlock, rc = -ENODATA);
4793 dt_read_unlock(env, obj);
4794 lfsck_object_put(env, obj);
4800 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
4801 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
4802 * layout EA from MDT to OST. On one hand, the OST no need to understand
4803 * the layout EA structure; on the other hand, it may cause trouble when
4804 * transfer large layout EA from MDT to OST via normal OUT RPC.
4806 * \ret > 0: unrecognized stripe
4807 * \ret = 0: recognized stripe
4808 * \ret < 0: other failures
4810 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
4811 struct lfsck_component *com,
4812 struct lu_fid *cfid,
4813 struct lu_fid *pfid, __u32 comp_id)
4815 struct lfsck_instance *lfsck = com->lc_lfsck;
4816 struct obd_device *obd = lfsck->li_obd;
4817 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4818 struct obd_export *exp = NULL;
4819 struct ptlrpc_request *req = NULL;
4820 struct lfsck_request *lr;
4821 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
4825 if (unlikely(fid_is_idif(pfid)))
4828 fld_range_set_any(range);
4829 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
4831 RETURN(rc == -ENOENT ? 1 : rc);
4833 if (unlikely(!fld_range_is_mdt(range)))
4836 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
4837 if (unlikely(exp == NULL))
4840 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
4841 GOTO(out, rc = -EOPNOTSUPP);
4843 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4845 GOTO(out, rc = -ENOMEM);
4847 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4849 ptlrpc_request_free(req);
4854 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4855 memset(lr, 0, sizeof(*lr));
4856 lr->lr_event = LE_PAIRS_VERIFY;
4857 lr->lr_active = LFSCK_TYPE_LAYOUT;
4858 lr->lr_fid = *cfid; /* OST-object itself FID. */
4859 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
4860 lr->lr_comp_id = comp_id;
4862 ptlrpc_request_set_replen(req);
4863 rc = ptlrpc_queue_wait(req);
4864 ptlrpc_req_finished(req);
4866 if (rc == -ENOENT || rc == -ENODATA)
4873 class_export_put(exp);
4878 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
4879 struct lfsck_component *com,
4880 struct lfsck_req_local *lrl)
4882 struct dt_object *obj;
4886 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lrl->lrl_fid);
4888 GOTO(log, rc = PTR_ERR(obj));
4890 dt_write_lock(env, obj, 0);
4891 if (unlikely(dt_object_exists(obj) == 0 ||
4892 lfsck_is_dead_obj(obj)))
4893 GOTO(unlock, rc = 0);
4895 rc = __lfsck_layout_update_pfid(env, obj, &lrl->lrl_ff_client.ff_parent,
4896 &lrl->lrl_ff_client.ff_layout,
4897 lrl->lrl_ff_client.ff_parent.f_ver);
4902 dt_write_unlock(env, obj);
4903 lfsck_object_put(env, obj);
4906 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
4907 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
4908 PFID(&lrl->lrl_fid), PFID(&lrl->lrl_ff_client.ff_parent), rc);
4915 static void lfsck_layout_slave_quit(const struct lu_env *env,
4916 struct lfsck_component *com);
4918 static int lfsck_layout_reset(const struct lu_env *env,
4919 struct lfsck_component *com, bool init)
4921 struct lfsck_layout *lo = com->lc_file_ram;
4924 down_write(&com->lc_sem);
4926 memset(lo, 0, com->lc_file_size);
4928 __u32 count = lo->ll_success_count;
4929 __u64 last_time = lo->ll_time_last_complete;
4931 memset(lo, 0, com->lc_file_size);
4932 lo->ll_success_count = count;
4933 lo->ll_time_last_complete = last_time;
4936 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
4937 lo->ll_status = LS_INIT;
4939 if (com->lc_lfsck->li_master) {
4940 struct lfsck_assistant_data *lad = com->lc_data;
4942 lad->lad_incomplete = 0;
4943 CFS_RESET_BITMAP(lad->lad_bitmap);
4946 rc = lfsck_layout_store(env, com);
4947 if (rc == 0 && com->lc_lfsck->li_master)
4948 rc = lfsck_load_sub_trace_files(env, com,
4949 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
4950 up_write(&com->lc_sem);
4952 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
4953 lfsck_lfsck2name(com->lc_lfsck), rc);
4958 static void lfsck_layout_fail(const struct lu_env *env,
4959 struct lfsck_component *com, bool new_checked)
4961 struct lfsck_layout *lo = com->lc_file_ram;
4963 down_write(&com->lc_sem);
4965 com->lc_new_checked++;
4966 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
4967 up_write(&com->lc_sem);
4970 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
4971 struct lfsck_component *com, bool init)
4973 struct lfsck_instance *lfsck = com->lc_lfsck;
4974 struct lfsck_layout *lo = com->lc_file_ram;
4978 rc = lfsck_checkpoint_generic(env, com);
4980 return rc > 0 ? 0 : rc;
4983 down_write(&com->lc_sem);
4985 lo->ll_pos_latest_start =
4986 lfsck->li_pos_checkpoint.lp_oit_cookie;
4988 lo->ll_pos_last_checkpoint =
4989 lfsck->li_pos_checkpoint.lp_oit_cookie;
4990 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4991 HALF_SEC - lfsck->li_time_last_checkpoint);
4992 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4993 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4994 com->lc_new_checked = 0;
4997 rc = lfsck_layout_store(env, com);
4998 up_write(&com->lc_sem);
5000 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
5001 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5002 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5007 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
5008 struct lfsck_component *com, bool init)
5010 struct lfsck_instance *lfsck = com->lc_lfsck;
5011 struct lfsck_layout *lo = com->lc_file_ram;
5014 if (com->lc_new_checked == 0 && !init)
5017 down_write(&com->lc_sem);
5019 lo->ll_pos_latest_start =
5020 lfsck->li_pos_checkpoint.lp_oit_cookie;
5022 lo->ll_pos_last_checkpoint =
5023 lfsck->li_pos_checkpoint.lp_oit_cookie;
5024 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5025 HALF_SEC - lfsck->li_time_last_checkpoint);
5026 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5027 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5028 com->lc_new_checked = 0;
5031 rc = lfsck_layout_store(env, com);
5032 up_write(&com->lc_sem);
5034 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
5035 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5036 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5041 static int lfsck_layout_prep(const struct lu_env *env,
5042 struct lfsck_component *com,
5043 struct lfsck_start *start)
5045 struct lfsck_instance *lfsck = com->lc_lfsck;
5046 struct lfsck_layout *lo = com->lc_file_ram;
5047 struct lfsck_position *pos = &com->lc_pos_start;
5049 fid_zero(&pos->lp_dir_parent);
5050 pos->lp_dir_cookie = 0;
5051 if (lo->ll_status == LS_COMPLETED ||
5052 lo->ll_status == LS_PARTIAL ||
5053 /* To handle orphan, must scan from the beginning. */
5054 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
5057 rc = lfsck_layout_reset(env, com, false);
5059 rc = lfsck_set_param(env, lfsck, start, true);
5062 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
5063 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
5069 down_write(&com->lc_sem);
5070 lo->ll_time_latest_start = cfs_time_current_sec();
5071 spin_lock(&lfsck->li_lock);
5072 if (lo->ll_flags & LF_SCANNED_ONCE) {
5073 if (!lfsck->li_drop_dryrun ||
5074 lo->ll_pos_first_inconsistent == 0) {
5075 lo->ll_status = LS_SCANNING_PHASE2;
5076 list_move_tail(&com->lc_link,
5077 &lfsck->li_list_double_scan);
5078 pos->lp_oit_cookie = 0;
5082 lo->ll_status = LS_SCANNING_PHASE1;
5083 lo->ll_run_time_phase1 = 0;
5084 lo->ll_run_time_phase2 = 0;
5085 lo->ll_objs_checked_phase1 = 0;
5086 lo->ll_objs_checked_phase2 = 0;
5087 lo->ll_objs_failed_phase1 = 0;
5088 lo->ll_objs_failed_phase2 = 0;
5089 for (i = 0; i < LLIT_MAX; i++)
5090 lo->ll_objs_repaired[i] = 0;
5092 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5093 fid_zero(&com->lc_fid_latest_scanned_phase2);
5096 lo->ll_status = LS_SCANNING_PHASE1;
5097 if (!lfsck->li_drop_dryrun ||
5098 lo->ll_pos_first_inconsistent == 0)
5099 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
5101 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5103 spin_unlock(&lfsck->li_lock);
5104 up_write(&com->lc_sem);
5109 static int lfsck_layout_slave_prep(const struct lu_env *env,
5110 struct lfsck_component *com,
5111 struct lfsck_start_param *lsp)
5113 struct lfsck_layout_slave_data *llsd = com->lc_data;
5114 struct lfsck_instance *lfsck = com->lc_lfsck;
5115 struct lfsck_layout *lo = com->lc_file_ram;
5116 struct lfsck_start *start = lsp->lsp_start;
5119 rc = lfsck_layout_prep(env, com, start);
5123 if (lo->ll_flags & LF_CRASHED_LASTID &&
5124 list_empty(&llsd->llsd_master_list)) {
5125 LASSERT(lfsck->li_out_notify != NULL);
5127 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5128 LE_LASTID_REBUILDING);
5131 if (!lsp->lsp_index_valid)
5134 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5135 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
5136 LASSERT(!llsd->llsd_rbtree_valid);
5138 write_lock(&llsd->llsd_rb_lock);
5139 rc = lfsck_rbtree_setup(env, com);
5140 write_unlock(&llsd->llsd_rb_lock);
5143 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
5144 "%llu]\n", lfsck_lfsck2name(lfsck),
5145 com->lc_pos_start.lp_oit_cookie);
5150 static int lfsck_layout_master_prep(const struct lu_env *env,
5151 struct lfsck_component *com,
5152 struct lfsck_start_param *lsp)
5157 rc = lfsck_layout_load_bitmap(env, com);
5159 rc = lfsck_layout_reset(env, com, false);
5161 rc = lfsck_set_param(env, com->lc_lfsck,
5162 lsp->lsp_start, true);
5168 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
5172 rc = lfsck_start_assistant(env, com, lsp);
5177 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
5178 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
5179 com->lc_pos_start.lp_oit_cookie);
5184 /* Pre-fetch the attribute for each stripe in the given layout EA. */
5185 static int lfsck_layout_scan_stripes(const struct lu_env *env,
5186 struct lfsck_component *com,
5187 struct dt_object *parent,
5188 struct lov_mds_md_v1 *lmm, __u32 comp_id)
5190 struct lfsck_thread_info *info = lfsck_env_info(env);
5191 struct lfsck_instance *lfsck = com->lc_lfsck;
5192 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5193 struct lfsck_layout *lo = com->lc_file_ram;
5194 struct lfsck_assistant_data *lad = com->lc_data;
5195 struct lfsck_assistant_object *lso = NULL;
5196 struct lov_ost_data_v1 *objs;
5197 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
5198 struct ptlrpc_thread *mthread = &lfsck->li_thread;
5199 struct ptlrpc_thread *athread = &lad->lad_thread;
5200 struct l_wait_info lwi = { 0 };
5208 lfsck_buf_init(&buf, &info->lti_ff, sizeof(struct filter_fid));
5209 magic = le32_to_cpu(lmm->lmm_magic);
5210 if (magic == LOV_MAGIC_V1) {
5211 objs = &lmm->lmm_objects[0];
5213 LASSERT(magic == LOV_MAGIC_V3);
5214 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5217 count = le16_to_cpu(lmm->lmm_stripe_count);
5218 for (i = 0; i < count; i++, objs++) {
5219 struct lu_fid *fid = &info->lti_fid;
5220 struct ost_id *oi = &info->lti_oi;
5221 struct lfsck_layout_req *llr;
5222 struct lfsck_tgt_desc *tgt = NULL;
5223 struct dt_object *cobj = NULL;
5225 bool wakeup = false;
5227 if (unlikely(lovea_slot_is_dummy(objs)))
5230 l_wait_event(mthread->t_ctl_waitq,
5231 lad->lad_prefetched < bk->lb_async_windows ||
5232 !thread_is_running(mthread) ||
5233 thread_is_stopped(athread),
5236 if (unlikely(!thread_is_running(mthread)) ||
5237 thread_is_stopped(athread))
5240 if (unlikely(lfsck_is_dead_obj(parent)))
5243 ostid_le_to_cpu(&objs->l_ost_oi, oi);
5244 index = le32_to_cpu(objs->l_ost_idx);
5245 rc = ostid_to_fid(fid, oi, index);
5247 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
5248 ": "DOSTID", idx %u, comp_id %u\n",
5249 lfsck_lfsck2name(lfsck),
5250 PFID(lfsck_dto2fid(parent)), POSTID(oi),
5255 tgt = lfsck_tgt_get(ltds, index);
5256 if (unlikely(tgt == NULL)) {
5257 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
5258 "did not join the layout LFSCK, comp_id %u\n",
5259 lfsck_lfsck2name(lfsck), index, comp_id);
5260 lfsck_lad_set_bitmap(env, com, index);
5264 /* There is potential deadlock race condition between object
5265 * destroy and layout LFSCK. Consider the following scenario:
5267 * 1) The LFSCK thread obtained the parent object firstly, at
5268 * that time, the parent object has not been destroyed yet.
5270 * 2) One RPC service thread destroyed the parent and all its
5271 * children objects. Because the LFSCK is referencing the
5272 * parent object, then the parent object will be marked as
5273 * dying in RAM. On the other hand, the parent object is
5274 * referencing all its children objects, then all children
5275 * objects will be marked as dying in RAM also.
5277 * 3) The LFSCK thread tries to find some child object with
5278 * the parent object referenced. Then it will find that the
5279 * child object is dying. According to the object visibility
5280 * rules: the object with dying flag cannot be returned to
5281 * others. So the LFSCK thread has to wait until the dying
5282 * object has been purged from RAM, then it can allocate a
5283 * new object (with the same FID) in RAM. Unfortunately, the
5284 * LFSCK thread itself is referencing the parent object, and
5285 * cause the parent object cannot be purged, then cause the
5286 * child object cannot be purged also. So the LFSCK thread
5287 * will fall into deadlock.
5289 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
5291 if (lfsck_is_dead_obj(parent)) {
5301 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_ASSISTANT_DIRECT)) {
5302 rc = dt_declare_attr_get(env, cobj);
5306 rc = dt_declare_xattr_get(env, cobj, &buf,
5313 struct lu_attr *attr = &info->lti_la;
5315 rc = dt_attr_get(env, parent, attr);
5319 lso = lfsck_assistant_object_init(env,
5320 lfsck_dto2fid(parent), attr,
5321 lfsck->li_pos_current.lp_oit_cookie, false);
5330 llr = lfsck_layout_assistant_req_init(lso, cobj, comp_id,
5338 spin_lock(&lad->lad_lock);
5339 if (lad->lad_assistant_status < 0) {
5340 spin_unlock(&lad->lad_lock);
5341 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
5343 RETURN(lad->lad_assistant_status);
5346 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
5347 if (lad->lad_prefetched == 0)
5350 lad->lad_prefetched++;
5351 spin_unlock(&lad->lad_lock);
5353 wake_up_all(&athread->t_ctl_waitq);
5356 down_write(&com->lc_sem);
5357 com->lc_new_checked++;
5359 lfsck_layout_record_failure(env, lfsck, lo);
5360 up_write(&com->lc_sem);
5362 if (cobj != NULL && !IS_ERR(cobj))
5363 lfsck_object_put(env, cobj);
5365 if (likely(tgt != NULL))
5368 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5376 lfsck_assistant_object_put(env, lso);
5381 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
5382 * the OST-object's attribute and generate an structure lfsck_layout_req on the
5383 * list ::lad_req_list.
5385 * For each request on above list, the lfsck_layout_assistant thread compares
5386 * the OST side attribute with local attribute, if inconsistent, then repair it.
5388 * All above processing is async mode with pipeline. */
5389 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
5390 struct lfsck_component *com,
5391 struct dt_object *obj)
5393 struct lfsck_thread_info *info = lfsck_env_info(env);
5394 struct ost_id *oi = &info->lti_oi;
5395 struct lfsck_layout *lo = com->lc_file_ram;
5396 struct lfsck_assistant_data *lad = com->lc_data;
5397 struct lfsck_instance *lfsck = com->lc_lfsck;
5398 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5399 struct thandle *handle = NULL;
5400 struct lu_buf *buf = &info->lti_big_buf;
5401 struct lov_mds_md_v1 *lmm = NULL;
5402 struct dt_device *dev = lfsck_obj2dev(obj);
5403 struct lustre_handle lh = { 0 };
5404 struct lu_buf ea_buf = { NULL };
5405 struct lov_comp_md_v1 *lcm = NULL;
5406 struct lov_comp_md_entry_v1 *lcme = NULL;
5411 bool locked = false;
5412 bool stripe = false;
5413 bool bad_oi = false;
5416 if (!S_ISREG(lfsck_object_type(obj)))
5419 if (lad->lad_assistant_status < 0)
5420 GOTO(out, rc = -ESRCH);
5422 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
5423 lmm_oi_cpu_to_le(oi, oi);
5424 dt_read_lock(env, obj, 0);
5429 if (dt_object_exists(obj) == 0 ||
5430 lfsck_is_dead_obj(obj))
5433 rc = lfsck_layout_get_lovea(env, obj, buf);
5435 /* Skip bad lov EA during the 1st cycle scanning, and
5436 * try to recover it via orphan in the 2nd scanning. */
5437 GOTO(out, rc = (rc == -EINVAL ? 0 : rc));
5441 magic = le32_to_cpu(lmm->lmm_magic);
5442 if (magic == LOV_MAGIC_COMP_V1) {
5446 count = le16_to_cpu(lcm->lcm_entry_count);
5447 for (i = 0; i < count; i++) {
5448 lcme = &lcm->lcm_entries[i];
5449 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5450 if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) != 0)
5454 GOTO(out, stripe = true);
5455 } else if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0) {
5456 GOTO(out, stripe = true);
5460 /* Inconsistent lmm_oi, should be repaired. */
5463 if (bk->lb_param & LPF_DRYRUN) {
5464 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5466 GOTO(out, stripe = true);
5469 if (!lustre_handle_is_used(&lh)) {
5470 dt_read_unlock(env, obj);
5472 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5473 MDS_INODELOCK_LAYOUT |
5474 MDS_INODELOCK_XATTR, LCK_EX);
5478 handle = dt_trans_create(env, dev);
5480 GOTO(out, rc = PTR_ERR(handle));
5482 lfsck_buf_init(&ea_buf, lmm, size);
5483 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5484 LU_XATTR_REPLACE, handle);
5488 rc = dt_trans_start_local(env, dev, handle);
5492 dt_write_lock(env, obj, 0);
5498 if (magic == LOV_MAGIC_COMP_V1) {
5501 for (i = 0; i < count; i++) {
5502 lcme = &lcm->lcm_entries[i];
5503 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5510 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5511 LU_XATTR_REPLACE, handle);
5515 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5517 GOTO(out, stripe = true);
5521 if (lustre_handle_is_used(&lh))
5522 dt_write_unlock(env, obj);
5524 dt_read_unlock(env, obj);
5527 if (handle != NULL && !IS_ERR(handle))
5528 dt_trans_stop(env, dev, handle);
5530 lfsck_ibits_unlock(&lh, LCK_EX);
5533 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
5534 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
5535 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
5536 PFID(lfsck_dto2fid(obj)), rc);
5539 if (magic == LOV_MAGIC_COMP_V1) {
5542 for (i = 0; i < count; i++) {
5543 lcme = &lcm->lcm_entries[i];
5544 if (!(le32_to_cpu(lcme->lcme_flags) &
5548 rc = lfsck_layout_scan_stripes(env, com, obj,
5549 (struct lov_mds_md_v1 *)(buf->lb_buf +
5550 le32_to_cpu(lcme->lcme_offset)),
5551 le32_to_cpu(lcme->lcme_id));
5554 rc = lfsck_layout_scan_stripes(env, com, obj, lmm, 0);
5557 down_write(&com->lc_sem);
5558 com->lc_new_checked++;
5560 lfsck_layout_record_failure(env, lfsck, lo);
5561 up_write(&com->lc_sem);
5567 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
5568 struct lfsck_component *com,
5569 struct dt_object *obj)
5571 struct lfsck_instance *lfsck = com->lc_lfsck;
5572 struct lfsck_layout *lo = com->lc_file_ram;
5573 const struct lu_fid *fid = lfsck_dto2fid(obj);
5574 struct lfsck_layout_slave_data *llsd = com->lc_data;
5575 struct lfsck_layout_seq *lls;
5581 LASSERT(llsd != NULL);
5583 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
5584 cfs_fail_val == lfsck_dev_idx(lfsck)) {
5585 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(1),
5587 struct ptlrpc_thread *thread = &lfsck->li_thread;
5589 l_wait_event(thread->t_ctl_waitq,
5590 !thread_is_running(thread),
5594 lfsck_rbtree_update_bitmap(env, com, fid, false);
5596 down_write(&com->lc_sem);
5597 if (fid_is_idif(fid))
5599 else if (!fid_is_norm(fid) ||
5600 !fid_is_for_ostobj(env, lfsck, obj, fid))
5601 GOTO(unlock, rc = 0);
5604 com->lc_new_checked++;
5606 lls = lfsck_layout_seq_lookup(llsd, seq);
5609 if (unlikely(lls == NULL))
5610 GOTO(unlock, rc = -ENOMEM);
5612 INIT_LIST_HEAD(&lls->lls_list);
5614 rc = lfsck_layout_lastid_load(env, com, lls);
5616 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5617 "load LAST_ID for %#llx: rc = %d\n",
5618 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
5619 lo->ll_objs_failed_phase1++;
5624 lfsck_layout_seq_insert(llsd, lls);
5627 if (unlikely(fid_is_last_id(fid)))
5628 GOTO(unlock, rc = 0);
5630 if (fid_is_idif(fid))
5631 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
5635 if (oid > lls->lls_lastid_known)
5636 lls->lls_lastid_known = oid;
5638 if (oid > lls->lls_lastid) {
5639 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
5640 /* OFD may create new objects during LFSCK scanning. */
5641 rc = lfsck_layout_lastid_reload(env, com, lls);
5642 if (unlikely(rc != 0)) {
5643 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5644 "reload LAST_ID for %#llx: rc = %d\n",
5645 lfsck_lfsck2name(com->lc_lfsck),
5651 if (oid <= lls->lls_lastid ||
5652 lo->ll_flags & LF_CRASHED_LASTID)
5653 GOTO(unlock, rc = 0);
5655 LASSERT(lfsck->li_out_notify != NULL);
5657 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5658 LE_LASTID_REBUILDING);
5659 lo->ll_flags |= LF_CRASHED_LASTID;
5661 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
5662 "LAST_ID file (2) for the sequence %#llx"
5663 ", old value %llu, known value %llu\n",
5664 lfsck_lfsck2name(lfsck), lls->lls_seq,
5665 lls->lls_lastid, oid);
5668 lls->lls_lastid = oid;
5672 GOTO(unlock, rc = 0);
5675 up_write(&com->lc_sem);
5680 static int lfsck_layout_exec_dir(const struct lu_env *env,
5681 struct lfsck_component *com,
5682 struct lfsck_assistant_object *lso,
5683 struct lu_dirent *ent, __u16 type)
5688 static int lfsck_layout_master_post(const struct lu_env *env,
5689 struct lfsck_component *com,
5690 int result, bool init)
5692 struct lfsck_instance *lfsck = com->lc_lfsck;
5693 struct lfsck_layout *lo = com->lc_file_ram;
5697 lfsck_post_generic(env, com, &result);
5699 down_write(&com->lc_sem);
5700 spin_lock(&lfsck->li_lock);
5702 lo->ll_pos_last_checkpoint =
5703 lfsck->li_pos_checkpoint.lp_oit_cookie;
5706 if (lo->ll_flags & LF_INCOMPLETE)
5707 lo->ll_status = LS_PARTIAL;
5709 lo->ll_status = LS_SCANNING_PHASE2;
5710 lo->ll_flags |= LF_SCANNED_ONCE;
5711 lo->ll_flags &= ~LF_UPGRADE;
5712 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5713 } else if (result == 0) {
5714 if (lfsck->li_status != 0)
5715 lo->ll_status = lfsck->li_status;
5717 lo->ll_status = LS_STOPPED;
5718 if (lo->ll_status != LS_PAUSED)
5719 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5721 lo->ll_status = LS_FAILED;
5722 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5724 spin_unlock(&lfsck->li_lock);
5727 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5728 HALF_SEC - lfsck->li_time_last_checkpoint);
5729 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5730 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5731 com->lc_new_checked = 0;
5734 rc = lfsck_layout_store(env, com);
5735 up_write(&com->lc_sem);
5737 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
5738 lfsck_lfsck2name(lfsck), rc);
5743 static int lfsck_layout_slave_post(const struct lu_env *env,
5744 struct lfsck_component *com,
5745 int result, bool init)
5747 struct lfsck_instance *lfsck = com->lc_lfsck;
5748 struct lfsck_layout *lo = com->lc_file_ram;
5752 down_write(&com->lc_sem);
5753 rc = lfsck_layout_lastid_store(env, com);
5757 LASSERT(lfsck->li_out_notify != NULL);
5759 spin_lock(&lfsck->li_lock);
5761 lo->ll_pos_last_checkpoint =
5762 lfsck->li_pos_checkpoint.lp_oit_cookie;
5765 lo->ll_status = LS_SCANNING_PHASE2;
5766 lo->ll_flags |= LF_SCANNED_ONCE;
5767 if (lo->ll_flags & LF_CRASHED_LASTID) {
5769 lo->ll_flags &= ~LF_CRASHED_LASTID;
5771 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
5772 "crashed LAST_ID files successfully\n",
5773 lfsck_lfsck2name(lfsck));
5775 lo->ll_flags &= ~LF_UPGRADE;
5776 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5777 } else if (result == 0) {
5778 if (lfsck->li_status != 0)
5779 lo->ll_status = lfsck->li_status;
5781 lo->ll_status = LS_STOPPED;
5782 if (lo->ll_status != LS_PAUSED)
5783 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5785 lo->ll_status = LS_FAILED;
5786 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5788 spin_unlock(&lfsck->li_lock);
5791 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5795 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5796 HALF_SEC - lfsck->li_time_last_checkpoint);
5797 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5798 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5799 com->lc_new_checked = 0;
5802 rc = lfsck_layout_store(env, com);
5803 up_write(&com->lc_sem);
5805 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
5807 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
5808 lfsck_lfsck2name(lfsck), rc);
5813 static void lfsck_layout_dump(const struct lu_env *env,
5814 struct lfsck_component *com, struct seq_file *m)
5816 struct lfsck_instance *lfsck = com->lc_lfsck;
5817 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5818 struct lfsck_layout *lo = com->lc_file_ram;
5821 down_read(&com->lc_sem);
5822 if (bk->lb_param & LPF_DRYRUN)
5823 prefix = "inconsistent";
5825 prefix = "repaired";
5827 seq_printf(m, "name: lfsck_layout\n"
5833 lfsck_status2name(lo->ll_status));
5835 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
5837 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
5839 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
5841 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
5843 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
5845 seq_printf(m, "latest_start_position: %llu\n"
5846 "last_checkpoint_position: %llu\n"
5847 "first_failure_position: %llu\n",
5848 lo->ll_pos_latest_start,
5849 lo->ll_pos_last_checkpoint,
5850 lo->ll_pos_first_inconsistent);
5852 seq_printf(m, "success_count: %u\n"
5853 "%s_dangling: %llu\n"
5854 "%s_unmatched_pair: %llu\n"
5855 "%s_multiple_referenced: %llu\n"
5857 "%s_inconsistent_owner: %llu\n"
5860 "failed_phase1: %llu\n"
5861 "failed_phase2: %llu\n",
5862 lo->ll_success_count,
5863 prefix, lo->ll_objs_repaired[LLIT_DANGLING - 1],
5864 prefix, lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
5865 prefix, lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
5866 prefix, lo->ll_objs_repaired[LLIT_ORPHAN - 1],
5867 prefix, lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
5868 prefix, lo->ll_objs_repaired[LLIT_OTHERS - 1],
5869 lo->ll_objs_skipped,
5870 lo->ll_objs_failed_phase1,
5871 lo->ll_objs_failed_phase2);
5873 if (lo->ll_status == LS_SCANNING_PHASE1) {
5875 cfs_duration_t duration = cfs_time_current() -
5876 lfsck->li_time_last_checkpoint;
5877 __u64 checked = lo->ll_objs_checked_phase1 +
5878 com->lc_new_checked;
5879 __u64 speed = checked;
5880 __u64 new_checked = com->lc_new_checked *
5881 msecs_to_jiffies(MSEC_PER_SEC);
5882 __u32 rtime = lo->ll_run_time_phase1 +
5883 cfs_duration_sec(duration + HALF_SEC);
5886 do_div(new_checked, duration);
5888 do_div(speed, rtime);
5889 seq_printf(m, "checked_phase1: %llu\n"
5890 "checked_phase2: %llu\n"
5891 "run_time_phase1: %u seconds\n"
5892 "run_time_phase2: %u seconds\n"
5893 "average_speed_phase1: %llu items/sec\n"
5894 "average_speed_phase2: N/A\n"
5895 "real-time_speed_phase1: %llu items/sec\n"
5896 "real-time_speed_phase2: N/A\n",
5898 lo->ll_objs_checked_phase2,
5900 lo->ll_run_time_phase2,
5904 if (likely(lfsck->li_di_oit)) {
5905 const struct dt_it_ops *iops =
5906 &lfsck->li_obj_oit->do_index_ops->dio_it;
5908 /* The low layer otable-based iteration position may NOT
5909 * exactly match the layout-based directory traversal
5910 * cookie. Generally, it is not a serious issue. But the
5911 * caller should NOT make assumption on that. */
5912 pos = iops->store(env, lfsck->li_di_oit);
5913 if (!lfsck->li_current_oit_processed)
5916 pos = lo->ll_pos_last_checkpoint;
5919 seq_printf(m, "current_position: %llu\n", pos);
5920 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
5921 cfs_duration_t duration = cfs_time_current() -
5922 com->lc_time_last_checkpoint;
5923 __u64 checked = lo->ll_objs_checked_phase2 +
5924 com->lc_new_checked;
5925 __u64 speed1 = lo->ll_objs_checked_phase1;
5926 __u64 speed2 = checked;
5927 __u64 new_checked = com->lc_new_checked *
5928 msecs_to_jiffies(MSEC_PER_SEC);
5929 __u32 rtime = lo->ll_run_time_phase2 +
5930 cfs_duration_sec(duration + HALF_SEC);
5933 do_div(new_checked, duration);
5934 if (lo->ll_run_time_phase1 != 0)
5935 do_div(speed1, lo->ll_run_time_phase1);
5937 do_div(speed2, rtime);
5938 seq_printf(m, "checked_phase1: %llu\n"
5939 "checked_phase2: %llu\n"
5940 "run_time_phase1: %u seconds\n"
5941 "run_time_phase2: %u seconds\n"
5942 "average_speed_phase1: %llu items/sec\n"
5943 "average_speed_phase2: %llu items/sec\n"
5944 "real-time_speed_phase1: N/A\n"
5945 "real-time_speed_phase2: %llu items/sec\n"
5946 "current_position: "DFID"\n",
5947 lo->ll_objs_checked_phase1,
5949 lo->ll_run_time_phase1,
5954 PFID(&com->lc_fid_latest_scanned_phase2));
5956 __u64 speed1 = lo->ll_objs_checked_phase1;
5957 __u64 speed2 = lo->ll_objs_checked_phase2;
5959 if (lo->ll_run_time_phase1 != 0)
5960 do_div(speed1, lo->ll_run_time_phase1);
5961 if (lo->ll_run_time_phase2 != 0)
5962 do_div(speed2, lo->ll_run_time_phase2);
5963 seq_printf(m, "checked_phase1: %llu\n"
5964 "checked_phase2: %llu\n"
5965 "run_time_phase1: %u seconds\n"
5966 "run_time_phase2: %u seconds\n"
5967 "average_speed_phase1: %llu items/sec\n"
5968 "average_speed_phase2: %llu objs/sec\n"
5969 "real-time_speed_phase1: N/A\n"
5970 "real-time_speed_phase2: N/A\n"
5971 "current_position: N/A\n",
5972 lo->ll_objs_checked_phase1,
5973 lo->ll_objs_checked_phase2,
5974 lo->ll_run_time_phase1,
5975 lo->ll_run_time_phase2,
5980 up_read(&com->lc_sem);
5983 static int lfsck_layout_master_double_scan(const struct lu_env *env,
5984 struct lfsck_component *com)
5986 struct lfsck_layout *lo = com->lc_file_ram;
5987 struct lfsck_assistant_data *lad = com->lc_data;
5988 struct lfsck_instance *lfsck = com->lc_lfsck;
5989 struct lfsck_tgt_descs *ltds;
5990 struct lfsck_tgt_desc *ltd;
5991 struct lfsck_tgt_desc *next;
5994 rc = lfsck_double_scan_generic(env, com, lo->ll_status);
5996 if (thread_is_stopped(&lad->lad_thread)) {
5997 LASSERT(list_empty(&lad->lad_req_list));
5998 LASSERT(list_empty(&lad->lad_ost_phase1_list));
5999 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
6001 ltds = &lfsck->li_ost_descs;
6002 spin_lock(<ds->ltd_lock);
6003 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6004 ltd_layout_phase_list) {
6005 list_del_init(<d->ltd_layout_phase_list);
6007 spin_unlock(<ds->ltd_lock);
6009 ltds = &lfsck->li_mdt_descs;
6010 spin_lock(<ds->ltd_lock);
6011 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6012 ltd_layout_phase_list) {
6013 list_del_init(<d->ltd_layout_phase_list);
6015 spin_unlock(<ds->ltd_lock);
6021 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
6022 struct lfsck_component *com)
6024 struct lfsck_instance *lfsck = com->lc_lfsck;
6025 struct lfsck_layout_slave_data *llsd = com->lc_data;
6026 struct lfsck_layout *lo = com->lc_file_ram;
6027 struct ptlrpc_thread *thread = &lfsck->li_thread;
6031 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
6032 lfsck_lfsck2name(lfsck));
6034 atomic_inc(&lfsck->li_double_scan_count);
6036 if (lo->ll_flags & LF_INCOMPLETE)
6039 com->lc_new_checked = 0;
6040 com->lc_new_scanned = 0;
6041 com->lc_time_last_checkpoint = cfs_time_current();
6042 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6043 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6046 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
6049 rc = lfsck_layout_slave_query_master(env, com);
6050 if (list_empty(&llsd->llsd_master_list)) {
6051 if (unlikely(!thread_is_running(thread)))
6062 rc = l_wait_event(thread->t_ctl_waitq,
6063 !thread_is_running(thread) ||
6064 lo->ll_flags & LF_INCOMPLETE ||
6065 list_empty(&llsd->llsd_master_list),
6067 if (unlikely(!thread_is_running(thread)))
6070 if (lo->ll_flags & LF_INCOMPLETE)
6073 if (rc == -ETIMEDOUT)
6076 GOTO(done, rc = (rc < 0 ? rc : 1));
6080 rc = lfsck_layout_double_scan_result(env, com, rc);
6081 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
6082 (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
6083 lfsck_layout_slave_quit(env, com);
6084 if (atomic_dec_and_test(&lfsck->li_double_scan_count))
6085 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6087 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
6088 "status %d: rc = %d\n",
6089 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
6094 static void lfsck_layout_master_data_release(const struct lu_env *env,
6095 struct lfsck_component *com)
6097 struct lfsck_assistant_data *lad = com->lc_data;
6098 struct lfsck_instance *lfsck = com->lc_lfsck;
6099 struct lfsck_tgt_descs *ltds;
6100 struct lfsck_tgt_desc *ltd;
6101 struct lfsck_tgt_desc *next;
6103 LASSERT(lad != NULL);
6104 LASSERT(thread_is_init(&lad->lad_thread) ||
6105 thread_is_stopped(&lad->lad_thread));
6106 LASSERT(list_empty(&lad->lad_req_list));
6108 com->lc_data = NULL;
6110 ltds = &lfsck->li_ost_descs;
6111 spin_lock(<ds->ltd_lock);
6112 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6113 ltd_layout_phase_list) {
6114 list_del_init(<d->ltd_layout_phase_list);
6116 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6117 ltd_layout_phase_list) {
6118 list_del_init(<d->ltd_layout_phase_list);
6120 list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
6122 list_del_init(<d->ltd_layout_list);
6124 spin_unlock(<ds->ltd_lock);
6126 ltds = &lfsck->li_mdt_descs;
6127 spin_lock(<ds->ltd_lock);
6128 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6129 ltd_layout_phase_list) {
6130 list_del_init(<d->ltd_layout_phase_list);
6132 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6133 ltd_layout_phase_list) {
6134 list_del_init(<d->ltd_layout_phase_list);
6136 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
6138 list_del_init(<d->ltd_layout_list);
6140 spin_unlock(<ds->ltd_lock);
6142 if (likely(lad->lad_bitmap != NULL))
6143 CFS_FREE_BITMAP(lad->lad_bitmap);
6148 static void lfsck_layout_slave_data_release(const struct lu_env *env,
6149 struct lfsck_component *com)
6151 struct lfsck_layout_slave_data *llsd = com->lc_data;
6153 lfsck_layout_slave_quit(env, com);
6154 com->lc_data = NULL;
6158 static void lfsck_layout_master_quit(const struct lu_env *env,
6159 struct lfsck_component *com)
6161 struct lfsck_assistant_data *lad = com->lc_data;
6162 struct lfsck_instance *lfsck = com->lc_lfsck;
6163 struct lfsck_tgt_descs *ltds;
6164 struct lfsck_tgt_desc *ltd;
6165 struct lfsck_tgt_desc *next;
6167 LASSERT(lad != NULL);
6169 lfsck_quit_generic(env, com);
6171 LASSERT(thread_is_init(&lad->lad_thread) ||
6172 thread_is_stopped(&lad->lad_thread));
6173 LASSERT(list_empty(&lad->lad_req_list));
6175 ltds = &lfsck->li_ost_descs;
6176 spin_lock(<ds->ltd_lock);
6177 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6178 ltd_layout_phase_list) {
6179 list_del_init(<d->ltd_layout_phase_list);
6181 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6182 ltd_layout_phase_list) {
6183 list_del_init(<d->ltd_layout_phase_list);
6185 spin_unlock(<ds->ltd_lock);
6187 ltds = &lfsck->li_mdt_descs;
6188 spin_lock(<ds->ltd_lock);
6189 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6190 ltd_layout_phase_list) {
6191 list_del_init(<d->ltd_layout_phase_list);
6193 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6194 ltd_layout_phase_list) {
6195 list_del_init(<d->ltd_layout_phase_list);
6197 spin_unlock(<ds->ltd_lock);
6200 static void lfsck_layout_slave_quit(const struct lu_env *env,
6201 struct lfsck_component *com)
6203 struct lfsck_layout_slave_data *llsd = com->lc_data;
6204 struct lfsck_layout_seq *lls;
6205 struct lfsck_layout_seq *next;
6206 struct lfsck_layout_slave_target *llst;
6208 LASSERT(llsd != NULL);
6210 down_write(&com->lc_sem);
6211 list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
6213 list_del_init(&lls->lls_list);
6214 lfsck_object_put(env, lls->lls_lastid_obj);
6217 up_write(&com->lc_sem);
6219 spin_lock(&llsd->llsd_lock);
6220 while (!list_empty(&llsd->llsd_master_list)) {
6221 llst = list_entry(llsd->llsd_master_list.next,
6222 struct lfsck_layout_slave_target, llst_list);
6223 list_del_init(&llst->llst_list);
6224 spin_unlock(&llsd->llsd_lock);
6225 lfsck_layout_llst_put(llst);
6226 spin_lock(&llsd->llsd_lock);
6228 spin_unlock(&llsd->llsd_lock);
6230 lfsck_rbtree_cleanup(env, com);
6233 static int lfsck_layout_master_in_notify(const struct lu_env *env,
6234 struct lfsck_component *com,
6235 struct lfsck_request *lr)
6237 struct lfsck_instance *lfsck = com->lc_lfsck;
6238 struct lfsck_layout *lo = com->lc_file_ram;
6239 struct lfsck_assistant_data *lad = com->lc_data;
6240 struct lfsck_tgt_descs *ltds;
6241 struct lfsck_tgt_desc *ltd;
6245 if (lr->lr_event == LE_PAIRS_VERIFY) {
6248 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
6255 CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
6256 "from %s %x, status %d, flags %x, flags2 %x\n",
6257 lfsck_lfsck2name(lfsck), lr->lr_event,
6258 (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
6259 lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
6261 if (lr->lr_event != LE_PHASE1_DONE &&
6262 lr->lr_event != LE_PHASE2_DONE &&
6263 lr->lr_event != LE_PEER_EXIT)
6266 if (lr->lr_flags & LEF_FROM_OST)
6267 ltds = &lfsck->li_ost_descs;
6269 ltds = &lfsck->li_mdt_descs;
6270 spin_lock(<ds->ltd_lock);
6271 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
6273 spin_unlock(<ds->ltd_lock);
6278 list_del_init(<d->ltd_layout_phase_list);
6279 switch (lr->lr_event) {
6280 case LE_PHASE1_DONE:
6281 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
6282 if (lr->lr_flags2 & LF_INCOMPLETE) {
6283 if (lr->lr_flags & LEF_FROM_OST)
6284 lfsck_lad_set_bitmap(env, com,
6287 lo->ll_flags |= LF_INCOMPLETE;
6289 ltd->ltd_layout_done = 1;
6290 list_del_init(<d->ltd_layout_list);
6295 if (lr->lr_flags & LEF_FROM_OST) {
6296 if (list_empty(<d->ltd_layout_list))
6297 list_add_tail(<d->ltd_layout_list,
6298 &lad->lad_ost_list);
6299 list_add_tail(<d->ltd_layout_phase_list,
6300 &lad->lad_ost_phase2_list);
6302 if (list_empty(<d->ltd_layout_list))
6303 list_add_tail(<d->ltd_layout_list,
6304 &lad->lad_mdt_list);
6305 list_add_tail(<d->ltd_layout_phase_list,
6306 &lad->lad_mdt_phase2_list);
6309 case LE_PHASE2_DONE:
6310 ltd->ltd_layout_done = 1;
6311 if (!list_empty(<d->ltd_layout_list)) {
6312 list_del_init(<d->ltd_layout_list);
6313 if (lr->lr_flags2 & LF_INCOMPLETE) {
6314 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
6322 ltd->ltd_layout_done = 1;
6323 list_del_init(<d->ltd_layout_list);
6324 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
6325 !(lr->lr_flags & LEF_FROM_OST))
6326 lo->ll_flags |= LF_INCOMPLETE;
6331 spin_unlock(<ds->ltd_lock);
6333 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
6334 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6336 memset(stop, 0, sizeof(*stop));
6337 stop->ls_status = lr->lr_status;
6338 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6339 lfsck_stop(env, lfsck->li_bottom, stop);
6340 } else if (lfsck_phase2_next_ready(lad)) {
6341 wake_up_all(&lad->lad_thread.t_ctl_waitq);
6347 static int lfsck_layout_slave_in_notify_local(const struct lu_env *env,
6348 struct lfsck_component *com,
6349 struct lfsck_req_local *lrl,
6354 switch (lrl->lrl_event) {
6355 case LEL_FID_ACCESSED:
6356 lfsck_rbtree_update_bitmap(env, com, &lrl->lrl_fid, true);
6358 case LEL_PAIRS_VERIFY_LOCAL: {
6361 lrl->lrl_status = LPVS_INIT;
6362 /* Firstly, if the MDT-object which is claimed via OST-object
6363 * local stored PFID xattr recognizes the OST-object, then it
6364 * must be that the client given PFID is wrong. */
6365 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6366 &lrl->lrl_ff_local.ff_parent,
6367 lrl->lrl_ff_local.ff_layout.ol_comp_id);
6371 lrl->lrl_status = LPVS_INCONSISTENT;
6372 /* The OST-object local stored PFID xattr is stale. We need to
6373 * check whether the MDT-object that is claimed via the client
6374 * given PFID information recognizes the OST-object or not. If
6375 * matches, then need to update the OST-object's PFID xattr. */
6376 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6377 &lrl->lrl_ff_client.ff_parent,
6378 lrl->lrl_ff_client.ff_layout.ol_comp_id);
6380 * We are not sure whether the client given PFID information
6381 * is correct or not, do nothing to avoid improper fixing.
6384 * The client given PFID information is also invalid, we can
6385 * NOT fix the OST-object inconsistency.
6388 lrl->lrl_status = LPVS_INCONSISTENT_TOFIX;
6389 rc = lfsck_layout_slave_repair_pfid(env, com, lrl);
6398 RETURN(-EOPNOTSUPP);
6401 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
6402 struct lfsck_component *com,
6403 struct lfsck_request *lr)
6405 struct lfsck_instance *lfsck = com->lc_lfsck;
6406 struct lfsck_layout_slave_data *llsd = com->lc_data;
6407 struct lfsck_layout_slave_target *llst;
6411 switch (lr->lr_event) {
6412 case LE_CONDITIONAL_DESTROY:
6413 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
6415 case LE_PHASE1_DONE: {
6416 if (lr->lr_flags2 & LF_INCOMPLETE) {
6417 struct lfsck_layout *lo = com->lc_file_ram;
6419 lo->ll_flags |= LF_INCOMPLETE;
6420 llst = lfsck_layout_llst_find_and_del(llsd,
6424 lfsck_layout_llst_put(llst);
6425 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6431 case LE_PHASE2_DONE:
6433 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
6434 "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
6435 lr->lr_event, lr->lr_index, lr->lr_status);
6441 llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
6445 lfsck_layout_llst_put(llst);
6446 if (list_empty(&llsd->llsd_master_list))
6447 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6449 if (lr->lr_event == LE_PEER_EXIT &&
6450 (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
6451 (list_empty(&llsd->llsd_master_list) &&
6452 (lr->lr_status == LS_STOPPED ||
6453 lr->lr_status == LS_CO_STOPPED)))) {
6454 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6456 memset(stop, 0, sizeof(*stop));
6457 stop->ls_status = lr->lr_status;
6458 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6459 lfsck_stop(env, lfsck->li_bottom, stop);
6465 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
6469 for (i = 0; i < LLIT_MAX; i++)
6470 *count += lo->ll_objs_repaired[i];
6473 static int lfsck_layout_query_all(const struct lu_env *env,
6474 struct lfsck_component *com,
6475 __u32 *mdts_count, __u32 *osts_count,
6478 struct lfsck_layout *lo = com->lc_file_ram;
6479 struct lfsck_tgt_descs *ltds;
6480 struct lfsck_tgt_desc *ltd;
6485 rc = lfsck_query_all(env, com);
6489 ltds = &com->lc_lfsck->li_mdt_descs;
6490 down_read(<ds->ltd_rw_sem);
6491 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
6492 ltd = lfsck_ltd2tgt(ltds, idx);
6493 LASSERT(ltd != NULL);
6495 mdts_count[ltd->ltd_layout_status]++;
6496 *repaired += ltd->ltd_layout_repaired;
6498 up_read(<ds->ltd_rw_sem);
6500 ltds = &com->lc_lfsck->li_ost_descs;
6501 down_read(<ds->ltd_rw_sem);
6502 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
6503 ltd = lfsck_ltd2tgt(ltds, idx);
6504 LASSERT(ltd != NULL);
6506 osts_count[ltd->ltd_layout_status]++;
6507 *repaired += ltd->ltd_layout_repaired;
6509 up_read(<ds->ltd_rw_sem);
6511 down_read(&com->lc_sem);
6512 mdts_count[lo->ll_status]++;
6513 lfsck_layout_repaired(lo, repaired);
6514 up_read(&com->lc_sem);
6519 static int lfsck_layout_query(const struct lu_env *env,
6520 struct lfsck_component *com,
6521 struct lfsck_request *req,
6522 struct lfsck_reply *rep,
6523 struct lfsck_query *que, int idx)
6525 struct lfsck_layout *lo = com->lc_file_ram;
6529 LASSERT(com->lc_lfsck->li_master);
6531 rc = lfsck_layout_query_all(env, com,
6532 que->lu_mdts_count[idx],
6533 que->lu_osts_count[idx],
6534 &que->lu_repaired[idx]);
6536 down_read(&com->lc_sem);
6537 rep->lr_status = lo->ll_status;
6538 if (req->lr_flags & LEF_QUERY_ALL)
6539 lfsck_layout_repaired(lo, &rep->lr_repaired);
6540 up_read(&com->lc_sem);
6546 /* with lfsck::li_lock held */
6547 static int lfsck_layout_slave_join(const struct lu_env *env,
6548 struct lfsck_component *com,
6549 struct lfsck_start_param *lsp)
6551 struct lfsck_instance *lfsck = com->lc_lfsck;
6552 struct lfsck_layout_slave_data *llsd = com->lc_data;
6553 struct lfsck_layout_slave_target *llst;
6554 struct lfsck_start *start = lsp->lsp_start;
6558 if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
6561 if (!lsp->lsp_index_valid)
6564 /* If someone is running the LFSCK without orphan handling,
6565 * it will not maintain the object accessing rbtree. So we
6566 * cannot join it for orphan handling. */
6567 if (!llsd->llsd_rbtree_valid)
6570 spin_unlock(&lfsck->li_lock);
6571 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
6572 spin_lock(&lfsck->li_lock);
6573 if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
6574 spin_unlock(&lfsck->li_lock);
6575 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
6578 lfsck_layout_llst_put(llst);
6579 spin_lock(&lfsck->li_lock);
6586 static struct lfsck_operations lfsck_layout_master_ops = {
6587 .lfsck_reset = lfsck_layout_reset,
6588 .lfsck_fail = lfsck_layout_fail,
6589 .lfsck_checkpoint = lfsck_layout_master_checkpoint,
6590 .lfsck_prep = lfsck_layout_master_prep,
6591 .lfsck_exec_oit = lfsck_layout_master_exec_oit,
6592 .lfsck_exec_dir = lfsck_layout_exec_dir,
6593 .lfsck_post = lfsck_layout_master_post,
6594 .lfsck_dump = lfsck_layout_dump,
6595 .lfsck_double_scan = lfsck_layout_master_double_scan,
6596 .lfsck_data_release = lfsck_layout_master_data_release,
6597 .lfsck_quit = lfsck_layout_master_quit,
6598 .lfsck_in_notify = lfsck_layout_master_in_notify,
6599 .lfsck_query = lfsck_layout_query,
6602 static struct lfsck_operations lfsck_layout_slave_ops = {
6603 .lfsck_reset = lfsck_layout_reset,
6604 .lfsck_fail = lfsck_layout_fail,
6605 .lfsck_checkpoint = lfsck_layout_slave_checkpoint,
6606 .lfsck_prep = lfsck_layout_slave_prep,
6607 .lfsck_exec_oit = lfsck_layout_slave_exec_oit,
6608 .lfsck_exec_dir = lfsck_layout_exec_dir,
6609 .lfsck_post = lfsck_layout_slave_post,
6610 .lfsck_dump = lfsck_layout_dump,
6611 .lfsck_double_scan = lfsck_layout_slave_double_scan,
6612 .lfsck_data_release = lfsck_layout_slave_data_release,
6613 .lfsck_quit = lfsck_layout_slave_quit,
6614 .lfsck_in_notify_local = lfsck_layout_slave_in_notify_local,
6615 .lfsck_in_notify = lfsck_layout_slave_in_notify,
6616 .lfsck_query = lfsck_layout_query,
6617 .lfsck_join = lfsck_layout_slave_join,
6620 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
6621 struct lfsck_component *com,
6622 struct lfsck_position *pos)
6624 struct lfsck_assistant_data *lad = com->lc_data;
6625 struct lfsck_layout_req *llr;
6627 if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
6631 if (list_empty(&lad->lad_req_list))
6634 llr = list_entry(lad->lad_req_list.next,
6635 struct lfsck_layout_req,
6637 pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
6640 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
6641 .la_handler_p1 = lfsck_layout_assistant_handler_p1,
6642 .la_handler_p2 = lfsck_layout_assistant_handler_p2,
6643 .la_fill_pos = lfsck_layout_assistant_fill_pos,
6644 .la_double_scan_result = lfsck_layout_double_scan_result,
6645 .la_req_fini = lfsck_layout_assistant_req_fini,
6646 .la_sync_failures = lfsck_layout_assistant_sync_failures,
6649 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
6651 struct lfsck_component *com;
6652 struct lfsck_layout *lo;
6653 struct dt_object *root = NULL;
6654 struct dt_object *obj;
6663 INIT_LIST_HEAD(&com->lc_link);
6664 INIT_LIST_HEAD(&com->lc_link_dir);
6665 init_rwsem(&com->lc_sem);
6666 atomic_set(&com->lc_ref, 1);
6667 com->lc_lfsck = lfsck;
6668 com->lc_type = LFSCK_TYPE_LAYOUT;
6669 if (lfsck->li_master) {
6670 com->lc_ops = &lfsck_layout_master_ops;
6671 com->lc_data = lfsck_assistant_data_init(
6672 &lfsck_layout_assistant_ops,
6674 if (com->lc_data == NULL)
6675 GOTO(out, rc = -ENOMEM);
6677 for (i = 0; i < LFSCK_STF_COUNT; i++)
6678 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6680 struct lfsck_layout_slave_data *llsd;
6682 com->lc_ops = &lfsck_layout_slave_ops;
6683 OBD_ALLOC_PTR(llsd);
6685 GOTO(out, rc = -ENOMEM);
6687 INIT_LIST_HEAD(&llsd->llsd_seq_list);
6688 INIT_LIST_HEAD(&llsd->llsd_master_list);
6689 spin_lock_init(&llsd->llsd_lock);
6690 llsd->llsd_rb_root = RB_ROOT;
6691 rwlock_init(&llsd->llsd_rb_lock);
6692 com->lc_data = llsd;
6694 com->lc_file_size = sizeof(*lo);
6695 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6696 if (com->lc_file_ram == NULL)
6697 GOTO(out, rc = -ENOMEM);
6699 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6700 if (com->lc_file_disk == NULL)
6701 GOTO(out, rc = -ENOMEM);
6703 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6705 GOTO(out, rc = PTR_ERR(root));
6707 if (unlikely(!dt_try_as_dir(env, root)))
6708 GOTO(out, rc = -ENOTDIR);
6710 obj = local_file_find_or_create(env, lfsck->li_los, root,
6712 S_IFREG | S_IRUGO | S_IWUSR);
6714 GOTO(out, rc = PTR_ERR(obj));
6717 rc = lfsck_layout_load(env, com);
6719 rc = lfsck_layout_reset(env, com, true);
6720 else if (rc == -ENOENT)
6721 rc = lfsck_layout_init(env, com);
6722 else if (lfsck->li_master)
6723 rc = lfsck_load_sub_trace_files(env, com,
6724 &dt_lfsck_layout_dangling_features,
6725 LFSCK_LAYOUT, false);
6730 lo = com->lc_file_ram;
6731 switch (lo->ll_status) {
6737 spin_lock(&lfsck->li_lock);
6738 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6739 spin_unlock(&lfsck->li_lock);
6742 CERROR("%s: unknown lfsck_layout status %d\n",
6743 lfsck_lfsck2name(lfsck), lo->ll_status);
6745 case LS_SCANNING_PHASE1:
6746 case LS_SCANNING_PHASE2:
6747 /* No need to store the status to disk right now.
6748 * If the system crashed before the status stored,
6749 * it will be loaded back when next time. */
6750 lo->ll_status = LS_CRASHED;
6751 if (!lfsck->li_master)
6752 lo->ll_flags |= LF_INCOMPLETE;
6759 spin_lock(&lfsck->li_lock);
6760 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6761 spin_unlock(&lfsck->li_lock);
6765 if (lo->ll_flags & LF_CRASHED_LASTID) {
6766 LASSERT(lfsck->li_out_notify != NULL);
6768 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6769 LE_LASTID_REBUILDING);
6775 if (root != NULL && !IS_ERR(root))
6776 lfsck_object_put(env, root);
6779 lfsck_component_cleanup(env, com);
6780 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
6781 lfsck_lfsck2name(lfsck), rc);
6787 struct lfsck_orphan_it {
6788 struct lfsck_component *loi_com;
6789 struct lfsck_rbtree_node *loi_lrn;
6790 struct lfsck_layout_slave_target *loi_llst;
6791 struct lu_fid loi_key;
6792 struct lu_orphan_rec_v2 loi_rec;
6794 unsigned int loi_over:1;
6797 static int lfsck_fid_match_idx(const struct lu_env *env,
6798 struct lfsck_instance *lfsck,
6799 const struct lu_fid *fid, int idx)
6801 struct seq_server_site *ss;
6802 struct lu_server_fld *sf;
6803 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
6806 /* All abnormal cases will be returned to MDT0. */
6807 if (!fid_is_norm(fid)) {
6814 ss = lfsck_dev_site(lfsck);
6815 if (unlikely(ss == NULL))
6818 sf = ss->ss_server_fld;
6819 LASSERT(sf != NULL);
6821 fld_range_set_any(range);
6822 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
6826 if (!fld_range_is_mdt(range))
6829 if (range->lsr_index == idx)
6835 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
6836 struct dt_object *obj)
6838 struct dt_device *dev = lfsck_obj2dev(obj);
6839 struct thandle *handle;
6843 handle = dt_trans_create(env, dev);
6847 rc = dt_declare_ref_del(env, obj, handle);
6851 rc = dt_declare_destroy(env, obj, handle);
6855 rc = dt_trans_start_local(env, dev, handle);
6859 dt_write_lock(env, obj, 0);
6860 rc = dt_ref_del(env, obj, handle);
6862 rc = dt_destroy(env, obj, handle);
6863 dt_write_unlock(env, obj);
6868 dt_trans_stop(env, dev, handle);
6870 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
6871 PFID(lfsck_dto2fid(obj)), rc);
6876 static int lfsck_orphan_index_lookup(const struct lu_env *env,
6877 struct dt_object *dt,
6879 const struct dt_key *key)
6884 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
6885 struct dt_object *dt,
6886 const struct dt_rec *rec,
6887 const struct dt_key *key,
6888 struct thandle *handle)
6893 static int lfsck_orphan_index_insert(const struct lu_env *env,
6894 struct dt_object *dt,
6895 const struct dt_rec *rec,
6896 const struct dt_key *key,
6897 struct thandle *handle,
6903 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
6904 struct dt_object *dt,
6905 const struct dt_key *key,
6906 struct thandle *handle)
6911 static int lfsck_orphan_index_delete(const struct lu_env *env,
6912 struct dt_object *dt,
6913 const struct dt_key *key,
6914 struct thandle *handle)
6919 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
6920 struct dt_object *dt,
6923 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
6924 struct lfsck_instance *lfsck;
6925 struct lfsck_component *com = NULL;
6926 struct lfsck_layout_slave_data *llsd;
6927 struct lfsck_orphan_it *it = NULL;
6928 struct lfsck_layout *lo;
6932 lfsck = lfsck_instance_find(dev, true, false);
6933 if (unlikely(lfsck == NULL))
6934 RETURN(ERR_PTR(-ENXIO));
6936 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
6937 if (unlikely(com == NULL))
6938 GOTO(out, rc = -ENOENT);
6940 lo = com->lc_file_ram;
6941 if (lo->ll_flags & LF_INCOMPLETE)
6942 GOTO(out, rc = -ESRCH);
6944 llsd = com->lc_data;
6945 if (!llsd->llsd_rbtree_valid)
6946 GOTO(out, rc = -ESRCH);
6950 GOTO(out, rc = -ENOMEM);
6952 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
6953 if (it->loi_llst == NULL)
6954 GOTO(out, rc = -ENXIO);
6956 if (dev->dd_record_fid_accessed) {
6957 /* The first iteration against the rbtree, scan the whole rbtree
6958 * to remove the nodes which do NOT need to be handled. */
6959 write_lock(&llsd->llsd_rb_lock);
6960 if (dev->dd_record_fid_accessed) {
6961 struct rb_node *node;
6962 struct rb_node *next;
6963 struct lfsck_rbtree_node *lrn;
6965 /* No need to record the fid accessing anymore. */
6966 dev->dd_record_fid_accessed = 0;
6968 node = rb_first(&llsd->llsd_rb_root);
6969 while (node != NULL) {
6970 next = rb_next(node);
6971 lrn = rb_entry(node, struct lfsck_rbtree_node,
6973 if (atomic_read(&lrn->lrn_known_count) <=
6974 atomic_read(&lrn->lrn_accessed_count)) {
6975 rb_erase(node, &llsd->llsd_rb_root);
6976 lfsck_rbtree_free(lrn);
6981 write_unlock(&llsd->llsd_rb_lock);
6984 /* read lock the rbtree when init, and unlock when fini */
6985 read_lock(&llsd->llsd_rb_lock);
6993 lfsck_component_put(env, com);
6995 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
6996 lfsck_lfsck2name(lfsck), rc);
6998 lfsck_instance_put(env, lfsck);
7003 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
7006 return (struct dt_it *)it;
7009 static void lfsck_orphan_it_fini(const struct lu_env *env,
7012 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7013 struct lfsck_component *com = it->loi_com;
7014 struct lfsck_layout_slave_data *llsd;
7015 struct lfsck_layout_slave_target *llst;
7018 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
7019 lfsck_lfsck2name(com->lc_lfsck));
7021 llsd = com->lc_data;
7022 read_unlock(&llsd->llsd_rb_lock);
7023 llst = it->loi_llst;
7024 LASSERT(llst != NULL);
7026 /* Save the key and hash for iterate next. */
7027 llst->llst_fid = it->loi_key;
7028 llst->llst_hash = it->loi_hash;
7029 lfsck_layout_llst_put(llst);
7030 lfsck_component_put(env, com);
7036 * \retval +1: the iteration finished
7037 * \retval 0: on success, not finished
7038 * \retval -ve: on error
7040 static int lfsck_orphan_it_next(const struct lu_env *env,
7043 struct lfsck_thread_info *info = lfsck_env_info(env);
7044 struct filter_fid *ff = &info->lti_ff;
7045 struct lu_attr *la = &info->lti_la;
7046 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7047 struct lu_fid *key = &it->loi_key;
7048 struct lu_orphan_rec_v2 *rec = &it->loi_rec;
7049 struct ost_layout *ol = &rec->lor_layout;
7050 struct lfsck_component *com = it->loi_com;
7051 struct lfsck_instance *lfsck = com->lc_lfsck;
7052 struct lfsck_layout_slave_data *llsd = com->lc_data;
7053 struct dt_object *obj;
7054 struct lfsck_rbtree_node *lrn;
7058 __u32 idx = it->loi_llst->llst_index;
7068 lrn = lfsck_rbtree_search(llsd, key, &exact);
7076 key->f_seq = lrn->lrn_seq;
7077 key->f_oid = lrn->lrn_first_oid;
7082 if (unlikely(key->f_oid == 0)) {
7089 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
7095 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
7096 atomic_read(&lrn->lrn_accessed_count))) {
7097 struct rb_node *next = rb_next(&lrn->lrn_node);
7099 while (next != NULL) {
7100 lrn = rb_entry(next, struct lfsck_rbtree_node,
7102 if (atomic_read(&lrn->lrn_known_count) >
7103 atomic_read(&lrn->lrn_accessed_count))
7105 next = rb_next(next);
7114 key->f_seq = lrn->lrn_seq;
7115 key->f_oid = lrn->lrn_first_oid;
7119 pos = key->f_oid - lrn->lrn_first_oid;
7122 pos = find_next_bit(lrn->lrn_known_bitmap,
7123 LFSCK_RBTREE_BITMAP_WIDTH, pos);
7124 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
7125 key->f_oid = lrn->lrn_first_oid + pos;
7126 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
7134 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
7139 key->f_oid = lrn->lrn_first_oid + pos;
7140 obj = lfsck_object_find_bottom(env, lfsck, key);
7143 if (rc == -ENOENT) {
7150 dt_read_lock(env, obj, 0);
7151 if (dt_object_exists(obj) == 0 ||
7152 lfsck_is_dead_obj(obj)) {
7153 dt_read_unlock(env, obj);
7154 lfsck_object_put(env, obj);
7159 rc = dt_attr_get(env, obj, la);
7163 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, ff, sizeof(*ff)),
7165 if (rc == -ENODATA) {
7166 /* For the pre-created OST-object, update the bitmap to avoid
7167 * others LFSCK (second phase) iteration to touch it again. */
7168 if (la->la_ctime == 0) {
7169 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
7170 atomic_inc(&lrn->lrn_accessed_count);
7172 /* For the race between repairing dangling referenced
7173 * MDT-object and unlink the file, it may left orphan
7174 * OST-object there. Destroy it now! */
7175 if (unlikely(!(la->la_mode & S_ISUID))) {
7176 dt_read_unlock(env, obj);
7177 lfsck_layout_destroy_orphan(env, obj);
7178 lfsck_object_put(env, obj);
7182 } else if (idx == 0) {
7183 /* If the orphan OST-object has no parent information,
7184 * regard it as referenced by the MDT-object on MDT0. */
7185 fid_zero(&rec->lor_rec.lor_fid);
7186 rec->lor_rec.lor_uid = la->la_uid;
7187 rec->lor_rec.lor_gid = la->la_gid;
7188 memset(ol, 0, sizeof(*ol));
7193 dt_read_unlock(env, obj);
7194 lfsck_object_put(env, obj);
7199 if (rc < sizeof(struct lu_fid))
7200 GOTO(out, rc = (rc < 0 ? rc : -EINVAL));
7202 fid_le_to_cpu(&rec->lor_rec.lor_fid, &ff->ff_parent);
7203 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
7204 * MDT-object's FID::f_ver, instead it is the OST-object index in its
7205 * parent MDT-object's layout EA. */
7206 save = rec->lor_rec.lor_fid.f_stripe_idx;
7207 rec->lor_rec.lor_fid.f_ver = 0;
7208 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_rec.lor_fid, idx);
7209 /* If the orphan OST-object does not claim the MDT, then next.
7211 * If we do not know whether it matches or not, then return it
7212 * to the MDT for further check. */
7214 dt_read_unlock(env, obj);
7215 lfsck_object_put(env, obj);
7220 rec->lor_rec.lor_fid.f_stripe_idx = save;
7221 rec->lor_rec.lor_uid = la->la_uid;
7222 rec->lor_rec.lor_gid = la->la_gid;
7223 ost_layout_le_to_cpu(ol, &ff->ff_layout);
7225 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u, "
7226 "stripe size %u, stripe count %u, COMP id %u, COMP start %llu, "
7227 "COMP end %llu\n", lfsck_lfsck2name(com->lc_lfsck), PFID(key),
7228 PFID(&rec->lor_rec.lor_fid), rec->lor_rec.lor_uid,
7229 rec->lor_rec.lor_gid, ol->ol_stripe_size, ol->ol_stripe_count,
7230 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end);
7235 dt_read_unlock(env, obj);
7236 lfsck_object_put(env, obj);
7244 * \retval +1: locate to the exactly position
7245 * \retval 0: cannot locate to the exactly position,
7246 * call next() to move to a valid position.
7247 * \retval -ve: on error
7249 static int lfsck_orphan_it_get(const struct lu_env *env,
7251 const struct dt_key *key)
7253 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7256 it->loi_key = *(struct lu_fid *)key;
7257 rc = lfsck_orphan_it_next(env, di);
7267 static void lfsck_orphan_it_put(const struct lu_env *env,
7272 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
7273 const struct dt_it *di)
7275 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7277 return (struct dt_key *)&it->loi_key;
7280 static int lfsck_orphan_it_key_size(const struct lu_env *env,
7281 const struct dt_it *di)
7283 return sizeof(struct lu_fid);
7286 static int lfsck_orphan_it_rec(const struct lu_env *env,
7287 const struct dt_it *di,
7291 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7293 *(struct lu_orphan_rec_v2 *)rec = it->loi_rec;
7298 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
7299 const struct dt_it *di)
7301 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7303 return it->loi_hash;
7307 * \retval +1: locate to the exactly position
7308 * \retval 0: cannot locate to the exactly position,
7309 * call next() to move to a valid position.
7310 * \retval -ve: on error
7312 static int lfsck_orphan_it_load(const struct lu_env *env,
7313 const struct dt_it *di,
7316 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7317 struct lfsck_layout_slave_target *llst = it->loi_llst;
7320 LASSERT(llst != NULL);
7322 if (hash != llst->llst_hash) {
7323 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
7324 "iteration does not match the one when fini "
7325 "%llu, to be reset.\n",
7326 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
7328 fid_zero(&llst->llst_fid);
7329 llst->llst_hash = 0;
7332 it->loi_key = llst->llst_fid;
7333 it->loi_hash = llst->llst_hash;
7334 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
7344 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
7345 const struct dt_it *di,
7351 const struct dt_index_operations lfsck_orphan_index_ops = {
7352 .dio_lookup = lfsck_orphan_index_lookup,
7353 .dio_declare_insert = lfsck_orphan_index_declare_insert,
7354 .dio_insert = lfsck_orphan_index_insert,
7355 .dio_declare_delete = lfsck_orphan_index_declare_delete,
7356 .dio_delete = lfsck_orphan_index_delete,
7358 .init = lfsck_orphan_it_init,
7359 .fini = lfsck_orphan_it_fini,
7360 .get = lfsck_orphan_it_get,
7361 .put = lfsck_orphan_it_put,
7362 .next = lfsck_orphan_it_next,
7363 .key = lfsck_orphan_it_key,
7364 .key_size = lfsck_orphan_it_key_size,
7365 .rec = lfsck_orphan_it_rec,
7366 .store = lfsck_orphan_it_store,
7367 .load = lfsck_orphan_it_load,
7368 .key_rec = lfsck_orphan_it_key_rec,