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 CDEBUG(D_LFSCK, "%s: layout LFSCK double scan: rc = %d\n",
1445 lfsck_lfsck2name(lfsck), rc);
1447 down_write(&com->lc_sem);
1448 lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1449 HALF_SEC - com->lc_time_last_checkpoint);
1450 lo->ll_time_last_checkpoint = cfs_time_current_sec();
1451 lo->ll_objs_checked_phase2 += com->lc_new_checked;
1454 if (lo->ll_flags & LF_INCOMPLETE) {
1455 lo->ll_status = LS_PARTIAL;
1457 if (lfsck->li_master) {
1458 struct lfsck_assistant_data *lad = com->lc_data;
1460 if (lad->lad_incomplete)
1461 lo->ll_status = LS_PARTIAL;
1463 lo->ll_status = LS_COMPLETED;
1465 lo->ll_status = LS_COMPLETED;
1468 lo->ll_flags &= ~LF_SCANNED_ONCE;
1469 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1470 lo->ll_flags &= ~LF_INCONSISTENT;
1471 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1472 lo->ll_success_count++;
1473 } else if (rc == 0) {
1474 if (lfsck->li_status != 0)
1475 lo->ll_status = lfsck->li_status;
1477 lo->ll_status = LS_STOPPED;
1479 lo->ll_status = LS_FAILED;
1482 rc = lfsck_layout_store(env, com);
1483 up_write(&com->lc_sem);
1485 CDEBUG(D_LFSCK, "%s: layout LFSCK double scan result %u: rc = %d\n",
1486 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
1491 static int lfsck_layout_trans_stop(const struct lu_env *env,
1492 struct dt_device *dev,
1493 struct thandle *handle, int result)
1497 /* XXX: If there is something worng or it needs to repair nothing,
1498 * then notify the lower to stop the modification. Currently,
1499 * we use th_result for such purpose, that may be replaced by
1500 * some rollback mechanism in the future. */
1501 handle->th_result = result;
1502 rc = dt_trans_stop(env, dev, handle);
1504 return result > 0 ? 0 : result;
1506 return rc == 0 ? 1 : rc;
1509 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1510 struct lfsck_component *com,
1511 const struct lu_fid *pfid,
1512 const struct lu_fid *cfid,
1513 __u32 comp_id, __u32 ea_off,
1516 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1517 struct lu_fid *rec = &lfsck_env_info(env)->lti_fid3;
1518 struct dt_device *dev;
1519 struct dt_object *obj;
1520 struct thandle *th = NULL;
1525 idx = lfsck_sub_trace_file_fid2idx(pfid);
1526 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1527 dev = lfsck_obj2dev(obj);
1529 fid_cpu_to_be(&key->lldk_fid, pfid);
1530 key->lldk_comp_id = cpu_to_be32(comp_id);
1531 key->lldk_ea_off = cpu_to_be32(ea_off);
1533 fid_cpu_to_be(rec, cfid);
1534 rec->f_ver = cpu_to_be32(ost_idx);
1536 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1538 th = dt_trans_create(env, dev);
1540 GOTO(unlock, rc = PTR_ERR(th));
1542 rc = dt_declare_insert(env, obj,
1543 (const struct dt_rec *)rec,
1544 (const struct dt_key *)key, th);
1548 rc = dt_trans_start_local(env, dev, th);
1552 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1553 (const struct dt_key *)key, th, 1);
1558 if (th && !IS_ERR(th))
1559 dt_trans_stop(env, dev, th);
1561 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1563 CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", comp_id = %u, "
1564 "ea_off = %u, ost_idx = %u, into the trace file for further "
1565 "dangling check: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1566 PFID(pfid), PFID(cfid), comp_id, ea_off, ost_idx, rc);
1571 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1572 struct lfsck_component *com,
1573 const struct lu_fid *fid,
1574 __u32 comp_id, __u32 ea_off)
1576 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1577 struct dt_device *dev;
1578 struct dt_object *obj;
1579 struct thandle *th = NULL;
1584 idx = lfsck_sub_trace_file_fid2idx(fid);
1585 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1586 dev = lfsck_obj2dev(obj);
1588 fid_cpu_to_be(&key->lldk_fid, fid);
1589 key->lldk_comp_id = cpu_to_be32(comp_id);
1590 key->lldk_ea_off = cpu_to_be32(ea_off);
1592 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1594 th = dt_trans_create(env, dev);
1596 GOTO(unlock, rc = PTR_ERR(th));
1598 rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1602 rc = dt_trans_start_local(env, dev, th);
1606 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1611 if (th && !IS_ERR(th))
1612 dt_trans_stop(env, dev, th);
1614 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1616 CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1617 ", comp_id = %u, ea_off = %u from the trace file: rc = %d\n",
1618 lfsck_lfsck2name(com->lc_lfsck), PFID(fid), comp_id, ea_off, rc);
1624 * Get the system default stripe size.
1626 * \param[in] env pointer to the thread context
1627 * \param[in] lfsck pointer to the lfsck instance
1628 * \param[out] size pointer to the default stripe size
1630 * \retval 0 for success
1631 * \retval negative error number on failure
1633 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1634 struct lfsck_instance *lfsck,
1637 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1638 struct dt_object *root;
1641 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1643 return PTR_ERR(root);
1645 /* Get the default stripe size via xattr_get on the backend root. */
1646 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1649 /* The lum->lmm_stripe_size is LE mode. The *size also
1650 * should be LE mode. So it is unnecessary to convert. */
1651 *size = lum->lmm_stripe_size;
1653 } else if (unlikely(rc == 0)) {
1657 lfsck_object_put(env, root);
1663 * \retval +1: repaired
1664 * \retval 0: did nothing
1665 * \retval -ve: on error
1667 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1668 struct lfsck_instance *lfsck,
1669 struct thandle *handle,
1670 struct dt_object *parent,
1671 const struct lu_fid *cfid,
1673 struct lov_mds_md_v1 *lmm,
1674 struct lov_ost_data_v1 *slot,
1675 int fl, __u32 ost_idx, int size)
1677 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1678 struct lu_buf ea_buf;
1685 magic = le32_to_cpu(lmm->lmm_magic);
1686 pattern = le32_to_cpu(lmm->lmm_pattern);
1687 count = le16_to_cpu(lmm->lmm_stripe_count);
1689 fid_to_ostid(cfid, oi);
1690 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1691 slot->l_ost_gen = cpu_to_le32(0);
1692 slot->l_ost_idx = cpu_to_le32(ost_idx);
1694 if (pattern & LOV_PATTERN_F_HOLE) {
1695 struct lov_ost_data_v1 *objs;
1698 if (magic == LOV_MAGIC_V1)
1699 objs = &lmm->lmm_objects[0];
1701 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1702 for (i = 0; i < count; i++, objs++) {
1703 if (lovea_slot_is_dummy(objs))
1707 /* If the @slot is the last dummy slot to be refilled,
1708 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1711 cpu_to_le32(pattern & ~LOV_PATTERN_F_HOLE);
1713 CDEBUG(D_LFSCK, "%s: remove layout HOLE for "DFID
1714 ": parent "DFID"\n", lfsck_lfsck2name(lfsck),
1715 PFID(cfid), PFID(lfsck_dto2fid(parent)));
1719 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
1720 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1727 static struct lov_ost_data_v1 *
1728 __lfsck_layout_new_v1_lovea(struct lov_mds_md_v1 *lmm,
1729 const struct lu_fid *pfid,
1730 __u32 stripe_size, __u32 ea_off,
1731 __u32 pattern, __u16 count)
1733 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1734 lmm->lmm_pattern = cpu_to_le32(pattern);
1735 fid_to_lmm_oi(pfid, &lmm->lmm_oi);
1736 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1737 lmm->lmm_stripe_size = cpu_to_le32(stripe_size);
1738 lmm->lmm_stripe_count = cpu_to_le16(count);
1739 lmm->lmm_layout_gen = cpu_to_le16(1);
1740 memset(&lmm->lmm_objects[0], 0,
1741 sizeof(struct lov_ost_data_v1) * count);
1743 return &lmm->lmm_objects[ea_off];
1746 static int lfsck_layout_new_v1_lovea(const struct lu_env *env,
1747 struct lfsck_instance *lfsck,
1748 struct ost_layout *ol,
1749 struct dt_object *parent,
1750 struct lu_buf *buf, __u32 ea_off,
1751 struct lov_mds_md_v1 **lmm,
1752 struct lov_ost_data_v1 **objs)
1755 __u32 stripe_size = ol->ol_stripe_size;
1756 __u32 pattern = LOV_PATTERN_RAID0;
1759 if (ol->ol_stripe_count != 0)
1760 count = ol->ol_stripe_count;
1764 size = lov_mds_md_size(count, LOV_MAGIC_V1);
1765 LASSERTF(buf->lb_len >= size,
1766 "buffer len %d is less than real size %d\n",
1767 (int)buf->lb_len, size);
1769 if (stripe_size == 0) {
1772 rc = lfsck_layout_get_def_stripesize(env, lfsck, &stripe_size);
1778 if (ol->ol_stripe_count > 1 ||
1779 (ol->ol_stripe_count == 0 && ea_off != 0)) {
1780 pattern |= LOV_PATTERN_F_HOLE;
1781 memset(&(*lmm)->lmm_objects[0], 0,
1782 count * sizeof(struct lov_ost_data_v1));
1785 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1786 stripe_size, ea_off, pattern, count);
1791 static int lfsck_layout_new_comp_lovea(const struct lu_env *env,
1792 struct ost_layout *ol,
1793 struct dt_object *parent,
1794 struct lu_buf *buf, __u32 ea_off,
1795 struct lov_mds_md_v1 **lmm,
1796 struct lov_ost_data_v1 **objs)
1798 struct lov_comp_md_v1 *lcm;
1799 struct lov_comp_md_entry_v1 *lcme;
1800 __u32 pattern = LOV_PATTERN_RAID0;
1801 __u32 offset = sizeof(*lcm) + sizeof(*lcme);
1802 int lcme_size = lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1803 int size = offset + lcme_size;
1805 LASSERTF(buf->lb_len >= size,
1806 "buffer len %d is less than real size %d\n",
1807 (int)buf->lb_len, size);
1810 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1811 lcm->lcm_size = cpu_to_le32(size);
1812 lcm->lcm_layout_gen = cpu_to_le32(1);
1814 lcm->lcm_entry_count = cpu_to_le16(1);
1816 lcme = &lcm->lcm_entries[0];
1817 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1818 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1819 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1820 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1821 lcme->lcme_offset = cpu_to_le32(offset);
1822 lcme->lcme_size = cpu_to_le32(lcme_size);
1823 if (ol->ol_stripe_count > 1)
1824 pattern |= LOV_PATTERN_F_HOLE;
1826 *lmm = buf->lb_buf + offset;
1827 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1828 ol->ol_stripe_size, ea_off,
1829 pattern, ol->ol_stripe_count);
1834 static int lfsck_layout_add_comp_comp(const struct lu_env *env,
1835 struct lfsck_instance *lfsck,
1836 struct thandle *handle,
1837 struct ost_layout *ol,
1838 struct dt_object *parent,
1839 const struct lu_fid *cfid,
1840 struct lu_buf *buf, __u32 ost_idx,
1841 __u32 ea_off, int pos)
1843 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1844 struct lov_comp_md_entry_v1 *lcme;
1845 struct lov_mds_md_v1 *lmm;
1846 struct lov_ost_data_v1 *objs;
1847 int added = sizeof(*lcme) +
1848 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1849 int size = le32_to_cpu(lcm->lcm_size) + added;
1853 __u32 pattern = LOV_PATTERN_RAID0;
1854 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1857 lu_buf_check_and_grow(buf, size);
1858 /* set the lcm again because lu_buf_check_and_grow() may
1859 * have reallocated the buf. */
1861 lcm->lcm_size = cpu_to_le32(size);
1862 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1863 lcm->lcm_entry_count = cpu_to_le16(count + 1);
1865 /* 1. Move the component bodies from [pos, count-1] to [pos+1, count]
1866 * with distance of 'added'. */
1869 for (i = pos; i < count; i++) {
1870 lcme = &lcm->lcm_entries[i];
1871 size += le32_to_cpu(lcme->lcme_size);
1874 offset = le32_to_cpu(lcm->lcm_entries[pos].lcme_offset);
1875 memmove(buf->lb_buf + offset + added,
1876 buf->lb_buf + offset, size);
1880 /* 2. Move the component header [0, pos-1] to [0, pos-1] with distance
1881 * of 'sizeof(struct lov_comp_md_entry_v1)' */
1883 for (i = 0; i < pos; i++) {
1884 lcme = &lcm->lcm_entries[i];
1885 size += le32_to_cpu(lcme->lcme_size);
1888 offset = le32_to_cpu(lcm->lcm_entries[0].lcme_offset);
1889 memmove(buf->lb_buf + offset + sizeof(*lcme),
1890 buf->lb_buf + offset, size);
1893 /* 3. Recalculate the enter offset for the component [pos, count-1] */
1894 for (i = count - 1; i >= pos; i--) {
1895 lcm->lcm_entries[i + 1] = lcm->lcm_entries[i];
1896 lcm->lcm_entries[i + 1].lcme_offset =
1897 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i + 1].
1898 lcme_offset) + added);
1901 /* 4. Recalculate the enter offset for the component [0, pos) */
1902 for (i = 0; i < pos; i++) {
1903 lcm->lcm_entries[i].lcme_offset =
1904 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i].
1905 lcme_offset) + sizeof(*lcme));
1908 offset = sizeof(*lcm) + sizeof(*lcme) * (count + 1) + size;
1909 /* 4. Insert the new component header (entry) at the slot 'pos'. */
1910 lcme = &lcm->lcm_entries[pos];
1911 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1912 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1913 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1914 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1915 lcme->lcme_offset = cpu_to_le32(offset);
1916 lcme->lcme_size = cpu_to_le32(lov_mds_md_size(ol->ol_stripe_count,
1919 if (ol->ol_stripe_count > 1)
1920 pattern |= LOV_PATTERN_F_HOLE;
1922 lmm = buf->lb_buf + offset;
1923 /* 5. Insert teh new component body at the 'offset'. */
1924 objs = __lfsck_layout_new_v1_lovea(lmm, lfsck_dto2fid(parent),
1925 ol->ol_stripe_size, ea_off,
1926 pattern, ol->ol_stripe_count);
1928 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1929 lmm, objs, LU_XATTR_REPLACE, ost_idx,
1930 le32_to_cpu(lcm->lcm_size));
1932 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant add new COMP for "
1933 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1934 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
1935 "comp_end %llu, %s LOV EA hole: rc = %d\n",
1936 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1937 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
1938 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
1939 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
1940 "with" : "without", rc);
1945 static int lfsck_layout_extend_v1v3_lovea(const struct lu_env *env,
1946 struct lfsck_instance *lfsck,
1947 struct thandle *handle,
1948 struct ost_layout *ol,
1949 struct dt_object *parent,
1950 const struct lu_fid *cfid,
1951 struct lu_buf *buf, __u32 ost_idx,
1954 struct lov_mds_md_v1 *lmm = buf->lb_buf;
1955 struct lov_ost_data_v1 *objs;
1956 __u16 count = le16_to_cpu(lmm->lmm_stripe_count);
1957 __u32 magic = le32_to_cpu(lmm->lmm_magic);
1963 /* The original LOVEA maybe re-generated via old filter_fid, at
1964 * that time, we do not know the stripe count and stripe size. */
1965 if (ol->ol_stripe_count > count)
1966 count = ol->ol_stripe_count;
1967 if (ol->ol_stripe_size != 0 &&
1968 ol->ol_stripe_size != le32_to_cpu(lmm->lmm_stripe_size))
1969 lmm->lmm_stripe_size = cpu_to_le32(ol->ol_stripe_size);
1971 if (magic == LOV_MAGIC_V1)
1972 objs = &lmm->lmm_objects[count];
1974 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[count];
1976 gap = ea_off - count;
1980 size = lov_mds_md_size(count, magic);
1981 LASSERTF(buf->lb_len >= size,
1982 "buffer len %d is less than real size %d\n",
1983 (int)buf->lb_len, size);
1986 memset(objs, 0, gap * sizeof(*objs));
1987 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1990 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1991 lmm->lmm_stripe_count = cpu_to_le16(count);
1994 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
1995 lmm, objs, LU_XATTR_REPLACE, ost_idx, size);
1997 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1998 DFID": parent "DFID", OST-index %u, stripe-index %u, "
1999 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2000 "comp_end %llu, %s LOV EA hole: rc = %d\n",
2001 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2002 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2003 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2004 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2005 "with" : "without", rc);
2011 * \retval +1: repaired
2012 * \retval 0: did nothing
2013 * \retval -ve: on error
2015 static int lfsck_layout_update_lovea(const struct lu_env *env,
2016 struct lfsck_instance *lfsck,
2017 struct thandle *handle,
2018 struct ost_layout *ol,
2019 struct dt_object *parent,
2020 const struct lu_fid *cfid,
2021 struct lu_buf *buf, int fl,
2022 __u32 ost_idx, __u32 ea_off)
2024 struct lov_mds_md_v1 *lmm = NULL;
2025 struct lov_ost_data_v1 *objs = NULL;
2029 if (ol->ol_comp_id != 0)
2030 rc = lfsck_layout_new_comp_lovea(env, ol, parent, buf, ea_off,
2033 rc = lfsck_layout_new_v1_lovea(env, lfsck, ol, parent, buf,
2034 ea_off, &lmm, &objs);
2037 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid,
2038 buf, lmm, objs, fl, ost_idx, rc);
2040 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant created layout EA for "
2041 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2042 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2043 "comp_end %llu, fl %d, %s LOV EA hole: rc = %d\n",
2044 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2045 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2046 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end, fl,
2047 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2048 "with" : "without", rc);
2053 static int __lfsck_layout_update_pfid(const struct lu_env *env,
2054 struct dt_object *child,
2055 const struct lu_fid *pfid,
2056 const struct ost_layout *ol, __u32 offset)
2058 struct dt_device *dev = lfsck_obj2dev(child);
2059 struct filter_fid *ff = &lfsck_env_info(env)->lti_ff;
2060 struct thandle *handle;
2061 struct lu_buf buf = { NULL };
2064 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2065 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2066 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2067 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2068 * parent MDT-object's layout EA. */
2069 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
2070 ost_layout_cpu_to_le(&ff->ff_layout, ol);
2071 lfsck_buf_init(&buf, ff, sizeof(*ff));
2073 handle = dt_trans_create(env, dev);
2075 RETURN(PTR_ERR(handle));
2077 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2081 rc = dt_trans_start_local(env, dev, handle);
2085 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2090 dt_trans_stop(env, dev, handle);
2096 * \retval +1: repaired
2097 * \retval 0: did nothing
2098 * \retval -ve: on error
2100 static int lfsck_layout_update_pfid(const struct lu_env *env,
2101 struct lfsck_component *com,
2102 struct dt_object *parent,
2103 struct lu_fid *cfid,
2104 struct dt_device *cdev,
2105 struct ost_layout *ol, __u32 ea_off)
2107 struct dt_object *child;
2111 child = lfsck_object_find_by_dev(env, cdev, cfid);
2113 RETURN(PTR_ERR(child));
2115 rc = __lfsck_layout_update_pfid(env, child,
2116 lu_object_fid(&parent->do_lu),
2118 lfsck_object_put(env, child);
2120 RETURN(rc == 0 ? 1 : rc);
2123 static int lfsck_lovea_size(struct ost_layout *ol, __u32 ea_off)
2125 if (ol->ol_comp_id != 0)
2126 return sizeof(struct lov_comp_md_v1) +
2127 sizeof(struct lov_comp_md_entry_v1) +
2128 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2130 if (ol->ol_stripe_count != 0)
2131 return lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2133 return lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2137 * This function will create the MDT-object with the given (partial) LOV EA.
2139 * Under some data corruption cases, the MDT-object of the file may be lost,
2140 * but its OST-objects, or some of them are there. The layout LFSCK needs to
2141 * re-create the MDT-object with the orphan OST-object(s) information.
2143 * On the other hand, the LFSCK may has created some OST-object for repairing
2144 * dangling LOV EA reference, but as the LFSCK processing, it may find that
2145 * the old OST-object is there and should replace the former new created OST
2146 * object. Unfortunately, some others have modified such newly created object.
2147 * To keep the data (both new and old), the LFSCK will create MDT-object with
2148 * new FID to reference the original OST-object.
2150 * \param[in] env pointer to the thread context
2151 * \param[in] com pointer to the lfsck component
2152 * \param[in] ltd pointer to target device descriptor
2153 * \param[in] rec pointer to the record for the orphan OST-object
2154 * \param[in] cfid pointer to FID for the orphan OST-object
2155 * \param[in] infix additional information, such as the FID for original
2156 * MDT-object and the stripe offset in the LOV EA
2157 * \param[in] type the type for describing why the orphan MDT-object is
2158 * created. The rules are as following:
2160 * type "C": Multiple OST-objects claim the same MDT-object and the
2161 * same slot in the layout EA. Then the LFSCK will create
2162 * new MDT-object(s) to hold the conflict OST-object(s).
2164 * type "N": The orphan OST-object does not know which one was the
2165 * real parent MDT-object, so the LFSCK uses new FID for
2166 * its parent MDT-object.
2168 * type "R": The orphan OST-object knows its parent MDT-object FID,
2169 * but does not know the position (the file name) in the
2172 * type "D": The MDT-object is a directory, it may knows its parent
2173 * but because there is no valid linkEA, the LFSCK cannot
2174 * know where to put it back to the namespace.
2175 * type "O": The MDT-object has no linkEA, and there is no name
2176 * entry that references the MDT-object.
2178 * type "P": The orphan object to be created was a parent directory
2179 * of some MDT-object which linkEA shows that the @orphan
2180 * object is missing.
2182 * The orphan name will be like:
2183 * ${FID}-${infix}-${type}-${conflict_version}
2185 * \param[in] ea_off the stripe offset in the LOV EA
2187 * \retval positive on repaired something
2188 * \retval 0 if needs to repair nothing
2189 * \retval negative error number on failure
2191 static int lfsck_layout_recreate_parent(const struct lu_env *env,
2192 struct lfsck_component *com,
2193 struct lfsck_tgt_desc *ltd,
2194 struct lu_orphan_rec_v2 *rec,
2195 struct lu_fid *cfid,
2200 struct lfsck_thread_info *info = lfsck_env_info(env);
2201 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
2202 char *name = info->lti_key;
2203 struct lu_attr *la = &info->lti_la2;
2204 struct dt_object_format *dof = &info->lti_dof;
2205 struct lfsck_instance *lfsck = com->lc_lfsck;
2206 struct ost_layout *ol = &rec->lor_layout;
2207 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
2208 struct lu_fid *tfid = &info->lti_fid3;
2209 struct dt_device *dev = lfsck->li_bottom;
2210 struct dt_object *lpf = lfsck->li_lpf_obj;
2211 struct dt_object *pobj = NULL;
2212 struct dt_object *cobj = NULL;
2213 struct thandle *th = NULL;
2214 struct lu_buf *ea_buf = &info->lti_big_buf;
2215 struct lu_buf lov_buf;
2216 struct lfsck_lock_handle *llh = &info->lti_llh;
2217 struct linkea_data ldata = { NULL };
2218 struct lu_buf linkea_buf;
2219 const struct lu_name *pname;
2225 if (unlikely(lpf == NULL))
2226 GOTO(log, rc = -ENXIO);
2228 /* We use two separated transactions to repair the inconsistency.
2230 * 1) create the MDT-object locally.
2231 * 2) update the OST-object's PFID EA if necessary.
2233 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
2234 * updated when the layout LFSCK run next time.
2236 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
2237 * when the layout LFSCK run next time. */
2239 if (fid_is_zero(pfid)) {
2240 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
2244 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2246 GOTO(log, rc = PTR_ERR(cobj));
2249 pobj = lfsck_object_find_by_dev(env, dev, pfid);
2251 GOTO(log, rc = PTR_ERR(pobj));
2253 LASSERT(infix != NULL);
2254 LASSERT(type != NULL);
2256 memset(la, 0, sizeof(*la));
2257 la->la_uid = rec->lor_rec.lor_uid;
2258 la->la_gid = rec->lor_rec.lor_gid;
2259 la->la_mode = S_IFREG | S_IRUSR;
2260 la->la_valid = LA_MODE | LA_UID | LA_GID;
2262 memset(dof, 0, sizeof(*dof));
2263 dof->dof_type = dt_mode_to_dft(S_IFREG);
2264 /* Because the dof->dof_reg.striped = 0, the LOD will not create
2265 * the stripe(s). The LFSCK will specify the LOV EA via
2266 * lfsck_layout_update_lovea(). */
2268 size = lfsck_lovea_size(ol, ea_off);
2269 if (ea_buf->lb_len < size) {
2270 lu_buf_realloc(ea_buf, size);
2271 if (ea_buf->lb_buf == NULL)
2272 GOTO(log, rc = -ENOMEM);
2277 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
2279 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2280 (const struct dt_key *)name);
2281 if (rc != 0 && rc != -ENOENT)
2285 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
2286 MDS_INODELOCK_UPDATE, LCK_PW);
2290 /* Re-check whether the name conflict with othrs after taken
2292 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2293 (const struct dt_key *)name);
2294 if (unlikely(rc == 0)) {
2302 pname = lfsck_name_get_const(env, name, strlen(name));
2303 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
2304 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
2308 /* The 1st transaction. */
2309 th = dt_trans_create(env, dev);
2311 GOTO(unlock, rc = PTR_ERR(th));
2313 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
2317 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
2318 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
2319 LU_XATTR_CREATE, th);
2323 dtrec->rec_fid = pfid;
2324 dtrec->rec_type = S_IFREG;
2325 rc = dt_declare_insert(env, lpf,
2326 (const struct dt_rec *)dtrec,
2327 (const struct dt_key *)name, th);
2331 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2332 ldata.ld_leh->leh_len);
2333 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
2334 XATTR_NAME_LINK, 0, th);
2338 rc = dt_trans_start_local(env, dev, th);
2342 dt_write_lock(env, pobj, 0);
2343 rc = dt_create(env, pobj, la, NULL, dof, th);
2345 rc = lfsck_layout_update_lovea(env, lfsck, th, ol, pobj, cfid,
2346 &lov_buf, LU_XATTR_CREATE, ltd->ltd_index, ea_off);
2347 dt_write_unlock(env, pobj);
2351 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
2352 (const struct dt_key *)name, th, 1);
2356 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
2357 if (rc == 0 && cobj != NULL) {
2358 dt_trans_stop(env, dev, th);
2361 /* The 2nd transaction. */
2362 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ol, ea_off);
2369 dt_trans_stop(env, dev, th);
2375 if (cobj != NULL && !IS_ERR(cobj))
2376 lfsck_object_put(env, cobj);
2377 if (pobj != NULL && !IS_ERR(pobj))
2378 lfsck_object_put(env, pobj);
2381 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2382 "recreate the lost MDT-object: parent "DFID
2383 ", child "DFID", OST-index %u, stripe-index %u, "
2384 "infix %s, type %s: rc = %d\n",
2385 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2386 ltd->ltd_index, ea_off, infix, type, rc);
2388 return rc >= 0 ? 1 : rc;
2391 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2392 struct lfsck_component *com,
2393 const struct lu_fid *fid,
2396 struct lfsck_thread_info *info = lfsck_env_info(env);
2397 struct lfsck_request *lr = &info->lti_lr;
2398 struct lfsck_instance *lfsck = com->lc_lfsck;
2399 struct lfsck_tgt_desc *ltd;
2400 struct ptlrpc_request *req;
2401 struct lfsck_request *tmp;
2402 struct obd_export *exp;
2406 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2407 if (unlikely(ltd == NULL))
2411 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2412 GOTO(put, rc = -EOPNOTSUPP);
2414 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2416 GOTO(put, rc = -ENOMEM);
2418 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2420 ptlrpc_request_free(req);
2425 memset(lr, 0, sizeof(*lr));
2426 lr->lr_event = LE_CONDITIONAL_DESTROY;
2427 lr->lr_active = LFSCK_TYPE_LAYOUT;
2430 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2432 ptlrpc_request_set_replen(req);
2434 rc = ptlrpc_queue_wait(req);
2435 ptlrpc_req_finished(req);
2445 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2446 struct lfsck_component *com,
2447 struct lfsck_request *lr)
2449 struct lfsck_thread_info *info = lfsck_env_info(env);
2450 struct lu_attr *la = &info->lti_la;
2451 union ldlm_policy_data *policy = &info->lti_policy;
2452 struct ldlm_res_id *resid = &info->lti_resid;
2453 struct lfsck_instance *lfsck = com->lc_lfsck;
2454 struct dt_device *dev = lfsck->li_bottom;
2455 struct lu_fid *fid = &lr->lr_fid;
2456 struct dt_object *obj;
2457 struct thandle *th = NULL;
2458 struct lustre_handle lh = { 0 };
2463 obj = lfsck_object_find_by_dev(env, dev, fid);
2465 RETURN(PTR_ERR(obj));
2467 dt_read_lock(env, obj, 0);
2468 if (dt_object_exists(obj) == 0 ||
2469 lfsck_is_dead_obj(obj)) {
2470 dt_read_unlock(env, obj);
2472 GOTO(put, rc = -ENOENT);
2475 /* Get obj's attr without lock firstly. */
2476 rc = dt_attr_get(env, obj, la);
2477 dt_read_unlock(env, obj);
2481 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2482 GOTO(put, rc = -ETXTBSY);
2484 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2485 LASSERT(lfsck->li_namespace != NULL);
2487 memset(policy, 0, sizeof(*policy));
2488 policy->l_extent.end = OBD_OBJECT_EOF;
2489 ost_fid_build_resid(fid, resid);
2490 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2491 policy, LCK_EX, &flags, ldlm_blocking_ast,
2492 ldlm_completion_ast, NULL, NULL, 0,
2493 LVB_T_NONE, NULL, &lh);
2495 GOTO(put, rc = -EIO);
2497 dt_write_lock(env, obj, 0);
2498 /* Get obj's attr within lock again. */
2499 rc = dt_attr_get(env, obj, la);
2503 if (la->la_ctime != 0)
2504 GOTO(unlock, rc = -ETXTBSY);
2506 th = dt_trans_create(env, dev);
2508 GOTO(unlock, rc = PTR_ERR(th));
2510 rc = dt_declare_ref_del(env, obj, th);
2514 rc = dt_declare_destroy(env, obj, th);
2518 rc = dt_trans_start_local(env, dev, th);
2522 rc = dt_ref_del(env, obj, th);
2526 rc = dt_destroy(env, obj, th);
2528 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2529 "OST-object "DFID" that was created for reparing "
2530 "dangling referenced case. But the original missing "
2531 "OST-object is found now.\n",
2532 lfsck_lfsck2name(lfsck), PFID(fid));
2537 dt_trans_stop(env, dev, th);
2540 dt_write_unlock(env, obj);
2541 ldlm_lock_decref(&lh, LCK_EX);
2544 lfsck_object_put(env, obj);
2550 * Some OST-object has occupied the specified layout EA slot.
2551 * Such OST-object may be generated by the LFSCK when repair
2552 * dangling referenced MDT-object, which can be indicated by
2553 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2554 * is true and such OST-object has not been modified yet, we
2555 * will replace it with the orphan OST-object; otherwise the
2556 * LFSCK will create new MDT-object to reference the orphan.
2558 * \retval +1: repaired
2559 * \retval 0: did nothing
2560 * \retval -ve: on error
2562 static int lfsck_layout_conflict_create(const struct lu_env *env,
2563 struct lfsck_component *com,
2564 struct lfsck_tgt_desc *ltd,
2565 struct lu_orphan_rec_v2 *rec,
2566 struct dt_object *parent,
2567 struct lu_fid *cfid,
2568 struct lu_buf *ea_buf,
2569 struct lov_mds_md_v1 *lmm,
2570 struct lov_ost_data_v1 *slot,
2571 __u32 ea_off, int lovea_size)
2573 struct lfsck_thread_info *info = lfsck_env_info(env);
2574 struct lu_fid *cfid2 = &info->lti_fid2;
2575 struct ost_id *oi = &info->lti_oi;
2576 struct dt_device *dev = lfsck_obj2dev(parent);
2577 struct thandle *th = NULL;
2578 struct lustre_handle lh = { 0 };
2579 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2583 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2584 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2588 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2589 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2593 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2594 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2599 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2601 /* If the conflict OST-obejct is not created for fixing dangling
2602 * referenced MDT-object in former LFSCK check/repair, or it has
2603 * been modified by others, then we cannot destroy it. Re-create
2604 * a new MDT-object for the orphan OST-object. */
2605 if (rc == -ETXTBSY) {
2606 /* No need the layout lock on the original parent. */
2607 lfsck_ibits_unlock(&lh, LCK_EX);
2609 fid_zero(&rec->lor_rec.lor_fid);
2610 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2611 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2613 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2614 info->lti_tmpbuf, "C", ea_off);
2619 if (rc != 0 && rc != -ENOENT)
2622 th = dt_trans_create(env, dev);
2624 GOTO(unlock, rc = PTR_ERR(th));
2626 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2627 LU_XATTR_REPLACE, th);
2631 rc = dt_trans_start_local(env, dev, th);
2635 dt_write_lock(env, parent, 0);
2636 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2637 rc = lfsck_layout_refill_lovea(env, com->lc_lfsck, th, parent, cfid,
2638 ea_buf, lmm, slot, LU_XATTR_REPLACE,
2639 ltd->ltd_index, lovea_size);
2640 dt_write_unlock(env, parent);
2645 dt_trans_stop(env, dev, th);
2648 lfsck_ibits_unlock(&lh, LCK_EX);
2651 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2652 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2653 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2654 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2655 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2658 return rc >= 0 ? 1 : rc;
2662 * \retval +1: repaired
2663 * \retval 0: did nothing
2664 * \retval -ve: on error
2666 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2667 struct lfsck_component *com,
2668 struct lfsck_tgt_desc *ltd,
2669 struct lu_orphan_rec_v2 *rec,
2670 struct dt_object *parent,
2671 struct lu_fid *cfid,
2672 __u32 ost_idx, __u32 ea_off)
2674 struct lfsck_thread_info *info = lfsck_env_info(env);
2675 struct lu_buf *buf = &info->lti_big_buf;
2676 struct lu_fid *fid = &info->lti_fid2;
2677 struct ost_id *oi = &info->lti_oi;
2678 struct lfsck_instance *lfsck = com->lc_lfsck;
2679 struct dt_device *dt = lfsck_obj2dev(parent);
2680 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2681 struct ost_layout *ol = &rec->lor_layout;
2682 struct lov_comp_md_v1 *lcm = NULL;
2683 struct lov_comp_md_entry_v1 *lcme = NULL;
2684 struct thandle *handle = NULL;
2686 struct lov_mds_md_v1 *lmm;
2687 struct lov_ost_data_v1 *objs;
2688 struct lustre_handle lh = { 0 };
2696 bool locked = false;
2699 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2700 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2703 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2704 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2705 "stripe-index %u, comp_id %u, comp_start %llu, "
2706 "comp_end %llu: rc = %d\n",
2707 lfsck_lfsck2name(lfsck), PFID(cfid),
2708 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off,
2709 ol->ol_comp_id, ol->ol_comp_start,
2710 ol->ol_comp_end, rc);
2717 dt_write_unlock(env, parent);
2721 if (handle != NULL) {
2722 dt_trans_stop(env, dt, handle);
2727 GOTO(unlock_layout, rc);
2730 if (buf->lb_len < lovea_size) {
2731 lu_buf_realloc(buf, lovea_size);
2732 if (buf->lb_buf == NULL)
2733 GOTO(unlock_layout, rc = -ENOMEM);
2736 if (!(bk->lb_param & LPF_DRYRUN)) {
2737 handle = dt_trans_create(env, dt);
2739 GOTO(unlock_layout, rc = PTR_ERR(handle));
2741 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2746 rc = dt_trans_start_local(env, dt, handle);
2751 dt_write_lock(env, parent, 0);
2753 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2754 if (rc == -ERANGE) {
2755 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2758 } else if (rc == -ENODATA || rc == 0) {
2759 lovea_size = lfsck_lovea_size(ol, ea_off);
2760 /* If the declared is not big enough, re-try. */
2761 if (buf->lb_len < lovea_size) {
2765 fl = LU_XATTR_CREATE;
2766 } else if (rc < 0) {
2767 GOTO(unlock_parent, rc);
2768 } else if (unlikely(buf->lb_len == 0)) {
2771 fl = LU_XATTR_REPLACE;
2775 if (fl == LU_XATTR_CREATE) {
2776 if (bk->lb_param & LPF_DRYRUN)
2777 GOTO(unlock_parent, rc = 1);
2779 LASSERT(buf->lb_len >= lovea_size);
2781 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2782 cfid, buf, fl, ost_idx, ea_off);
2784 GOTO(unlock_parent, rc);
2788 rc1 = lfsck_layout_verify_header(parent, lmm);
2790 /* If the LOV EA crashed, the rebuild it. */
2791 if (rc1 == -EINVAL) {
2792 if (bk->lb_param & LPF_DRYRUN)
2793 GOTO(unlock_parent, rc = 1);
2795 LASSERT(buf->lb_len >= lovea_size);
2797 rc = lfsck_layout_update_lovea(env, lfsck, handle, ol, parent,
2798 cfid, buf, fl, ost_idx, ea_off);
2800 GOTO(unlock_parent, rc);
2803 /* For other unknown magic/pattern, keep the current LOV EA. */
2805 GOTO(unlock_parent, rc = rc1);
2807 magic = le32_to_cpu(lmm->lmm_magic);
2808 if (magic == LOV_MAGIC_COMP_V1) {
2812 if (bk->lb_param & LPF_DRYRUN)
2813 GOTO(unlock_parent, rc = 1);
2816 count = le16_to_cpu(lcm->lcm_entry_count);
2817 for (i = 0; i < count; i++) {
2818 lcme = &lcm->lcm_entries[i];
2819 start = le64_to_cpu(lcme->lcme_extent.e_start);
2820 end = le64_to_cpu(lcme->lcme_extent.e_end);
2822 if (end <= ol->ol_comp_start)
2825 if (start >= ol->ol_comp_end)
2828 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
2829 magic = le32_to_cpu(lmm->lmm_magic);
2830 flags = le32_to_cpu(lcme->lcme_flags);
2834 rc = lfsck_layout_add_comp_comp(env, lfsck, handle, ol, parent,
2835 cfid, buf, ost_idx, ea_off, i);
2837 GOTO(unlock_parent, rc);
2841 count = le16_to_cpu(lmm->lmm_stripe_count);
2843 GOTO(unlock_parent, rc = -EINVAL);
2846 /* Exceed the current end of MDT-object layout EA. Then extend it. */
2847 if (count <= ea_off) {
2848 if (bk->lb_param & LPF_DRYRUN)
2849 GOTO(unlock_parent, rc = 1);
2851 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2852 /* If the declared is not big enough, re-try. */
2853 if (buf->lb_len < lovea_size) {
2858 if (lcme && !(flags & LCME_FL_INIT))
2859 lcme->lcme_flags = cpu_to_le32(flags | LCME_FL_INIT);
2861 rc = lfsck_layout_extend_v1v3_lovea(env, lfsck, handle, ol,
2862 parent, cfid, buf, ost_idx, ea_off);
2864 GOTO(unlock_parent, rc);
2867 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2869 if (magic == LOV_MAGIC_V1) {
2870 objs = &lmm->lmm_objects[0];
2872 LASSERT(magic == LOV_MAGIC_V3);
2873 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2876 for (i = 0; i < count; i++, objs++) {
2877 /* The MDT-object was created via lfsck_layout_recover_create()
2878 * by others before, and we fill the dummy layout EA. */
2879 if ((lcme && !(flags & LCME_FL_INIT)) ||
2880 lovea_slot_is_dummy(objs)) {
2884 if (bk->lb_param & LPF_DRYRUN)
2885 GOTO(unlock_parent, rc = 1);
2887 lmm->lmm_layout_gen =
2888 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2892 if (le32_to_cpu(lmm->lmm_stripe_size) !=
2893 ol->ol_stripe_size ||
2894 le16_to_cpu(lmm->lmm_stripe_count) !=
2895 ol->ol_stripe_count ||
2896 le64_to_cpu(lcme->lcme_extent.e_start) !=
2897 ol->ol_comp_start ||
2898 le64_to_cpu(lcme->lcme_extent.e_end) !=
2900 CDEBUG(D_LFSCK, "%s: found invalid "
2901 "component for "DFID ": parent "DFID
2902 ", stripe-index %u, stripe_size %u, "
2903 "stripe_count %u, comp_id %u, "
2904 "comp_start %llu, comp_end %llu, "
2905 "cur_stripe_size %u, "
2906 "cur_stripe_count %u, "
2907 "cur_comp_start %llu, "
2908 "cur_comp_end %llu\n",
2909 lfsck_lfsck2name(lfsck), PFID(cfid),
2910 PFID(lfsck_dto2fid(parent)), ea_off,
2912 ol->ol_stripe_count, ol->ol_comp_id,
2913 ol->ol_comp_start, ol->ol_comp_end,
2914 le32_to_cpu(lmm->lmm_stripe_size),
2915 le16_to_cpu(lmm->lmm_stripe_count),
2916 le64_to_cpu(lcme->lcme_extent.e_start),
2917 le64_to_cpu(lcme->lcme_extent.e_end));
2919 GOTO(unlock_parent, rc = -EINVAL);
2922 le32_add_cpu(&lcm->lcm_layout_gen, 1);
2923 lovea_size = le32_to_cpu(lcm->lcm_size);
2924 if (!(flags & LCME_FL_INIT))
2925 lcme->lcme_flags = cpu_to_le32(flags |
2929 LASSERTF(buf->lb_len >= lovea_size,
2930 "buffer len %d is less than real size %d\n",
2931 (int)buf->lb_len, (int)lovea_size);
2933 rc = lfsck_layout_refill_lovea(env, lfsck, handle,
2934 parent, cfid, buf, lmm, objs,
2935 fl, ost_idx, lovea_size);
2937 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2938 "dummy layout slot for "DFID": parent "DFID
2939 ", OST-index %u, stripe-index %u: rc = %d\n",
2940 lfsck_lfsck2name(lfsck), PFID(cfid),
2941 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2943 GOTO(unlock_parent, rc);
2946 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2947 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2949 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2950 "invalid layout EA at the slot %d, index %u\n",
2951 lfsck_lfsck2name(lfsck),
2952 PFID(lfsck_dto2fid(parent)), i,
2953 le32_to_cpu(objs->l_ost_idx));
2955 GOTO(unlock_parent, rc);
2958 /* It should be rare case, the slot is there, but the LFSCK
2959 * does not handle it during the first-phase cycle scanning. */
2960 if (unlikely(lu_fid_eq(fid, cfid))) {
2962 GOTO(unlock_parent, rc = 0);
2964 /* Rare case that the OST-object index
2965 * does not match the parent MDT-object
2966 * layout EA. We trust the later one. */
2967 if (bk->lb_param & LPF_DRYRUN)
2968 GOTO(unlock_parent, rc = 1);
2970 dt_write_unlock(env, parent);
2972 dt_trans_stop(env, dt, handle);
2973 lfsck_ibits_unlock(&lh, LCK_EX);
2974 rc = lfsck_layout_update_pfid(env, com, parent,
2978 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2979 "updated OST-object's pfid for "DFID
2980 ": parent "DFID", OST-index %u, "
2981 "stripe-index %u: rc = %d\n",
2982 lfsck_lfsck2name(lfsck), PFID(cfid),
2983 PFID(lfsck_dto2fid(parent)),
2984 ltd->ltd_index, i, rc);
2991 /* The MDT-object exists, but related layout EA slot is occupied
2993 if (bk->lb_param & LPF_DRYRUN)
2994 GOTO(unlock_parent, rc = 1);
2996 dt_write_unlock(env, parent);
2998 dt_trans_stop(env, dt, handle);
2999 lfsck_ibits_unlock(&lh, LCK_EX);
3000 if (magic == LOV_MAGIC_V1)
3001 objs = &lmm->lmm_objects[ea_off];
3003 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
3004 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
3005 buf, lmm, objs, ea_off, lovea_size);
3011 dt_write_unlock(env, parent);
3015 dt_trans_stop(env, dt, handle);
3018 lfsck_ibits_unlock(&lh, LCK_EX);
3023 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
3024 struct lfsck_component *com,
3025 struct lfsck_tgt_desc *ltd,
3026 struct lu_orphan_rec_v2 *rec,
3027 struct lu_fid *cfid)
3029 struct lfsck_layout *lo = com->lc_file_ram;
3030 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
3031 struct dt_object *parent = NULL;
3032 __u32 ea_off = pfid->f_stripe_idx;
3036 if (!fid_is_sane(cfid))
3037 GOTO(out, rc = -EINVAL);
3040 if (fid_is_zero(pfid)) {
3041 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3046 if (!fid_is_sane(pfid))
3047 GOTO(out, rc = -EINVAL);
3049 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
3051 GOTO(out, rc = PTR_ERR(parent));
3053 if (unlikely(dt_object_remote(parent) != 0))
3054 GOTO(put, rc = -EXDEV);
3056 if (dt_object_exists(parent) == 0) {
3057 lfsck_object_put(env, parent);
3058 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3063 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
3064 GOTO(put, rc = -EISDIR);
3066 /* The orphan OST-object claims to be the parent's stripe, then
3067 * related dangling record in the trace file is meaningless. */
3068 rc = lfsck_layout_del_dangling_rec(env, com, pfid,
3069 rec->lor_layout.ol_comp_id, ea_off);
3070 if (rc && rc != -ENOENT)
3073 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
3074 ltd->ltd_index, ea_off);
3080 lfsck_object_put(env, parent);
3082 /* The layout EA is changed, need to be reloaded next time. */
3083 dt_object_put_nocache(env, parent);
3086 down_write(&com->lc_sem);
3087 com->lc_new_scanned++;
3088 com->lc_new_checked++;
3090 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
3092 } else if (rc < 0) {
3093 lo->ll_objs_failed_phase2++;
3095 up_write(&com->lc_sem);
3100 static int lfsck_layout_scan_orphan(const struct lu_env *env,
3101 struct lfsck_component *com,
3102 struct lfsck_tgt_desc *ltd)
3104 struct lfsck_assistant_data *lad = com->lc_data;
3105 struct lfsck_instance *lfsck = com->lc_lfsck;
3106 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3107 struct lfsck_thread_info *info = lfsck_env_info(env);
3108 struct lu_fid *fid = &info->lti_fid;
3109 struct dt_object *obj;
3110 const struct dt_it_ops *iops;
3115 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
3116 "scanning for OST%04x\n",
3117 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3119 if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
3120 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
3121 "scanning for OST%04x\n",
3122 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3127 fid->f_seq = fid_idif_seq(0, ltd->ltd_index);
3128 fid->f_oid = fid->f_ver = 0;
3130 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
3131 if (unlikely(IS_ERR(obj)))
3132 GOTO(log, rc = PTR_ERR(obj));
3134 rc = obj->do_ops->do_index_try(env, obj,
3135 &dt_lfsck_layout_orphan_features);
3139 iops = &obj->do_index_ops->dio_it;
3140 di = iops->init(env, obj, 0);
3142 GOTO(put, rc = PTR_ERR(di));
3144 rc = iops->load(env, di, 0);
3146 /* -ESRCH means that the orphan OST-objects rbtree has been
3147 * cleanup because of the OSS server restart or other errors. */
3148 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
3153 rc = iops->next(env, di);
3165 struct lu_orphan_rec_v2 *rec = &info->lti_rec;
3167 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3168 unlikely(!thread_is_running(&lfsck->li_thread)))
3171 key = iops->key(env, di);
3172 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
3173 /* Remote target OST may be runnning old LFSCK */
3174 memset(rec, 0, sizeof(*rec));
3175 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
3177 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
3178 &com->lc_fid_latest_scanned_phase2);
3179 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3182 lfsck_control_speed_by_self(com);
3184 rc = iops->next(env, di);
3185 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
3192 iops->fini(env, di);
3194 lfsck_object_put(env, obj);
3197 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
3198 "scanning for OST%04x: rc = %d\n",
3199 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
3201 return rc > 0 ? 0 : rc;
3204 static int lfsck_lmm2layout(struct lov_mds_md_v1 *lmm, struct ost_layout *ol,
3207 __u32 magic = le32_to_cpu(lmm->lmm_magic);
3211 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3) {
3212 ol->ol_stripe_size = lmm->lmm_stripe_size;
3213 ol->ol_stripe_count = lmm->lmm_stripe_count;
3214 ol->ol_comp_start = 0;
3215 ol->ol_comp_end = 0;
3217 } else if (magic == LOV_MAGIC_COMP_V1) {
3218 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
3219 struct lov_comp_md_entry_v1 *lcme = NULL;
3220 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3223 for (i = 0; i < count; i++) {
3224 lcme = &lcm->lcm_entries[i];
3225 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3226 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3233 /* The comp has been removed, do nothing. */
3237 lmm = (void *)lmm + le32_to_cpu(lcme->lcme_offset);
3238 ol->ol_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
3239 ol->ol_stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
3240 ol->ol_comp_start = le64_to_cpu(lcme->lcme_extent.e_start);
3241 ol->ol_comp_end = le64_to_cpu(lcme->lcme_extent.e_end);
3242 ol->ol_comp_id = le32_to_cpu(lcme->lcme_id);
3244 GOTO(out, rc = -EINVAL);
3254 * Repair the MDT-object with dangling LOV EA reference.
3256 * we need to repair the inconsistency according to the users' requirement:
3258 * 1) Keep the inconsistency there and report the inconsistency case,
3259 * then give the chance to the application to find related issues,
3260 * and the users can make the decision about how to handle it with
3261 * more human knownledge. (by default)
3263 * 2) Re-create the missing OST-object with the FID/owner information.
3265 * \param[in] env pointer to the thread context
3266 * \param[in] com the layout LFSCK component
3267 * \param[in] parent the MDT-object with dangling LOV EA reference
3268 * \param[in] child the OST-object to be created
3269 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3270 * \param[in] ea_off the offset of the OST-object in the LOV EA
3271 * \param[in] ost_idx the index of OST on which the OST-object resides
3273 * \retval +1 for repair successfully
3274 * \retval 0 for did nothing
3275 * \retval negative error number on failure
3277 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
3278 struct lfsck_component *com,
3279 struct dt_object *parent,
3280 struct dt_object *child,
3281 __u32 comp_id, __u32 ea_off,
3282 __u32 ost_idx, bool log)
3284 struct lfsck_thread_info *info = lfsck_env_info(env);
3285 struct filter_fid *ff = &info->lti_ff;
3286 struct ost_layout *ol = &ff->ff_layout;
3287 struct dt_object_format *dof = &info->lti_dof;
3288 struct lu_attr *la = &info->lti_la;
3289 struct lfsck_instance *lfsck = com->lc_lfsck;
3290 struct dt_device *dev = lfsck_obj2dev(child);
3291 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3292 const struct lu_fid *cfid = lfsck_dto2fid(child);
3293 struct lu_buf *tbuf = &info->lti_big_buf;
3294 struct thandle *handle;
3296 struct lustre_handle lh = { 0 };
3300 if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
3303 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3304 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3309 rc = dt_attr_get(env, parent, la);
3313 la->la_mode = S_IFREG | 0666;
3314 la->la_atime = la->la_mtime = la->la_ctime = 0;
3315 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
3316 LA_ATIME | LA_MTIME | LA_CTIME;
3317 memset(dof, 0, sizeof(*dof));
3318 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
3319 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
3320 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3321 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3322 * parent MDT-object's layout EA. */
3323 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
3325 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3329 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, comp_id);
3333 buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
3334 handle = dt_trans_create(env, dev);
3336 GOTO(unlock1, rc = PTR_ERR(handle));
3338 rc = dt_declare_create(env, child, la, NULL, dof, handle);
3342 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
3343 LU_XATTR_CREATE, handle);
3347 rc = dt_trans_start_local(env, dev, handle);
3351 dt_read_lock(env, parent, 0);
3352 if (unlikely(lfsck_is_dead_obj(parent)))
3353 GOTO(unlock2, rc = 0);
3355 if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
3356 struct ost_id *oi = &info->lti_oi;
3357 struct lu_fid *tfid = &info->lti_fid2;
3358 struct lu_buf *lovea = &info->lti_big_buf;
3359 struct lov_mds_md_v1 *lmm;
3360 struct lov_ost_data_v1 *objs;
3365 rc = lfsck_layout_get_lovea(env, parent, lovea);
3369 lmm = lovea->lb_buf;
3370 magic = le32_to_cpu(lmm->lmm_magic);
3371 if (magic == LOV_MAGIC_COMP_V1) {
3372 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3373 struct lov_comp_md_entry_v1 *lcme;
3374 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3377 for (i = 0; i < count; i++) {
3378 lcme = &lcm->lcm_entries[i];
3379 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3380 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3383 lmm = lovea->lb_buf +
3384 le32_to_cpu(lcme->lcme_offset);
3385 magic = le32_to_cpu(lmm->lmm_magic);
3390 /* Someone removed the component, do nothing. */
3391 GOTO(unlock2, rc = 0);
3395 count = le16_to_cpu(lmm->lmm_stripe_count);
3396 /* Someone changed the LOV EA, do nothing. */
3397 if (count <= ea_off)
3398 GOTO(unlock2, rc = 0);
3400 if (magic == LOV_MAGIC_V1) {
3401 objs = &lmm->lmm_objects[ea_off];
3403 LASSERT(magic == LOV_MAGIC_V3);
3405 objs = &((struct lov_mds_md_v3 *)lmm)->\
3406 lmm_objects[ea_off];
3409 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3410 idx2 = le32_to_cpu(objs->l_ost_idx);
3411 rc = ostid_to_fid(tfid, oi, idx2);
3412 /* Someone changed the LOV EA, do nothing. */
3413 if (rc != 0 || !lu_fid_eq(tfid, cfid))
3417 rc = dt_create(env, child, la, NULL, dof, handle);
3421 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
3427 dt_read_unlock(env, parent);
3430 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3433 lfsck_ibits_unlock(&lh, LCK_EX);
3437 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3438 "dangling reference for: parent "DFID", child "
3439 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: "
3441 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3442 comp_id, ea_off, ost_idx,
3443 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3444 "Create the lost OST-object as required" :
3445 "Keep the MDT-object there by default", rc);
3451 * Repair the MDT-object with dangling LOV EA reference.
3453 * Prepare parameters and call __lfsck_layout_repair_dangling()
3454 * to repair the dangling LOV EA reference.
3456 * \param[in] env pointer to the thread context
3457 * \param[in] com the layout LFSCK component
3458 * \param[in] pfid the MDT-object's FID
3459 * \param[in] cfid the FID for the OST-object to be created
3460 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3461 * \param[in] ea_off the offset of the OST-object in the LOV EA
3462 * \param[in] ost_idx the index of OST on which the OST-object resides
3464 * \retval +1 for repair successfully
3465 * \retval 0 for did nothing
3466 * \retval negative error number on failure
3468 static int lfsck_layout_repair_dangling(const struct lu_env *env,
3469 struct lfsck_component *com,
3470 const struct lu_fid *pfid,
3471 const struct lu_fid *cfid,
3472 __u32 comp_id, __u32 ea_off,
3475 struct lfsck_instance *lfsck = com->lc_lfsck;
3476 struct dt_object *parent = NULL;
3477 struct dt_object *child = NULL;
3478 struct lfsck_tgt_desc *ltd;
3482 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3484 GOTO(log, rc = PTR_ERR(parent));
3486 /* The MDT-object has been removed. */
3487 if (dt_object_exists(parent) == 0)
3490 ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
3491 if (unlikely(ltd == NULL))
3492 GOTO(log, rc = -ENODEV);
3494 child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
3496 GOTO(log, rc = PTR_ERR(child));
3498 /* The OST-object has been created. */
3499 if (unlikely(dt_object_exists(child) != 0))
3502 rc = __lfsck_layout_repair_dangling(env, com, parent, child,
3503 comp_id, ea_off, ost_idx, false);
3508 if (child != NULL && !IS_ERR(child))
3509 lfsck_object_put(env, child);
3511 if (parent != NULL && !IS_ERR(parent))
3512 lfsck_object_put(env, parent);
3515 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3516 "dangling reference for: parent "DFID", child "
3517 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: rc = %d\n",
3518 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3519 comp_id, ea_off, ost_idx,
3520 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3521 "Create the lost OST-object as required" :
3522 "Keep the MDT-object there by default", rc);
3527 /* If the OST-object does not recognize the MDT-object as its parent, and
3528 * there is no other MDT-object claims as its parent, then just trust the
3529 * given MDT-object as its parent. So update the OST-object filter_fid. */
3530 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
3531 struct lfsck_component *com,
3532 struct dt_object *parent,
3533 struct lfsck_layout_req *llr,
3536 struct lfsck_thread_info *info = lfsck_env_info(env);
3537 struct filter_fid *ff = &info->lti_ff;
3538 struct ost_layout *ol = &ff->ff_layout;
3539 struct dt_object *child = llr->llr_child;
3540 struct dt_device *dev = lfsck_obj2dev(child);
3541 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
3542 struct lu_buf *tbuf = &info->lti_big_buf;
3543 struct thandle *handle;
3545 struct lustre_handle lh = { 0 };
3549 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3550 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3555 ff->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3556 ff->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3557 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3558 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3559 * parent MDT-object's layout EA. */
3560 ff->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3562 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3566 rc = lfsck_lmm2layout(tbuf->lb_buf, ol, llr->llr_comp_id);
3570 buf = lfsck_buf_get(env, ff, sizeof(*ff));
3572 handle = dt_trans_create(env, dev);
3574 GOTO(unlock1, rc = PTR_ERR(handle));
3576 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3580 rc = dt_attr_get(env, parent, la);
3584 la->la_valid = LA_UID | LA_GID;
3585 rc = dt_declare_attr_set(env, child, la, handle);
3589 rc = dt_trans_start_local(env, dev, handle);
3593 dt_write_lock(env, parent, 0);
3594 if (unlikely(lfsck_is_dead_obj(parent)))
3595 GOTO(unlock2, rc = 1);
3597 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3601 /* Get the latest parent's owner. */
3602 rc = dt_attr_get(env, parent, la);
3606 la->la_valid = LA_UID | LA_GID;
3607 rc = dt_attr_set(env, child, la, handle);
3612 dt_write_unlock(env, parent);
3615 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3618 lfsck_ibits_unlock(&lh, LCK_EX);
3622 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3623 "unmatched MDT-OST pair for: parent "DFID
3624 ", child "DFID", comp_id %u, OST-index %u, "
3625 "stripe-index %u, owner %u/%u: rc = %d\n",
3626 lfsck_lfsck2name(com->lc_lfsck),
3627 PFID(lfsck_dto2fid(parent)),
3628 PFID(lfsck_dto2fid(child)),
3629 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3630 la->la_uid, la->la_gid, rc);
3635 /* If there are more than one MDT-objects claim as the OST-object's parent,
3636 * and the OST-object only recognizes one of them, then we need to generate
3637 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3638 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3639 struct lfsck_component *com,
3640 struct dt_object *parent,
3641 struct lfsck_layout_req *llr,
3644 struct lfsck_thread_info *info = lfsck_env_info(env);
3645 struct dt_allocation_hint *hint = &info->lti_hint;
3646 struct dt_object_format *dof = &info->lti_dof;
3647 struct ost_id *oi = &info->lti_oi;
3648 struct lu_buf *buf = &info->lti_big_buf;
3649 struct lfsck_instance *lfsck = com->lc_lfsck;
3650 struct dt_device *dev;
3651 struct lu_device *d =
3652 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3653 struct lu_object *o;
3654 struct lu_object *n;
3655 struct dt_object *child = NULL;
3656 struct thandle *handle = NULL;
3657 struct lov_mds_md_v1 *lmm;
3658 struct lov_ost_data_v1 *objs;
3659 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3661 struct lustre_handle lh = { 0 };
3667 /* We use two separated transactions to repair the inconsistency.
3669 * 1) create the child (OST-object).
3670 * 2) update the parent LOV EA according to the child's FID.
3672 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3673 * handled as orphan when the layout LFSCK run next time.
3675 * If 1) failed, but 2) succeed, then such OST-object will be re-created
3676 * as dangling referened case when the layout LFSCK run next time. */
3678 /* The 1st transaction. */
3679 o = lu_object_anon(env, d, NULL);
3681 GOTO(log, rc = PTR_ERR(o));
3683 n = lu_object_locate(o->lo_header, d->ld_type);
3684 if (unlikely(n == NULL)) {
3685 lu_object_put_nocache(env, o);
3687 GOTO(log, rc = -EINVAL);
3690 child = container_of(n, struct dt_object, do_lu);
3691 memset(hint, 0, sizeof(*hint));
3692 rc = dt_attr_get(env, parent, la);
3696 la->la_valid = LA_UID | LA_GID;
3697 memset(dof, 0, sizeof(*dof));
3699 dev = lfsck_obj2dev(child);
3700 handle = dt_trans_create(env, dev);
3702 GOTO(log, rc = PTR_ERR(handle));
3704 rc = dt_declare_create(env, child, la, hint, dof, handle);
3708 rc = dt_trans_start_local(env, dev, handle);
3712 rc = dt_create(env, child, la, hint, dof, handle);
3713 dt_trans_stop(env, dev, handle);
3718 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3719 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3724 /* The 2nd transaction. */
3726 /* XXX: Generally, we should use bottom device (OSD) to update parent
3727 * LOV EA. But because the LOD-object still references the wrong
3728 * OSP-object that should be detached after the parent's LOV EA
3729 * refreshed. Unfortunately, there is no suitable API for that.
3730 * So we have to make the LOD to re-load the OSP-object(s) via
3731 * replacing the LOV EA against the LOD-object.
3733 * Once the DNE2 patches have been landed, we can replace the
3734 * LOD device with the OSD device. LU-6230. */
3736 dev = lfsck->li_next;
3737 parent = lfsck_object_locate(dev, parent);
3739 GOTO(log, rc = PTR_ERR(parent));
3741 handle = dt_trans_create(env, dev);
3743 GOTO(log, rc = PTR_ERR(handle));
3745 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3746 LU_XATTR_REPLACE, handle);
3750 rc = dt_trans_start_local(env, dev, handle);
3754 dt_write_lock(env, parent, 0);
3755 if (unlikely(lfsck_is_dead_obj(parent)))
3756 GOTO(unlock, rc = 0);
3758 rc = lfsck_layout_get_lovea(env, parent, buf);
3759 if (unlikely(!rc || rc == -ENODATA))
3760 GOTO(unlock, rc = 0);
3763 magic = le32_to_cpu(lmm->lmm_magic);
3764 if (magic == LOV_MAGIC_COMP_V1) {
3765 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3766 struct lov_comp_md_entry_v1 *lcme;
3767 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3770 LASSERT(llr->llr_comp_id != 0);
3772 for (i = 0; i < count; i++) {
3773 lcme = &lcm->lcm_entries[i];
3774 if (le32_to_cpu(lcme->lcme_id) == llr->llr_comp_id) {
3775 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3778 le32_add_cpu(&lcm->lcm_layout_gen, 1);
3780 le32_to_cpu(lcme->lcme_offset);
3781 magic = le32_to_cpu(lmm->lmm_magic);
3786 GOTO(unlock, rc = 0);
3790 if (magic == LOV_MAGIC_V1) {
3791 objs = &lmm->lmm_objects[llr->llr_lov_idx];
3793 LASSERT(magic == LOV_MAGIC_V3);
3795 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
3798 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3799 index = le32_to_cpu(objs->l_ost_idx);
3800 rc = ostid_to_fid(&tfid, oi, index);
3801 /* Someone changed layout during the LFSCK, no need to repair then. */
3802 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
3803 GOTO(unlock, rc = 0);
3805 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3806 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3807 ostid_cpu_to_le(oi, &objs->l_ost_oi);
3808 objs->l_ost_gen = cpu_to_le32(0);
3809 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3810 rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3811 LU_XATTR_REPLACE, handle);
3813 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3816 dt_write_unlock(env, parent);
3820 dt_trans_stop(env, dev, handle);
3823 lfsck_ibits_unlock(&lh, LCK_EX);
3825 lfsck_object_put(env, child);
3828 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3829 "multiple references for: parent "DFID", comp_id %u, "
3830 "OST-index %u, stripe-index %u, owner %u/%u: rc = %d\n",
3831 lfsck_lfsck2name(lfsck), PFID(pfid),
3832 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3833 la->la_uid, la->la_gid, rc);
3838 /* If the MDT-object and the OST-object have different owner information,
3839 * then trust the MDT-object, because the normal chown/chgrp handle order
3840 * is from MDT to OST, and it is possible that some chown/chgrp operation
3841 * is partly done. */
3842 static int lfsck_layout_repair_owner(const struct lu_env *env,
3843 struct lfsck_component *com,
3844 struct dt_object *parent,
3845 struct lfsck_layout_req *llr,
3846 struct lu_attr *pla,
3847 const struct lu_attr *cla)
3849 struct lfsck_thread_info *info = lfsck_env_info(env);
3850 struct lu_attr *tla = &info->lti_la2;
3851 struct dt_object *child = llr->llr_child;
3852 struct dt_device *dev = lfsck_obj2dev(child);
3853 struct thandle *handle;
3857 tla->la_uid = pla->la_uid;
3858 tla->la_gid = pla->la_gid;
3859 tla->la_valid = LA_UID | LA_GID;
3860 handle = dt_trans_create(env, dev);
3862 GOTO(log, rc = PTR_ERR(handle));
3864 rc = dt_declare_attr_set(env, child, tla, handle);
3868 rc = dt_trans_start_local(env, dev, handle);
3872 /* Use the dt_object lock to serialize with destroy and attr_set. */
3873 dt_read_lock(env, parent, 0);
3874 if (unlikely(lfsck_is_dead_obj(parent)))
3875 GOTO(unlock, rc = 1);
3877 /* Get the latest parent's owner. */
3878 rc = dt_attr_get(env, parent, pla);
3882 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3883 if (unlikely(tla->la_uid != pla->la_uid ||
3884 tla->la_gid != pla->la_gid))
3887 rc = dt_attr_set(env, child, tla, handle);
3892 dt_read_unlock(env, parent);
3895 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3899 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3900 "inconsistent file owner for: parent "DFID", child "DFID
3901 ", OST-index %u, stripe-index %u, old owner %u/%u, "
3902 "new owner %u/%u: rc = %d\n",
3903 lfsck_lfsck2name(com->lc_lfsck),
3904 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3905 llr->llr_ost_idx, llr->llr_lov_idx,
3906 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3911 /* Check whether the OST-object correctly back points to the
3912 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3913 static int lfsck_layout_check_parent(const struct lu_env *env,
3914 struct lfsck_component *com,
3915 struct lfsck_assistant_object *lso,
3916 struct filter_fid *ff,
3917 const struct lu_fid *cfid,
3918 const struct lu_attr *cla,
3919 struct lfsck_layout_req *llr)
3921 struct lfsck_thread_info *info = lfsck_env_info(env);
3922 struct lu_buf *buf = &info->lti_big_buf;
3923 struct lu_fid *pfid = &info->lti_fid;
3924 struct dt_object *tobj;
3925 struct lov_mds_md_v1 *lmm;
3926 struct lov_ost_data_v1 *objs;
3927 struct lustre_handle lh = { 0 };
3935 *pfid = ff->ff_parent;
3936 idx = pfid->f_stripe_idx;
3939 if (unlikely(!fid_is_sane(pfid)))
3940 RETURN(LLIT_UNMATCHED_PAIR);
3942 if (lu_fid_eq(pfid, &lso->lso_fid)) {
3943 if (likely(llr->llr_lov_idx == idx))
3946 RETURN(LLIT_UNMATCHED_PAIR);
3949 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3951 RETURN(PTR_ERR(tobj));
3953 if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3954 !S_ISREG(lfsck_object_type(tobj)))
3955 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3957 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3958 * remote one on another MDT. Then check whether the given OST-object
3959 * is in such layout. If yes, it is multiple referenced, otherwise it
3960 * is unmatched referenced case. */
3961 rc = lfsck_layout_get_lovea(env, tobj, buf);
3962 if (rc == 0 || rc == -ENOENT)
3963 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3969 magic = le32_to_cpu(lmm->lmm_magic);
3970 if (magic == LOV_MAGIC_COMP_V1) {
3971 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3972 struct lov_comp_md_entry_v1 *lcme;
3974 if (ff->ff_layout.ol_comp_id == 0)
3975 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3977 count = le16_to_cpu(lcm->lcm_entry_count);
3978 for (i = 0; i < count; i++) {
3979 lcme = &lcm->lcm_entries[i];
3980 if (le32_to_cpu(lcme->lcme_id) ==
3981 ff->ff_layout.ol_comp_id) {
3983 le32_to_cpu(lcme->lcme_offset);
3984 magic = le32_to_cpu(lmm->lmm_magic);
3985 if (!(le32_to_cpu(lcme->lcme_flags) &
3987 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3993 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3997 if (magic == LOV_MAGIC_V1) {
3998 objs = &lmm->lmm_objects[0];
4000 LASSERT(magic == LOV_MAGIC_V3);
4001 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4004 count = le16_to_cpu(lmm->lmm_stripe_count);
4005 for (i = 0; i < count; i++, objs++) {
4006 struct lu_fid *tfid = &info->lti_fid2;
4007 struct ost_id *oi = &info->lti_oi;
4010 if (lovea_slot_is_dummy(objs))
4013 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4014 idx2 = le32_to_cpu(objs->l_ost_idx);
4015 rc = ostid_to_fid(tfid, oi, idx2);
4017 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
4018 "invalid layout EA at the slot %d, index %u\n",
4019 lfsck_lfsck2name(com->lc_lfsck),
4020 PFID(pfid), i, idx2);
4022 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4025 if (lu_fid_eq(cfid, tfid)) {
4026 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
4027 MDS_INODELOCK_UPDATE |
4028 MDS_INODELOCK_LAYOUT |
4029 MDS_INODELOCK_XATTR,
4034 dt_read_lock(env, tobj, 0);
4036 /* For local MDT-object, re-check existence
4037 * after taken the lock. */
4038 if (!dt_object_remote(tobj)) {
4039 if (dt_object_exists(tobj) == 0 ||
4040 lfsck_is_dead_obj(tobj))
4041 rc = LLIT_UNMATCHED_PAIR;
4043 rc = LLIT_MULTIPLE_REFERENCED;
4048 /* For migration case, the new MDT-object and old
4049 * MDT-object may reference the same OST-object at
4050 * some migration internal time.
4052 * For remote MDT-object, the local MDT may not know
4053 * whether it has been removed or not. Try checking
4054 * for a non-existent xattr to check if this object
4055 * has been been removed or not. */
4056 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
4058 if (unlikely(rc == -ENOENT || rc >= 0))
4059 rc = LLIT_UNMATCHED_PAIR;
4060 else if (rc == -ENODATA)
4061 rc = LLIT_MULTIPLE_REFERENCED;
4067 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4070 if (lustre_handle_is_used(&lh)) {
4071 dt_read_unlock(env, tobj);
4072 lfsck_ibits_unlock(&lh, LCK_EX);
4076 lfsck_object_put(env, tobj);
4081 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
4082 struct lfsck_component *com,
4083 struct lfsck_assistant_req *lar)
4085 struct lfsck_layout_req *llr =
4086 container_of0(lar, struct lfsck_layout_req, llr_lar);
4087 struct lfsck_assistant_object *lso = lar->lar_parent;
4088 struct lfsck_layout *lo = com->lc_file_ram;
4089 struct lfsck_thread_info *info = lfsck_env_info(env);
4090 struct filter_fid *ff = &info->lti_ff;
4091 struct lu_buf buf = { .lb_buf = ff,
4092 .lb_len = sizeof(*ff) };
4093 struct dt_object *parent = NULL;
4094 struct dt_object *child = llr->llr_child;
4095 struct lu_attr *pla = &lso->lso_attr;
4096 struct lu_attr *cla = &info->lti_la;
4097 struct lfsck_instance *lfsck = com->lc_lfsck;
4098 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4099 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
4106 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ASSISTANT_DIRECT, cfs_fail_val);
4108 rc = dt_attr_get(env, child, cla);
4109 if (rc == -ENOENT) {
4110 parent = lfsck_assistant_object_load(env, lfsck, lso);
4111 if (IS_ERR(parent)) {
4112 rc = PTR_ERR(parent);
4114 RETURN(rc == -ENOENT ? 0 : rc);
4117 type = LLIT_DANGLING;
4124 lfsck_buf_init(&buf, ff, sizeof(*ff));
4125 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
4126 if (unlikely(rc > 0 && rc < sizeof(struct lu_fid))) {
4127 type = LLIT_UNMATCHED_PAIR;
4131 if (rc < 0 && rc != -ENODATA)
4134 if (rc == 0 || rc == -ENODATA)
4135 GOTO(check_owner, rc = 0);
4137 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
4138 rc = lfsck_layout_check_parent(env, com, lso, ff,
4139 lu_object_fid(&child->do_lu), cla, llr);
4149 /* Someone may has changed the owner after the parent attr pre-loaded.
4150 * It can be handled later inside the lfsck_layout_repair_owner(). */
4151 if (unlikely(cla->la_uid != pla->la_uid ||
4152 cla->la_gid != pla->la_gid)) {
4153 type = LLIT_INCONSISTENT_OWNER;
4158 if (type == LLIT_NONE)
4161 if (bk->lb_param & LPF_DRYRUN)
4164 if (parent == NULL) {
4165 parent = lfsck_assistant_object_load(env, lfsck, lso);
4166 if (IS_ERR(parent)) {
4167 rc = PTR_ERR(parent);
4178 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
4179 rc = lfsck_layout_ins_dangling_rec(env, com,
4180 lfsck_dto2fid(parent), lfsck_dto2fid(child),
4181 llr->llr_comp_id, llr->llr_lov_idx,
4184 rc = __lfsck_layout_repair_dangling(env, com, parent,
4191 case LLIT_UNMATCHED_PAIR:
4192 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
4195 case LLIT_MULTIPLE_REFERENCED:
4196 rc = lfsck_layout_repair_multiple_references(env, com, parent,
4199 case LLIT_INCONSISTENT_OWNER:
4200 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
4210 down_write(&com->lc_sem);
4212 struct lfsck_assistant_data *lad = com->lc_data;
4214 if (unlikely(lad->lad_exit)) {
4216 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
4217 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
4218 rc == -EHOSTUNREACH) {
4219 /* If cannot touch the target server,
4220 * mark the LFSCK as INCOMPLETE. */
4221 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
4222 "talk with OST %x: rc = %d\n",
4223 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
4224 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
4225 lo->ll_objs_skipped++;
4228 lfsck_layout_record_failure(env, lfsck, lo);
4230 } else if (rc > 0 && (type != LLIT_DANGLING ||
4231 !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
4232 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
4233 "unknown type = %d\n", type);
4235 lo->ll_objs_repaired[type - 1]++;
4236 if (bk->lb_param & LPF_DRYRUN &&
4237 unlikely(lo->ll_pos_first_inconsistent == 0))
4238 lo->ll_pos_first_inconsistent =
4239 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
4242 up_write(&com->lc_sem);
4244 if (parent != NULL && !IS_ERR(parent))
4245 lfsck_object_put(env, parent);
4251 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
4252 struct lfsck_component *com,
4253 struct dt_object *obj, bool first)
4255 struct lfsck_instance *lfsck = com->lc_lfsck;
4256 struct ptlrpc_thread *thread = &lfsck->li_thread;
4257 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4258 struct lfsck_layout *lo = com->lc_file_ram;
4259 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
4262 struct lfsck_layout_dangling_key *parent =
4263 &lfsck_env_info(env)->lti_lldk;
4264 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
4269 di = iops->init(env, obj, 0);
4271 RETURN(PTR_ERR(di));
4274 lldk_cpu_to_be(parent, &lo->ll_lldk_latest_scanned_phase2);
4276 memset(parent, 0, sizeof(*parent));
4277 rc = iops->get(env, di, (const struct dt_key *)parent);
4282 /* The start one either has been processed or does not exist,
4284 rc = iops->next(env, di);
4290 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
4291 unlikely(!thread_is_running(thread)))
4294 key = iops->key(env, di);
4303 lldk_be_to_cpu(parent,
4304 (const struct lfsck_layout_dangling_key *)key);
4305 if (!fid_is_sane(&parent->lldk_fid)) {
4310 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
4312 fid_be_to_cpu(cfid, cfid);
4313 ost_idx = cfid->f_ver;
4315 if (!fid_is_sane(cfid)) {
4320 rc = lfsck_layout_repair_dangling(env, com,
4321 &parent->lldk_fid, cfid,
4322 parent->lldk_comp_id,
4323 parent->lldk_ea_off, ost_idx);
4327 down_write(&com->lc_sem);
4328 com->lc_new_checked++;
4329 com->lc_new_scanned++;
4331 lo->ll_lldk_latest_scanned_phase2 = *parent;
4334 lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
4336 lo->ll_objs_failed_phase2++;
4337 up_write(&com->lc_sem);
4339 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4342 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
4343 cfs_time_current())) &&
4344 com->lc_new_checked != 0) {
4345 down_write(&com->lc_sem);
4346 lo->ll_run_time_phase2 +=
4347 cfs_duration_sec(cfs_time_current() +
4348 HALF_SEC - com->lc_time_last_checkpoint);
4349 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4350 lo->ll_objs_checked_phase2 += com->lc_new_checked;
4351 com->lc_new_checked = 0;
4352 lfsck_layout_store(env, com);
4353 up_write(&com->lc_sem);
4355 com->lc_time_last_checkpoint = cfs_time_current();
4356 com->lc_time_next_checkpoint =
4357 com->lc_time_last_checkpoint +
4358 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4361 lfsck_control_speed_by_self(com);
4362 if (unlikely(!thread_is_running(thread)))
4365 rc = iops->next(env, di);
4374 iops->fini(env, di);
4379 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
4380 struct lfsck_component *com)
4382 struct lfsck_assistant_data *lad = com->lc_data;
4383 struct lfsck_instance *lfsck = com->lc_lfsck;
4384 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4385 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4386 struct lfsck_tgt_desc *ltd;
4390 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
4391 lfsck_lfsck2name(lfsck));
4393 spin_lock(<ds->ltd_lock);
4394 while (!list_empty(&lad->lad_ost_phase2_list)) {
4395 ltd = list_entry(lad->lad_ost_phase2_list.next,
4396 struct lfsck_tgt_desc,
4397 ltd_layout_phase_list);
4398 list_del_init(<d->ltd_layout_phase_list);
4399 if (bk->lb_param & LPF_OST_ORPHAN) {
4400 spin_unlock(<ds->ltd_lock);
4401 rc = lfsck_layout_scan_orphan(env, com, ltd);
4402 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
4405 if (unlikely(lad->lad_exit ||
4406 !thread_is_running(&lfsck->li_thread)))
4408 spin_lock(<ds->ltd_lock);
4412 if (list_empty(&lad->lad_ost_phase1_list))
4416 spin_unlock(<ds->ltd_lock);
4418 if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
4419 struct lfsck_layout *lo = com->lc_file_ram;
4422 com->lc_new_checked = 0;
4423 com->lc_new_scanned = 0;
4424 com->lc_time_last_checkpoint = cfs_time_current();
4425 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4426 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4428 i = lfsck_sub_trace_file_fid2idx(
4429 &lo->ll_lldk_latest_scanned_phase2.lldk_fid);
4430 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4431 com->lc_sub_trace_objs[i].lsto_obj, true);
4432 while (rc > 0 && ++i < LFSCK_STF_COUNT)
4433 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4434 com->lc_sub_trace_objs[i].lsto_obj, false);
4436 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
4437 "at the No. %d trace file: rc = %d\n",
4438 lfsck_lfsck2name(lfsck), i, rc);
4441 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
4442 lfsck_lfsck2name(lfsck), rc);
4448 lfsck_layout_slave_async_interpret(const struct lu_env *env,
4449 struct ptlrpc_request *req,
4452 struct lfsck_layout_slave_async_args *llsaa = args;
4453 struct obd_export *exp = llsaa->llsaa_exp;
4454 struct lfsck_component *com = llsaa->llsaa_com;
4455 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
4456 struct lfsck_layout_slave_data *llsd = com->lc_data;
4457 struct lfsck_reply *lr = NULL;
4461 /* It is probably caused by network trouble, or target crash,
4462 * it will try several times (depends on the obd_timeout, and
4463 * will not less than 3 times). But to make the LFSCK can go
4464 * ahead, we should not try for ever. After some try but still
4465 * hit failure, it will assume that the target exit the LFSCK
4466 * prcoessing and stop try. */
4467 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
4468 int max_try = max_t(int, obd_timeout / 30, 3);
4470 if (++(llst->llst_failures) > max_try)
4476 llst->llst_failures = 0;
4477 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
4478 if (lr->lr_status != LS_SCANNING_PHASE1 &&
4479 lr->lr_status != LS_SCANNING_PHASE2)
4484 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
4485 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
4486 llst->llst_index, lr != NULL ? lr->lr_status : rc,
4487 llst->llst_failures);
4489 lfsck_layout_llst_del(llsd, llst);
4492 lfsck_layout_llst_put(llst);
4493 lfsck_component_put(env, com);
4494 class_export_put(exp);
4499 static int lfsck_layout_async_query(const struct lu_env *env,
4500 struct lfsck_component *com,
4501 struct obd_export *exp,
4502 struct lfsck_layout_slave_target *llst,
4503 struct lfsck_request *lr,
4504 struct ptlrpc_request_set *set)
4506 struct lfsck_layout_slave_async_args *llsaa;
4507 struct ptlrpc_request *req;
4508 struct lfsck_request *tmp;
4512 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
4516 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
4518 ptlrpc_request_free(req);
4522 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4524 ptlrpc_request_set_replen(req);
4526 llsaa = ptlrpc_req_async_args(req);
4527 llsaa->llsaa_exp = exp;
4528 llsaa->llsaa_com = lfsck_component_get(com);
4529 llsaa->llsaa_llst = llst;
4530 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
4531 req->rq_allow_intr = 1;
4532 req->rq_no_delay = 1;
4533 ptlrpc_set_add_req(set, req);
4538 static int lfsck_layout_async_notify(const struct lu_env *env,
4539 struct obd_export *exp,
4540 struct lfsck_request *lr,
4541 struct ptlrpc_request_set *set)
4543 struct ptlrpc_request *req;
4544 struct lfsck_request *tmp;
4548 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4552 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4554 ptlrpc_request_free(req);
4558 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4560 ptlrpc_request_set_replen(req);
4561 req->rq_allow_intr = 1;
4562 req->rq_no_delay = 1;
4563 ptlrpc_set_add_req(set, req);
4569 lfsck_layout_slave_query_master(const struct lu_env *env,
4570 struct lfsck_component *com)
4572 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4573 struct lfsck_instance *lfsck = com->lc_lfsck;
4574 struct lfsck_layout_slave_data *llsd = com->lc_data;
4575 struct lfsck_layout_slave_target *llst;
4576 struct obd_export *exp;
4577 struct ptlrpc_request_set *set;
4582 set = ptlrpc_prep_set();
4584 GOTO(log, rc = -ENOMEM);
4586 memset(lr, 0, sizeof(*lr));
4587 lr->lr_event = LE_QUERY;
4588 lr->lr_active = LFSCK_TYPE_LAYOUT;
4590 llsd->llsd_touch_gen++;
4591 spin_lock(&llsd->llsd_lock);
4592 while (!list_empty(&llsd->llsd_master_list)) {
4593 llst = list_entry(llsd->llsd_master_list.next,
4594 struct lfsck_layout_slave_target,
4596 if (llst->llst_gen == llsd->llsd_touch_gen)
4599 llst->llst_gen = llsd->llsd_touch_gen;
4600 list_move_tail(&llst->llst_list,
4601 &llsd->llsd_master_list);
4602 atomic_inc(&llst->llst_ref);
4603 spin_unlock(&llsd->llsd_lock);
4605 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4608 lfsck_layout_llst_del(llsd, llst);
4609 lfsck_layout_llst_put(llst);
4610 spin_lock(&llsd->llsd_lock);
4614 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
4616 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4617 "query %s for layout: rc = %d\n",
4618 lfsck_lfsck2name(lfsck),
4619 exp->exp_obd->obd_name, rc);
4622 lfsck_layout_llst_put(llst);
4623 class_export_put(exp);
4625 spin_lock(&llsd->llsd_lock);
4627 spin_unlock(&llsd->llsd_lock);
4629 rc = ptlrpc_set_wait(set);
4630 ptlrpc_set_destroy(set);
4632 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4635 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4636 lfsck_lfsck2name(com->lc_lfsck), rc);
4642 lfsck_layout_slave_notify_master(const struct lu_env *env,
4643 struct lfsck_component *com,
4644 enum lfsck_events event, int result)
4646 struct lfsck_layout *lo = com->lc_file_ram;
4647 struct lfsck_instance *lfsck = com->lc_lfsck;
4648 struct lfsck_layout_slave_data *llsd = com->lc_data;
4649 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4650 struct lfsck_layout_slave_target *llst;
4651 struct obd_export *exp;
4652 struct ptlrpc_request_set *set;
4656 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4657 lfsck_lfsck2name(com->lc_lfsck));
4659 set = ptlrpc_prep_set();
4663 memset(lr, 0, sizeof(*lr));
4664 lr->lr_event = event;
4665 lr->lr_flags = LEF_FROM_OST;
4666 lr->lr_status = result;
4667 lr->lr_index = lfsck_dev_idx(lfsck);
4668 lr->lr_active = LFSCK_TYPE_LAYOUT;
4669 lr->lr_flags2 = lo->ll_flags;
4670 llsd->llsd_touch_gen++;
4671 spin_lock(&llsd->llsd_lock);
4672 while (!list_empty(&llsd->llsd_master_list)) {
4673 llst = list_entry(llsd->llsd_master_list.next,
4674 struct lfsck_layout_slave_target,
4676 if (llst->llst_gen == llsd->llsd_touch_gen)
4679 llst->llst_gen = llsd->llsd_touch_gen;
4680 list_move_tail(&llst->llst_list,
4681 &llsd->llsd_master_list);
4682 atomic_inc(&llst->llst_ref);
4683 spin_unlock(&llsd->llsd_lock);
4685 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4688 lfsck_layout_llst_del(llsd, llst);
4689 lfsck_layout_llst_put(llst);
4690 spin_lock(&llsd->llsd_lock);
4694 rc = lfsck_layout_async_notify(env, exp, lr, set);
4696 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4697 "notify %s for layout: rc = %d\n",
4698 lfsck_lfsck2name(lfsck),
4699 exp->exp_obd->obd_name, rc);
4701 lfsck_layout_llst_put(llst);
4702 class_export_put(exp);
4703 spin_lock(&llsd->llsd_lock);
4705 spin_unlock(&llsd->llsd_lock);
4707 ptlrpc_set_wait(set);
4708 ptlrpc_set_destroy(set);
4714 * \ret -ENODATA: unrecognized stripe
4715 * \ret = 0 : recognized stripe
4716 * \ret < 0 : other failures
4718 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4719 struct lfsck_component *com,
4720 struct lu_fid *cfid,
4721 struct lu_fid *pfid, __u32 comp_id)
4723 struct lfsck_thread_info *info = lfsck_env_info(env);
4724 struct lu_buf *buf = &info->lti_big_buf;
4725 struct ost_id *oi = &info->lti_oi;
4726 struct dt_object *obj;
4727 struct lov_mds_md_v1 *lmm;
4728 struct lov_ost_data_v1 *objs;
4729 __u32 idx = pfid->f_stripe_idx;
4737 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4739 RETURN(PTR_ERR(obj));
4741 dt_read_lock(env, obj, 0);
4742 if (unlikely(dt_object_exists(obj) == 0 ||
4743 lfsck_is_dead_obj(obj)))
4744 GOTO(unlock, rc = -ENOENT);
4746 if (!S_ISREG(lfsck_object_type(obj)))
4747 GOTO(unlock, rc = -ENODATA);
4749 rc = lfsck_layout_get_lovea(env, obj, buf);
4754 GOTO(unlock, rc = -ENODATA);
4757 magic = le32_to_cpu(lmm->lmm_magic);
4758 if (magic == LOV_MAGIC_COMP_V1) {
4759 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4760 struct lov_comp_md_entry_v1 *lcme;
4763 GOTO(unlock, rc = -ENODATA);
4765 count = le16_to_cpu(lcm->lcm_entry_count);
4766 for (i = 0; i < count; i++) {
4767 lcme = &lcm->lcm_entries[i];
4768 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
4770 le32_to_cpu(lcme->lcme_offset);
4771 magic = le32_to_cpu(lmm->lmm_magic);
4772 if (!(le32_to_cpu(lcme->lcme_flags) &
4774 GOTO(unlock, rc = -ENODATA);
4780 GOTO(unlock, rc = -ENODATA);
4784 if (magic == LOV_MAGIC_V1) {
4785 objs = &lmm->lmm_objects[0];
4787 LASSERT(magic == LOV_MAGIC_V3);
4788 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4791 fid_to_ostid(cfid, oi);
4792 count = le16_to_cpu(lmm->lmm_stripe_count);
4793 for (i = 0; i < count; i++, objs++) {
4796 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
4797 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
4798 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
4801 GOTO(unlock, rc = -ENODATA);
4804 dt_read_unlock(env, obj);
4805 lfsck_object_put(env, obj);
4811 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
4812 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
4813 * layout EA from MDT to OST. On one hand, the OST no need to understand
4814 * the layout EA structure; on the other hand, it may cause trouble when
4815 * transfer large layout EA from MDT to OST via normal OUT RPC.
4817 * \ret > 0: unrecognized stripe
4818 * \ret = 0: recognized stripe
4819 * \ret < 0: other failures
4821 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
4822 struct lfsck_component *com,
4823 struct lu_fid *cfid,
4824 struct lu_fid *pfid, __u32 comp_id)
4826 struct lfsck_instance *lfsck = com->lc_lfsck;
4827 struct obd_device *obd = lfsck->li_obd;
4828 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4829 struct obd_export *exp = NULL;
4830 struct ptlrpc_request *req = NULL;
4831 struct lfsck_request *lr;
4832 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
4836 if (unlikely(fid_is_idif(pfid)))
4839 fld_range_set_any(range);
4840 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
4842 RETURN(rc == -ENOENT ? 1 : rc);
4844 if (unlikely(!fld_range_is_mdt(range)))
4847 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
4848 if (unlikely(exp == NULL))
4851 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
4852 GOTO(out, rc = -EOPNOTSUPP);
4854 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4856 GOTO(out, rc = -ENOMEM);
4858 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4860 ptlrpc_request_free(req);
4865 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4866 memset(lr, 0, sizeof(*lr));
4867 lr->lr_event = LE_PAIRS_VERIFY;
4868 lr->lr_active = LFSCK_TYPE_LAYOUT;
4869 lr->lr_fid = *cfid; /* OST-object itself FID. */
4870 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
4871 lr->lr_comp_id = comp_id;
4873 ptlrpc_request_set_replen(req);
4874 rc = ptlrpc_queue_wait(req);
4875 ptlrpc_req_finished(req);
4877 if (rc == -ENOENT || rc == -ENODATA)
4884 class_export_put(exp);
4889 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
4890 struct lfsck_component *com,
4891 struct lfsck_req_local *lrl)
4893 struct dt_object *obj;
4897 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lrl->lrl_fid);
4899 GOTO(log, rc = PTR_ERR(obj));
4901 dt_write_lock(env, obj, 0);
4902 if (unlikely(dt_object_exists(obj) == 0 ||
4903 lfsck_is_dead_obj(obj)))
4904 GOTO(unlock, rc = 0);
4906 rc = __lfsck_layout_update_pfid(env, obj, &lrl->lrl_ff_client.ff_parent,
4907 &lrl->lrl_ff_client.ff_layout,
4908 lrl->lrl_ff_client.ff_parent.f_ver);
4913 dt_write_unlock(env, obj);
4914 lfsck_object_put(env, obj);
4917 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
4918 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
4919 PFID(&lrl->lrl_fid), PFID(&lrl->lrl_ff_client.ff_parent), rc);
4926 static void lfsck_layout_slave_quit(const struct lu_env *env,
4927 struct lfsck_component *com);
4929 static int lfsck_layout_reset(const struct lu_env *env,
4930 struct lfsck_component *com, bool init)
4932 struct lfsck_layout *lo = com->lc_file_ram;
4935 down_write(&com->lc_sem);
4937 memset(lo, 0, com->lc_file_size);
4939 __u32 count = lo->ll_success_count;
4940 __u64 last_time = lo->ll_time_last_complete;
4942 memset(lo, 0, com->lc_file_size);
4943 lo->ll_success_count = count;
4944 lo->ll_time_last_complete = last_time;
4947 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
4948 lo->ll_status = LS_INIT;
4950 if (com->lc_lfsck->li_master) {
4951 struct lfsck_assistant_data *lad = com->lc_data;
4953 lad->lad_incomplete = 0;
4954 CFS_RESET_BITMAP(lad->lad_bitmap);
4957 rc = lfsck_layout_store(env, com);
4958 if (rc == 0 && com->lc_lfsck->li_master)
4959 rc = lfsck_load_sub_trace_files(env, com,
4960 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
4961 up_write(&com->lc_sem);
4963 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
4964 lfsck_lfsck2name(com->lc_lfsck), rc);
4969 static void lfsck_layout_fail(const struct lu_env *env,
4970 struct lfsck_component *com, bool new_checked)
4972 struct lfsck_layout *lo = com->lc_file_ram;
4974 down_write(&com->lc_sem);
4976 com->lc_new_checked++;
4977 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
4978 up_write(&com->lc_sem);
4981 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
4982 struct lfsck_component *com, bool init)
4984 struct lfsck_instance *lfsck = com->lc_lfsck;
4985 struct lfsck_layout *lo = com->lc_file_ram;
4989 rc = lfsck_checkpoint_generic(env, com);
4991 return rc > 0 ? 0 : rc;
4994 down_write(&com->lc_sem);
4996 lo->ll_pos_latest_start =
4997 lfsck->li_pos_checkpoint.lp_oit_cookie;
4999 lo->ll_pos_last_checkpoint =
5000 lfsck->li_pos_checkpoint.lp_oit_cookie;
5001 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5002 HALF_SEC - lfsck->li_time_last_checkpoint);
5003 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5004 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5005 com->lc_new_checked = 0;
5008 rc = lfsck_layout_store(env, com);
5009 up_write(&com->lc_sem);
5011 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
5012 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5013 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5018 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
5019 struct lfsck_component *com, bool init)
5021 struct lfsck_instance *lfsck = com->lc_lfsck;
5022 struct lfsck_layout *lo = com->lc_file_ram;
5025 if (com->lc_new_checked == 0 && !init)
5028 down_write(&com->lc_sem);
5030 lo->ll_pos_latest_start =
5031 lfsck->li_pos_checkpoint.lp_oit_cookie;
5033 lo->ll_pos_last_checkpoint =
5034 lfsck->li_pos_checkpoint.lp_oit_cookie;
5035 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5036 HALF_SEC - lfsck->li_time_last_checkpoint);
5037 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5038 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5039 com->lc_new_checked = 0;
5042 rc = lfsck_layout_store(env, com);
5043 up_write(&com->lc_sem);
5045 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
5046 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5047 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5052 static int lfsck_layout_prep(const struct lu_env *env,
5053 struct lfsck_component *com,
5054 struct lfsck_start *start)
5056 struct lfsck_instance *lfsck = com->lc_lfsck;
5057 struct lfsck_layout *lo = com->lc_file_ram;
5058 struct lfsck_position *pos = &com->lc_pos_start;
5060 fid_zero(&pos->lp_dir_parent);
5061 pos->lp_dir_cookie = 0;
5062 if (lo->ll_status == LS_COMPLETED ||
5063 lo->ll_status == LS_PARTIAL ||
5064 /* To handle orphan, must scan from the beginning. */
5065 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
5068 rc = lfsck_layout_reset(env, com, false);
5070 rc = lfsck_set_param(env, lfsck, start, true);
5073 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
5074 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
5080 down_write(&com->lc_sem);
5081 lo->ll_time_latest_start = cfs_time_current_sec();
5082 spin_lock(&lfsck->li_lock);
5083 if (lo->ll_flags & LF_SCANNED_ONCE) {
5084 if (!lfsck->li_drop_dryrun ||
5085 lo->ll_pos_first_inconsistent == 0) {
5086 lo->ll_status = LS_SCANNING_PHASE2;
5087 list_move_tail(&com->lc_link,
5088 &lfsck->li_list_double_scan);
5089 pos->lp_oit_cookie = 0;
5093 lo->ll_status = LS_SCANNING_PHASE1;
5094 lo->ll_run_time_phase1 = 0;
5095 lo->ll_run_time_phase2 = 0;
5096 lo->ll_objs_checked_phase1 = 0;
5097 lo->ll_objs_checked_phase2 = 0;
5098 lo->ll_objs_failed_phase1 = 0;
5099 lo->ll_objs_failed_phase2 = 0;
5100 for (i = 0; i < LLIT_MAX; i++)
5101 lo->ll_objs_repaired[i] = 0;
5103 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5104 fid_zero(&com->lc_fid_latest_scanned_phase2);
5107 lo->ll_status = LS_SCANNING_PHASE1;
5108 if (!lfsck->li_drop_dryrun ||
5109 lo->ll_pos_first_inconsistent == 0)
5110 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
5112 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5114 spin_unlock(&lfsck->li_lock);
5115 up_write(&com->lc_sem);
5120 static int lfsck_layout_slave_prep(const struct lu_env *env,
5121 struct lfsck_component *com,
5122 struct lfsck_start_param *lsp)
5124 struct lfsck_layout_slave_data *llsd = com->lc_data;
5125 struct lfsck_instance *lfsck = com->lc_lfsck;
5126 struct lfsck_layout *lo = com->lc_file_ram;
5127 struct lfsck_start *start = lsp->lsp_start;
5130 rc = lfsck_layout_prep(env, com, start);
5134 if (lo->ll_flags & LF_CRASHED_LASTID &&
5135 list_empty(&llsd->llsd_master_list)) {
5136 LASSERT(lfsck->li_out_notify != NULL);
5138 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5139 LE_LASTID_REBUILDING);
5142 if (!lsp->lsp_index_valid)
5145 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5146 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
5147 LASSERT(!llsd->llsd_rbtree_valid);
5149 write_lock(&llsd->llsd_rb_lock);
5150 rc = lfsck_rbtree_setup(env, com);
5151 write_unlock(&llsd->llsd_rb_lock);
5154 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
5155 "%llu]\n", lfsck_lfsck2name(lfsck),
5156 com->lc_pos_start.lp_oit_cookie);
5161 static int lfsck_layout_master_prep(const struct lu_env *env,
5162 struct lfsck_component *com,
5163 struct lfsck_start_param *lsp)
5168 rc = lfsck_layout_load_bitmap(env, com);
5170 rc = lfsck_layout_reset(env, com, false);
5172 rc = lfsck_set_param(env, com->lc_lfsck,
5173 lsp->lsp_start, true);
5179 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
5183 rc = lfsck_start_assistant(env, com, lsp);
5188 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
5189 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
5190 com->lc_pos_start.lp_oit_cookie);
5195 /* Pre-fetch the attribute for each stripe in the given layout EA. */
5196 static int lfsck_layout_scan_stripes(const struct lu_env *env,
5197 struct lfsck_component *com,
5198 struct dt_object *parent,
5199 struct lov_mds_md_v1 *lmm, __u32 comp_id)
5201 struct lfsck_thread_info *info = lfsck_env_info(env);
5202 struct lfsck_instance *lfsck = com->lc_lfsck;
5203 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5204 struct lfsck_layout *lo = com->lc_file_ram;
5205 struct lfsck_assistant_data *lad = com->lc_data;
5206 struct lfsck_assistant_object *lso = NULL;
5207 struct lov_ost_data_v1 *objs;
5208 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
5209 struct ptlrpc_thread *mthread = &lfsck->li_thread;
5210 struct ptlrpc_thread *athread = &lad->lad_thread;
5211 struct l_wait_info lwi = { 0 };
5219 lfsck_buf_init(&buf, &info->lti_ff, sizeof(struct filter_fid));
5220 magic = le32_to_cpu(lmm->lmm_magic);
5221 if (magic == LOV_MAGIC_V1) {
5222 objs = &lmm->lmm_objects[0];
5224 LASSERT(magic == LOV_MAGIC_V3);
5225 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5228 count = le16_to_cpu(lmm->lmm_stripe_count);
5229 for (i = 0; i < count; i++, objs++) {
5230 struct lu_fid *fid = &info->lti_fid;
5231 struct ost_id *oi = &info->lti_oi;
5232 struct lfsck_layout_req *llr;
5233 struct lfsck_tgt_desc *tgt = NULL;
5234 struct dt_object *cobj = NULL;
5236 bool wakeup = false;
5238 if (unlikely(lovea_slot_is_dummy(objs)))
5241 l_wait_event(mthread->t_ctl_waitq,
5242 lad->lad_prefetched < bk->lb_async_windows ||
5243 !thread_is_running(mthread) ||
5244 thread_is_stopped(athread),
5247 if (unlikely(!thread_is_running(mthread)) ||
5248 thread_is_stopped(athread))
5251 if (unlikely(lfsck_is_dead_obj(parent)))
5254 ostid_le_to_cpu(&objs->l_ost_oi, oi);
5255 index = le32_to_cpu(objs->l_ost_idx);
5256 rc = ostid_to_fid(fid, oi, index);
5258 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
5259 ": "DOSTID", idx %u, comp_id %u\n",
5260 lfsck_lfsck2name(lfsck),
5261 PFID(lfsck_dto2fid(parent)), POSTID(oi),
5266 tgt = lfsck_tgt_get(ltds, index);
5267 if (unlikely(tgt == NULL)) {
5268 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
5269 "did not join the layout LFSCK, comp_id %u\n",
5270 lfsck_lfsck2name(lfsck), index, comp_id);
5271 lfsck_lad_set_bitmap(env, com, index);
5275 /* There is potential deadlock race condition between object
5276 * destroy and layout LFSCK. Consider the following scenario:
5278 * 1) The LFSCK thread obtained the parent object firstly, at
5279 * that time, the parent object has not been destroyed yet.
5281 * 2) One RPC service thread destroyed the parent and all its
5282 * children objects. Because the LFSCK is referencing the
5283 * parent object, then the parent object will be marked as
5284 * dying in RAM. On the other hand, the parent object is
5285 * referencing all its children objects, then all children
5286 * objects will be marked as dying in RAM also.
5288 * 3) The LFSCK thread tries to find some child object with
5289 * the parent object referenced. Then it will find that the
5290 * child object is dying. According to the object visibility
5291 * rules: the object with dying flag cannot be returned to
5292 * others. So the LFSCK thread has to wait until the dying
5293 * object has been purged from RAM, then it can allocate a
5294 * new object (with the same FID) in RAM. Unfortunately, the
5295 * LFSCK thread itself is referencing the parent object, and
5296 * cause the parent object cannot be purged, then cause the
5297 * child object cannot be purged also. So the LFSCK thread
5298 * will fall into deadlock.
5300 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
5302 if (lfsck_is_dead_obj(parent)) {
5312 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_ASSISTANT_DIRECT)) {
5313 rc = dt_declare_attr_get(env, cobj);
5317 rc = dt_declare_xattr_get(env, cobj, &buf,
5324 struct lu_attr *attr = &info->lti_la;
5326 rc = dt_attr_get(env, parent, attr);
5330 lso = lfsck_assistant_object_init(env,
5331 lfsck_dto2fid(parent), attr,
5332 lfsck->li_pos_current.lp_oit_cookie, false);
5341 llr = lfsck_layout_assistant_req_init(lso, cobj, comp_id,
5349 spin_lock(&lad->lad_lock);
5350 if (lad->lad_assistant_status < 0) {
5351 spin_unlock(&lad->lad_lock);
5352 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
5354 RETURN(lad->lad_assistant_status);
5357 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
5358 if (lad->lad_prefetched == 0)
5361 lad->lad_prefetched++;
5362 spin_unlock(&lad->lad_lock);
5364 wake_up_all(&athread->t_ctl_waitq);
5367 down_write(&com->lc_sem);
5368 com->lc_new_checked++;
5370 lfsck_layout_record_failure(env, lfsck, lo);
5371 up_write(&com->lc_sem);
5373 if (cobj != NULL && !IS_ERR(cobj))
5374 lfsck_object_put(env, cobj);
5376 if (likely(tgt != NULL))
5379 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5387 lfsck_assistant_object_put(env, lso);
5392 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
5393 * the OST-object's attribute and generate an structure lfsck_layout_req on the
5394 * list ::lad_req_list.
5396 * For each request on above list, the lfsck_layout_assistant thread compares
5397 * the OST side attribute with local attribute, if inconsistent, then repair it.
5399 * All above processing is async mode with pipeline. */
5400 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
5401 struct lfsck_component *com,
5402 struct dt_object *obj)
5404 struct lfsck_thread_info *info = lfsck_env_info(env);
5405 struct ost_id *oi = &info->lti_oi;
5406 struct lfsck_layout *lo = com->lc_file_ram;
5407 struct lfsck_assistant_data *lad = com->lc_data;
5408 struct lfsck_instance *lfsck = com->lc_lfsck;
5409 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5410 struct thandle *handle = NULL;
5411 struct lu_buf *buf = &info->lti_big_buf;
5412 struct lov_mds_md_v1 *lmm = NULL;
5413 struct dt_device *dev = lfsck_obj2dev(obj);
5414 struct lustre_handle lh = { 0 };
5415 struct lu_buf ea_buf = { NULL };
5416 struct lov_comp_md_v1 *lcm = NULL;
5417 struct lov_comp_md_entry_v1 *lcme = NULL;
5422 bool locked = false;
5423 bool stripe = false;
5424 bool bad_oi = false;
5427 if (!S_ISREG(lfsck_object_type(obj)))
5430 if (lad->lad_assistant_status < 0)
5431 GOTO(out, rc = -ESRCH);
5433 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
5434 lmm_oi_cpu_to_le(oi, oi);
5435 dt_read_lock(env, obj, 0);
5440 if (dt_object_exists(obj) == 0 ||
5441 lfsck_is_dead_obj(obj))
5444 rc = lfsck_layout_get_lovea(env, obj, buf);
5446 /* Skip bad lov EA during the 1st cycle scanning, and
5447 * try to recover it via orphan in the 2nd scanning. */
5448 GOTO(out, rc = (rc == -EINVAL ? 0 : rc));
5452 magic = le32_to_cpu(lmm->lmm_magic);
5453 if (magic == LOV_MAGIC_COMP_V1) {
5457 count = le16_to_cpu(lcm->lcm_entry_count);
5458 for (i = 0; i < count; i++) {
5459 lcme = &lcm->lcm_entries[i];
5460 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5461 if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) != 0)
5465 GOTO(out, stripe = true);
5466 } else if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0) {
5467 GOTO(out, stripe = true);
5471 /* Inconsistent lmm_oi, should be repaired. */
5474 if (bk->lb_param & LPF_DRYRUN) {
5475 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5477 GOTO(out, stripe = true);
5480 if (!lustre_handle_is_used(&lh)) {
5481 dt_read_unlock(env, obj);
5483 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5484 MDS_INODELOCK_LAYOUT |
5485 MDS_INODELOCK_XATTR, LCK_EX);
5489 handle = dt_trans_create(env, dev);
5491 GOTO(out, rc = PTR_ERR(handle));
5493 lfsck_buf_init(&ea_buf, lmm, size);
5494 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5495 LU_XATTR_REPLACE, handle);
5499 rc = dt_trans_start_local(env, dev, handle);
5503 dt_write_lock(env, obj, 0);
5509 if (magic == LOV_MAGIC_COMP_V1) {
5512 for (i = 0; i < count; i++) {
5513 lcme = &lcm->lcm_entries[i];
5514 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5521 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5522 LU_XATTR_REPLACE, handle);
5526 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5528 GOTO(out, stripe = true);
5532 if (lustre_handle_is_used(&lh))
5533 dt_write_unlock(env, obj);
5535 dt_read_unlock(env, obj);
5538 if (handle != NULL && !IS_ERR(handle))
5539 dt_trans_stop(env, dev, handle);
5541 lfsck_ibits_unlock(&lh, LCK_EX);
5544 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
5545 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
5546 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
5547 PFID(lfsck_dto2fid(obj)), rc);
5550 if (magic == LOV_MAGIC_COMP_V1) {
5553 for (i = 0; i < count; i++) {
5554 lcme = &lcm->lcm_entries[i];
5555 if (!(le32_to_cpu(lcme->lcme_flags) &
5559 rc = lfsck_layout_scan_stripes(env, com, obj,
5560 (struct lov_mds_md_v1 *)(buf->lb_buf +
5561 le32_to_cpu(lcme->lcme_offset)),
5562 le32_to_cpu(lcme->lcme_id));
5565 rc = lfsck_layout_scan_stripes(env, com, obj, lmm, 0);
5568 down_write(&com->lc_sem);
5569 com->lc_new_checked++;
5571 lfsck_layout_record_failure(env, lfsck, lo);
5572 up_write(&com->lc_sem);
5578 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
5579 struct lfsck_component *com,
5580 struct dt_object *obj)
5582 struct lfsck_instance *lfsck = com->lc_lfsck;
5583 struct lfsck_layout *lo = com->lc_file_ram;
5584 const struct lu_fid *fid = lfsck_dto2fid(obj);
5585 struct lfsck_layout_slave_data *llsd = com->lc_data;
5586 struct lfsck_layout_seq *lls;
5592 LASSERT(llsd != NULL);
5594 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
5595 cfs_fail_val == lfsck_dev_idx(lfsck)) {
5596 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(1),
5598 struct ptlrpc_thread *thread = &lfsck->li_thread;
5600 l_wait_event(thread->t_ctl_waitq,
5601 !thread_is_running(thread),
5605 lfsck_rbtree_update_bitmap(env, com, fid, false);
5607 down_write(&com->lc_sem);
5608 if (fid_is_idif(fid))
5610 else if (!fid_is_norm(fid) ||
5611 !fid_is_for_ostobj(env, lfsck, obj, fid))
5612 GOTO(unlock, rc = 0);
5615 com->lc_new_checked++;
5617 lls = lfsck_layout_seq_lookup(llsd, seq);
5620 if (unlikely(lls == NULL))
5621 GOTO(unlock, rc = -ENOMEM);
5623 INIT_LIST_HEAD(&lls->lls_list);
5625 rc = lfsck_layout_lastid_load(env, com, lls);
5627 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5628 "load LAST_ID for %#llx: rc = %d\n",
5629 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
5630 lo->ll_objs_failed_phase1++;
5635 lfsck_layout_seq_insert(llsd, lls);
5638 if (unlikely(fid_is_last_id(fid)))
5639 GOTO(unlock, rc = 0);
5641 if (fid_is_idif(fid))
5642 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
5646 if (oid > lls->lls_lastid_known)
5647 lls->lls_lastid_known = oid;
5649 if (oid > lls->lls_lastid) {
5650 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
5651 /* OFD may create new objects during LFSCK scanning. */
5652 rc = lfsck_layout_lastid_reload(env, com, lls);
5653 if (unlikely(rc != 0)) {
5654 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5655 "reload LAST_ID for %#llx: rc = %d\n",
5656 lfsck_lfsck2name(com->lc_lfsck),
5662 if (oid <= lls->lls_lastid ||
5663 lo->ll_flags & LF_CRASHED_LASTID)
5664 GOTO(unlock, rc = 0);
5666 LASSERT(lfsck->li_out_notify != NULL);
5668 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5669 LE_LASTID_REBUILDING);
5670 lo->ll_flags |= LF_CRASHED_LASTID;
5672 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
5673 "LAST_ID file (2) for the sequence %#llx"
5674 ", old value %llu, known value %llu\n",
5675 lfsck_lfsck2name(lfsck), lls->lls_seq,
5676 lls->lls_lastid, oid);
5679 lls->lls_lastid = oid;
5683 GOTO(unlock, rc = 0);
5686 up_write(&com->lc_sem);
5691 static int lfsck_layout_exec_dir(const struct lu_env *env,
5692 struct lfsck_component *com,
5693 struct lfsck_assistant_object *lso,
5694 struct lu_dirent *ent, __u16 type)
5699 static int lfsck_layout_master_post(const struct lu_env *env,
5700 struct lfsck_component *com,
5701 int result, bool init)
5703 struct lfsck_instance *lfsck = com->lc_lfsck;
5704 struct lfsck_layout *lo = com->lc_file_ram;
5708 lfsck_post_generic(env, com, &result);
5710 down_write(&com->lc_sem);
5711 spin_lock(&lfsck->li_lock);
5713 lo->ll_pos_last_checkpoint =
5714 lfsck->li_pos_checkpoint.lp_oit_cookie;
5717 if (lo->ll_flags & LF_INCOMPLETE)
5718 lo->ll_status = LS_PARTIAL;
5720 lo->ll_status = LS_SCANNING_PHASE2;
5721 lo->ll_flags |= LF_SCANNED_ONCE;
5722 lo->ll_flags &= ~LF_UPGRADE;
5723 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5724 } else if (result == 0) {
5725 if (lfsck->li_status != 0)
5726 lo->ll_status = lfsck->li_status;
5728 lo->ll_status = LS_STOPPED;
5729 if (lo->ll_status != LS_PAUSED)
5730 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5732 lo->ll_status = LS_FAILED;
5733 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5735 spin_unlock(&lfsck->li_lock);
5738 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5739 HALF_SEC - lfsck->li_time_last_checkpoint);
5740 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5741 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5742 com->lc_new_checked = 0;
5745 rc = lfsck_layout_store(env, com);
5746 up_write(&com->lc_sem);
5748 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
5749 lfsck_lfsck2name(lfsck), rc);
5754 static int lfsck_layout_slave_post(const struct lu_env *env,
5755 struct lfsck_component *com,
5756 int result, bool init)
5758 struct lfsck_instance *lfsck = com->lc_lfsck;
5759 struct lfsck_layout *lo = com->lc_file_ram;
5763 down_write(&com->lc_sem);
5764 rc = lfsck_layout_lastid_store(env, com);
5768 LASSERT(lfsck->li_out_notify != NULL);
5770 spin_lock(&lfsck->li_lock);
5772 lo->ll_pos_last_checkpoint =
5773 lfsck->li_pos_checkpoint.lp_oit_cookie;
5776 lo->ll_status = LS_SCANNING_PHASE2;
5777 lo->ll_flags |= LF_SCANNED_ONCE;
5778 if (lo->ll_flags & LF_CRASHED_LASTID) {
5780 lo->ll_flags &= ~LF_CRASHED_LASTID;
5782 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
5783 "crashed LAST_ID files successfully\n",
5784 lfsck_lfsck2name(lfsck));
5786 lo->ll_flags &= ~LF_UPGRADE;
5787 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5788 } else if (result == 0) {
5789 if (lfsck->li_status != 0)
5790 lo->ll_status = lfsck->li_status;
5792 lo->ll_status = LS_STOPPED;
5793 if (lo->ll_status != LS_PAUSED)
5794 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5796 lo->ll_status = LS_FAILED;
5797 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5799 spin_unlock(&lfsck->li_lock);
5802 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5806 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5807 HALF_SEC - lfsck->li_time_last_checkpoint);
5808 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5809 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5810 com->lc_new_checked = 0;
5813 rc = lfsck_layout_store(env, com);
5814 up_write(&com->lc_sem);
5816 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
5818 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
5819 lfsck_lfsck2name(lfsck), rc);
5824 static void lfsck_layout_dump(const struct lu_env *env,
5825 struct lfsck_component *com, struct seq_file *m)
5827 struct lfsck_instance *lfsck = com->lc_lfsck;
5828 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5829 struct lfsck_layout *lo = com->lc_file_ram;
5832 down_read(&com->lc_sem);
5833 if (bk->lb_param & LPF_DRYRUN)
5834 prefix = "inconsistent";
5836 prefix = "repaired";
5838 seq_printf(m, "name: lfsck_layout\n"
5844 lfsck_status2name(lo->ll_status));
5846 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
5848 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
5850 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
5852 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
5854 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
5856 seq_printf(m, "latest_start_position: %llu\n"
5857 "last_checkpoint_position: %llu\n"
5858 "first_failure_position: %llu\n",
5859 lo->ll_pos_latest_start,
5860 lo->ll_pos_last_checkpoint,
5861 lo->ll_pos_first_inconsistent);
5863 seq_printf(m, "success_count: %u\n"
5864 "%s_dangling: %llu\n"
5865 "%s_unmatched_pair: %llu\n"
5866 "%s_multiple_referenced: %llu\n"
5868 "%s_inconsistent_owner: %llu\n"
5871 "failed_phase1: %llu\n"
5872 "failed_phase2: %llu\n",
5873 lo->ll_success_count,
5874 prefix, lo->ll_objs_repaired[LLIT_DANGLING - 1],
5875 prefix, lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
5876 prefix, lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
5877 prefix, lo->ll_objs_repaired[LLIT_ORPHAN - 1],
5878 prefix, lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
5879 prefix, lo->ll_objs_repaired[LLIT_OTHERS - 1],
5880 lo->ll_objs_skipped,
5881 lo->ll_objs_failed_phase1,
5882 lo->ll_objs_failed_phase2);
5884 if (lo->ll_status == LS_SCANNING_PHASE1) {
5886 cfs_duration_t duration = cfs_time_current() -
5887 lfsck->li_time_last_checkpoint;
5888 __u64 checked = lo->ll_objs_checked_phase1 +
5889 com->lc_new_checked;
5890 __u64 speed = checked;
5891 __u64 new_checked = com->lc_new_checked *
5892 msecs_to_jiffies(MSEC_PER_SEC);
5893 __u32 rtime = lo->ll_run_time_phase1 +
5894 cfs_duration_sec(duration + HALF_SEC);
5897 new_checked = div64_s64(new_checked, duration);
5899 speed = div64_s64(speed, rtime);
5900 seq_printf(m, "checked_phase1: %llu\n"
5901 "checked_phase2: %llu\n"
5902 "run_time_phase1: %u seconds\n"
5903 "run_time_phase2: %u seconds\n"
5904 "average_speed_phase1: %llu items/sec\n"
5905 "average_speed_phase2: N/A\n"
5906 "real-time_speed_phase1: %llu items/sec\n"
5907 "real-time_speed_phase2: N/A\n",
5909 lo->ll_objs_checked_phase2,
5911 lo->ll_run_time_phase2,
5915 if (likely(lfsck->li_di_oit)) {
5916 const struct dt_it_ops *iops =
5917 &lfsck->li_obj_oit->do_index_ops->dio_it;
5919 /* The low layer otable-based iteration position may NOT
5920 * exactly match the layout-based directory traversal
5921 * cookie. Generally, it is not a serious issue. But the
5922 * caller should NOT make assumption on that. */
5923 pos = iops->store(env, lfsck->li_di_oit);
5924 if (!lfsck->li_current_oit_processed)
5927 pos = lo->ll_pos_last_checkpoint;
5930 seq_printf(m, "current_position: %llu\n", pos);
5931 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
5932 cfs_duration_t duration = cfs_time_current() -
5933 com->lc_time_last_checkpoint;
5934 __u64 checked = lo->ll_objs_checked_phase2 +
5935 com->lc_new_checked;
5936 __u64 speed1 = lo->ll_objs_checked_phase1;
5937 __u64 speed2 = checked;
5938 __u64 new_checked = com->lc_new_checked *
5939 msecs_to_jiffies(MSEC_PER_SEC);
5940 __u32 rtime = lo->ll_run_time_phase2 +
5941 cfs_duration_sec(duration + HALF_SEC);
5944 new_checked = div64_s64(new_checked, duration);
5945 if (lo->ll_run_time_phase1 != 0)
5946 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
5948 speed2 = div64_s64(speed2, rtime);
5949 seq_printf(m, "checked_phase1: %llu\n"
5950 "checked_phase2: %llu\n"
5951 "run_time_phase1: %u seconds\n"
5952 "run_time_phase2: %u seconds\n"
5953 "average_speed_phase1: %llu items/sec\n"
5954 "average_speed_phase2: %llu items/sec\n"
5955 "real-time_speed_phase1: N/A\n"
5956 "real-time_speed_phase2: %llu items/sec\n"
5957 "current_position: "DFID"\n",
5958 lo->ll_objs_checked_phase1,
5960 lo->ll_run_time_phase1,
5965 PFID(&com->lc_fid_latest_scanned_phase2));
5967 __u64 speed1 = lo->ll_objs_checked_phase1;
5968 __u64 speed2 = lo->ll_objs_checked_phase2;
5970 if (lo->ll_run_time_phase1 != 0)
5971 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
5972 if (lo->ll_run_time_phase2 != 0)
5973 speed2 = div64_s64(speed2, lo->ll_run_time_phase2);
5974 seq_printf(m, "checked_phase1: %llu\n"
5975 "checked_phase2: %llu\n"
5976 "run_time_phase1: %u seconds\n"
5977 "run_time_phase2: %u seconds\n"
5978 "average_speed_phase1: %llu items/sec\n"
5979 "average_speed_phase2: %llu objs/sec\n"
5980 "real-time_speed_phase1: N/A\n"
5981 "real-time_speed_phase2: N/A\n"
5982 "current_position: N/A\n",
5983 lo->ll_objs_checked_phase1,
5984 lo->ll_objs_checked_phase2,
5985 lo->ll_run_time_phase1,
5986 lo->ll_run_time_phase2,
5991 up_read(&com->lc_sem);
5994 static int lfsck_layout_master_double_scan(const struct lu_env *env,
5995 struct lfsck_component *com)
5997 struct lfsck_layout *lo = com->lc_file_ram;
5998 struct lfsck_assistant_data *lad = com->lc_data;
5999 struct lfsck_instance *lfsck = com->lc_lfsck;
6000 struct lfsck_tgt_descs *ltds;
6001 struct lfsck_tgt_desc *ltd;
6002 struct lfsck_tgt_desc *next;
6005 rc = lfsck_double_scan_generic(env, com, lo->ll_status);
6007 if (thread_is_stopped(&lad->lad_thread)) {
6008 LASSERT(list_empty(&lad->lad_req_list));
6009 LASSERT(list_empty(&lad->lad_ost_phase1_list));
6010 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
6012 ltds = &lfsck->li_ost_descs;
6013 spin_lock(<ds->ltd_lock);
6014 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6015 ltd_layout_phase_list) {
6016 list_del_init(<d->ltd_layout_phase_list);
6018 spin_unlock(<ds->ltd_lock);
6020 ltds = &lfsck->li_mdt_descs;
6021 spin_lock(<ds->ltd_lock);
6022 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6023 ltd_layout_phase_list) {
6024 list_del_init(<d->ltd_layout_phase_list);
6026 spin_unlock(<ds->ltd_lock);
6032 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
6033 struct lfsck_component *com)
6035 struct lfsck_instance *lfsck = com->lc_lfsck;
6036 struct lfsck_layout_slave_data *llsd = com->lc_data;
6037 struct lfsck_layout *lo = com->lc_file_ram;
6038 struct ptlrpc_thread *thread = &lfsck->li_thread;
6042 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
6043 lfsck_lfsck2name(lfsck));
6045 atomic_inc(&lfsck->li_double_scan_count);
6047 if (lo->ll_flags & LF_INCOMPLETE)
6050 com->lc_new_checked = 0;
6051 com->lc_new_scanned = 0;
6052 com->lc_time_last_checkpoint = cfs_time_current();
6053 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6054 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
6057 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
6060 rc = lfsck_layout_slave_query_master(env, com);
6061 if (list_empty(&llsd->llsd_master_list)) {
6062 if (unlikely(!thread_is_running(thread)))
6073 rc = l_wait_event(thread->t_ctl_waitq,
6074 !thread_is_running(thread) ||
6075 lo->ll_flags & LF_INCOMPLETE ||
6076 list_empty(&llsd->llsd_master_list),
6078 if (unlikely(!thread_is_running(thread)))
6081 if (lo->ll_flags & LF_INCOMPLETE)
6084 if (rc == -ETIMEDOUT)
6087 GOTO(done, rc = (rc < 0 ? rc : 1));
6091 rc = lfsck_layout_double_scan_result(env, com, rc);
6092 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
6093 (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
6094 lfsck_layout_slave_quit(env, com);
6095 if (atomic_dec_and_test(&lfsck->li_double_scan_count))
6096 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6098 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
6099 "status %d: rc = %d\n",
6100 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
6105 static void lfsck_layout_master_data_release(const struct lu_env *env,
6106 struct lfsck_component *com)
6108 struct lfsck_assistant_data *lad = com->lc_data;
6109 struct lfsck_instance *lfsck = com->lc_lfsck;
6110 struct lfsck_tgt_descs *ltds;
6111 struct lfsck_tgt_desc *ltd;
6112 struct lfsck_tgt_desc *next;
6114 LASSERT(lad != NULL);
6115 LASSERT(thread_is_init(&lad->lad_thread) ||
6116 thread_is_stopped(&lad->lad_thread));
6117 LASSERT(list_empty(&lad->lad_req_list));
6119 com->lc_data = NULL;
6121 ltds = &lfsck->li_ost_descs;
6122 spin_lock(<ds->ltd_lock);
6123 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6124 ltd_layout_phase_list) {
6125 list_del_init(<d->ltd_layout_phase_list);
6127 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6128 ltd_layout_phase_list) {
6129 list_del_init(<d->ltd_layout_phase_list);
6131 list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
6133 list_del_init(<d->ltd_layout_list);
6135 spin_unlock(<ds->ltd_lock);
6137 ltds = &lfsck->li_mdt_descs;
6138 spin_lock(<ds->ltd_lock);
6139 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6140 ltd_layout_phase_list) {
6141 list_del_init(<d->ltd_layout_phase_list);
6143 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6144 ltd_layout_phase_list) {
6145 list_del_init(<d->ltd_layout_phase_list);
6147 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
6149 list_del_init(<d->ltd_layout_list);
6151 spin_unlock(<ds->ltd_lock);
6153 if (likely(lad->lad_bitmap != NULL))
6154 CFS_FREE_BITMAP(lad->lad_bitmap);
6159 static void lfsck_layout_slave_data_release(const struct lu_env *env,
6160 struct lfsck_component *com)
6162 struct lfsck_layout_slave_data *llsd = com->lc_data;
6164 lfsck_layout_slave_quit(env, com);
6165 com->lc_data = NULL;
6169 static void lfsck_layout_master_quit(const struct lu_env *env,
6170 struct lfsck_component *com)
6172 struct lfsck_assistant_data *lad = com->lc_data;
6173 struct lfsck_instance *lfsck = com->lc_lfsck;
6174 struct lfsck_tgt_descs *ltds;
6175 struct lfsck_tgt_desc *ltd;
6176 struct lfsck_tgt_desc *next;
6178 LASSERT(lad != NULL);
6180 lfsck_quit_generic(env, com);
6182 LASSERT(thread_is_init(&lad->lad_thread) ||
6183 thread_is_stopped(&lad->lad_thread));
6184 LASSERT(list_empty(&lad->lad_req_list));
6186 ltds = &lfsck->li_ost_descs;
6187 spin_lock(<ds->ltd_lock);
6188 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6189 ltd_layout_phase_list) {
6190 list_del_init(<d->ltd_layout_phase_list);
6192 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6193 ltd_layout_phase_list) {
6194 list_del_init(<d->ltd_layout_phase_list);
6196 spin_unlock(<ds->ltd_lock);
6198 ltds = &lfsck->li_mdt_descs;
6199 spin_lock(<ds->ltd_lock);
6200 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6201 ltd_layout_phase_list) {
6202 list_del_init(<d->ltd_layout_phase_list);
6204 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6205 ltd_layout_phase_list) {
6206 list_del_init(<d->ltd_layout_phase_list);
6208 spin_unlock(<ds->ltd_lock);
6211 static void lfsck_layout_slave_quit(const struct lu_env *env,
6212 struct lfsck_component *com)
6214 struct lfsck_layout_slave_data *llsd = com->lc_data;
6215 struct lfsck_layout_seq *lls;
6216 struct lfsck_layout_seq *next;
6217 struct lfsck_layout_slave_target *llst;
6219 LASSERT(llsd != NULL);
6221 down_write(&com->lc_sem);
6222 list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
6224 list_del_init(&lls->lls_list);
6225 lfsck_object_put(env, lls->lls_lastid_obj);
6228 up_write(&com->lc_sem);
6230 spin_lock(&llsd->llsd_lock);
6231 while (!list_empty(&llsd->llsd_master_list)) {
6232 llst = list_entry(llsd->llsd_master_list.next,
6233 struct lfsck_layout_slave_target, llst_list);
6234 list_del_init(&llst->llst_list);
6235 spin_unlock(&llsd->llsd_lock);
6236 lfsck_layout_llst_put(llst);
6237 spin_lock(&llsd->llsd_lock);
6239 spin_unlock(&llsd->llsd_lock);
6241 lfsck_rbtree_cleanup(env, com);
6244 static int lfsck_layout_master_in_notify(const struct lu_env *env,
6245 struct lfsck_component *com,
6246 struct lfsck_request *lr)
6248 struct lfsck_instance *lfsck = com->lc_lfsck;
6249 struct lfsck_layout *lo = com->lc_file_ram;
6250 struct lfsck_assistant_data *lad = com->lc_data;
6251 struct lfsck_tgt_descs *ltds;
6252 struct lfsck_tgt_desc *ltd;
6256 if (lr->lr_event == LE_PAIRS_VERIFY) {
6259 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
6266 CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
6267 "from %s %x, status %d, flags %x, flags2 %x\n",
6268 lfsck_lfsck2name(lfsck), lr->lr_event,
6269 (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
6270 lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
6272 if (lr->lr_event != LE_PHASE1_DONE &&
6273 lr->lr_event != LE_PHASE2_DONE &&
6274 lr->lr_event != LE_PEER_EXIT)
6277 if (lr->lr_flags & LEF_FROM_OST)
6278 ltds = &lfsck->li_ost_descs;
6280 ltds = &lfsck->li_mdt_descs;
6281 spin_lock(<ds->ltd_lock);
6282 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
6284 spin_unlock(<ds->ltd_lock);
6289 list_del_init(<d->ltd_layout_phase_list);
6290 switch (lr->lr_event) {
6291 case LE_PHASE1_DONE:
6292 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
6293 if (lr->lr_flags2 & LF_INCOMPLETE) {
6294 if (lr->lr_flags & LEF_FROM_OST)
6295 lfsck_lad_set_bitmap(env, com,
6298 lo->ll_flags |= LF_INCOMPLETE;
6300 ltd->ltd_layout_done = 1;
6301 list_del_init(<d->ltd_layout_list);
6306 if (lr->lr_flags & LEF_FROM_OST) {
6307 if (list_empty(<d->ltd_layout_list))
6308 list_add_tail(<d->ltd_layout_list,
6309 &lad->lad_ost_list);
6310 list_add_tail(<d->ltd_layout_phase_list,
6311 &lad->lad_ost_phase2_list);
6313 if (list_empty(<d->ltd_layout_list))
6314 list_add_tail(<d->ltd_layout_list,
6315 &lad->lad_mdt_list);
6316 list_add_tail(<d->ltd_layout_phase_list,
6317 &lad->lad_mdt_phase2_list);
6320 case LE_PHASE2_DONE:
6321 ltd->ltd_layout_done = 1;
6322 if (!list_empty(<d->ltd_layout_list))
6323 list_del_init(<d->ltd_layout_list);
6325 if (lr->lr_flags2 & LF_INCOMPLETE) {
6326 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
6333 ltd->ltd_layout_done = 1;
6334 list_del_init(<d->ltd_layout_list);
6335 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
6336 !(lr->lr_flags & LEF_FROM_OST))
6337 lo->ll_flags |= LF_INCOMPLETE;
6342 spin_unlock(<ds->ltd_lock);
6344 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
6345 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6347 memset(stop, 0, sizeof(*stop));
6348 stop->ls_status = lr->lr_status;
6349 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6350 lfsck_stop(env, lfsck->li_bottom, stop);
6351 } else if (lfsck_phase2_next_ready(lad)) {
6352 wake_up_all(&lad->lad_thread.t_ctl_waitq);
6358 static int lfsck_layout_slave_in_notify_local(const struct lu_env *env,
6359 struct lfsck_component *com,
6360 struct lfsck_req_local *lrl,
6365 switch (lrl->lrl_event) {
6366 case LEL_FID_ACCESSED:
6367 lfsck_rbtree_update_bitmap(env, com, &lrl->lrl_fid, true);
6369 case LEL_PAIRS_VERIFY_LOCAL: {
6372 lrl->lrl_status = LPVS_INIT;
6373 /* Firstly, if the MDT-object which is claimed via OST-object
6374 * local stored PFID xattr recognizes the OST-object, then it
6375 * must be that the client given PFID is wrong. */
6376 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6377 &lrl->lrl_ff_local.ff_parent,
6378 lrl->lrl_ff_local.ff_layout.ol_comp_id);
6382 lrl->lrl_status = LPVS_INCONSISTENT;
6383 /* The OST-object local stored PFID xattr is stale. We need to
6384 * check whether the MDT-object that is claimed via the client
6385 * given PFID information recognizes the OST-object or not. If
6386 * matches, then need to update the OST-object's PFID xattr. */
6387 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6388 &lrl->lrl_ff_client.ff_parent,
6389 lrl->lrl_ff_client.ff_layout.ol_comp_id);
6391 * We are not sure whether the client given PFID information
6392 * is correct or not, do nothing to avoid improper fixing.
6395 * The client given PFID information is also invalid, we can
6396 * NOT fix the OST-object inconsistency.
6399 lrl->lrl_status = LPVS_INCONSISTENT_TOFIX;
6400 rc = lfsck_layout_slave_repair_pfid(env, com, lrl);
6409 RETURN(-EOPNOTSUPP);
6412 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
6413 struct lfsck_component *com,
6414 struct lfsck_request *lr)
6416 struct lfsck_instance *lfsck = com->lc_lfsck;
6417 struct lfsck_layout_slave_data *llsd = com->lc_data;
6418 struct lfsck_layout_slave_target *llst;
6422 switch (lr->lr_event) {
6423 case LE_CONDITIONAL_DESTROY:
6424 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
6426 case LE_PHASE1_DONE: {
6427 if (lr->lr_flags2 & LF_INCOMPLETE) {
6428 struct lfsck_layout *lo = com->lc_file_ram;
6430 lo->ll_flags |= LF_INCOMPLETE;
6431 llst = lfsck_layout_llst_find_and_del(llsd,
6435 lfsck_layout_llst_put(llst);
6436 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6442 case LE_PHASE2_DONE:
6444 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
6445 "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
6446 lr->lr_event, lr->lr_index, lr->lr_status);
6452 llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
6456 lfsck_layout_llst_put(llst);
6457 if (list_empty(&llsd->llsd_master_list))
6458 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6460 if (lr->lr_event == LE_PEER_EXIT &&
6461 (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
6462 (list_empty(&llsd->llsd_master_list) &&
6463 (lr->lr_status == LS_STOPPED ||
6464 lr->lr_status == LS_CO_STOPPED)))) {
6465 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6467 memset(stop, 0, sizeof(*stop));
6468 stop->ls_status = lr->lr_status;
6469 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6470 lfsck_stop(env, lfsck->li_bottom, stop);
6476 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
6480 for (i = 0; i < LLIT_MAX; i++)
6481 *count += lo->ll_objs_repaired[i];
6484 static int lfsck_layout_query_all(const struct lu_env *env,
6485 struct lfsck_component *com,
6486 __u32 *mdts_count, __u32 *osts_count,
6489 struct lfsck_layout *lo = com->lc_file_ram;
6490 struct lfsck_tgt_descs *ltds;
6491 struct lfsck_tgt_desc *ltd;
6496 rc = lfsck_query_all(env, com);
6500 ltds = &com->lc_lfsck->li_mdt_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 mdts_count[ltd->ltd_layout_status]++;
6507 *repaired += ltd->ltd_layout_repaired;
6509 up_read(<ds->ltd_rw_sem);
6511 ltds = &com->lc_lfsck->li_ost_descs;
6512 down_read(<ds->ltd_rw_sem);
6513 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
6514 ltd = lfsck_ltd2tgt(ltds, idx);
6515 LASSERT(ltd != NULL);
6517 osts_count[ltd->ltd_layout_status]++;
6518 *repaired += ltd->ltd_layout_repaired;
6520 up_read(<ds->ltd_rw_sem);
6522 down_read(&com->lc_sem);
6523 mdts_count[lo->ll_status]++;
6524 lfsck_layout_repaired(lo, repaired);
6525 up_read(&com->lc_sem);
6530 static int lfsck_layout_query(const struct lu_env *env,
6531 struct lfsck_component *com,
6532 struct lfsck_request *req,
6533 struct lfsck_reply *rep,
6534 struct lfsck_query *que, int idx)
6536 struct lfsck_layout *lo = com->lc_file_ram;
6540 LASSERT(com->lc_lfsck->li_master);
6542 rc = lfsck_layout_query_all(env, com,
6543 que->lu_mdts_count[idx],
6544 que->lu_osts_count[idx],
6545 &que->lu_repaired[idx]);
6547 down_read(&com->lc_sem);
6548 rep->lr_status = lo->ll_status;
6549 if (req->lr_flags & LEF_QUERY_ALL)
6550 lfsck_layout_repaired(lo, &rep->lr_repaired);
6551 up_read(&com->lc_sem);
6557 /* with lfsck::li_lock held */
6558 static int lfsck_layout_slave_join(const struct lu_env *env,
6559 struct lfsck_component *com,
6560 struct lfsck_start_param *lsp)
6562 struct lfsck_instance *lfsck = com->lc_lfsck;
6563 struct lfsck_layout_slave_data *llsd = com->lc_data;
6564 struct lfsck_layout_slave_target *llst;
6565 struct lfsck_start *start = lsp->lsp_start;
6569 if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
6572 if (!lsp->lsp_index_valid)
6575 /* If someone is running the LFSCK without orphan handling,
6576 * it will not maintain the object accessing rbtree. So we
6577 * cannot join it for orphan handling. */
6578 if (!llsd->llsd_rbtree_valid)
6581 spin_unlock(&lfsck->li_lock);
6582 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
6583 spin_lock(&lfsck->li_lock);
6584 if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
6585 spin_unlock(&lfsck->li_lock);
6586 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
6589 lfsck_layout_llst_put(llst);
6590 spin_lock(&lfsck->li_lock);
6597 static struct lfsck_operations lfsck_layout_master_ops = {
6598 .lfsck_reset = lfsck_layout_reset,
6599 .lfsck_fail = lfsck_layout_fail,
6600 .lfsck_checkpoint = lfsck_layout_master_checkpoint,
6601 .lfsck_prep = lfsck_layout_master_prep,
6602 .lfsck_exec_oit = lfsck_layout_master_exec_oit,
6603 .lfsck_exec_dir = lfsck_layout_exec_dir,
6604 .lfsck_post = lfsck_layout_master_post,
6605 .lfsck_dump = lfsck_layout_dump,
6606 .lfsck_double_scan = lfsck_layout_master_double_scan,
6607 .lfsck_data_release = lfsck_layout_master_data_release,
6608 .lfsck_quit = lfsck_layout_master_quit,
6609 .lfsck_in_notify = lfsck_layout_master_in_notify,
6610 .lfsck_query = lfsck_layout_query,
6613 static struct lfsck_operations lfsck_layout_slave_ops = {
6614 .lfsck_reset = lfsck_layout_reset,
6615 .lfsck_fail = lfsck_layout_fail,
6616 .lfsck_checkpoint = lfsck_layout_slave_checkpoint,
6617 .lfsck_prep = lfsck_layout_slave_prep,
6618 .lfsck_exec_oit = lfsck_layout_slave_exec_oit,
6619 .lfsck_exec_dir = lfsck_layout_exec_dir,
6620 .lfsck_post = lfsck_layout_slave_post,
6621 .lfsck_dump = lfsck_layout_dump,
6622 .lfsck_double_scan = lfsck_layout_slave_double_scan,
6623 .lfsck_data_release = lfsck_layout_slave_data_release,
6624 .lfsck_quit = lfsck_layout_slave_quit,
6625 .lfsck_in_notify_local = lfsck_layout_slave_in_notify_local,
6626 .lfsck_in_notify = lfsck_layout_slave_in_notify,
6627 .lfsck_query = lfsck_layout_query,
6628 .lfsck_join = lfsck_layout_slave_join,
6631 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
6632 struct lfsck_component *com,
6633 struct lfsck_position *pos)
6635 struct lfsck_assistant_data *lad = com->lc_data;
6636 struct lfsck_layout_req *llr;
6638 if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
6642 if (list_empty(&lad->lad_req_list))
6645 llr = list_entry(lad->lad_req_list.next,
6646 struct lfsck_layout_req,
6648 pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
6651 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
6652 .la_handler_p1 = lfsck_layout_assistant_handler_p1,
6653 .la_handler_p2 = lfsck_layout_assistant_handler_p2,
6654 .la_fill_pos = lfsck_layout_assistant_fill_pos,
6655 .la_double_scan_result = lfsck_layout_double_scan_result,
6656 .la_req_fini = lfsck_layout_assistant_req_fini,
6657 .la_sync_failures = lfsck_layout_assistant_sync_failures,
6660 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
6662 struct lfsck_component *com;
6663 struct lfsck_layout *lo;
6664 struct dt_object *root = NULL;
6665 struct dt_object *obj;
6674 INIT_LIST_HEAD(&com->lc_link);
6675 INIT_LIST_HEAD(&com->lc_link_dir);
6676 init_rwsem(&com->lc_sem);
6677 atomic_set(&com->lc_ref, 1);
6678 com->lc_lfsck = lfsck;
6679 com->lc_type = LFSCK_TYPE_LAYOUT;
6680 if (lfsck->li_master) {
6681 com->lc_ops = &lfsck_layout_master_ops;
6682 com->lc_data = lfsck_assistant_data_init(
6683 &lfsck_layout_assistant_ops,
6685 if (com->lc_data == NULL)
6686 GOTO(out, rc = -ENOMEM);
6688 for (i = 0; i < LFSCK_STF_COUNT; i++)
6689 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6691 struct lfsck_layout_slave_data *llsd;
6693 com->lc_ops = &lfsck_layout_slave_ops;
6694 OBD_ALLOC_PTR(llsd);
6696 GOTO(out, rc = -ENOMEM);
6698 INIT_LIST_HEAD(&llsd->llsd_seq_list);
6699 INIT_LIST_HEAD(&llsd->llsd_master_list);
6700 spin_lock_init(&llsd->llsd_lock);
6701 llsd->llsd_rb_root = RB_ROOT;
6702 rwlock_init(&llsd->llsd_rb_lock);
6703 com->lc_data = llsd;
6705 com->lc_file_size = sizeof(*lo);
6706 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6707 if (com->lc_file_ram == NULL)
6708 GOTO(out, rc = -ENOMEM);
6710 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6711 if (com->lc_file_disk == NULL)
6712 GOTO(out, rc = -ENOMEM);
6714 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6716 GOTO(out, rc = PTR_ERR(root));
6718 if (unlikely(!dt_try_as_dir(env, root)))
6719 GOTO(out, rc = -ENOTDIR);
6721 obj = local_file_find_or_create(env, lfsck->li_los, root,
6723 S_IFREG | S_IRUGO | S_IWUSR);
6725 GOTO(out, rc = PTR_ERR(obj));
6728 rc = lfsck_layout_load(env, com);
6730 rc = lfsck_layout_reset(env, com, true);
6731 else if (rc == -ENOENT)
6732 rc = lfsck_layout_init(env, com);
6733 else if (lfsck->li_master)
6734 rc = lfsck_load_sub_trace_files(env, com,
6735 &dt_lfsck_layout_dangling_features,
6736 LFSCK_LAYOUT, false);
6741 lo = com->lc_file_ram;
6742 switch (lo->ll_status) {
6748 spin_lock(&lfsck->li_lock);
6749 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6750 spin_unlock(&lfsck->li_lock);
6753 CERROR("%s: unknown lfsck_layout status %d\n",
6754 lfsck_lfsck2name(lfsck), lo->ll_status);
6756 case LS_SCANNING_PHASE1:
6757 case LS_SCANNING_PHASE2:
6758 /* No need to store the status to disk right now.
6759 * If the system crashed before the status stored,
6760 * it will be loaded back when next time. */
6761 lo->ll_status = LS_CRASHED;
6762 if (!lfsck->li_master)
6763 lo->ll_flags |= LF_INCOMPLETE;
6770 spin_lock(&lfsck->li_lock);
6771 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6772 spin_unlock(&lfsck->li_lock);
6776 if (lo->ll_flags & LF_CRASHED_LASTID) {
6777 LASSERT(lfsck->li_out_notify != NULL);
6779 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6780 LE_LASTID_REBUILDING);
6786 if (root != NULL && !IS_ERR(root))
6787 lfsck_object_put(env, root);
6790 lfsck_component_cleanup(env, com);
6791 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
6792 lfsck_lfsck2name(lfsck), rc);
6798 struct lfsck_orphan_it {
6799 struct lfsck_component *loi_com;
6800 struct lfsck_rbtree_node *loi_lrn;
6801 struct lfsck_layout_slave_target *loi_llst;
6802 struct lu_fid loi_key;
6803 struct lu_orphan_rec_v2 loi_rec;
6805 unsigned int loi_over:1;
6808 static int lfsck_fid_match_idx(const struct lu_env *env,
6809 struct lfsck_instance *lfsck,
6810 const struct lu_fid *fid, int idx)
6812 struct seq_server_site *ss;
6813 struct lu_server_fld *sf;
6814 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
6817 /* All abnormal cases will be returned to MDT0. */
6818 if (!fid_is_norm(fid)) {
6825 ss = lfsck_dev_site(lfsck);
6826 if (unlikely(ss == NULL))
6829 sf = ss->ss_server_fld;
6830 LASSERT(sf != NULL);
6832 fld_range_set_any(range);
6833 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
6837 if (!fld_range_is_mdt(range))
6840 if (range->lsr_index == idx)
6846 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
6847 struct dt_object *obj)
6849 struct dt_device *dev = lfsck_obj2dev(obj);
6850 struct thandle *handle;
6854 handle = dt_trans_create(env, dev);
6858 rc = dt_declare_ref_del(env, obj, handle);
6862 rc = dt_declare_destroy(env, obj, handle);
6866 rc = dt_trans_start_local(env, dev, handle);
6870 dt_write_lock(env, obj, 0);
6871 rc = dt_ref_del(env, obj, handle);
6873 rc = dt_destroy(env, obj, handle);
6874 dt_write_unlock(env, obj);
6879 dt_trans_stop(env, dev, handle);
6881 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
6882 PFID(lfsck_dto2fid(obj)), rc);
6887 static int lfsck_orphan_index_lookup(const struct lu_env *env,
6888 struct dt_object *dt,
6890 const struct dt_key *key)
6895 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
6896 struct dt_object *dt,
6897 const struct dt_rec *rec,
6898 const struct dt_key *key,
6899 struct thandle *handle)
6904 static int lfsck_orphan_index_insert(const struct lu_env *env,
6905 struct dt_object *dt,
6906 const struct dt_rec *rec,
6907 const struct dt_key *key,
6908 struct thandle *handle,
6914 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
6915 struct dt_object *dt,
6916 const struct dt_key *key,
6917 struct thandle *handle)
6922 static int lfsck_orphan_index_delete(const struct lu_env *env,
6923 struct dt_object *dt,
6924 const struct dt_key *key,
6925 struct thandle *handle)
6930 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
6931 struct dt_object *dt,
6934 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
6935 struct lfsck_instance *lfsck;
6936 struct lfsck_component *com = NULL;
6937 struct lfsck_layout_slave_data *llsd;
6938 struct lfsck_orphan_it *it = NULL;
6939 struct lfsck_layout *lo;
6943 lfsck = lfsck_instance_find(dev, true, false);
6944 if (unlikely(lfsck == NULL))
6945 RETURN(ERR_PTR(-ENXIO));
6947 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
6948 if (unlikely(com == NULL))
6949 GOTO(out, rc = -ENOENT);
6951 lo = com->lc_file_ram;
6952 if (lo->ll_flags & LF_INCOMPLETE)
6953 GOTO(out, rc = -ESRCH);
6955 llsd = com->lc_data;
6956 if (!llsd->llsd_rbtree_valid)
6957 GOTO(out, rc = -ESRCH);
6961 GOTO(out, rc = -ENOMEM);
6963 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
6964 if (it->loi_llst == NULL)
6965 GOTO(out, rc = -ENXIO);
6967 if (dev->dd_record_fid_accessed) {
6968 /* The first iteration against the rbtree, scan the whole rbtree
6969 * to remove the nodes which do NOT need to be handled. */
6970 write_lock(&llsd->llsd_rb_lock);
6971 if (dev->dd_record_fid_accessed) {
6972 struct rb_node *node;
6973 struct rb_node *next;
6974 struct lfsck_rbtree_node *lrn;
6976 /* No need to record the fid accessing anymore. */
6977 dev->dd_record_fid_accessed = 0;
6979 node = rb_first(&llsd->llsd_rb_root);
6980 while (node != NULL) {
6981 next = rb_next(node);
6982 lrn = rb_entry(node, struct lfsck_rbtree_node,
6984 if (atomic_read(&lrn->lrn_known_count) <=
6985 atomic_read(&lrn->lrn_accessed_count)) {
6986 rb_erase(node, &llsd->llsd_rb_root);
6987 lfsck_rbtree_free(lrn);
6992 write_unlock(&llsd->llsd_rb_lock);
6995 /* read lock the rbtree when init, and unlock when fini */
6996 read_lock(&llsd->llsd_rb_lock);
7004 lfsck_component_put(env, com);
7006 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
7007 lfsck_lfsck2name(lfsck), rc);
7009 lfsck_instance_put(env, lfsck);
7014 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
7017 return (struct dt_it *)it;
7020 static void lfsck_orphan_it_fini(const struct lu_env *env,
7023 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7024 struct lfsck_component *com = it->loi_com;
7025 struct lfsck_layout_slave_data *llsd;
7026 struct lfsck_layout_slave_target *llst;
7029 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
7030 lfsck_lfsck2name(com->lc_lfsck));
7032 llsd = com->lc_data;
7033 read_unlock(&llsd->llsd_rb_lock);
7034 llst = it->loi_llst;
7035 LASSERT(llst != NULL);
7037 /* Save the key and hash for iterate next. */
7038 llst->llst_fid = it->loi_key;
7039 llst->llst_hash = it->loi_hash;
7040 lfsck_layout_llst_put(llst);
7041 lfsck_component_put(env, com);
7047 * \retval +1: the iteration finished
7048 * \retval 0: on success, not finished
7049 * \retval -ve: on error
7051 static int lfsck_orphan_it_next(const struct lu_env *env,
7054 struct lfsck_thread_info *info = lfsck_env_info(env);
7055 struct filter_fid *ff = &info->lti_ff;
7056 struct lu_attr *la = &info->lti_la;
7057 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7058 struct lu_fid *key = &it->loi_key;
7059 struct lu_orphan_rec_v2 *rec = &it->loi_rec;
7060 struct ost_layout *ol = &rec->lor_layout;
7061 struct lfsck_component *com = it->loi_com;
7062 struct lfsck_instance *lfsck = com->lc_lfsck;
7063 struct lfsck_layout_slave_data *llsd = com->lc_data;
7064 struct dt_object *obj;
7065 struct lfsck_rbtree_node *lrn;
7069 __u32 idx = it->loi_llst->llst_index;
7079 lrn = lfsck_rbtree_search(llsd, key, &exact);
7087 key->f_seq = lrn->lrn_seq;
7088 key->f_oid = lrn->lrn_first_oid;
7093 if (unlikely(key->f_oid == 0)) {
7100 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
7106 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
7107 atomic_read(&lrn->lrn_accessed_count))) {
7108 struct rb_node *next = rb_next(&lrn->lrn_node);
7110 while (next != NULL) {
7111 lrn = rb_entry(next, struct lfsck_rbtree_node,
7113 if (atomic_read(&lrn->lrn_known_count) >
7114 atomic_read(&lrn->lrn_accessed_count))
7116 next = rb_next(next);
7125 key->f_seq = lrn->lrn_seq;
7126 key->f_oid = lrn->lrn_first_oid;
7130 pos = key->f_oid - lrn->lrn_first_oid;
7133 pos = find_next_bit(lrn->lrn_known_bitmap,
7134 LFSCK_RBTREE_BITMAP_WIDTH, pos);
7135 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
7136 key->f_oid = lrn->lrn_first_oid + pos;
7137 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
7145 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
7150 key->f_oid = lrn->lrn_first_oid + pos;
7151 obj = lfsck_object_find_bottom(env, lfsck, key);
7154 if (rc == -ENOENT) {
7161 dt_read_lock(env, obj, 0);
7162 if (dt_object_exists(obj) == 0 ||
7163 lfsck_is_dead_obj(obj)) {
7164 dt_read_unlock(env, obj);
7165 lfsck_object_put(env, obj);
7170 rc = dt_attr_get(env, obj, la);
7174 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, ff, sizeof(*ff)),
7176 if (rc == -ENODATA) {
7177 /* For the pre-created OST-object, update the bitmap to avoid
7178 * others LFSCK (second phase) iteration to touch it again. */
7179 if (la->la_ctime == 0) {
7180 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
7181 atomic_inc(&lrn->lrn_accessed_count);
7183 /* For the race between repairing dangling referenced
7184 * MDT-object and unlink the file, it may left orphan
7185 * OST-object there. Destroy it now! */
7186 if (unlikely(!(la->la_mode & S_ISUID))) {
7187 dt_read_unlock(env, obj);
7188 lfsck_layout_destroy_orphan(env, obj);
7189 lfsck_object_put(env, obj);
7193 } else if (idx == 0) {
7194 /* If the orphan OST-object has no parent information,
7195 * regard it as referenced by the MDT-object on MDT0. */
7196 fid_zero(&rec->lor_rec.lor_fid);
7197 rec->lor_rec.lor_uid = la->la_uid;
7198 rec->lor_rec.lor_gid = la->la_gid;
7199 memset(ol, 0, sizeof(*ol));
7204 dt_read_unlock(env, obj);
7205 lfsck_object_put(env, obj);
7210 if (rc < sizeof(struct lu_fid))
7211 GOTO(out, rc = (rc < 0 ? rc : -EINVAL));
7213 fid_le_to_cpu(&rec->lor_rec.lor_fid, &ff->ff_parent);
7214 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
7215 * MDT-object's FID::f_ver, instead it is the OST-object index in its
7216 * parent MDT-object's layout EA. */
7217 save = rec->lor_rec.lor_fid.f_stripe_idx;
7218 rec->lor_rec.lor_fid.f_ver = 0;
7219 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_rec.lor_fid, idx);
7220 /* If the orphan OST-object does not claim the MDT, then next.
7222 * If we do not know whether it matches or not, then return it
7223 * to the MDT for further check. */
7225 dt_read_unlock(env, obj);
7226 lfsck_object_put(env, obj);
7231 rec->lor_rec.lor_fid.f_stripe_idx = save;
7232 rec->lor_rec.lor_uid = la->la_uid;
7233 rec->lor_rec.lor_gid = la->la_gid;
7234 ost_layout_le_to_cpu(ol, &ff->ff_layout);
7236 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u, "
7237 "stripe size %u, stripe count %u, COMP id %u, COMP start %llu, "
7238 "COMP end %llu\n", lfsck_lfsck2name(com->lc_lfsck), PFID(key),
7239 PFID(&rec->lor_rec.lor_fid), rec->lor_rec.lor_uid,
7240 rec->lor_rec.lor_gid, ol->ol_stripe_size, ol->ol_stripe_count,
7241 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end);
7246 dt_read_unlock(env, obj);
7247 lfsck_object_put(env, obj);
7255 * \retval +1: locate to the exactly position
7256 * \retval 0: cannot locate to the exactly position,
7257 * call next() to move to a valid position.
7258 * \retval -ve: on error
7260 static int lfsck_orphan_it_get(const struct lu_env *env,
7262 const struct dt_key *key)
7264 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7267 it->loi_key = *(struct lu_fid *)key;
7268 rc = lfsck_orphan_it_next(env, di);
7278 static void lfsck_orphan_it_put(const struct lu_env *env,
7283 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
7284 const struct dt_it *di)
7286 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7288 return (struct dt_key *)&it->loi_key;
7291 static int lfsck_orphan_it_key_size(const struct lu_env *env,
7292 const struct dt_it *di)
7294 return sizeof(struct lu_fid);
7297 static int lfsck_orphan_it_rec(const struct lu_env *env,
7298 const struct dt_it *di,
7302 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7304 *(struct lu_orphan_rec_v2 *)rec = it->loi_rec;
7309 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
7310 const struct dt_it *di)
7312 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7314 return it->loi_hash;
7318 * \retval +1: locate to the exactly position
7319 * \retval 0: cannot locate to the exactly position,
7320 * call next() to move to a valid position.
7321 * \retval -ve: on error
7323 static int lfsck_orphan_it_load(const struct lu_env *env,
7324 const struct dt_it *di,
7327 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7328 struct lfsck_layout_slave_target *llst = it->loi_llst;
7331 LASSERT(llst != NULL);
7333 if (hash != llst->llst_hash) {
7334 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
7335 "iteration does not match the one when fini "
7336 "%llu, to be reset.\n",
7337 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
7339 fid_zero(&llst->llst_fid);
7340 llst->llst_hash = 0;
7343 it->loi_key = llst->llst_fid;
7344 it->loi_hash = llst->llst_hash;
7345 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
7355 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
7356 const struct dt_it *di,
7362 const struct dt_index_operations lfsck_orphan_index_ops = {
7363 .dio_lookup = lfsck_orphan_index_lookup,
7364 .dio_declare_insert = lfsck_orphan_index_declare_insert,
7365 .dio_insert = lfsck_orphan_index_insert,
7366 .dio_declare_delete = lfsck_orphan_index_declare_delete,
7367 .dio_delete = lfsck_orphan_index_delete,
7369 .init = lfsck_orphan_it_init,
7370 .fini = lfsck_orphan_it_fini,
7371 .get = lfsck_orphan_it_get,
7372 .put = lfsck_orphan_it_put,
7373 .next = lfsck_orphan_it_next,
7374 .key = lfsck_orphan_it_key,
7375 .key_size = lfsck_orphan_it_key_size,
7376 .rec = lfsck_orphan_it_rec,
7377 .store = lfsck_orphan_it_store,
7378 .load = lfsck_orphan_it_load,
7379 .key_rec = lfsck_orphan_it_key_rec,