4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License version 2 for more details. A copy is
14 * included in the COPYING file that accompanied this code.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 * Copyright (c) 2014, 2016, Intel Corporation.
26 * lustre/lfsck/lfsck_layout.c
28 * Author: Fan, Yong <fan.yong@intel.com>
32 # define EXPORT_SYMTAB
34 #define DEBUG_SUBSYSTEM S_LFSCK
36 #include <linux/bitops.h>
37 #include <linux/rbtree.h>
39 #include <lustre/lustre_idl.h>
40 #include <lu_object.h>
41 #include <dt_object.h>
42 #include <lustre_fid.h>
43 #include <lustre_lib.h>
44 #include <lustre_net.h>
45 #include <lustre/lustre_user.h>
46 #include <md_object.h>
47 #include <obd_class.h>
49 #include "lfsck_internal.h"
51 #define LFSCK_LAYOUT_MAGIC_V1 0xB173AE14
52 #define LFSCK_LAYOUT_MAGIC_V2 0xB1734D76
54 #define LFSCK_LAYOUT_MAGIC LFSCK_LAYOUT_MAGIC_V2
56 struct lfsck_layout_seq {
57 struct list_head lls_list;
60 __u64 lls_lastid_known;
61 struct dt_object *lls_lastid_obj;
62 unsigned int lls_dirty:1;
65 struct lfsck_layout_slave_target {
66 /* link into lfsck_layout_slave_data::llsd_master_list. */
67 struct list_head llst_list;
68 /* The position for next record in the rbtree for iteration. */
69 struct lu_fid llst_fid;
70 /* Dummy hash for iteration against the rbtree. */
75 /* How many times we have failed to get the master status. */
79 struct lfsck_layout_slave_data {
80 /* list for lfsck_layout_seq */
81 struct list_head llsd_seq_list;
83 /* list for the masters involve layout verification. */
84 struct list_head llsd_master_list;
87 struct dt_object *llsd_rb_obj;
88 struct rb_root llsd_rb_root;
89 rwlock_t llsd_rb_lock;
90 unsigned int llsd_rbtree_valid:1;
93 struct lfsck_layout_slave_async_args {
94 struct obd_export *llsaa_exp;
95 struct lfsck_component *llsaa_com;
96 struct lfsck_layout_slave_target *llsaa_llst;
100 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
102 if (atomic_dec_and_test(&llst->llst_ref)) {
103 LASSERT(list_empty(&llst->llst_list));
110 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
112 struct lfsck_layout_slave_target *llst;
113 struct lfsck_layout_slave_target *tmp;
120 INIT_LIST_HEAD(&llst->llst_list);
122 llst->llst_index = index;
123 atomic_set(&llst->llst_ref, 1);
125 spin_lock(&llsd->llsd_lock);
126 list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
127 if (tmp->llst_index == index) {
133 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
134 spin_unlock(&llsd->llsd_lock);
143 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
144 struct lfsck_layout_slave_target *llst)
148 spin_lock(&llsd->llsd_lock);
149 if (!list_empty(&llst->llst_list)) {
150 list_del_init(&llst->llst_list);
153 spin_unlock(&llsd->llsd_lock);
156 lfsck_layout_llst_put(llst);
159 static inline struct lfsck_layout_slave_target *
160 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
161 __u32 index, bool unlink)
163 struct lfsck_layout_slave_target *llst;
165 spin_lock(&llsd->llsd_lock);
166 list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
167 if (llst->llst_index == index) {
169 list_del_init(&llst->llst_list);
171 atomic_inc(&llst->llst_ref);
172 spin_unlock(&llsd->llsd_lock);
177 spin_unlock(&llsd->llsd_lock);
182 static struct lfsck_layout_req *
183 lfsck_layout_assistant_req_init(struct lfsck_assistant_object *lso,
184 struct dt_object *child, __u32 ost_idx,
187 struct lfsck_layout_req *llr;
191 return ERR_PTR(-ENOMEM);
193 INIT_LIST_HEAD(&llr->llr_lar.lar_list);
194 llr->llr_lar.lar_parent = lfsck_assistant_object_get(lso);
195 llr->llr_child = child;
196 llr->llr_ost_idx = ost_idx;
197 llr->llr_lov_idx = lov_idx;
202 static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
203 struct lfsck_assistant_req *lar)
205 struct lfsck_layout_req *llr =
206 container_of0(lar, struct lfsck_layout_req, llr_lar);
208 lfsck_object_put(env, llr->llr_child);
209 lfsck_assistant_object_put(env, lar->lar_parent);
214 lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
215 struct ptlrpc_request *req,
219 struct lfsck_async_interpret_args *laia = args;
220 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
222 ltd->ltd_synced_failures = 1;
223 atomic_dec(laia->laia_count);
230 * Notify remote LFSCK instances about former failures.
232 * The local LFSCK instance has recorded which OSTs have ever failed to respond
233 * some LFSCK verification requests (maybe because of network issues or the OST
234 * itself trouble). During the respond gap, the OST may missed some OST-objects
235 * verification, then the OST cannot know whether related OST-objects have been
236 * referenced by related MDT-objects or not, then in the second-stage scanning,
237 * these OST-objects will be regarded as orphan, if the OST-object contains bad
238 * parent FID for back reference, then it will misguide the LFSCK to make wrong
239 * fixing for the fake orphan.
241 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
242 * it will scan the bitmap for the ever failed OSTs, and notify them that they
243 * have ever missed some OST-object verification and should skip the handling
244 * for orphan OST-objects on all MDTs that are in the layout LFSCK.
246 * \param[in] env pointer to the thread context
247 * \param[in] com pointer to the lfsck component
248 * \param[in] lr pointer to the lfsck request
250 static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
251 struct lfsck_component *com,
252 struct lfsck_request *lr)
254 struct lfsck_async_interpret_args *laia =
255 &lfsck_env_info(env)->lti_laia2;
256 struct lfsck_assistant_data *lad = com->lc_data;
257 struct lfsck_layout *lo = com->lc_file_ram;
258 struct lfsck_instance *lfsck = com->lc_lfsck;
259 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
260 struct lfsck_tgt_desc *ltd;
261 struct ptlrpc_request_set *set;
267 if (!lad->lad_incomplete || lo->ll_flags & LF_INCOMPLETE)
270 /* If the MDT has ever failed to verfiy some OST-objects,
271 * then sync failures with them firstly. */
272 lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
274 atomic_set(&count, 0);
275 memset(laia, 0, sizeof(*laia));
276 laia->laia_count = &count;
277 set = ptlrpc_prep_set();
279 GOTO(out, rc = -ENOMEM);
281 down_read(<ds->ltd_rw_sem);
282 cfs_foreach_bit(lad->lad_bitmap, idx) {
283 ltd = lfsck_ltd2tgt(ltds, idx);
284 LASSERT(ltd != NULL);
286 laia->laia_ltd = ltd;
287 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
288 lfsck_layout_assistant_sync_failures_interpret,
291 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
292 "notify target %x for %s phase1 done: "
293 "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
294 ltd->ltd_index, lad->lad_name, rc);
301 up_read(<ds->ltd_rw_sem);
303 if (rc == 0 && atomic_read(&count) > 0)
304 rc = ptlrpc_set_wait(set);
306 ptlrpc_set_destroy(set);
308 if (rc == 0 && atomic_read(&count) > 0)
315 /* If failed to sync failures with the OSTs, then have to
316 * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
317 * subsequent orphan OST-object handling. */
318 lo->ll_flags |= LF_INCOMPLETE;
320 lr->lr_flags2 = lo->ll_flags;
323 static int lfsck_layout_get_lovea(const struct lu_env *env,
324 struct dt_object *obj, struct lu_buf *buf)
329 rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
331 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
335 lu_buf_realloc(buf, rc);
336 if (buf->lb_buf == NULL)
348 if (unlikely(buf->lb_buf == NULL)) {
349 lu_buf_alloc(buf, rc);
350 if (buf->lb_buf == NULL)
359 static int lfsck_layout_verify_header(struct lov_mds_md_v1 *lmm)
364 magic = le32_to_cpu(lmm->lmm_magic);
365 /* If magic crashed, keep it there. Sometime later, during OST-object
366 * orphan handling, if some OST-object(s) back-point to it, it can be
367 * verified and repaired. */
368 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
372 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
373 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
378 CDEBUG(D_LFSCK, "%s LOV EA magic %u on "DOSTID"\n",
379 rc == -EINVAL ? "Unknown" : "Unsupported",
385 pattern = le32_to_cpu(lmm->lmm_pattern);
386 /* XXX: currently, we only support LOV_PATTERN_RAID0. */
387 if (lov_pattern(pattern) != LOV_PATTERN_RAID0) {
390 lmm_oi_le_to_cpu(&oi, &lmm->lmm_oi);
391 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u on "DOSTID"\n",
392 pattern, POSTID(&oi));
400 #define LFSCK_RBTREE_BITMAP_SIZE PAGE_SIZE
401 #define LFSCK_RBTREE_BITMAP_WIDTH (LFSCK_RBTREE_BITMAP_SIZE << 3)
402 #define LFSCK_RBTREE_BITMAP_MASK (LFSCK_RBTREE_BITMAP_WIDTH - 1)
404 struct lfsck_rbtree_node {
405 struct rb_node lrn_node;
408 atomic_t lrn_known_count;
409 atomic_t lrn_accessed_count;
410 void *lrn_known_bitmap;
411 void *lrn_accessed_bitmap;
414 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
415 __u64 seq, __u32 oid)
417 if (seq < lrn->lrn_seq)
420 if (seq > lrn->lrn_seq)
423 if (oid < lrn->lrn_first_oid)
426 if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
432 /* The caller should hold llsd->llsd_rb_lock. */
433 static struct lfsck_rbtree_node *
434 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
435 const struct lu_fid *fid, bool *exact)
437 struct rb_node *node = llsd->llsd_rb_root.rb_node;
438 struct rb_node *prev = NULL;
439 struct lfsck_rbtree_node *lrn = NULL;
445 while (node != NULL) {
447 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
448 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
450 node = node->rb_left;
452 node = node->rb_right;
460 /* If there is no exactly matched one, then to the next valid one. */
463 /* The rbtree is empty. */
470 node = rb_next(prev);
472 /* The end of the rbtree. */
476 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
481 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
482 const struct lu_fid *fid)
484 struct lfsck_rbtree_node *lrn;
488 return ERR_PTR(-ENOMEM);
490 OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
491 if (lrn->lrn_known_bitmap == NULL) {
494 return ERR_PTR(-ENOMEM);
497 OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
498 if (lrn->lrn_accessed_bitmap == NULL) {
499 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
502 return ERR_PTR(-ENOMEM);
505 RB_CLEAR_NODE(&lrn->lrn_node);
506 lrn->lrn_seq = fid_seq(fid);
507 lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
508 atomic_set(&lrn->lrn_known_count, 0);
509 atomic_set(&lrn->lrn_accessed_count, 0);
514 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
516 OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
517 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
521 /* The caller should hold lock. */
522 static struct lfsck_rbtree_node *
523 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
524 struct lfsck_rbtree_node *lrn)
526 struct rb_node **pos = &llsd->llsd_rb_root.rb_node;
527 struct rb_node *parent = NULL;
528 struct lfsck_rbtree_node *tmp;
531 while (*pos != NULL) {
533 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
534 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
536 pos = &(*pos)->rb_left;
538 pos = &(*pos)->rb_right;
543 rb_link_node(&lrn->lrn_node, parent, pos);
544 rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
549 extern const struct dt_index_operations lfsck_orphan_index_ops;
551 static int lfsck_rbtree_setup(const struct lu_env *env,
552 struct lfsck_component *com)
554 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
555 struct lfsck_instance *lfsck = com->lc_lfsck;
556 struct dt_device *dev = lfsck->li_bottom;
557 struct lfsck_layout_slave_data *llsd = com->lc_data;
558 struct dt_object *obj;
560 fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
561 fid->f_oid = lfsck_dev_idx(lfsck);
563 obj = dt_locate(env, dev, fid);
565 RETURN(PTR_ERR(obj));
567 /* Generate an in-RAM object to stand for the layout rbtree.
568 * Scanning the layout rbtree will be via the iteration over
569 * the object. In the future, the rbtree may be written onto
570 * disk with the object.
572 * Mark the object to be as exist. */
573 obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
574 obj->do_index_ops = &lfsck_orphan_index_ops;
575 llsd->llsd_rb_obj = obj;
576 llsd->llsd_rbtree_valid = 1;
577 dev->dd_record_fid_accessed = 1;
579 CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
580 lfsck_lfsck2name(lfsck));
585 static void lfsck_rbtree_cleanup(const struct lu_env *env,
586 struct lfsck_component *com)
588 struct lfsck_instance *lfsck = com->lc_lfsck;
589 struct lfsck_layout_slave_data *llsd = com->lc_data;
590 struct rb_node *node = rb_first(&llsd->llsd_rb_root);
591 struct rb_node *next;
592 struct lfsck_rbtree_node *lrn;
594 lfsck->li_bottom->dd_record_fid_accessed = 0;
595 /* Invalid the rbtree, then no others will use it. */
596 write_lock(&llsd->llsd_rb_lock);
597 llsd->llsd_rbtree_valid = 0;
598 write_unlock(&llsd->llsd_rb_lock);
600 while (node != NULL) {
601 next = rb_next(node);
602 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
603 rb_erase(node, &llsd->llsd_rb_root);
604 lfsck_rbtree_free(lrn);
608 if (llsd->llsd_rb_obj != NULL) {
609 lfsck_object_put(env, llsd->llsd_rb_obj);
610 llsd->llsd_rb_obj = NULL;
613 CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
614 lfsck_lfsck2name(lfsck));
617 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
618 struct lfsck_component *com,
619 const struct lu_fid *fid,
622 struct lfsck_layout_slave_data *llsd = com->lc_data;
623 struct lfsck_rbtree_node *lrn;
629 if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
632 if (!fid_is_idif(fid) && !fid_is_norm(fid))
635 read_lock(&llsd->llsd_rb_lock);
636 if (!llsd->llsd_rbtree_valid)
637 GOTO(unlock, rc = 0);
639 lrn = lfsck_rbtree_search(llsd, fid, NULL);
641 struct lfsck_rbtree_node *tmp;
645 read_unlock(&llsd->llsd_rb_lock);
646 tmp = lfsck_rbtree_new(env, fid);
648 GOTO(out, rc = PTR_ERR(tmp));
651 write_lock(&llsd->llsd_rb_lock);
652 if (!llsd->llsd_rbtree_valid) {
653 lfsck_rbtree_free(tmp);
654 GOTO(unlock, rc = 0);
657 lrn = lfsck_rbtree_insert(llsd, tmp);
659 lfsck_rbtree_free(tmp);
662 idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
663 /* Any accessed object must be a known object. */
664 if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
665 atomic_inc(&lrn->lrn_known_count);
666 if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
667 atomic_inc(&lrn->lrn_accessed_count);
669 GOTO(unlock, rc = 0);
673 write_unlock(&llsd->llsd_rb_lock);
675 read_unlock(&llsd->llsd_rb_lock);
677 if (rc != 0 && accessed) {
678 struct lfsck_layout *lo = com->lc_file_ram;
680 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
681 "bitmap, and will cause incorrect LFSCK OST-object "
682 "handling, so disable it to cancel orphan handling "
683 "for related device. rc = %d\n",
684 lfsck_lfsck2name(com->lc_lfsck), rc);
686 lo->ll_flags |= LF_INCOMPLETE;
687 lfsck_rbtree_cleanup(env, com);
691 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
692 const struct lfsck_layout *src)
696 des->ll_magic = le32_to_cpu(src->ll_magic);
697 des->ll_status = le32_to_cpu(src->ll_status);
698 des->ll_flags = le32_to_cpu(src->ll_flags);
699 des->ll_success_count = le32_to_cpu(src->ll_success_count);
700 des->ll_run_time_phase1 = le32_to_cpu(src->ll_run_time_phase1);
701 des->ll_run_time_phase2 = le32_to_cpu(src->ll_run_time_phase2);
702 des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
703 des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
704 des->ll_time_last_checkpoint =
705 le64_to_cpu(src->ll_time_last_checkpoint);
706 des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
707 des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
708 des->ll_pos_first_inconsistent =
709 le64_to_cpu(src->ll_pos_first_inconsistent);
710 des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
711 des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
712 des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
713 des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
714 for (i = 0; i < LLIT_MAX; i++)
715 des->ll_objs_repaired[i] =
716 le64_to_cpu(src->ll_objs_repaired[i]);
717 des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
718 des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
721 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
722 const struct lfsck_layout *src)
726 des->ll_magic = cpu_to_le32(src->ll_magic);
727 des->ll_status = cpu_to_le32(src->ll_status);
728 des->ll_flags = cpu_to_le32(src->ll_flags);
729 des->ll_success_count = cpu_to_le32(src->ll_success_count);
730 des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
731 des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
732 des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
733 des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
734 des->ll_time_last_checkpoint =
735 cpu_to_le64(src->ll_time_last_checkpoint);
736 des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
737 des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
738 des->ll_pos_first_inconsistent =
739 cpu_to_le64(src->ll_pos_first_inconsistent);
740 des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
741 des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
742 des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
743 des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
744 for (i = 0; i < LLIT_MAX; i++)
745 des->ll_objs_repaired[i] =
746 cpu_to_le64(src->ll_objs_repaired[i]);
747 des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
748 des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
752 * Load the OST bitmap from the lfsck_layout trace file.
754 * \param[in] env pointer to the thread context
755 * \param[in] com pointer to the lfsck component
757 * \retval 0 for success
758 * \retval negative error number on failure or data corruption
760 static int lfsck_layout_load_bitmap(const struct lu_env *env,
761 struct lfsck_component *com)
763 struct dt_object *obj = com->lc_obj;
764 struct lfsck_assistant_data *lad = com->lc_data;
765 struct lfsck_layout *lo = com->lc_file_ram;
766 struct cfs_bitmap *bitmap = lad->lad_bitmap;
767 loff_t pos = com->lc_file_size;
773 if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
775 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
777 nbits = lo->ll_bitmap_size;
779 if (unlikely(nbits < BITS_PER_LONG))
780 nbits = BITS_PER_LONG;
782 if (nbits > bitmap->size) {
783 __u32 new_bits = bitmap->size;
784 struct cfs_bitmap *new_bitmap;
786 while (new_bits < nbits)
789 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
790 if (new_bitmap == NULL)
793 lad->lad_bitmap = new_bitmap;
794 CFS_FREE_BITMAP(bitmap);
798 if (lo->ll_bitmap_size == 0) {
799 lad->lad_incomplete = 0;
800 CFS_RESET_BITMAP(bitmap);
805 size = (lo->ll_bitmap_size + 7) >> 3;
806 rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
808 RETURN(rc >= 0 ? -EINVAL : rc);
810 if (cfs_bitmap_check_empty(bitmap))
811 lad->lad_incomplete = 0;
813 lad->lad_incomplete = 1;
819 * Load the layout LFSCK trace file from disk.
821 * The layout LFSCK trace file records the layout LFSCK status information
822 * and other statistics, such as how many objects have been scanned, and how
823 * many objects have been repaired, and etc. It also contains the bitmap for
824 * failed OSTs during the layout LFSCK. All these information will be loaded
825 * from disk to RAM when the layout LFSCK component setup.
827 * \param[in] env pointer to the thread context
828 * \param[in] com pointer to the lfsck component
830 * \retval positive number for file data corruption, the caller
831 * should reset the layout LFSCK trace file
832 * \retval 0 for success
833 * \retval negative error number on failure
835 static int lfsck_layout_load(const struct lu_env *env,
836 struct lfsck_component *com)
838 struct lfsck_layout *lo = com->lc_file_ram;
839 ssize_t size = com->lc_file_size;
843 rc = dt_read(env, com->lc_obj,
844 lfsck_buf_get(env, com->lc_file_disk, size), &pos);
848 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
849 lfsck_lfsck2name(com->lc_lfsck), rc);
851 } else if (rc != size) {
852 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
853 lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
857 lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
858 if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
859 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
860 "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
861 lo->ll_magic, LFSCK_LAYOUT_MAGIC);
869 * Store the layout LFSCK trace file on disk.
871 * The layout LFSCK trace file records the layout LFSCK status information
872 * and other statistics, such as how many objects have been scanned, and how
873 * many objects have been repaired, and etc. It also contains the bitmap for
874 * failed OSTs during the layout LFSCK. All these information will be synced
875 * from RAM to disk periodically.
877 * \param[in] env pointer to the thread context
878 * \param[in] com pointer to the lfsck component
880 * \retval 0 for success
881 * \retval negative error number on failure
883 static int lfsck_layout_store(const struct lu_env *env,
884 struct lfsck_component *com)
886 struct dt_object *obj = com->lc_obj;
887 struct lfsck_instance *lfsck = com->lc_lfsck;
888 struct lfsck_layout *lo_ram = com->lc_file_ram;
889 struct lfsck_layout *lo = com->lc_file_disk;
891 struct dt_device *dev = lfsck_obj2dev(obj);
892 struct cfs_bitmap *bitmap = NULL;
894 ssize_t size = com->lc_file_size;
899 if (lfsck->li_master) {
900 struct lfsck_assistant_data *lad = com->lc_data;
902 bitmap = lad->lad_bitmap;
903 nbits = bitmap->size;
906 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
909 lo_ram->ll_bitmap_size = nbits;
910 lfsck_layout_cpu_to_le(lo, lo_ram);
911 th = dt_trans_create(env, dev);
913 GOTO(log, rc = PTR_ERR(th));
915 rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
920 if (bitmap != NULL) {
921 rc = dt_declare_record_write(env, obj,
922 lfsck_buf_get(env, bitmap->data, nbits >> 3),
928 rc = dt_trans_start_local(env, dev, th);
933 rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
937 if (bitmap != NULL) {
939 rc = dt_record_write(env, obj,
940 lfsck_buf_get(env, bitmap->data, nbits >> 3),
947 dt_trans_stop(env, dev, th);
951 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
952 lfsck_lfsck2name(lfsck), rc);
957 static int lfsck_layout_init(const struct lu_env *env,
958 struct lfsck_component *com)
960 struct lfsck_layout *lo = com->lc_file_ram;
963 memset(lo, 0, com->lc_file_size);
964 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
965 lo->ll_status = LS_INIT;
966 down_write(&com->lc_sem);
967 rc = lfsck_layout_store(env, com);
968 up_write(&com->lc_sem);
973 static int fid_is_for_ostobj(const struct lu_env *env,
974 struct lfsck_instance *lfsck,
975 struct dt_object *obj, const struct lu_fid *fid)
977 struct seq_server_site *ss = lfsck_dev_site(lfsck);
978 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
979 struct lustre_mdt_attrs *lma;
982 fld_range_set_any(range);
983 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
985 if (fld_range_is_ost(range))
991 lma = &lfsck_env_info(env)->lti_lma;
992 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
994 if (rc == sizeof(*lma)) {
995 lustre_lma_swab(lma);
997 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1000 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1005 static struct lfsck_layout_seq *
1006 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1008 struct lfsck_layout_seq *lls;
1010 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1011 if (lls->lls_seq == seq)
1014 if (lls->lls_seq > seq)
1022 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1023 struct lfsck_layout_seq *lls)
1025 struct lfsck_layout_seq *tmp;
1026 struct list_head *pos = &llsd->llsd_seq_list;
1028 list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1029 if (lls->lls_seq < tmp->lls_seq) {
1030 pos = &tmp->lls_list;
1034 list_add_tail(&lls->lls_list, pos);
1038 lfsck_layout_lastid_create(const struct lu_env *env,
1039 struct lfsck_instance *lfsck,
1040 struct dt_object *obj)
1042 struct lfsck_thread_info *info = lfsck_env_info(env);
1043 struct lu_attr *la = &info->lti_la;
1044 struct dt_object_format *dof = &info->lti_dof;
1045 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1046 struct dt_device *dt = lfsck_obj2dev(obj);
1053 if (bk->lb_param & LPF_DRYRUN)
1056 memset(la, 0, sizeof(*la));
1057 la->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1058 la->la_valid = LA_MODE | LA_UID | LA_GID;
1059 memset(dof, 0, sizeof(*dof));
1060 dof->dof_type = dt_mode_to_dft(S_IFREG);
1062 th = dt_trans_create(env, dt);
1064 GOTO(log, rc = PTR_ERR(th));
1066 rc = dt_declare_create(env, obj, la, NULL, dof, th);
1070 rc = dt_declare_record_write(env, obj,
1071 lfsck_buf_get(env, &lastid,
1077 rc = dt_trans_start_local(env, dt, th);
1081 dt_write_lock(env, obj, 0);
1082 if (likely(dt_object_exists(obj) == 0)) {
1083 rc = dt_create(env, obj, la, NULL, dof, th);
1085 rc = dt_record_write(env, obj,
1086 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1089 dt_write_unlock(env, obj);
1094 dt_trans_stop(env, dt, th);
1097 CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1099 lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1105 lfsck_layout_lastid_reload(const struct lu_env *env,
1106 struct lfsck_component *com,
1107 struct lfsck_layout_seq *lls)
1113 dt_read_lock(env, lls->lls_lastid_obj, 0);
1114 rc = dt_record_read(env, lls->lls_lastid_obj,
1115 lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1116 dt_read_unlock(env, lls->lls_lastid_obj);
1117 if (unlikely(rc != 0))
1120 lastid = le64_to_cpu(lastid);
1121 if (lastid < lls->lls_lastid_known) {
1122 struct lfsck_instance *lfsck = com->lc_lfsck;
1123 struct lfsck_layout *lo = com->lc_file_ram;
1125 lls->lls_lastid = lls->lls_lastid_known;
1127 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1128 LASSERT(lfsck->li_out_notify != NULL);
1130 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1131 LE_LASTID_REBUILDING);
1132 lo->ll_flags |= LF_CRASHED_LASTID;
1134 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1135 "LAST_ID file (1) for the sequence %#llx"
1136 ", old value %llu, known value %llu\n",
1137 lfsck_lfsck2name(lfsck), lls->lls_seq,
1138 lastid, lls->lls_lastid);
1140 } else if (lastid >= lls->lls_lastid) {
1141 lls->lls_lastid = lastid;
1149 lfsck_layout_lastid_store(const struct lu_env *env,
1150 struct lfsck_component *com)
1152 struct lfsck_instance *lfsck = com->lc_lfsck;
1153 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1154 struct dt_device *dt = lfsck->li_bottom;
1155 struct lfsck_layout_slave_data *llsd = com->lc_data;
1156 struct lfsck_layout_seq *lls;
1162 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1165 if (!lls->lls_dirty)
1168 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1169 "<seq> %#llx as <oid> %llu\n",
1170 lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1172 if (bk->lb_param & LPF_DRYRUN) {
1177 th = dt_trans_create(env, dt);
1180 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1181 "the LAST_ID for <seq> %#llx(1): rc = %d\n",
1182 lfsck_lfsck2name(com->lc_lfsck),
1187 lastid = cpu_to_le64(lls->lls_lastid);
1188 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1189 lfsck_buf_get(env, &lastid,
1195 rc = dt_trans_start_local(env, dt, th);
1199 dt_write_lock(env, lls->lls_lastid_obj, 0);
1200 rc = dt_record_write(env, lls->lls_lastid_obj,
1201 lfsck_buf_get(env, &lastid,
1202 sizeof(lastid)), &pos, th);
1203 dt_write_unlock(env, lls->lls_lastid_obj);
1208 dt_trans_stop(env, dt, th);
1211 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1212 "the LAST_ID for <seq> %#llx(2): rc = %d\n",
1213 lfsck_lfsck2name(com->lc_lfsck),
1222 lfsck_layout_lastid_load(const struct lu_env *env,
1223 struct lfsck_component *com,
1224 struct lfsck_layout_seq *lls)
1226 struct lfsck_instance *lfsck = com->lc_lfsck;
1227 struct lfsck_layout *lo = com->lc_file_ram;
1228 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
1229 struct dt_object *obj;
1234 lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1235 obj = dt_locate(env, lfsck->li_bottom, fid);
1237 RETURN(PTR_ERR(obj));
1239 /* LAST_ID crashed, to be rebuilt */
1240 if (dt_object_exists(obj) == 0) {
1241 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1242 LASSERT(lfsck->li_out_notify != NULL);
1244 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1245 LE_LASTID_REBUILDING);
1246 lo->ll_flags |= LF_CRASHED_LASTID;
1248 CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1249 "LAST_ID file for sequence %#llx\n",
1250 lfsck_lfsck2name(lfsck), lls->lls_seq);
1252 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1254 struct l_wait_info lwi = LWI_TIMEOUT(
1255 cfs_time_seconds(cfs_fail_val),
1258 /* Some others may changed the cfs_fail_val
1259 * as zero after above check, re-check it for
1260 * sure to avoid falling into wait for ever. */
1261 if (likely(lwi.lwi_timeout > 0)) {
1262 struct ptlrpc_thread *thread =
1265 up_write(&com->lc_sem);
1266 l_wait_event(thread->t_ctl_waitq,
1267 !thread_is_running(thread),
1269 down_write(&com->lc_sem);
1274 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1276 dt_read_lock(env, obj, 0);
1277 rc = dt_read(env, obj,
1278 lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1280 dt_read_unlock(env, obj);
1281 if (rc != 0 && rc != sizeof(__u64))
1282 GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1284 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1285 LASSERT(lfsck->li_out_notify != NULL);
1287 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1288 LE_LASTID_REBUILDING);
1289 lo->ll_flags |= LF_CRASHED_LASTID;
1291 CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1292 "LAST_ID file for the sequence %#llx"
1294 lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1297 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1305 lfsck_object_put(env, obj);
1307 lls->lls_lastid_obj = obj;
1312 static void lfsck_layout_record_failure(const struct lu_env *env,
1313 struct lfsck_instance *lfsck,
1314 struct lfsck_layout *lo)
1318 lo->ll_objs_failed_phase1++;
1319 cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1321 if (lo->ll_pos_first_inconsistent == 0 ||
1322 lo->ll_pos_first_inconsistent < cookie) {
1323 lo->ll_pos_first_inconsistent = cookie;
1325 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1326 "inconsistency at the pos [%llu]\n",
1327 lfsck_lfsck2name(lfsck),
1328 lo->ll_pos_first_inconsistent);
1332 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1333 struct lfsck_component *com,
1336 struct lfsck_instance *lfsck = com->lc_lfsck;
1337 struct lfsck_layout *lo = com->lc_file_ram;
1339 down_write(&com->lc_sem);
1340 lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1341 HALF_SEC - com->lc_time_last_checkpoint);
1342 lo->ll_time_last_checkpoint = cfs_time_current_sec();
1343 lo->ll_objs_checked_phase2 += com->lc_new_checked;
1346 if (lo->ll_flags & LF_INCOMPLETE) {
1347 lo->ll_status = LS_PARTIAL;
1349 if (lfsck->li_master) {
1350 struct lfsck_assistant_data *lad = com->lc_data;
1352 if (lad->lad_incomplete)
1353 lo->ll_status = LS_PARTIAL;
1355 lo->ll_status = LS_COMPLETED;
1357 lo->ll_status = LS_COMPLETED;
1360 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1361 lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1362 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1363 lo->ll_success_count++;
1364 } else if (rc == 0) {
1365 if (lfsck->li_status != 0)
1366 lo->ll_status = lfsck->li_status;
1368 lo->ll_status = LS_STOPPED;
1370 lo->ll_status = LS_FAILED;
1373 rc = lfsck_layout_store(env, com);
1374 up_write(&com->lc_sem);
1379 static int lfsck_layout_trans_stop(const struct lu_env *env,
1380 struct dt_device *dev,
1381 struct thandle *handle, int result)
1385 /* XXX: If there is something worng or it needs to repair nothing,
1386 * then notify the lower to stop the modification. Currently,
1387 * we use th_result for such purpose, that may be replaced by
1388 * some rollback mechanism in the future. */
1389 handle->th_result = result;
1390 rc = dt_trans_stop(env, dev, handle);
1392 return result > 0 ? 0 : result;
1394 return rc == 0 ? 1 : rc;
1398 * Get the system default stripe size.
1400 * \param[in] env pointer to the thread context
1401 * \param[in] lfsck pointer to the lfsck instance
1402 * \param[out] size pointer to the default stripe size
1404 * \retval 0 for success
1405 * \retval negative error number on failure
1407 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1408 struct lfsck_instance *lfsck,
1411 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1412 struct dt_object *root;
1415 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1417 return PTR_ERR(root);
1419 /* Get the default stripe size via xattr_get on the backend root. */
1420 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1423 /* The lum->lmm_stripe_size is LE mode. The *size also
1424 * should be LE mode. So it is unnecessary to convert. */
1425 *size = lum->lmm_stripe_size;
1427 } else if (unlikely(rc == 0)) {
1431 lfsck_object_put(env, root);
1437 * \retval +1: repaired
1438 * \retval 0: did nothing
1439 * \retval -ve: on error
1441 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1442 struct thandle *handle,
1443 struct dt_object *parent,
1444 struct lu_fid *cfid,
1446 struct lov_ost_data_v1 *slot,
1447 int fl, __u32 ost_idx)
1449 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1450 struct lov_mds_md_v1 *lmm = buf->lb_buf;
1451 struct lu_buf ea_buf;
1457 magic = le32_to_cpu(lmm->lmm_magic);
1458 count = le16_to_cpu(lmm->lmm_stripe_count);
1460 fid_to_ostid(cfid, oi);
1461 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1462 slot->l_ost_gen = cpu_to_le32(0);
1463 slot->l_ost_idx = cpu_to_le32(ost_idx);
1465 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
1466 struct lov_ost_data_v1 *objs;
1469 if (magic == LOV_MAGIC_V1)
1470 objs = &lmm->lmm_objects[0];
1472 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1473 for (i = 0; i < count; i++, objs++) {
1474 if (objs != slot && lovea_slot_is_dummy(objs))
1478 /* If the @slot is the last dummy slot to be refilled,
1479 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1481 lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
1484 lfsck_buf_init(&ea_buf, lmm, lov_mds_md_size(count, magic));
1485 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1493 * \retval +1: repaired
1494 * \retval 0: did nothing
1495 * \retval -ve: on error
1497 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1498 struct lfsck_instance *lfsck,
1499 struct thandle *handle,
1500 struct dt_object *parent,
1501 struct lu_fid *cfid,
1502 struct lu_buf *buf, int fl,
1503 __u32 ost_idx, __u32 ea_off, bool reset)
1505 struct lov_mds_md_v1 *lmm = buf->lb_buf;
1506 struct lov_ost_data_v1 *objs;
1512 if (fl == LU_XATTR_CREATE || reset) {
1513 __u32 pattern = LOV_PATTERN_RAID0;
1516 LASSERT(buf->lb_len >= lov_mds_md_size(count, LOV_MAGIC_V1));
1518 if (ea_off != 0 || reset) {
1519 pattern |= LOV_PATTERN_F_HOLE;
1523 memset(lmm, 0, buf->lb_len);
1524 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1525 lmm->lmm_pattern = cpu_to_le32(pattern);
1526 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1527 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1529 rc = lfsck_layout_get_def_stripesize(env, lfsck,
1530 &lmm->lmm_stripe_size);
1534 objs = &lmm->lmm_objects[ea_off];
1536 __u32 magic = le32_to_cpu(lmm->lmm_magic);
1539 count = le16_to_cpu(lmm->lmm_stripe_count);
1540 if (magic == LOV_MAGIC_V1)
1541 objs = &lmm->lmm_objects[count];
1543 objs = &((struct lov_mds_md_v3 *)lmm)->
1546 gap = ea_off - count;
1549 LASSERT(buf->lb_len >= lov_mds_md_size(count, magic));
1552 memset(objs, 0, gap * sizeof(*objs));
1553 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1557 lmm->lmm_layout_gen =
1558 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1562 lmm->lmm_stripe_count = cpu_to_le16(count);
1563 rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1566 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1567 DFID": parent "DFID", OST-index %u, stripe-index %u, fl %d, "
1568 "reset %s, %s LOV EA hole: rc = %d\n",
1569 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1570 ost_idx, ea_off, fl, reset ? "yes" : "no",
1571 hole ? "with" : "without", rc);
1576 static int __lfsck_layout_update_pfid(const struct lu_env *env,
1577 struct dt_object *child,
1578 const struct lu_fid *pfid, __u32 offset)
1580 struct dt_device *dev = lfsck_obj2dev(child);
1581 struct filter_fid *ff = &lfsck_env_info(env)->lti_new_pfid;
1582 struct thandle *handle;
1583 struct lu_buf buf = { NULL };
1586 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1587 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1588 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1589 * MDT-object's FID::f_ver, instead it is the OST-object index in its
1590 * parent MDT-object's layout EA. */
1591 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
1592 lfsck_buf_init(&buf, ff, sizeof(struct filter_fid));
1594 handle = dt_trans_create(env, dev);
1596 RETURN(PTR_ERR(handle));
1598 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1602 rc = dt_trans_start_local(env, dev, handle);
1606 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1611 dt_trans_stop(env, dev, handle);
1617 * \retval +1: repaired
1618 * \retval 0: did nothing
1619 * \retval -ve: on error
1621 static int lfsck_layout_update_pfid(const struct lu_env *env,
1622 struct lfsck_component *com,
1623 struct dt_object *parent,
1624 struct lu_fid *cfid,
1625 struct dt_device *cdev, __u32 ea_off)
1627 struct dt_object *child;
1631 child = lfsck_object_find_by_dev(env, cdev, cfid);
1633 RETURN(PTR_ERR(child));
1635 rc = __lfsck_layout_update_pfid(env, child,
1636 lu_object_fid(&parent->do_lu), ea_off);
1637 lfsck_object_put(env, child);
1639 RETURN(rc == 0 ? 1 : rc);
1643 * This function will create the MDT-object with the given (partial) LOV EA.
1645 * Under some data corruption cases, the MDT-object of the file may be lost,
1646 * but its OST-objects, or some of them are there. The layout LFSCK needs to
1647 * re-create the MDT-object with the orphan OST-object(s) information.
1649 * On the other hand, the LFSCK may has created some OST-object for repairing
1650 * dangling LOV EA reference, but as the LFSCK processing, it may find that
1651 * the old OST-object is there and should replace the former new created OST
1652 * object. Unfortunately, some others have modified such newly created object.
1653 * To keep the data (both new and old), the LFSCK will create MDT-object with
1654 * new FID to reference the original OST-object.
1656 * \param[in] env pointer to the thread context
1657 * \param[in] com pointer to the lfsck component
1658 * \param[in] ltd pointer to target device descriptor
1659 * \param[in] rec pointer to the record for the orphan OST-object
1660 * \param[in] cfid pointer to FID for the orphan OST-object
1661 * \param[in] infix additional information, such as the FID for original
1662 * MDT-object and the stripe offset in the LOV EA
1663 * \param[in] type the type for describing why the orphan MDT-object is
1664 * created. The rules are as following:
1666 * type "C": Multiple OST-objects claim the same MDT-object and the
1667 * same slot in the layout EA. Then the LFSCK will create
1668 * new MDT-object(s) to hold the conflict OST-object(s).
1670 * type "N": The orphan OST-object does not know which one was the
1671 * real parent MDT-object, so the LFSCK uses new FID for
1672 * its parent MDT-object.
1674 * type "R": The orphan OST-object knows its parent MDT-object FID,
1675 * but does not know the position (the file name) in the
1678 * type "D": The MDT-object is a directory, it may knows its parent
1679 * but because there is no valid linkEA, the LFSCK cannot
1680 * know where to put it back to the namespace.
1681 * type "O": The MDT-object has no linkEA, and there is no name
1682 * entry that references the MDT-object.
1684 * type "P": The orphan object to be created was a parent directory
1685 * of some MDT-object which linkEA shows that the @orphan
1686 * object is missing.
1688 * The orphan name will be like:
1689 * ${FID}-${infix}-${type}-${conflict_version}
1691 * \param[in] ea_off the stripe offset in the LOV EA
1693 * \retval positive on repaired something
1694 * \retval 0 if needs to repair nothing
1695 * \retval negative error number on failure
1697 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1698 struct lfsck_component *com,
1699 struct lfsck_tgt_desc *ltd,
1700 struct lu_orphan_rec *rec,
1701 struct lu_fid *cfid,
1706 struct lfsck_thread_info *info = lfsck_env_info(env);
1707 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
1708 char *name = info->lti_key;
1709 struct lu_attr *la = &info->lti_la2;
1710 struct dt_object_format *dof = &info->lti_dof;
1711 struct lfsck_instance *lfsck = com->lc_lfsck;
1712 struct lu_fid *pfid = &rec->lor_fid;
1713 struct lu_fid *tfid = &info->lti_fid3;
1714 struct dt_device *dev = lfsck->li_bottom;
1715 struct dt_object *lpf = lfsck->li_lpf_obj;
1716 struct dt_object *pobj = NULL;
1717 struct dt_object *cobj = NULL;
1718 struct thandle *th = NULL;
1719 struct lu_buf *ea_buf = &info->lti_big_buf;
1720 struct lu_buf lov_buf;
1721 struct lfsck_lock_handle *llh = &info->lti_llh;
1722 struct linkea_data ldata = { NULL };
1723 struct lu_buf linkea_buf;
1724 const struct lu_name *pname;
1730 if (unlikely(lpf == NULL))
1731 GOTO(log, rc = -ENXIO);
1733 /* We use two separated transactions to repair the inconsistency.
1735 * 1) create the MDT-object locally.
1736 * 2) update the OST-object's PFID EA if necessary.
1738 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
1739 * updated when the layout LFSCK run next time.
1741 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
1742 * when the layout LFSCK run next time. */
1744 if (fid_is_zero(pfid)) {
1745 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1749 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1751 GOTO(log, rc = PTR_ERR(cobj));
1754 pobj = lfsck_object_find_by_dev(env, dev, pfid);
1756 GOTO(log, rc = PTR_ERR(pobj));
1758 LASSERT(infix != NULL);
1759 LASSERT(type != NULL);
1761 memset(la, 0, sizeof(*la));
1762 la->la_uid = rec->lor_uid;
1763 la->la_gid = rec->lor_gid;
1764 la->la_mode = S_IFREG | S_IRUSR;
1765 la->la_valid = LA_MODE | LA_UID | LA_GID;
1767 memset(dof, 0, sizeof(*dof));
1768 dof->dof_type = dt_mode_to_dft(S_IFREG);
1769 /* Because the dof->dof_reg.striped = 0, the LOD will not create
1770 * the stripe(s). The LFSCK will specify the LOV EA via
1771 * lfsck_layout_extend_lovea(). */
1773 size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1774 if (ea_buf->lb_len < size) {
1775 lu_buf_realloc(ea_buf, size);
1776 if (ea_buf->lb_buf == NULL)
1777 GOTO(log, rc = -ENOMEM);
1782 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
1784 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1785 (const struct dt_key *)name);
1786 if (rc != 0 && rc != -ENOENT)
1790 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
1791 MDS_INODELOCK_UPDATE, LCK_PW);
1795 /* Re-check whether the name conflict with othrs after taken
1797 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1798 (const struct dt_key *)name);
1799 if (unlikely(rc == 0)) {
1807 pname = lfsck_name_get_const(env, name, strlen(name));
1808 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
1809 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
1813 /* The 1st transaction. */
1814 th = dt_trans_create(env, dev);
1816 GOTO(unlock, rc = PTR_ERR(th));
1818 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1822 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
1823 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
1824 LU_XATTR_REPLACE, th);
1828 dtrec->rec_fid = pfid;
1829 dtrec->rec_type = S_IFREG;
1830 rc = dt_declare_insert(env, lpf,
1831 (const struct dt_rec *)dtrec,
1832 (const struct dt_key *)name, th);
1836 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1837 ldata.ld_leh->leh_len);
1838 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
1839 XATTR_NAME_LINK, 0, th);
1843 rc = dt_trans_start_local(env, dev, th);
1847 dt_write_lock(env, pobj, 0);
1848 rc = dt_create(env, pobj, la, NULL, dof, th);
1850 rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
1851 &lov_buf, 0, ltd->ltd_index, ea_off, true);
1852 dt_write_unlock(env, pobj);
1856 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
1857 (const struct dt_key *)name, th, 1);
1861 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
1862 if (rc == 0 && cobj != NULL) {
1863 dt_trans_stop(env, dev, th);
1866 /* The 2nd transaction. */
1867 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ea_off);
1874 dt_trans_stop(env, dev, th);
1880 if (cobj != NULL && !IS_ERR(cobj))
1881 lfsck_object_put(env, cobj);
1882 if (pobj != NULL && !IS_ERR(pobj))
1883 lfsck_object_put(env, pobj);
1886 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
1887 "recreate the lost MDT-object: parent "DFID
1888 ", child "DFID", OST-index %u, stripe-index %u, "
1889 "infix %s, type %s: rc = %d\n",
1890 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
1891 ltd->ltd_index, ea_off, infix, type, rc);
1893 return rc >= 0 ? 1 : rc;
1896 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
1897 struct lfsck_component *com,
1898 const struct lu_fid *fid,
1901 struct lfsck_thread_info *info = lfsck_env_info(env);
1902 struct lfsck_request *lr = &info->lti_lr;
1903 struct lfsck_instance *lfsck = com->lc_lfsck;
1904 struct lfsck_tgt_desc *ltd;
1905 struct ptlrpc_request *req;
1906 struct lfsck_request *tmp;
1907 struct obd_export *exp;
1911 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
1912 if (unlikely(ltd == NULL))
1916 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
1917 GOTO(put, rc = -EOPNOTSUPP);
1919 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
1921 GOTO(put, rc = -ENOMEM);
1923 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1925 ptlrpc_request_free(req);
1930 memset(lr, 0, sizeof(*lr));
1931 lr->lr_event = LE_CONDITIONAL_DESTROY;
1932 lr->lr_active = LFSCK_TYPE_LAYOUT;
1935 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1937 ptlrpc_request_set_replen(req);
1939 rc = ptlrpc_queue_wait(req);
1940 ptlrpc_req_finished(req);
1950 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
1951 struct lfsck_component *com,
1952 struct lfsck_request *lr)
1954 struct lfsck_thread_info *info = lfsck_env_info(env);
1955 struct lu_attr *la = &info->lti_la;
1956 union ldlm_policy_data *policy = &info->lti_policy;
1957 struct ldlm_res_id *resid = &info->lti_resid;
1958 struct lfsck_instance *lfsck = com->lc_lfsck;
1959 struct dt_device *dev = lfsck->li_bottom;
1960 struct lu_fid *fid = &lr->lr_fid;
1961 struct dt_object *obj;
1962 struct thandle *th = NULL;
1963 struct lustre_handle lh = { 0 };
1968 obj = lfsck_object_find_by_dev(env, dev, fid);
1970 RETURN(PTR_ERR(obj));
1972 dt_read_lock(env, obj, 0);
1973 if (dt_object_exists(obj) == 0 ||
1974 lfsck_is_dead_obj(obj)) {
1975 dt_read_unlock(env, obj);
1977 GOTO(put, rc = -ENOENT);
1980 /* Get obj's attr without lock firstly. */
1981 rc = dt_attr_get(env, obj, la);
1982 dt_read_unlock(env, obj);
1986 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
1987 GOTO(put, rc = -ETXTBSY);
1989 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
1990 LASSERT(lfsck->li_namespace != NULL);
1992 memset(policy, 0, sizeof(*policy));
1993 policy->l_extent.end = OBD_OBJECT_EOF;
1994 ost_fid_build_resid(fid, resid);
1995 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
1996 policy, LCK_EX, &flags, ldlm_blocking_ast,
1997 ldlm_completion_ast, NULL, NULL, 0,
1998 LVB_T_NONE, NULL, &lh);
2000 GOTO(put, rc = -EIO);
2002 dt_write_lock(env, obj, 0);
2003 /* Get obj's attr within lock again. */
2004 rc = dt_attr_get(env, obj, la);
2008 if (la->la_ctime != 0)
2009 GOTO(unlock, rc = -ETXTBSY);
2011 th = dt_trans_create(env, dev);
2013 GOTO(unlock, rc = PTR_ERR(th));
2015 rc = dt_declare_ref_del(env, obj, th);
2019 rc = dt_declare_destroy(env, obj, th);
2023 rc = dt_trans_start_local(env, dev, th);
2027 rc = dt_ref_del(env, obj, th);
2031 rc = dt_destroy(env, obj, th);
2033 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2034 "OST-object "DFID" that was created for reparing "
2035 "dangling referenced case. But the original missing "
2036 "OST-object is found now.\n",
2037 lfsck_lfsck2name(lfsck), PFID(fid));
2042 dt_trans_stop(env, dev, th);
2045 dt_write_unlock(env, obj);
2046 ldlm_lock_decref(&lh, LCK_EX);
2049 lfsck_object_put(env, obj);
2055 * Some OST-object has occupied the specified layout EA slot.
2056 * Such OST-object may be generated by the LFSCK when repair
2057 * dangling referenced MDT-object, which can be indicated by
2058 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2059 * is true and such OST-object has not been modified yet, we
2060 * will replace it with the orphan OST-object; otherwise the
2061 * LFSCK will create new MDT-object to reference the orphan.
2063 * \retval +1: repaired
2064 * \retval 0: did nothing
2065 * \retval -ve: on error
2067 static int lfsck_layout_conflict_create(const struct lu_env *env,
2068 struct lfsck_component *com,
2069 struct lfsck_tgt_desc *ltd,
2070 struct lu_orphan_rec *rec,
2071 struct dt_object *parent,
2072 struct lu_fid *cfid,
2073 struct lu_buf *ea_buf,
2074 struct lov_ost_data_v1 *slot,
2077 struct lfsck_thread_info *info = lfsck_env_info(env);
2078 struct lu_fid *cfid2 = &info->lti_fid2;
2079 struct ost_id *oi = &info->lti_oi;
2080 struct lov_mds_md_v1 *lmm = ea_buf->lb_buf;
2081 struct dt_device *dev = lfsck_obj2dev(parent);
2082 struct thandle *th = NULL;
2083 struct lustre_handle lh = { 0 };
2084 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2088 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2089 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2093 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2094 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2098 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2099 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2104 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2106 /* If the conflict OST-obejct is not created for fixing dangling
2107 * referenced MDT-object in former LFSCK check/repair, or it has
2108 * been modified by others, then we cannot destroy it. Re-create
2109 * a new MDT-object for the orphan OST-object. */
2110 if (rc == -ETXTBSY) {
2111 /* No need the layout lock on the original parent. */
2112 lfsck_ibits_unlock(&lh, LCK_EX);
2114 fid_zero(&rec->lor_fid);
2115 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2116 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2118 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2119 info->lti_tmpbuf, "C", ea_off);
2124 if (rc != 0 && rc != -ENOENT)
2127 th = dt_trans_create(env, dev);
2129 GOTO(unlock, rc = PTR_ERR(th));
2131 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2132 LU_XATTR_REPLACE, th);
2136 rc = dt_trans_start_local(env, dev, th);
2140 dt_write_lock(env, parent, 0);
2141 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2142 rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2143 LU_XATTR_REPLACE, ltd->ltd_index);
2144 dt_write_unlock(env, parent);
2149 dt_trans_stop(env, dev, th);
2152 lfsck_ibits_unlock(&lh, LCK_EX);
2155 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2156 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2157 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2158 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2159 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2162 return rc >= 0 ? 1 : rc;
2166 * \retval +1: repaired
2167 * \retval 0: did nothing
2168 * \retval -ve: on error
2170 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2171 struct lfsck_component *com,
2172 struct lfsck_tgt_desc *ltd,
2173 struct lu_orphan_rec *rec,
2174 struct dt_object *parent,
2175 struct lu_fid *cfid,
2176 __u32 ost_idx, __u32 ea_off)
2178 struct lfsck_thread_info *info = lfsck_env_info(env);
2179 struct lu_buf *buf = &info->lti_big_buf;
2180 struct lu_fid *fid = &info->lti_fid2;
2181 struct ost_id *oi = &info->lti_oi;
2182 struct lfsck_instance *lfsck = com->lc_lfsck;
2183 struct dt_device *dt = lfsck_obj2dev(parent);
2184 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2185 struct thandle *handle = NULL;
2187 struct lov_mds_md_v1 *lmm;
2188 struct lov_ost_data_v1 *objs;
2189 struct lustre_handle lh = { 0 };
2196 bool locked = false;
2199 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2200 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2203 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2204 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2205 "stripe-index %u: rc = %d\n",
2206 lfsck_lfsck2name(lfsck), PFID(cfid),
2207 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off, rc);
2214 dt_write_unlock(env, parent);
2218 if (handle != NULL) {
2219 dt_trans_stop(env, dt, handle);
2224 GOTO(unlock_layout, rc);
2227 if (buf->lb_len < lovea_size) {
2228 lu_buf_realloc(buf, lovea_size);
2229 if (buf->lb_buf == NULL)
2230 GOTO(unlock_layout, rc = -ENOMEM);
2233 if (!(bk->lb_param & LPF_DRYRUN)) {
2234 handle = dt_trans_create(env, dt);
2236 GOTO(unlock_layout, rc = PTR_ERR(handle));
2238 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2243 rc = dt_trans_start_local(env, dt, handle);
2248 dt_write_lock(env, parent, 0);
2250 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2251 if (rc == -ERANGE) {
2252 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2255 } else if (rc == -ENODATA || rc == 0) {
2256 lovea_size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2257 /* If the declared is not big enough, re-try. */
2258 if (buf->lb_len < lovea_size) {
2262 fl = LU_XATTR_CREATE;
2263 } else if (rc < 0) {
2264 GOTO(unlock_parent, rc);
2265 } else if (unlikely(buf->lb_len == 0)) {
2268 fl = LU_XATTR_REPLACE;
2272 if (fl == LU_XATTR_CREATE) {
2273 if (bk->lb_param & LPF_DRYRUN)
2274 GOTO(unlock_parent, rc = 1);
2276 LASSERT(buf->lb_len >= lovea_size);
2278 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2279 buf, fl, ost_idx, ea_off, false);
2281 GOTO(unlock_parent, rc);
2285 rc1 = lfsck_layout_verify_header(lmm);
2287 /* If the LOV EA crashed, the rebuild it. */
2288 if (rc1 == -EINVAL) {
2289 if (bk->lb_param & LPF_DRYRUN)
2290 GOTO(unlock_parent, rc = 1);
2292 LASSERT(buf->lb_len >= lovea_size);
2294 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2295 buf, fl, ost_idx, ea_off, true);
2297 GOTO(unlock_parent, rc);
2300 /* For other unknown magic/pattern, keep the current LOV EA. */
2302 GOTO(unlock_parent, rc = rc1);
2304 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2305 * been verified in lfsck_layout_verify_header() already. If some
2306 * new magic introduced in the future, then layout LFSCK needs to
2307 * be updated also. */
2308 magic = le32_to_cpu(lmm->lmm_magic);
2309 if (magic == LOV_MAGIC_V1) {
2310 objs = &lmm->lmm_objects[0];
2312 LASSERT(magic == LOV_MAGIC_V3);
2313 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2316 count = le16_to_cpu(lmm->lmm_stripe_count);
2318 GOTO(unlock_parent, rc = -EINVAL);
2321 /* Exceed the current end of MDT-object layout EA. Then extend it. */
2322 if (count <= ea_off) {
2323 if (bk->lb_param & LPF_DRYRUN)
2324 GOTO(unlock_parent, rc = 1);
2326 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2327 /* If the declared is not big enough, re-try. */
2328 if (buf->lb_len < lovea_size) {
2333 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2334 buf, fl, ost_idx, ea_off, false);
2336 GOTO(unlock_parent, rc);
2339 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2341 for (i = 0; i < count; i++, objs++) {
2342 /* The MDT-object was created via lfsck_layout_recover_create()
2343 * by others before, and we fill the dummy layout EA. */
2344 if (lovea_slot_is_dummy(objs)) {
2348 if (bk->lb_param & LPF_DRYRUN)
2349 GOTO(unlock_parent, rc = 1);
2351 lmm->lmm_layout_gen =
2352 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2353 rc = lfsck_layout_refill_lovea(env, handle, parent,
2354 cfid, buf, objs, fl,
2357 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2358 "dummy layout slot for "DFID": parent "DFID
2359 ", OST-index %u, stripe-index %u: rc = %d\n",
2360 lfsck_lfsck2name(lfsck), PFID(cfid),
2361 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2363 GOTO(unlock_parent, rc);
2366 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2367 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2369 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2370 "invalid layout EA at the slot %d, index %u\n",
2371 lfsck_lfsck2name(lfsck),
2372 PFID(lfsck_dto2fid(parent)), i,
2373 le32_to_cpu(objs->l_ost_idx));
2375 GOTO(unlock_parent, rc);
2378 /* It should be rare case, the slot is there, but the LFSCK
2379 * does not handle it during the first-phase cycle scanning. */
2380 if (unlikely(lu_fid_eq(fid, cfid))) {
2382 GOTO(unlock_parent, rc = 0);
2384 /* Rare case that the OST-object index
2385 * does not match the parent MDT-object
2386 * layout EA. We trust the later one. */
2387 if (bk->lb_param & LPF_DRYRUN)
2388 GOTO(unlock_parent, rc = 1);
2390 dt_write_unlock(env, parent);
2392 dt_trans_stop(env, dt, handle);
2393 lfsck_ibits_unlock(&lh, LCK_EX);
2394 rc = lfsck_layout_update_pfid(env, com, parent,
2395 cfid, ltd->ltd_tgt, i);
2397 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2398 "updated OST-object's pfid for "DFID
2399 ": parent "DFID", OST-index %u, "
2400 "stripe-index %u: rc = %d\n",
2401 lfsck_lfsck2name(lfsck), PFID(cfid),
2402 PFID(lfsck_dto2fid(parent)),
2403 ltd->ltd_index, i, rc);
2410 /* The MDT-object exists, but related layout EA slot is occupied
2412 if (bk->lb_param & LPF_DRYRUN)
2413 GOTO(unlock_parent, rc = 1);
2415 dt_write_unlock(env, parent);
2417 dt_trans_stop(env, dt, handle);
2418 lfsck_ibits_unlock(&lh, LCK_EX);
2419 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2420 objs = &lmm->lmm_objects[ea_off];
2422 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2423 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2430 dt_write_unlock(env, parent);
2434 dt_trans_stop(env, dt, handle);
2437 lfsck_ibits_unlock(&lh, LCK_EX);
2442 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2443 struct lfsck_component *com,
2444 struct lfsck_tgt_desc *ltd,
2445 struct lu_orphan_rec *rec,
2446 struct lu_fid *cfid)
2448 struct lfsck_layout *lo = com->lc_file_ram;
2449 struct lu_fid *pfid = &rec->lor_fid;
2450 struct dt_object *parent = NULL;
2451 __u32 ea_off = pfid->f_stripe_idx;
2455 if (!fid_is_sane(cfid))
2456 GOTO(out, rc = -EINVAL);
2458 if (fid_is_zero(pfid)) {
2459 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2465 if (!fid_is_sane(pfid))
2466 GOTO(out, rc = -EINVAL);
2468 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2470 GOTO(out, rc = PTR_ERR(parent));
2472 if (unlikely(dt_object_remote(parent) != 0))
2473 GOTO(put, rc = -EXDEV);
2475 if (dt_object_exists(parent) == 0) {
2476 lfsck_object_put(env, parent);
2477 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2482 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2483 GOTO(put, rc = -EISDIR);
2485 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2486 ltd->ltd_index, ea_off);
2492 lfsck_object_put(env, parent);
2494 /* The layout EA is changed, need to be reloaded next time. */
2495 lu_object_put_nocache(env, &parent->do_lu);
2498 down_write(&com->lc_sem);
2499 com->lc_new_scanned++;
2500 com->lc_new_checked++;
2502 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2504 } else if (rc < 0) {
2505 lo->ll_objs_failed_phase2++;
2507 up_write(&com->lc_sem);
2512 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2513 struct lfsck_component *com,
2514 struct lfsck_tgt_desc *ltd)
2516 struct lfsck_assistant_data *lad = com->lc_data;
2517 struct lfsck_instance *lfsck = com->lc_lfsck;
2518 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2519 struct lfsck_thread_info *info = lfsck_env_info(env);
2520 struct ost_id *oi = &info->lti_oi;
2521 struct lu_fid *fid = &info->lti_fid;
2522 struct dt_object *obj;
2523 const struct dt_it_ops *iops;
2528 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
2529 "scanning for OST%04x\n",
2530 lfsck_lfsck2name(lfsck), ltd->ltd_index);
2532 if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
2533 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
2534 "scanning for OST%04x\n",
2535 lfsck_lfsck2name(lfsck), ltd->ltd_index);
2540 ostid_set_seq(oi, FID_SEQ_IDIF);
2541 ostid_set_id(oi, 0);
2542 rc = ostid_to_fid(fid, oi, ltd->ltd_index);
2546 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2547 if (unlikely(IS_ERR(obj)))
2548 GOTO(log, rc = PTR_ERR(obj));
2550 rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_orphan_features);
2554 iops = &obj->do_index_ops->dio_it;
2555 di = iops->init(env, obj, 0);
2557 GOTO(put, rc = PTR_ERR(di));
2559 rc = iops->load(env, di, 0);
2561 /* -ESRCH means that the orphan OST-objects rbtree has been
2562 * cleanup because of the OSS server restart or other errors. */
2563 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
2568 rc = iops->next(env, di);
2580 struct lu_orphan_rec *rec = &info->lti_rec;
2582 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
2583 unlikely(!thread_is_running(&lfsck->li_thread)))
2586 key = iops->key(env, di);
2587 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2588 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2590 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2591 &com->lc_fid_latest_scanned_phase2);
2592 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2595 lfsck_control_speed_by_self(com);
2597 rc = iops->next(env, di);
2598 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2605 iops->fini(env, di);
2607 lfsck_object_put(env, obj);
2610 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
2611 "scanning for OST%04x: rc = %d\n",
2612 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2614 return rc > 0 ? 0 : rc;
2617 /* For the MDT-object with dangling reference, we need to repare the
2618 * inconsistency according to the LFSCK sponsor's requirement:
2620 * 1) Keep the inconsistency there and report the inconsistency case,
2621 * then give the chance to the application to find related issues,
2622 * and the users can make the decision about how to handle it with
2623 * more human knownledge. (by default)
2625 * 2) Re-create the missing OST-object with the FID/owner information. */
2626 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2627 struct lfsck_component *com,
2628 struct dt_object *parent,
2629 struct lfsck_layout_req *llr,
2632 struct lfsck_thread_info *info = lfsck_env_info(env);
2633 struct filter_fid *pfid = &info->lti_new_pfid;
2634 struct dt_object_format *dof = &info->lti_dof;
2635 struct dt_object *child = llr->llr_child;
2636 struct dt_device *dev = lfsck_obj2dev(child);
2637 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
2638 struct thandle *handle;
2640 struct lustre_handle lh = { 0 };
2645 if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ)
2653 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2654 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2659 rc = dt_attr_get(env, parent, la);
2663 la->la_mode = S_IFREG | 0666;
2664 la->la_atime = la->la_mtime = la->la_ctime = 0;
2665 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2666 LA_ATIME | LA_MTIME | LA_CTIME;
2667 memset(dof, 0, sizeof(*dof));
2668 pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2669 pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2670 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2671 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2672 * parent MDT-object's layout EA. */
2673 pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2674 buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2676 handle = dt_trans_create(env, dev);
2678 GOTO(unlock1, rc = PTR_ERR(handle));
2680 rc = dt_declare_create(env, child, la, NULL, dof, handle);
2684 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2685 LU_XATTR_CREATE, handle);
2689 rc = dt_trans_start_local(env, dev, handle);
2693 dt_read_lock(env, parent, 0);
2694 if (unlikely(lfsck_is_dead_obj(parent)))
2695 GOTO(unlock2, rc = 1);
2697 rc = dt_create(env, child, la, NULL, dof, handle);
2701 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2707 dt_read_unlock(env, parent);
2710 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2713 lfsck_ibits_unlock(&lh, LCK_EX);
2717 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2718 "dangling reference for: parent "DFID", child "DFID
2719 ", OST-index %u, stripe-index %u, owner %u/%u. %s: "
2720 "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
2721 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2722 llr->llr_ost_idx, llr->llr_lov_idx,
2723 la->la_uid, la->la_gid,
2724 create ? "Create the lost OST-object as required" :
2725 "Keep the MDT-object there by default", rc);
2730 /* If the OST-object does not recognize the MDT-object as its parent, and
2731 * there is no other MDT-object claims as its parent, then just trust the
2732 * given MDT-object as its parent. So update the OST-object filter_fid. */
2733 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2734 struct lfsck_component *com,
2735 struct dt_object *parent,
2736 struct lfsck_layout_req *llr,
2739 struct lfsck_thread_info *info = lfsck_env_info(env);
2740 struct filter_fid *pfid = &info->lti_new_pfid;
2741 struct dt_object *child = llr->llr_child;
2742 struct dt_device *dev = lfsck_obj2dev(child);
2743 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
2744 struct thandle *handle;
2746 struct lustre_handle lh = { 0 };
2750 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2751 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2756 pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
2757 pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
2758 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2759 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2760 * parent MDT-object's layout EA. */
2761 pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
2762 buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
2764 handle = dt_trans_create(env, dev);
2766 GOTO(unlock1, rc = PTR_ERR(handle));
2768 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2772 rc = dt_attr_get(env, parent, la);
2776 la->la_valid = LA_UID | LA_GID;
2777 rc = dt_declare_attr_set(env, child, la, handle);
2781 rc = dt_trans_start_local(env, dev, handle);
2785 dt_write_lock(env, parent, 0);
2786 if (unlikely(lfsck_is_dead_obj(parent)))
2787 GOTO(unlock2, rc = 1);
2789 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
2793 /* Get the latest parent's owner. */
2794 rc = dt_attr_get(env, parent, la);
2798 la->la_valid = LA_UID | LA_GID;
2799 rc = dt_attr_set(env, child, la, handle);
2804 dt_write_unlock(env, parent);
2807 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2810 lfsck_ibits_unlock(&lh, LCK_EX);
2814 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
2815 "unmatched MDT-OST pair for: parent "DFID
2816 ", child "DFID", OST-index %u, stripe-index %u, "
2817 "owner %u/%u: rc = %d\n",
2818 lfsck_lfsck2name(com->lc_lfsck),
2819 PFID(lfsck_dto2fid(parent)),
2820 PFID(lfsck_dto2fid(child)),
2821 llr->llr_ost_idx, llr->llr_lov_idx,
2822 la->la_uid, la->la_gid, rc);
2827 /* If there are more than one MDT-objects claim as the OST-object's parent,
2828 * and the OST-object only recognizes one of them, then we need to generate
2829 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
2830 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
2831 struct lfsck_component *com,
2832 struct dt_object *parent,
2833 struct lfsck_layout_req *llr,
2837 struct lfsck_thread_info *info = lfsck_env_info(env);
2838 struct dt_allocation_hint *hint = &info->lti_hint;
2839 struct dt_object_format *dof = &info->lti_dof;
2840 struct ost_id *oi = &info->lti_oi;
2841 struct lfsck_instance *lfsck = com->lc_lfsck;
2842 struct dt_device *dev;
2843 struct lu_device *d =
2844 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
2845 struct lu_object *o;
2846 struct lu_object *n;
2847 struct dt_object *child = NULL;
2848 struct thandle *handle = NULL;
2849 struct lov_mds_md_v1 *lmm;
2850 struct lov_ost_data_v1 *objs;
2851 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2853 struct lustre_handle lh = { 0 };
2854 struct lu_buf ea_buf;
2860 /* We use two separated transactions to repair the inconsistency.
2862 * 1) create the child (OST-object).
2863 * 2) update the parent LOV EA according to the child's FID.
2865 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
2866 * handled as orphan when the layout LFSCK run next time.
2868 * If 1) failed, but 2) succeed, then such OST-object will be re-created
2869 * as dangling referened case when the layout LFSCK run next time. */
2871 /* The 1st transaction. */
2872 o = lu_object_anon(env, d, NULL);
2874 GOTO(log, rc = PTR_ERR(o));
2876 n = lu_object_locate(o->lo_header, d->ld_type);
2877 if (unlikely(n == NULL)) {
2878 lu_object_put_nocache(env, o);
2880 GOTO(log, rc = -EINVAL);
2883 child = container_of(n, struct dt_object, do_lu);
2884 memset(hint, 0, sizeof(*hint));
2885 rc = dt_attr_get(env, parent, la);
2889 la->la_valid = LA_UID | LA_GID;
2890 memset(dof, 0, sizeof(*dof));
2892 dev = lfsck_obj2dev(child);
2893 handle = dt_trans_create(env, dev);
2895 GOTO(log, rc = PTR_ERR(handle));
2897 rc = dt_declare_create(env, child, la, hint, dof, handle);
2901 rc = dt_trans_start_local(env, dev, handle);
2905 rc = dt_create(env, child, la, hint, dof, handle);
2906 dt_trans_stop(env, dev, handle);
2911 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2912 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2917 /* The 2nd transaction. */
2919 /* XXX: Generally, we should use bottom device (OSD) to update parent
2920 * LOV EA. But because the LOD-object still references the wrong
2921 * OSP-object that should be detached after the parent's LOV EA
2922 * refreshed. Unfortunately, there is no suitable API for that.
2923 * So we have to make the LOD to re-load the OSP-object(s) via
2924 * replacing the LOV EA against the LOD-object.
2926 * Once the DNE2 patches have been landed, we can replace the
2927 * LOD device with the OSD device. LU-6230. */
2929 dev = lfsck->li_next;
2930 parent = lfsck_object_locate(dev, parent);
2932 GOTO(log, rc = PTR_ERR(parent));
2934 handle = dt_trans_create(env, dev);
2936 GOTO(log, rc = PTR_ERR(handle));
2938 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2939 LU_XATTR_REPLACE, handle);
2943 rc = dt_trans_start_local(env, dev, handle);
2947 dt_write_lock(env, parent, 0);
2948 if (unlikely(lfsck_is_dead_obj(parent)))
2949 GOTO(unlock, rc = 0);
2951 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2952 if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
2953 GOTO(unlock, rc = 0);
2956 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2957 * been verified in lfsck_layout_verify_header() already. If some
2958 * new magic introduced in the future, then layout LFSCK needs to
2959 * be updated also. */
2960 magic = le32_to_cpu(lmm->lmm_magic);
2961 if (magic == LOV_MAGIC_V1) {
2962 objs = &lmm->lmm_objects[llr->llr_lov_idx];
2964 LASSERT(magic == LOV_MAGIC_V3);
2966 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
2969 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2970 index = le32_to_cpu(objs->l_ost_idx);
2971 rc = ostid_to_fid(&tfid, oi, index);
2972 /* Someone changed layout during the LFSCK, no need to repair then. */
2973 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
2974 GOTO(unlock, rc = 0);
2976 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2977 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
2978 ostid_cpu_to_le(oi, &objs->l_ost_oi);
2979 objs->l_ost_gen = cpu_to_le32(0);
2980 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
2981 lfsck_buf_init(&ea_buf, lmm,
2982 lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
2984 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV,
2985 LU_XATTR_REPLACE, handle);
2987 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2990 dt_write_unlock(env, parent);
2994 dt_trans_stop(env, dev, handle);
2997 lfsck_ibits_unlock(&lh, LCK_EX);
2999 lfsck_object_put(env, child);
3002 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3003 "multiple references for: parent "DFID", OST-index %u, "
3004 "stripe-index %u, owner %u/%u: rc = %d\n",
3005 lfsck_lfsck2name(lfsck), PFID(pfid), llr->llr_ost_idx,
3006 llr->llr_lov_idx, la->la_uid, la->la_gid, rc);
3011 /* If the MDT-object and the OST-object have different owner information,
3012 * then trust the MDT-object, because the normal chown/chgrp handle order
3013 * is from MDT to OST, and it is possible that some chown/chgrp operation
3014 * is partly done. */
3015 static int lfsck_layout_repair_owner(const struct lu_env *env,
3016 struct lfsck_component *com,
3017 struct dt_object *parent,
3018 struct lfsck_layout_req *llr,
3019 struct lu_attr *pla,
3020 const struct lu_attr *cla)
3022 struct lfsck_thread_info *info = lfsck_env_info(env);
3023 struct lu_attr *tla = &info->lti_la2;
3024 struct dt_object *child = llr->llr_child;
3025 struct dt_device *dev = lfsck_obj2dev(child);
3026 struct thandle *handle;
3030 tla->la_uid = pla->la_uid;
3031 tla->la_gid = pla->la_gid;
3032 tla->la_valid = LA_UID | LA_GID;
3033 handle = dt_trans_create(env, dev);
3035 GOTO(log, rc = PTR_ERR(handle));
3037 rc = dt_declare_attr_set(env, child, tla, handle);
3041 rc = dt_trans_start_local(env, dev, handle);
3045 /* Use the dt_object lock to serialize with destroy and attr_set. */
3046 dt_read_lock(env, parent, 0);
3047 if (unlikely(lfsck_is_dead_obj(parent)))
3048 GOTO(unlock, rc = 1);
3050 /* Get the latest parent's owner. */
3051 rc = dt_attr_get(env, parent, pla);
3055 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3056 if (unlikely(tla->la_uid != pla->la_uid ||
3057 tla->la_gid != pla->la_gid))
3060 rc = dt_attr_set(env, child, tla, handle);
3065 dt_read_unlock(env, parent);
3068 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3072 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3073 "inconsistent file owner for: parent "DFID", child "DFID
3074 ", OST-index %u, stripe-index %u, old owner %u/%u, "
3075 "new owner %u/%u: rc = %d\n",
3076 lfsck_lfsck2name(com->lc_lfsck),
3077 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3078 llr->llr_ost_idx, llr->llr_lov_idx,
3079 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3084 /* Check whether the OST-object correctly back points to the
3085 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3086 static int lfsck_layout_check_parent(const struct lu_env *env,
3087 struct lfsck_component *com,
3088 struct lfsck_assistant_object *lso,
3089 const struct lu_fid *pfid,
3090 const struct lu_fid *cfid,
3091 const struct lu_attr *cla,
3092 struct lfsck_layout_req *llr,
3093 struct lu_buf *lov_ea, __u32 idx)
3095 struct lfsck_thread_info *info = lfsck_env_info(env);
3096 struct lu_buf *buf = &info->lti_big_buf;
3097 struct dt_object *tobj;
3098 struct lov_mds_md_v1 *lmm;
3099 struct lov_ost_data_v1 *objs;
3100 struct lustre_handle lh = { 0 };
3107 if (unlikely(!fid_is_sane(pfid)))
3108 RETURN(LLIT_UNMATCHED_PAIR);
3110 if (lu_fid_eq(pfid, &lso->lso_fid)) {
3111 if (likely(llr->llr_lov_idx == idx))
3114 RETURN(LLIT_UNMATCHED_PAIR);
3117 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3119 RETURN(PTR_ERR(tobj));
3121 if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3122 !S_ISREG(lfsck_object_type(tobj)))
3123 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3125 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3126 * remote one on another MDT. Then check whether the given OST-object
3127 * is in such layout. If yes, it is multiple referenced, otherwise it
3128 * is unmatched referenced case. */
3129 rc = lfsck_layout_get_lovea(env, tobj, buf);
3130 if (rc == 0 || rc == -ENOENT)
3131 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3137 magic = le32_to_cpu(lmm->lmm_magic);
3138 if (magic == LOV_MAGIC_V1) {
3139 objs = &lmm->lmm_objects[0];
3141 LASSERT(magic == LOV_MAGIC_V3);
3142 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3145 count = le16_to_cpu(lmm->lmm_stripe_count);
3146 for (i = 0; i < count; i++, objs++) {
3147 struct lu_fid *tfid = &info->lti_fid2;
3148 struct ost_id *oi = &info->lti_oi;
3151 if (lovea_slot_is_dummy(objs))
3154 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3155 idx2 = le32_to_cpu(objs->l_ost_idx);
3156 rc = ostid_to_fid(tfid, oi, idx2);
3158 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3159 "invalid layout EA at the slot %d, index %u\n",
3160 lfsck_lfsck2name(com->lc_lfsck),
3161 PFID(pfid), i, idx2);
3163 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3166 if (lu_fid_eq(cfid, tfid)) {
3167 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
3168 MDS_INODELOCK_UPDATE |
3169 MDS_INODELOCK_LAYOUT |
3170 MDS_INODELOCK_XATTR,
3175 dt_read_lock(env, tobj, 0);
3177 /* For local MDT-object, re-check existence
3178 * after taken the lock. */
3179 if (!dt_object_remote(tobj)) {
3180 if (dt_object_exists(tobj) == 0 ||
3181 lfsck_is_dead_obj(tobj)) {
3182 rc = LLIT_UNMATCHED_PAIR;
3185 rc = LLIT_MULTIPLE_REFERENCED;
3191 /* For migration case, the new MDT-object and old
3192 * MDT-object may reference the same OST-object at
3193 * some migration internal time.
3195 * For remote MDT-object, the local MDT may not know
3196 * whether it has been removed or not. Try checking
3197 * for a non-existent xattr to check if this object
3198 * has been been removed or not. */
3199 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
3201 if (unlikely(rc == -ENOENT || rc >= 0)) {
3202 rc = LLIT_UNMATCHED_PAIR;
3203 } else if (rc == -ENODATA) {
3205 rc = LLIT_MULTIPLE_REFERENCED;
3212 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3215 if (lustre_handle_is_used(&lh)) {
3216 dt_read_unlock(env, tobj);
3217 lfsck_ibits_unlock(&lh, LCK_EX);
3221 lfsck_object_put(env, tobj);
3226 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
3227 struct lfsck_component *com,
3228 struct lfsck_assistant_req *lar)
3230 struct lfsck_layout_req *llr =
3231 container_of0(lar, struct lfsck_layout_req, llr_lar);
3232 struct lfsck_assistant_object *lso = lar->lar_parent;
3233 struct lfsck_layout *lo = com->lc_file_ram;
3234 struct lfsck_thread_info *info = lfsck_env_info(env);
3235 struct filter_fid_old *pea = &info->lti_old_pfid;
3236 struct lu_fid *pfid = &info->lti_fid;
3237 struct lu_buf buf = { NULL };
3238 struct dt_object *parent = NULL;
3239 struct dt_object *child = llr->llr_child;
3240 struct lu_attr *pla = &lso->lso_attr;
3241 struct lu_attr *cla = &info->lti_la;
3242 struct lfsck_instance *lfsck = com->lc_lfsck;
3243 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3244 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
3252 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ASSISTANT_DIRECT, cfs_fail_val);
3254 rc = dt_attr_get(env, child, cla);
3255 if (rc == -ENOENT) {
3256 parent = lfsck_assistant_object_load(env, lfsck, lso);
3257 if (IS_ERR(parent)) {
3258 rc = PTR_ERR(parent);
3260 RETURN(rc == -ENOENT ? 0 : rc);
3263 type = LLIT_DANGLING;
3270 lfsck_buf_init(&buf, pea, sizeof(struct filter_fid_old));
3271 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
3272 if (unlikely(rc > 0 && rc != sizeof(struct filter_fid_old) &&
3273 rc != sizeof(struct filter_fid))) {
3274 type = LLIT_UNMATCHED_PAIR;
3278 if (rc < 0 && rc != -ENODATA)
3281 if (rc == 0 || rc == -ENODATA)
3282 GOTO(check_owner, rc = 0);
3284 fid_le_to_cpu(pfid, &pea->ff_parent);
3285 /* Currently, the filter_fid::ff_parent::f_ver is not the
3286 * real parent MDT-object's FID::f_ver, instead it is the
3287 * OST-object index in its parent MDT-object's layout EA. */
3288 idx = pfid->f_stripe_idx;
3290 rc = lfsck_layout_check_parent(env, com, lso, pfid,
3291 lu_object_fid(&child->do_lu),
3292 cla, llr, &buf, idx);
3302 /* Someone may has changed the owner after the parent attr pre-loaded.
3303 * It can be handled later inside the lfsck_layout_repair_owner(). */
3304 if (unlikely(cla->la_uid != pla->la_uid ||
3305 cla->la_gid != pla->la_gid)) {
3306 type = LLIT_INCONSISTENT_OWNER;
3311 if (type == LLIT_NONE)
3314 if (bk->lb_param & LPF_DRYRUN)
3317 if (parent == NULL) {
3318 parent = lfsck_assistant_object_load(env, lfsck, lso);
3319 if (IS_ERR(parent)) {
3320 rc = PTR_ERR(parent);
3331 rc = lfsck_layout_repair_dangling(env, com, parent, llr, pla);
3333 case LLIT_UNMATCHED_PAIR:
3334 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
3337 case LLIT_MULTIPLE_REFERENCED:
3338 rc = lfsck_layout_repair_multiple_references(env, com, parent,
3341 case LLIT_INCONSISTENT_OWNER:
3342 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
3352 down_write(&com->lc_sem);
3354 struct lfsck_assistant_data *lad = com->lc_data;
3356 if (unlikely(lad->lad_exit)) {
3358 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
3359 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
3360 rc == -EHOSTUNREACH) {
3361 /* If cannot touch the target server,
3362 * mark the LFSCK as INCOMPLETE. */
3363 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
3364 "talk with OST %x: rc = %d\n",
3365 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
3366 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
3367 lo->ll_objs_skipped++;
3370 lfsck_layout_record_failure(env, lfsck, lo);
3372 } else if (rc > 0) {
3373 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
3374 "unknown type = %d\n", type);
3376 lo->ll_objs_repaired[type - 1]++;
3377 if (bk->lb_param & LPF_DRYRUN &&
3378 unlikely(lo->ll_pos_first_inconsistent == 0))
3379 lo->ll_pos_first_inconsistent =
3380 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
3383 up_write(&com->lc_sem);
3385 if (parent != NULL && !IS_ERR(parent))
3386 lfsck_object_put(env, parent);
3391 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
3392 struct lfsck_component *com)
3394 struct lfsck_assistant_data *lad = com->lc_data;
3395 struct lfsck_instance *lfsck = com->lc_lfsck;
3396 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3397 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
3398 struct lfsck_tgt_desc *ltd;
3402 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
3403 lfsck_lfsck2name(lfsck));
3405 spin_lock(<ds->ltd_lock);
3406 while (!list_empty(&lad->lad_ost_phase2_list)) {
3407 ltd = list_entry(lad->lad_ost_phase2_list.next,
3408 struct lfsck_tgt_desc,
3409 ltd_layout_phase_list);
3410 list_del_init(<d->ltd_layout_phase_list);
3411 if (bk->lb_param & LPF_ALL_TGT) {
3412 spin_unlock(<ds->ltd_lock);
3413 rc = lfsck_layout_scan_orphan(env, com, ltd);
3414 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3417 if (unlikely(lad->lad_exit ||
3418 !thread_is_running(&lfsck->li_thread)))
3420 spin_lock(<ds->ltd_lock);
3424 if (list_empty(&lad->lad_ost_phase1_list))
3428 spin_unlock(<ds->ltd_lock);
3430 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
3431 lfsck_lfsck2name(lfsck), rc);
3437 lfsck_layout_slave_async_interpret(const struct lu_env *env,
3438 struct ptlrpc_request *req,
3441 struct lfsck_layout_slave_async_args *llsaa = args;
3442 struct obd_export *exp = llsaa->llsaa_exp;
3443 struct lfsck_component *com = llsaa->llsaa_com;
3444 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
3445 struct lfsck_layout_slave_data *llsd = com->lc_data;
3446 struct lfsck_reply *lr = NULL;
3450 /* It is probably caused by network trouble, or target crash,
3451 * it will try several times (depends on the obd_timeout, and
3452 * will not less than 3 times). But to make the LFSCK can go
3453 * ahead, we should not try for ever. After some try but still
3454 * hit failure, it will assume that the target exit the LFSCK
3455 * prcoessing and stop try. */
3456 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
3457 int max_try = max_t(int, obd_timeout / 30, 3);
3459 if (++(llst->llst_failures) > max_try)
3465 llst->llst_failures = 0;
3466 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
3467 if (lr->lr_status != LS_SCANNING_PHASE1 &&
3468 lr->lr_status != LS_SCANNING_PHASE2)
3473 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
3474 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
3475 llst->llst_index, lr != NULL ? lr->lr_status : rc,
3476 llst->llst_failures);
3478 lfsck_layout_llst_del(llsd, llst);
3481 lfsck_layout_llst_put(llst);
3482 lfsck_component_put(env, com);
3483 class_export_put(exp);
3488 static int lfsck_layout_async_query(const struct lu_env *env,
3489 struct lfsck_component *com,
3490 struct obd_export *exp,
3491 struct lfsck_layout_slave_target *llst,
3492 struct lfsck_request *lr,
3493 struct ptlrpc_request_set *set)
3495 struct lfsck_layout_slave_async_args *llsaa;
3496 struct ptlrpc_request *req;
3497 struct lfsck_request *tmp;
3501 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
3505 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
3507 ptlrpc_request_free(req);
3511 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3513 ptlrpc_request_set_replen(req);
3515 llsaa = ptlrpc_req_async_args(req);
3516 llsaa->llsaa_exp = exp;
3517 llsaa->llsaa_com = lfsck_component_get(com);
3518 llsaa->llsaa_llst = llst;
3519 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
3520 req->rq_allow_intr = 1;
3521 ptlrpc_set_add_req(set, req);
3526 static int lfsck_layout_async_notify(const struct lu_env *env,
3527 struct obd_export *exp,
3528 struct lfsck_request *lr,
3529 struct ptlrpc_request_set *set)
3531 struct ptlrpc_request *req;
3532 struct lfsck_request *tmp;
3536 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3540 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3542 ptlrpc_request_free(req);
3546 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3548 ptlrpc_request_set_replen(req);
3549 req->rq_allow_intr = 1;
3550 ptlrpc_set_add_req(set, req);
3556 lfsck_layout_slave_query_master(const struct lu_env *env,
3557 struct lfsck_component *com)
3559 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
3560 struct lfsck_instance *lfsck = com->lc_lfsck;
3561 struct lfsck_layout_slave_data *llsd = com->lc_data;
3562 struct lfsck_layout_slave_target *llst;
3563 struct obd_export *exp;
3564 struct ptlrpc_request_set *set;
3569 set = ptlrpc_prep_set();
3571 GOTO(log, rc = -ENOMEM);
3573 memset(lr, 0, sizeof(*lr));
3574 lr->lr_event = LE_QUERY;
3575 lr->lr_active = LFSCK_TYPE_LAYOUT;
3577 llsd->llsd_touch_gen++;
3578 spin_lock(&llsd->llsd_lock);
3579 while (!list_empty(&llsd->llsd_master_list)) {
3580 llst = list_entry(llsd->llsd_master_list.next,
3581 struct lfsck_layout_slave_target,
3583 if (llst->llst_gen == llsd->llsd_touch_gen)
3586 llst->llst_gen = llsd->llsd_touch_gen;
3587 list_move_tail(&llst->llst_list,
3588 &llsd->llsd_master_list);
3589 atomic_inc(&llst->llst_ref);
3590 spin_unlock(&llsd->llsd_lock);
3592 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3595 lfsck_layout_llst_del(llsd, llst);
3596 lfsck_layout_llst_put(llst);
3597 spin_lock(&llsd->llsd_lock);
3601 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
3603 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3604 "query %s for layout: rc = %d\n",
3605 lfsck_lfsck2name(lfsck),
3606 exp->exp_obd->obd_name, rc);
3609 lfsck_layout_llst_put(llst);
3610 class_export_put(exp);
3612 spin_lock(&llsd->llsd_lock);
3614 spin_unlock(&llsd->llsd_lock);
3616 rc = ptlrpc_set_wait(set);
3617 ptlrpc_set_destroy(set);
3619 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
3622 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
3623 lfsck_lfsck2name(com->lc_lfsck), rc);
3629 lfsck_layout_slave_notify_master(const struct lu_env *env,
3630 struct lfsck_component *com,
3631 enum lfsck_events event, int result)
3633 struct lfsck_layout *lo = com->lc_file_ram;
3634 struct lfsck_instance *lfsck = com->lc_lfsck;
3635 struct lfsck_layout_slave_data *llsd = com->lc_data;
3636 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
3637 struct lfsck_layout_slave_target *llst;
3638 struct obd_export *exp;
3639 struct ptlrpc_request_set *set;
3643 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
3644 lfsck_lfsck2name(com->lc_lfsck));
3646 set = ptlrpc_prep_set();
3650 memset(lr, 0, sizeof(*lr));
3651 lr->lr_event = event;
3652 lr->lr_flags = LEF_FROM_OST;
3653 lr->lr_status = result;
3654 lr->lr_index = lfsck_dev_idx(lfsck);
3655 lr->lr_active = LFSCK_TYPE_LAYOUT;
3656 lr->lr_flags2 = lo->ll_flags;
3657 llsd->llsd_touch_gen++;
3658 spin_lock(&llsd->llsd_lock);
3659 while (!list_empty(&llsd->llsd_master_list)) {
3660 llst = list_entry(llsd->llsd_master_list.next,
3661 struct lfsck_layout_slave_target,
3663 if (llst->llst_gen == llsd->llsd_touch_gen)
3666 llst->llst_gen = llsd->llsd_touch_gen;
3667 list_move_tail(&llst->llst_list,
3668 &llsd->llsd_master_list);
3669 atomic_inc(&llst->llst_ref);
3670 spin_unlock(&llsd->llsd_lock);
3672 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
3675 lfsck_layout_llst_del(llsd, llst);
3676 lfsck_layout_llst_put(llst);
3677 spin_lock(&llsd->llsd_lock);
3681 rc = lfsck_layout_async_notify(env, exp, lr, set);
3683 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
3684 "notify %s for layout: rc = %d\n",
3685 lfsck_lfsck2name(lfsck),
3686 exp->exp_obd->obd_name, rc);
3688 lfsck_layout_llst_put(llst);
3689 class_export_put(exp);
3690 spin_lock(&llsd->llsd_lock);
3692 spin_unlock(&llsd->llsd_lock);
3694 ptlrpc_set_wait(set);
3695 ptlrpc_set_destroy(set);
3701 * \ret -ENODATA: unrecognized stripe
3702 * \ret = 0 : recognized stripe
3703 * \ret < 0 : other failures
3705 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
3706 struct lfsck_component *com,
3707 struct lu_fid *cfid,
3708 struct lu_fid *pfid)
3710 struct lfsck_thread_info *info = lfsck_env_info(env);
3711 struct lu_buf *buf = &info->lti_big_buf;
3712 struct ost_id *oi = &info->lti_oi;
3713 struct dt_object *obj;
3714 struct lov_mds_md_v1 *lmm;
3715 struct lov_ost_data_v1 *objs;
3716 __u32 idx = pfid->f_stripe_idx;
3724 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3726 RETURN(PTR_ERR(obj));
3728 dt_read_lock(env, obj, 0);
3729 if (unlikely(dt_object_exists(obj) == 0 ||
3730 lfsck_is_dead_obj(obj)))
3731 GOTO(unlock, rc = -ENOENT);
3733 if (!S_ISREG(lfsck_object_type(obj)))
3734 GOTO(unlock, rc = -ENODATA);
3736 rc = lfsck_layout_get_lovea(env, obj, buf);
3741 GOTO(unlock, rc = -ENODATA);
3744 rc = lfsck_layout_verify_header(lmm);
3748 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3749 * been verified in lfsck_layout_verify_header() already. If some
3750 * new magic introduced in the future, then layout LFSCK needs to
3751 * be updated also. */
3752 magic = le32_to_cpu(lmm->lmm_magic);
3753 if (magic == LOV_MAGIC_V1) {
3754 objs = &lmm->lmm_objects[0];
3756 LASSERT(magic == LOV_MAGIC_V3);
3757 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3760 fid_to_ostid(cfid, oi);
3761 count = le16_to_cpu(lmm->lmm_stripe_count);
3762 for (i = 0; i < count; i++, objs++) {
3765 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
3766 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
3767 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
3770 GOTO(unlock, rc = -ENODATA);
3773 dt_read_unlock(env, obj);
3774 lfsck_object_put(env, obj);
3780 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
3781 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
3782 * layout EA from MDT to OST. On one hand, the OST no need to understand
3783 * the layout EA structure; on the other hand, it may cause trouble when
3784 * transfer large layout EA from MDT to OST via normal OUT RPC.
3786 * \ret > 0: unrecognized stripe
3787 * \ret = 0: recognized stripe
3788 * \ret < 0: other failures
3790 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
3791 struct lfsck_component *com,
3792 struct lu_fid *cfid,
3793 struct lu_fid *pfid)
3795 struct lfsck_instance *lfsck = com->lc_lfsck;
3796 struct obd_device *obd = lfsck->li_obd;
3797 struct seq_server_site *ss = lfsck_dev_site(lfsck);
3798 struct obd_export *exp = NULL;
3799 struct ptlrpc_request *req = NULL;
3800 struct lfsck_request *lr;
3801 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
3805 if (unlikely(fid_is_idif(pfid)))
3808 fld_range_set_any(range);
3809 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
3811 RETURN(rc == -ENOENT ? 1 : rc);
3813 if (unlikely(!fld_range_is_mdt(range)))
3816 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
3817 if (unlikely(exp == NULL))
3820 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
3821 GOTO(out, rc = -EOPNOTSUPP);
3823 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3825 GOTO(out, rc = -ENOMEM);
3827 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3829 ptlrpc_request_free(req);
3834 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3835 memset(lr, 0, sizeof(*lr));
3836 lr->lr_event = LE_PAIRS_VERIFY;
3837 lr->lr_active = LFSCK_TYPE_LAYOUT;
3838 lr->lr_fid = *cfid; /* OST-object itself FID. */
3839 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
3841 ptlrpc_request_set_replen(req);
3842 rc = ptlrpc_queue_wait(req);
3843 ptlrpc_req_finished(req);
3845 if (rc == -ENOENT || rc == -ENODATA)
3852 class_export_put(exp);
3857 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
3858 struct lfsck_component *com,
3859 struct lfsck_request *lr)
3861 struct dt_object *obj;
3865 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lr->lr_fid);
3867 GOTO(log, rc = PTR_ERR(obj));
3869 dt_write_lock(env, obj, 0);
3870 if (unlikely(dt_object_exists(obj) == 0 ||
3871 lfsck_is_dead_obj(obj)))
3872 GOTO(unlock, rc = 0);
3874 rc = __lfsck_layout_update_pfid(env, obj, &lr->lr_fid2,
3880 dt_write_unlock(env, obj);
3881 lfsck_object_put(env, obj);
3884 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
3885 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
3886 PFID(&lr->lr_fid), PFID(&lr->lr_fid2), rc);
3893 static void lfsck_layout_slave_quit(const struct lu_env *env,
3894 struct lfsck_component *com);
3896 static int lfsck_layout_reset(const struct lu_env *env,
3897 struct lfsck_component *com, bool init)
3899 struct lfsck_layout *lo = com->lc_file_ram;
3902 down_write(&com->lc_sem);
3904 memset(lo, 0, com->lc_file_size);
3906 __u32 count = lo->ll_success_count;
3907 __u64 last_time = lo->ll_time_last_complete;
3909 memset(lo, 0, com->lc_file_size);
3910 lo->ll_success_count = count;
3911 lo->ll_time_last_complete = last_time;
3914 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
3915 lo->ll_status = LS_INIT;
3917 if (com->lc_lfsck->li_master) {
3918 struct lfsck_assistant_data *lad = com->lc_data;
3920 lad->lad_incomplete = 0;
3921 CFS_RESET_BITMAP(lad->lad_bitmap);
3924 rc = lfsck_layout_store(env, com);
3925 up_write(&com->lc_sem);
3927 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
3928 lfsck_lfsck2name(com->lc_lfsck), rc);
3933 static void lfsck_layout_fail(const struct lu_env *env,
3934 struct lfsck_component *com, bool new_checked)
3936 struct lfsck_layout *lo = com->lc_file_ram;
3938 down_write(&com->lc_sem);
3940 com->lc_new_checked++;
3941 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
3942 up_write(&com->lc_sem);
3945 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
3946 struct lfsck_component *com, bool init)
3948 struct lfsck_instance *lfsck = com->lc_lfsck;
3949 struct lfsck_layout *lo = com->lc_file_ram;
3953 rc = lfsck_checkpoint_generic(env, com);
3955 return rc > 0 ? 0 : rc;
3958 down_write(&com->lc_sem);
3960 lo->ll_pos_latest_start =
3961 lfsck->li_pos_checkpoint.lp_oit_cookie;
3963 lo->ll_pos_last_checkpoint =
3964 lfsck->li_pos_checkpoint.lp_oit_cookie;
3965 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3966 HALF_SEC - lfsck->li_time_last_checkpoint);
3967 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3968 lo->ll_objs_checked_phase1 += com->lc_new_checked;
3969 com->lc_new_checked = 0;
3972 rc = lfsck_layout_store(env, com);
3973 up_write(&com->lc_sem);
3975 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
3976 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
3977 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
3982 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
3983 struct lfsck_component *com, bool init)
3985 struct lfsck_instance *lfsck = com->lc_lfsck;
3986 struct lfsck_layout *lo = com->lc_file_ram;
3989 if (com->lc_new_checked == 0 && !init)
3992 down_write(&com->lc_sem);
3994 lo->ll_pos_latest_start =
3995 lfsck->li_pos_checkpoint.lp_oit_cookie;
3997 lo->ll_pos_last_checkpoint =
3998 lfsck->li_pos_checkpoint.lp_oit_cookie;
3999 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4000 HALF_SEC - lfsck->li_time_last_checkpoint);
4001 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4002 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4003 com->lc_new_checked = 0;
4006 rc = lfsck_layout_store(env, com);
4007 up_write(&com->lc_sem);
4009 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
4010 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
4011 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
4016 static int lfsck_layout_prep(const struct lu_env *env,
4017 struct lfsck_component *com,
4018 struct lfsck_start *start)
4020 struct lfsck_instance *lfsck = com->lc_lfsck;
4021 struct lfsck_layout *lo = com->lc_file_ram;
4022 struct lfsck_position *pos = &com->lc_pos_start;
4024 fid_zero(&pos->lp_dir_parent);
4025 pos->lp_dir_cookie = 0;
4026 if (lo->ll_status == LS_COMPLETED ||
4027 lo->ll_status == LS_PARTIAL ||
4028 /* To handle orphan, must scan from the beginning. */
4029 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
4032 rc = lfsck_layout_reset(env, com, false);
4034 rc = lfsck_set_param(env, lfsck, start, true);
4037 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
4038 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4044 down_write(&com->lc_sem);
4045 lo->ll_time_latest_start = cfs_time_current_sec();
4046 spin_lock(&lfsck->li_lock);
4047 if (lo->ll_flags & LF_SCANNED_ONCE) {
4048 if (!lfsck->li_drop_dryrun ||
4049 lo->ll_pos_first_inconsistent == 0) {
4050 lo->ll_status = LS_SCANNING_PHASE2;
4051 list_move_tail(&com->lc_link,
4052 &lfsck->li_list_double_scan);
4053 pos->lp_oit_cookie = 0;
4057 lo->ll_status = LS_SCANNING_PHASE1;
4058 lo->ll_run_time_phase1 = 0;
4059 lo->ll_run_time_phase2 = 0;
4060 lo->ll_objs_checked_phase1 = 0;
4061 lo->ll_objs_checked_phase2 = 0;
4062 lo->ll_objs_failed_phase1 = 0;
4063 lo->ll_objs_failed_phase2 = 0;
4064 for (i = 0; i < LLIT_MAX; i++)
4065 lo->ll_objs_repaired[i] = 0;
4067 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4068 fid_zero(&com->lc_fid_latest_scanned_phase2);
4071 lo->ll_status = LS_SCANNING_PHASE1;
4072 if (!lfsck->li_drop_dryrun ||
4073 lo->ll_pos_first_inconsistent == 0)
4074 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
4076 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4078 spin_unlock(&lfsck->li_lock);
4079 up_write(&com->lc_sem);
4084 static int lfsck_layout_slave_prep(const struct lu_env *env,
4085 struct lfsck_component *com,
4086 struct lfsck_start_param *lsp)
4088 struct lfsck_layout_slave_data *llsd = com->lc_data;
4089 struct lfsck_instance *lfsck = com->lc_lfsck;
4090 struct lfsck_layout *lo = com->lc_file_ram;
4091 struct lfsck_start *start = lsp->lsp_start;
4094 rc = lfsck_layout_prep(env, com, start);
4098 if (lo->ll_flags & LF_CRASHED_LASTID &&
4099 list_empty(&llsd->llsd_master_list)) {
4100 LASSERT(lfsck->li_out_notify != NULL);
4102 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4103 LE_LASTID_REBUILDING);
4106 if (!lsp->lsp_index_valid)
4109 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4110 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
4111 LASSERT(!llsd->llsd_rbtree_valid);
4113 write_lock(&llsd->llsd_rb_lock);
4114 rc = lfsck_rbtree_setup(env, com);
4115 write_unlock(&llsd->llsd_rb_lock);
4118 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
4119 "%llu]\n", lfsck_lfsck2name(lfsck),
4120 com->lc_pos_start.lp_oit_cookie);
4125 static int lfsck_layout_master_prep(const struct lu_env *env,
4126 struct lfsck_component *com,
4127 struct lfsck_start_param *lsp)
4132 rc = lfsck_layout_load_bitmap(env, com);
4134 rc = lfsck_layout_reset(env, com, false);
4136 rc = lfsck_set_param(env, com->lc_lfsck,
4137 lsp->lsp_start, true);
4143 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
4147 rc = lfsck_start_assistant(env, com, lsp);
4152 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
4153 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
4154 com->lc_pos_start.lp_oit_cookie);
4159 /* Pre-fetch the attribute for each stripe in the given layout EA. */
4160 static int lfsck_layout_scan_stripes(const struct lu_env *env,
4161 struct lfsck_component *com,
4162 struct dt_object *parent,
4163 struct lov_mds_md_v1 *lmm)
4165 struct lfsck_thread_info *info = lfsck_env_info(env);
4166 struct lfsck_instance *lfsck = com->lc_lfsck;
4167 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4168 struct lfsck_layout *lo = com->lc_file_ram;
4169 struct lfsck_assistant_data *lad = com->lc_data;
4170 struct lfsck_assistant_object *lso = NULL;
4171 struct lov_ost_data_v1 *objs;
4172 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4173 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4174 struct ptlrpc_thread *athread = &lad->lad_thread;
4175 struct l_wait_info lwi = { 0 };
4183 lfsck_buf_init(&buf, &info->lti_old_pfid,
4184 sizeof(struct filter_fid_old));
4185 count = le16_to_cpu(lmm->lmm_stripe_count);
4186 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4187 * been verified in lfsck_layout_verify_header() already. If some
4188 * new magic introduced in the future, then layout LFSCK needs to
4189 * be updated also. */
4190 magic = le32_to_cpu(lmm->lmm_magic);
4191 if (magic == LOV_MAGIC_V1) {
4192 objs = &lmm->lmm_objects[0];
4194 LASSERT(magic == LOV_MAGIC_V3);
4195 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4198 for (i = 0; i < count; i++, objs++) {
4199 struct lu_fid *fid = &info->lti_fid;
4200 struct ost_id *oi = &info->lti_oi;
4201 struct lfsck_layout_req *llr;
4202 struct lfsck_tgt_desc *tgt = NULL;
4203 struct dt_object *cobj = NULL;
4205 bool wakeup = false;
4207 if (unlikely(lovea_slot_is_dummy(objs)))
4210 l_wait_event(mthread->t_ctl_waitq,
4211 lad->lad_prefetched < bk->lb_async_windows ||
4212 !thread_is_running(mthread) ||
4213 thread_is_stopped(athread),
4216 if (unlikely(!thread_is_running(mthread)) ||
4217 thread_is_stopped(athread))
4220 if (unlikely(lfsck_is_dead_obj(parent)))
4223 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4224 index = le32_to_cpu(objs->l_ost_idx);
4225 rc = ostid_to_fid(fid, oi, index);
4227 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
4228 ": "DOSTID", idx:%u\n", lfsck_lfsck2name(lfsck),
4229 PFID(lfsck_dto2fid(parent)), POSTID(oi), index);
4233 tgt = lfsck_tgt_get(ltds, index);
4234 if (unlikely(tgt == NULL)) {
4235 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
4236 "did not join the layout LFSCK\n",
4237 lfsck_lfsck2name(lfsck), index);
4238 lfsck_lad_set_bitmap(env, com, index);
4242 /* There is potential deadlock race condition between object
4243 * destroy and layout LFSCK. Consider the following scenario:
4245 * 1) The LFSCK thread obtained the parent object firstly, at
4246 * that time, the parent object has not been destroyed yet.
4248 * 2) One RPC service thread destroyed the parent and all its
4249 * children objects. Because the LFSCK is referencing the
4250 * parent object, then the parent object will be marked as
4251 * dying in RAM. On the other hand, the parent object is
4252 * referencing all its children objects, then all children
4253 * objects will be marked as dying in RAM also.
4255 * 3) The LFSCK thread tries to find some child object with
4256 * the parent object referenced. Then it will find that the
4257 * child object is dying. According to the object visibility
4258 * rules: the object with dying flag cannot be returned to
4259 * others. So the LFSCK thread has to wait until the dying
4260 * object has been purged from RAM, then it can allocate a
4261 * new object (with the same FID) in RAM. Unfortunately, the
4262 * LFSCK thread itself is referencing the parent object, and
4263 * cause the parent object cannot be purged, then cause the
4264 * child object cannot be purged also. So the LFSCK thread
4265 * will fall into deadlock.
4267 * We introduce non-blocked version lu_object_find() to allow
4268 * the LFSCK thread to return failure immediately (instead of
4269 * wait) when it finds dying (child) object, then the LFSCK
4270 * thread can check whether the parent object is dying or not.
4271 * So avoid above deadlock. LU-5395 */
4272 cobj = lfsck_object_find_by_dev_nowait(env, tgt->ltd_tgt, fid);
4274 if (lfsck_is_dead_obj(parent)) {
4284 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_ASSISTANT_DIRECT)) {
4285 rc = dt_declare_attr_get(env, cobj);
4289 rc = dt_declare_xattr_get(env, cobj, &buf,
4296 struct lu_attr *attr = &info->lti_la;
4298 rc = dt_attr_get(env, parent, attr);
4302 lso = lfsck_assistant_object_init(env,
4303 lfsck_dto2fid(parent), attr,
4304 lfsck->li_pos_current.lp_oit_cookie, false);
4313 llr = lfsck_layout_assistant_req_init(lso, cobj, index, i);
4320 spin_lock(&lad->lad_lock);
4321 if (lad->lad_assistant_status < 0) {
4322 spin_unlock(&lad->lad_lock);
4323 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
4325 RETURN(lad->lad_assistant_status);
4328 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
4329 if (lad->lad_prefetched == 0)
4332 lad->lad_prefetched++;
4333 spin_unlock(&lad->lad_lock);
4335 wake_up_all(&athread->t_ctl_waitq);
4338 down_write(&com->lc_sem);
4339 com->lc_new_checked++;
4341 lfsck_layout_record_failure(env, lfsck, lo);
4342 up_write(&com->lc_sem);
4344 if (cobj != NULL && !IS_ERR(cobj))
4345 lfsck_object_put(env, cobj);
4347 if (likely(tgt != NULL))
4350 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4358 lfsck_assistant_object_put(env, lso);
4363 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
4364 * the OST-object's attribute and generate an structure lfsck_layout_req on the
4365 * list ::lad_req_list.
4367 * For each request on above list, the lfsck_layout_assistant thread compares
4368 * the OST side attribute with local attribute, if inconsistent, then repair it.
4370 * All above processing is async mode with pipeline. */
4371 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
4372 struct lfsck_component *com,
4373 struct dt_object *obj)
4375 struct lfsck_thread_info *info = lfsck_env_info(env);
4376 struct ost_id *oi = &info->lti_oi;
4377 struct lfsck_layout *lo = com->lc_file_ram;
4378 struct lfsck_assistant_data *lad = com->lc_data;
4379 struct lfsck_instance *lfsck = com->lc_lfsck;
4380 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4381 struct thandle *handle = NULL;
4382 struct lu_buf *buf = &info->lti_big_buf;
4383 struct lov_mds_md_v1 *lmm = NULL;
4384 struct dt_device *dev = lfsck_obj2dev(obj);
4385 struct lustre_handle lh = { 0 };
4386 struct lu_buf ea_buf = { NULL };
4389 bool locked = false;
4390 bool stripe = false;
4391 bool bad_oi = false;
4394 if (!S_ISREG(lfsck_object_type(obj)))
4397 if (lad->lad_assistant_status < 0)
4398 GOTO(out, rc = -ESRCH);
4400 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
4401 lmm_oi_cpu_to_le(oi, oi);
4402 dt_read_lock(env, obj, 0);
4406 if (dt_object_exists(obj) == 0 ||
4407 lfsck_is_dead_obj(obj))
4410 rc = lfsck_layout_get_lovea(env, obj, buf);
4416 rc = lfsck_layout_verify_header(lmm);
4417 /* If the LOV EA crashed, then it is possible to be rebuilt later
4418 * when handle orphan OST-objects. */
4422 if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
4423 GOTO(out, stripe = true);
4425 /* Inconsistent lmm_oi, should be repaired. */
4429 if (bk->lb_param & LPF_DRYRUN) {
4430 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4432 GOTO(out, stripe = true);
4435 if (!lustre_handle_is_used(&lh)) {
4436 dt_read_unlock(env, obj);
4438 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4439 MDS_INODELOCK_LAYOUT |
4440 MDS_INODELOCK_XATTR, LCK_EX);
4444 handle = dt_trans_create(env, dev);
4446 GOTO(out, rc = PTR_ERR(handle));
4448 lfsck_buf_init(&ea_buf, lmm, size);
4449 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4450 LU_XATTR_REPLACE, handle);
4454 rc = dt_trans_start_local(env, dev, handle);
4458 dt_write_lock(env, obj, 0);
4464 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4465 LU_XATTR_REPLACE, handle);
4469 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4471 GOTO(out, stripe = true);
4475 if (lustre_handle_is_used(&lh))
4476 dt_write_unlock(env, obj);
4478 dt_read_unlock(env, obj);
4481 if (handle != NULL && !IS_ERR(handle))
4482 dt_trans_stop(env, dev, handle);
4484 lfsck_ibits_unlock(&lh, LCK_EX);
4487 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
4488 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
4489 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
4490 PFID(lfsck_dto2fid(obj)), rc);
4493 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
4495 down_write(&com->lc_sem);
4496 com->lc_new_checked++;
4498 lfsck_layout_record_failure(env, lfsck, lo);
4499 up_write(&com->lc_sem);
4505 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
4506 struct lfsck_component *com,
4507 struct dt_object *obj)
4509 struct lfsck_instance *lfsck = com->lc_lfsck;
4510 struct lfsck_layout *lo = com->lc_file_ram;
4511 const struct lu_fid *fid = lfsck_dto2fid(obj);
4512 struct lfsck_layout_slave_data *llsd = com->lc_data;
4513 struct lfsck_layout_seq *lls;
4519 LASSERT(llsd != NULL);
4521 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
4522 cfs_fail_val == lfsck_dev_idx(lfsck)) {
4523 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(1),
4525 struct ptlrpc_thread *thread = &lfsck->li_thread;
4527 l_wait_event(thread->t_ctl_waitq,
4528 !thread_is_running(thread),
4532 lfsck_rbtree_update_bitmap(env, com, fid, false);
4534 down_write(&com->lc_sem);
4535 if (fid_is_idif(fid))
4537 else if (!fid_is_norm(fid) ||
4538 !fid_is_for_ostobj(env, lfsck, obj, fid))
4539 GOTO(unlock, rc = 0);
4542 com->lc_new_checked++;
4544 lls = lfsck_layout_seq_lookup(llsd, seq);
4547 if (unlikely(lls == NULL))
4548 GOTO(unlock, rc = -ENOMEM);
4550 INIT_LIST_HEAD(&lls->lls_list);
4552 rc = lfsck_layout_lastid_load(env, com, lls);
4554 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4555 "load LAST_ID for %#llx: rc = %d\n",
4556 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
4557 lo->ll_objs_failed_phase1++;
4562 lfsck_layout_seq_insert(llsd, lls);
4565 if (unlikely(fid_is_last_id(fid)))
4566 GOTO(unlock, rc = 0);
4568 if (fid_is_idif(fid))
4569 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
4573 if (oid > lls->lls_lastid_known)
4574 lls->lls_lastid_known = oid;
4576 if (oid > lls->lls_lastid) {
4577 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
4578 /* OFD may create new objects during LFSCK scanning. */
4579 rc = lfsck_layout_lastid_reload(env, com, lls);
4580 if (unlikely(rc != 0)) {
4581 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4582 "reload LAST_ID for %#llx: rc = %d\n",
4583 lfsck_lfsck2name(com->lc_lfsck),
4589 if (oid <= lls->lls_lastid ||
4590 lo->ll_flags & LF_CRASHED_LASTID)
4591 GOTO(unlock, rc = 0);
4593 LASSERT(lfsck->li_out_notify != NULL);
4595 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4596 LE_LASTID_REBUILDING);
4597 lo->ll_flags |= LF_CRASHED_LASTID;
4599 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
4600 "LAST_ID file (2) for the sequence %#llx"
4601 ", old value %llu, known value %llu\n",
4602 lfsck_lfsck2name(lfsck), lls->lls_seq,
4603 lls->lls_lastid, oid);
4606 lls->lls_lastid = oid;
4610 GOTO(unlock, rc = 0);
4613 up_write(&com->lc_sem);
4618 static int lfsck_layout_exec_dir(const struct lu_env *env,
4619 struct lfsck_component *com,
4620 struct lfsck_assistant_object *lso,
4621 struct lu_dirent *ent, __u16 type)
4626 static int lfsck_layout_master_post(const struct lu_env *env,
4627 struct lfsck_component *com,
4628 int result, bool init)
4630 struct lfsck_instance *lfsck = com->lc_lfsck;
4631 struct lfsck_layout *lo = com->lc_file_ram;
4635 lfsck_post_generic(env, com, &result);
4637 down_write(&com->lc_sem);
4638 spin_lock(&lfsck->li_lock);
4640 lo->ll_pos_last_checkpoint =
4641 lfsck->li_pos_checkpoint.lp_oit_cookie;
4644 if (lo->ll_flags & LF_INCOMPLETE)
4645 lo->ll_status = LS_PARTIAL;
4647 lo->ll_status = LS_SCANNING_PHASE2;
4648 lo->ll_flags |= LF_SCANNED_ONCE;
4649 lo->ll_flags &= ~LF_UPGRADE;
4650 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4651 } else if (result == 0) {
4652 if (lfsck->li_status != 0)
4653 lo->ll_status = lfsck->li_status;
4655 lo->ll_status = LS_STOPPED;
4656 if (lo->ll_status != LS_PAUSED)
4657 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4659 lo->ll_status = LS_FAILED;
4660 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4662 spin_unlock(&lfsck->li_lock);
4665 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4666 HALF_SEC - lfsck->li_time_last_checkpoint);
4667 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4668 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4669 com->lc_new_checked = 0;
4672 rc = lfsck_layout_store(env, com);
4673 up_write(&com->lc_sem);
4675 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
4676 lfsck_lfsck2name(lfsck), rc);
4681 static int lfsck_layout_slave_post(const struct lu_env *env,
4682 struct lfsck_component *com,
4683 int result, bool init)
4685 struct lfsck_instance *lfsck = com->lc_lfsck;
4686 struct lfsck_layout *lo = com->lc_file_ram;
4690 down_write(&com->lc_sem);
4691 rc = lfsck_layout_lastid_store(env, com);
4695 LASSERT(lfsck->li_out_notify != NULL);
4697 spin_lock(&lfsck->li_lock);
4699 lo->ll_pos_last_checkpoint =
4700 lfsck->li_pos_checkpoint.lp_oit_cookie;
4703 lo->ll_status = LS_SCANNING_PHASE2;
4704 lo->ll_flags |= LF_SCANNED_ONCE;
4705 if (lo->ll_flags & LF_CRASHED_LASTID) {
4707 lo->ll_flags &= ~LF_CRASHED_LASTID;
4709 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
4710 "crashed LAST_ID files successfully\n",
4711 lfsck_lfsck2name(lfsck));
4713 lo->ll_flags &= ~LF_UPGRADE;
4714 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4715 } else if (result == 0) {
4716 if (lfsck->li_status != 0)
4717 lo->ll_status = lfsck->li_status;
4719 lo->ll_status = LS_STOPPED;
4720 if (lo->ll_status != LS_PAUSED)
4721 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4723 lo->ll_status = LS_FAILED;
4724 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4726 spin_unlock(&lfsck->li_lock);
4729 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4733 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4734 HALF_SEC - lfsck->li_time_last_checkpoint);
4735 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4736 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4737 com->lc_new_checked = 0;
4740 rc = lfsck_layout_store(env, com);
4741 up_write(&com->lc_sem);
4743 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
4745 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
4746 lfsck_lfsck2name(lfsck), rc);
4751 static void lfsck_layout_dump(const struct lu_env *env,
4752 struct lfsck_component *com, struct seq_file *m)
4754 struct lfsck_instance *lfsck = com->lc_lfsck;
4755 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4756 struct lfsck_layout *lo = com->lc_file_ram;
4758 down_read(&com->lc_sem);
4759 seq_printf(m, "name: lfsck_layout\n"
4765 lfsck_status2name(lo->ll_status));
4767 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
4769 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4771 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
4773 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
4775 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
4777 seq_printf(m, "latest_start_position: %llu\n"
4778 "last_checkpoint_position: %llu\n"
4779 "first_failure_position: %llu\n",
4780 lo->ll_pos_latest_start,
4781 lo->ll_pos_last_checkpoint,
4782 lo->ll_pos_first_inconsistent);
4784 seq_printf(m, "success_count: %u\n"
4785 "repaired_dangling: %llu\n"
4786 "repaired_unmatched_pair: %llu\n"
4787 "repaired_multiple_referenced: %llu\n"
4788 "repaired_orphan: %llu\n"
4789 "repaired_inconsistent_owner: %llu\n"
4790 "repaired_others: %llu\n"
4792 "failed_phase1: %llu\n"
4793 "failed_phase2: %llu\n",
4794 lo->ll_success_count,
4795 lo->ll_objs_repaired[LLIT_DANGLING - 1],
4796 lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
4797 lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
4798 lo->ll_objs_repaired[LLIT_ORPHAN - 1],
4799 lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
4800 lo->ll_objs_repaired[LLIT_OTHERS - 1],
4801 lo->ll_objs_skipped,
4802 lo->ll_objs_failed_phase1,
4803 lo->ll_objs_failed_phase2);
4805 if (lo->ll_status == LS_SCANNING_PHASE1) {
4807 const struct dt_it_ops *iops;
4808 cfs_duration_t duration = cfs_time_current() -
4809 lfsck->li_time_last_checkpoint;
4810 __u64 checked = lo->ll_objs_checked_phase1 +
4811 com->lc_new_checked;
4812 __u64 speed = checked;
4813 __u64 new_checked = com->lc_new_checked *
4814 msecs_to_jiffies(MSEC_PER_SEC);
4815 __u32 rtime = lo->ll_run_time_phase1 +
4816 cfs_duration_sec(duration + HALF_SEC);
4819 do_div(new_checked, duration);
4821 do_div(speed, rtime);
4822 seq_printf(m, "checked_phase1: %llu\n"
4823 "checked_phase2: %llu\n"
4824 "run_time_phase1: %u seconds\n"
4825 "run_time_phase2: %u seconds\n"
4826 "average_speed_phase1: %llu items/sec\n"
4827 "average_speed_phase2: N/A\n"
4828 "real-time_speed_phase1: %llu items/sec\n"
4829 "real-time_speed_phase2: N/A\n",
4831 lo->ll_objs_checked_phase2,
4833 lo->ll_run_time_phase2,
4837 LASSERT(lfsck->li_di_oit != NULL);
4839 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4841 /* The low layer otable-based iteration position may NOT
4842 * exactly match the layout-based directory traversal
4843 * cookie. Generally, it is not a serious issue. But the
4844 * caller should NOT make assumption on that. */
4845 pos = iops->store(env, lfsck->li_di_oit);
4846 if (!lfsck->li_current_oit_processed)
4848 seq_printf(m, "current_position: %llu\n", pos);
4850 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
4851 cfs_duration_t duration = cfs_time_current() -
4852 com->lc_time_last_checkpoint;
4853 __u64 checked = lo->ll_objs_checked_phase2 +
4854 com->lc_new_checked;
4855 __u64 speed1 = lo->ll_objs_checked_phase1;
4856 __u64 speed2 = checked;
4857 __u64 new_checked = com->lc_new_checked *
4858 msecs_to_jiffies(MSEC_PER_SEC);
4859 __u32 rtime = lo->ll_run_time_phase2 +
4860 cfs_duration_sec(duration + HALF_SEC);
4863 do_div(new_checked, duration);
4864 if (lo->ll_run_time_phase1 != 0)
4865 do_div(speed1, lo->ll_run_time_phase1);
4867 do_div(speed2, rtime);
4868 seq_printf(m, "checked_phase1: %llu\n"
4869 "checked_phase2: %llu\n"
4870 "run_time_phase1: %u seconds\n"
4871 "run_time_phase2: %u seconds\n"
4872 "average_speed_phase1: %llu items/sec\n"
4873 "average_speed_phase2: %llu items/sec\n"
4874 "real-time_speed_phase1: N/A\n"
4875 "real-time_speed_phase2: %llu items/sec\n"
4876 "current_position: "DFID"\n",
4877 lo->ll_objs_checked_phase1,
4879 lo->ll_run_time_phase1,
4884 PFID(&com->lc_fid_latest_scanned_phase2));
4886 __u64 speed1 = lo->ll_objs_checked_phase1;
4887 __u64 speed2 = lo->ll_objs_checked_phase2;
4889 if (lo->ll_run_time_phase1 != 0)
4890 do_div(speed1, lo->ll_run_time_phase1);
4891 if (lo->ll_run_time_phase2 != 0)
4892 do_div(speed2, lo->ll_run_time_phase2);
4893 seq_printf(m, "checked_phase1: %llu\n"
4894 "checked_phase2: %llu\n"
4895 "run_time_phase1: %u seconds\n"
4896 "run_time_phase2: %u seconds\n"
4897 "average_speed_phase1: %llu items/sec\n"
4898 "average_speed_phase2: %llu objs/sec\n"
4899 "real-time_speed_phase1: N/A\n"
4900 "real-time_speed_phase2: N/A\n"
4901 "current_position: N/A\n",
4902 lo->ll_objs_checked_phase1,
4903 lo->ll_objs_checked_phase2,
4904 lo->ll_run_time_phase1,
4905 lo->ll_run_time_phase2,
4910 up_read(&com->lc_sem);
4913 static int lfsck_layout_master_double_scan(const struct lu_env *env,
4914 struct lfsck_component *com)
4916 struct lfsck_layout *lo = com->lc_file_ram;
4917 struct lfsck_assistant_data *lad = com->lc_data;
4918 struct lfsck_instance *lfsck = com->lc_lfsck;
4919 struct lfsck_tgt_descs *ltds;
4920 struct lfsck_tgt_desc *ltd;
4921 struct lfsck_tgt_desc *next;
4924 rc = lfsck_double_scan_generic(env, com, lo->ll_status);
4926 if (thread_is_stopped(&lad->lad_thread)) {
4927 LASSERT(list_empty(&lad->lad_req_list));
4928 LASSERT(list_empty(&lad->lad_ost_phase1_list));
4929 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4931 ltds = &lfsck->li_ost_descs;
4932 spin_lock(<ds->ltd_lock);
4933 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
4934 ltd_layout_phase_list) {
4935 list_del_init(<d->ltd_layout_phase_list);
4937 spin_unlock(<ds->ltd_lock);
4939 ltds = &lfsck->li_mdt_descs;
4940 spin_lock(<ds->ltd_lock);
4941 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4942 ltd_layout_phase_list) {
4943 list_del_init(<d->ltd_layout_phase_list);
4945 spin_unlock(<ds->ltd_lock);
4951 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
4952 struct lfsck_component *com)
4954 struct lfsck_instance *lfsck = com->lc_lfsck;
4955 struct lfsck_layout_slave_data *llsd = com->lc_data;
4956 struct lfsck_layout *lo = com->lc_file_ram;
4957 struct ptlrpc_thread *thread = &lfsck->li_thread;
4961 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
4962 lfsck_lfsck2name(lfsck));
4964 atomic_inc(&lfsck->li_double_scan_count);
4966 if (lo->ll_flags & LF_INCOMPLETE)
4969 com->lc_new_checked = 0;
4970 com->lc_new_scanned = 0;
4971 com->lc_time_last_checkpoint = cfs_time_current();
4972 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4973 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
4976 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
4979 rc = lfsck_layout_slave_query_master(env, com);
4980 if (list_empty(&llsd->llsd_master_list)) {
4981 if (unlikely(!thread_is_running(thread)))
4992 rc = l_wait_event(thread->t_ctl_waitq,
4993 !thread_is_running(thread) ||
4994 lo->ll_flags & LF_INCOMPLETE ||
4995 list_empty(&llsd->llsd_master_list),
4997 if (unlikely(!thread_is_running(thread)))
5000 if (lo->ll_flags & LF_INCOMPLETE)
5003 if (rc == -ETIMEDOUT)
5006 GOTO(done, rc = (rc < 0 ? rc : 1));
5010 rc = lfsck_layout_double_scan_result(env, com, rc);
5011 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
5012 (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
5013 lfsck_layout_slave_quit(env, com);
5014 if (atomic_dec_and_test(&lfsck->li_double_scan_count))
5015 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5017 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
5018 "status %d: rc = %d\n",
5019 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
5024 static void lfsck_layout_master_data_release(const struct lu_env *env,
5025 struct lfsck_component *com)
5027 struct lfsck_assistant_data *lad = com->lc_data;
5028 struct lfsck_instance *lfsck = com->lc_lfsck;
5029 struct lfsck_tgt_descs *ltds;
5030 struct lfsck_tgt_desc *ltd;
5031 struct lfsck_tgt_desc *next;
5033 LASSERT(lad != NULL);
5034 LASSERT(thread_is_init(&lad->lad_thread) ||
5035 thread_is_stopped(&lad->lad_thread));
5036 LASSERT(list_empty(&lad->lad_req_list));
5038 com->lc_data = NULL;
5040 ltds = &lfsck->li_ost_descs;
5041 spin_lock(<ds->ltd_lock);
5042 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5043 ltd_layout_phase_list) {
5044 list_del_init(<d->ltd_layout_phase_list);
5046 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5047 ltd_layout_phase_list) {
5048 list_del_init(<d->ltd_layout_phase_list);
5050 list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
5052 list_del_init(<d->ltd_layout_list);
5054 spin_unlock(<ds->ltd_lock);
5056 ltds = &lfsck->li_mdt_descs;
5057 spin_lock(<ds->ltd_lock);
5058 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5059 ltd_layout_phase_list) {
5060 list_del_init(<d->ltd_layout_phase_list);
5062 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5063 ltd_layout_phase_list) {
5064 list_del_init(<d->ltd_layout_phase_list);
5066 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
5068 list_del_init(<d->ltd_layout_list);
5070 spin_unlock(<ds->ltd_lock);
5072 if (likely(lad->lad_bitmap != NULL))
5073 CFS_FREE_BITMAP(lad->lad_bitmap);
5078 static void lfsck_layout_slave_data_release(const struct lu_env *env,
5079 struct lfsck_component *com)
5081 struct lfsck_layout_slave_data *llsd = com->lc_data;
5083 lfsck_layout_slave_quit(env, com);
5084 com->lc_data = NULL;
5088 static void lfsck_layout_master_quit(const struct lu_env *env,
5089 struct lfsck_component *com)
5091 struct lfsck_assistant_data *lad = com->lc_data;
5092 struct lfsck_instance *lfsck = com->lc_lfsck;
5093 struct lfsck_tgt_descs *ltds;
5094 struct lfsck_tgt_desc *ltd;
5095 struct lfsck_tgt_desc *next;
5097 LASSERT(lad != NULL);
5099 lfsck_quit_generic(env, com);
5101 LASSERT(thread_is_init(&lad->lad_thread) ||
5102 thread_is_stopped(&lad->lad_thread));
5103 LASSERT(list_empty(&lad->lad_req_list));
5105 ltds = &lfsck->li_ost_descs;
5106 spin_lock(<ds->ltd_lock);
5107 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5108 ltd_layout_phase_list) {
5109 list_del_init(<d->ltd_layout_phase_list);
5111 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5112 ltd_layout_phase_list) {
5113 list_del_init(<d->ltd_layout_phase_list);
5115 spin_unlock(<ds->ltd_lock);
5117 ltds = &lfsck->li_mdt_descs;
5118 spin_lock(<ds->ltd_lock);
5119 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5120 ltd_layout_phase_list) {
5121 list_del_init(<d->ltd_layout_phase_list);
5123 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5124 ltd_layout_phase_list) {
5125 list_del_init(<d->ltd_layout_phase_list);
5127 spin_unlock(<ds->ltd_lock);
5130 static void lfsck_layout_slave_quit(const struct lu_env *env,
5131 struct lfsck_component *com)
5133 struct lfsck_layout_slave_data *llsd = com->lc_data;
5134 struct lfsck_layout_seq *lls;
5135 struct lfsck_layout_seq *next;
5136 struct lfsck_layout_slave_target *llst;
5138 LASSERT(llsd != NULL);
5140 down_write(&com->lc_sem);
5141 list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
5143 list_del_init(&lls->lls_list);
5144 lfsck_object_put(env, lls->lls_lastid_obj);
5147 up_write(&com->lc_sem);
5149 spin_lock(&llsd->llsd_lock);
5150 while (!list_empty(&llsd->llsd_master_list)) {
5151 llst = list_entry(llsd->llsd_master_list.next,
5152 struct lfsck_layout_slave_target, llst_list);
5153 list_del_init(&llst->llst_list);
5154 spin_unlock(&llsd->llsd_lock);
5155 lfsck_layout_llst_put(llst);
5156 spin_lock(&llsd->llsd_lock);
5158 spin_unlock(&llsd->llsd_lock);
5160 lfsck_rbtree_cleanup(env, com);
5163 static int lfsck_layout_master_in_notify(const struct lu_env *env,
5164 struct lfsck_component *com,
5165 struct lfsck_request *lr,
5168 struct lfsck_instance *lfsck = com->lc_lfsck;
5169 struct lfsck_layout *lo = com->lc_file_ram;
5170 struct lfsck_assistant_data *lad = com->lc_data;
5171 struct lfsck_tgt_descs *ltds;
5172 struct lfsck_tgt_desc *ltd;
5176 if (lr->lr_event == LE_PAIRS_VERIFY) {
5179 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
5185 CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
5186 "from %s %x, status %d, flags %x, flags2 %x\n",
5187 lfsck_lfsck2name(lfsck), lr->lr_event,
5188 (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
5189 lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
5191 if (lr->lr_event != LE_PHASE1_DONE &&
5192 lr->lr_event != LE_PHASE2_DONE &&
5193 lr->lr_event != LE_PEER_EXIT)
5196 if (lr->lr_flags & LEF_FROM_OST)
5197 ltds = &lfsck->li_ost_descs;
5199 ltds = &lfsck->li_mdt_descs;
5200 spin_lock(<ds->ltd_lock);
5201 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5203 spin_unlock(<ds->ltd_lock);
5208 list_del_init(<d->ltd_layout_phase_list);
5209 switch (lr->lr_event) {
5210 case LE_PHASE1_DONE:
5211 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
5212 if (lr->lr_flags2 & LF_INCOMPLETE) {
5213 if (lr->lr_flags & LEF_FROM_OST)
5214 lfsck_lad_set_bitmap(env, com,
5217 lo->ll_flags |= LF_INCOMPLETE;
5219 ltd->ltd_layout_done = 1;
5220 list_del_init(<d->ltd_layout_list);
5225 if (lr->lr_flags & LEF_FROM_OST) {
5226 if (list_empty(<d->ltd_layout_list))
5227 list_add_tail(<d->ltd_layout_list,
5228 &lad->lad_ost_list);
5229 list_add_tail(<d->ltd_layout_phase_list,
5230 &lad->lad_ost_phase2_list);
5232 if (list_empty(<d->ltd_layout_list))
5233 list_add_tail(<d->ltd_layout_list,
5234 &lad->lad_mdt_list);
5235 list_add_tail(<d->ltd_layout_phase_list,
5236 &lad->lad_mdt_phase2_list);
5239 case LE_PHASE2_DONE:
5240 ltd->ltd_layout_done = 1;
5241 if (!list_empty(<d->ltd_layout_list)) {
5242 list_del_init(<d->ltd_layout_list);
5243 if (lr->lr_flags2 & LF_INCOMPLETE) {
5244 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
5252 ltd->ltd_layout_done = 1;
5253 list_del_init(<d->ltd_layout_list);
5254 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
5255 !(lr->lr_flags & LEF_FROM_OST))
5256 lo->ll_flags |= LF_INCOMPLETE;
5261 spin_unlock(<ds->ltd_lock);
5263 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5264 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5266 memset(stop, 0, sizeof(*stop));
5267 stop->ls_status = lr->lr_status;
5268 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5269 lfsck_stop(env, lfsck->li_bottom, stop);
5270 } else if (lfsck_phase2_next_ready(lad)) {
5271 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5277 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
5278 struct lfsck_component *com,
5279 struct lfsck_request *lr,
5282 struct lfsck_instance *lfsck = com->lc_lfsck;
5283 struct lfsck_layout_slave_data *llsd = com->lc_data;
5284 struct lfsck_layout_slave_target *llst;
5288 switch (lr->lr_event) {
5289 case LE_FID_ACCESSED:
5290 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
5292 case LE_CONDITIONAL_DESTROY:
5293 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
5295 case LE_PAIRS_VERIFY: {
5296 lr->lr_status = LPVS_INIT;
5297 /* Firstly, if the MDT-object which is claimed via OST-object
5298 * local stored PFID xattr recognizes the OST-object, then it
5299 * must be that the client given PFID is wrong. */
5300 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5305 lr->lr_status = LPVS_INCONSISTENT;
5306 /* The OST-object local stored PFID xattr is stale. We need to
5307 * check whether the MDT-object that is claimed via the client
5308 * given PFID information recognizes the OST-object or not. If
5309 * matches, then need to update the OST-object's PFID xattr. */
5310 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5313 * We are not sure whether the client given PFID information
5314 * is correct or not, do nothing to avoid improper fixing.
5317 * The client given PFID information is also invalid, we can
5318 * NOT fix the OST-object inconsistency.
5323 lr->lr_status = LPVS_INCONSISTENT_TOFIX;
5324 rc = lfsck_layout_slave_repair_pfid(env, com, lr);
5328 case LE_PHASE1_DONE: {
5329 if (lr->lr_flags2 & LF_INCOMPLETE) {
5330 struct lfsck_layout *lo = com->lc_file_ram;
5332 lo->ll_flags |= LF_INCOMPLETE;
5333 llst = lfsck_layout_llst_find_and_del(llsd,
5337 lfsck_layout_llst_put(llst);
5338 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5344 case LE_PHASE2_DONE:
5346 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
5347 "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
5348 lr->lr_event, lr->lr_index, lr->lr_status);
5354 llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
5358 lfsck_layout_llst_put(llst);
5359 if (list_empty(&llsd->llsd_master_list))
5360 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5362 if (lr->lr_event == LE_PEER_EXIT &&
5363 (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
5364 (list_empty(&llsd->llsd_master_list) &&
5365 (lr->lr_status == LS_STOPPED ||
5366 lr->lr_status == LS_CO_STOPPED)))) {
5367 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5369 memset(stop, 0, sizeof(*stop));
5370 stop->ls_status = lr->lr_status;
5371 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5372 lfsck_stop(env, lfsck->li_bottom, stop);
5378 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
5382 for (i = 0; i < LLIT_MAX; i++)
5383 *count += lo->ll_objs_repaired[i];
5386 static int lfsck_layout_query_all(const struct lu_env *env,
5387 struct lfsck_component *com,
5388 __u32 *mdts_count, __u32 *osts_count,
5391 struct lfsck_layout *lo = com->lc_file_ram;
5392 struct lfsck_tgt_descs *ltds;
5393 struct lfsck_tgt_desc *ltd;
5398 rc = lfsck_query_all(env, com);
5402 ltds = &com->lc_lfsck->li_mdt_descs;
5403 down_read(<ds->ltd_rw_sem);
5404 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5405 ltd = lfsck_ltd2tgt(ltds, idx);
5406 LASSERT(ltd != NULL);
5408 mdts_count[ltd->ltd_layout_status]++;
5409 *repaired += ltd->ltd_layout_repaired;
5411 up_read(<ds->ltd_rw_sem);
5413 ltds = &com->lc_lfsck->li_ost_descs;
5414 down_read(<ds->ltd_rw_sem);
5415 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5416 ltd = lfsck_ltd2tgt(ltds, idx);
5417 LASSERT(ltd != NULL);
5419 osts_count[ltd->ltd_layout_status]++;
5420 *repaired += ltd->ltd_layout_repaired;
5422 up_read(<ds->ltd_rw_sem);
5424 down_read(&com->lc_sem);
5425 mdts_count[lo->ll_status]++;
5426 lfsck_layout_repaired(lo, repaired);
5427 up_read(&com->lc_sem);
5432 static int lfsck_layout_query(const struct lu_env *env,
5433 struct lfsck_component *com,
5434 struct lfsck_request *req,
5435 struct lfsck_reply *rep,
5436 struct lfsck_query *que, int idx)
5438 struct lfsck_layout *lo = com->lc_file_ram;
5442 LASSERT(com->lc_lfsck->li_master);
5444 rc = lfsck_layout_query_all(env, com,
5445 que->lu_mdts_count[idx],
5446 que->lu_osts_count[idx],
5447 &que->lu_repaired[idx]);
5449 down_read(&com->lc_sem);
5450 rep->lr_status = lo->ll_status;
5451 if (req->lr_flags & LEF_QUERY_ALL)
5452 lfsck_layout_repaired(lo, &rep->lr_repaired);
5453 up_read(&com->lc_sem);
5459 /* with lfsck::li_lock held */
5460 static int lfsck_layout_slave_join(const struct lu_env *env,
5461 struct lfsck_component *com,
5462 struct lfsck_start_param *lsp)
5464 struct lfsck_instance *lfsck = com->lc_lfsck;
5465 struct lfsck_layout_slave_data *llsd = com->lc_data;
5466 struct lfsck_layout_slave_target *llst;
5467 struct lfsck_start *start = lsp->lsp_start;
5471 if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
5474 if (!lsp->lsp_index_valid)
5477 /* If someone is running the LFSCK without orphan handling,
5478 * it will not maintain the object accessing rbtree. So we
5479 * cannot join it for orphan handling. */
5480 if (!llsd->llsd_rbtree_valid)
5483 spin_unlock(&lfsck->li_lock);
5484 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5485 spin_lock(&lfsck->li_lock);
5486 if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
5487 spin_unlock(&lfsck->li_lock);
5488 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
5491 lfsck_layout_llst_put(llst);
5492 spin_lock(&lfsck->li_lock);
5499 static struct lfsck_operations lfsck_layout_master_ops = {
5500 .lfsck_reset = lfsck_layout_reset,
5501 .lfsck_fail = lfsck_layout_fail,
5502 .lfsck_checkpoint = lfsck_layout_master_checkpoint,
5503 .lfsck_prep = lfsck_layout_master_prep,
5504 .lfsck_exec_oit = lfsck_layout_master_exec_oit,
5505 .lfsck_exec_dir = lfsck_layout_exec_dir,
5506 .lfsck_post = lfsck_layout_master_post,
5507 .lfsck_dump = lfsck_layout_dump,
5508 .lfsck_double_scan = lfsck_layout_master_double_scan,
5509 .lfsck_data_release = lfsck_layout_master_data_release,
5510 .lfsck_quit = lfsck_layout_master_quit,
5511 .lfsck_in_notify = lfsck_layout_master_in_notify,
5512 .lfsck_query = lfsck_layout_query,
5515 static struct lfsck_operations lfsck_layout_slave_ops = {
5516 .lfsck_reset = lfsck_layout_reset,
5517 .lfsck_fail = lfsck_layout_fail,
5518 .lfsck_checkpoint = lfsck_layout_slave_checkpoint,
5519 .lfsck_prep = lfsck_layout_slave_prep,
5520 .lfsck_exec_oit = lfsck_layout_slave_exec_oit,
5521 .lfsck_exec_dir = lfsck_layout_exec_dir,
5522 .lfsck_post = lfsck_layout_slave_post,
5523 .lfsck_dump = lfsck_layout_dump,
5524 .lfsck_double_scan = lfsck_layout_slave_double_scan,
5525 .lfsck_data_release = lfsck_layout_slave_data_release,
5526 .lfsck_quit = lfsck_layout_slave_quit,
5527 .lfsck_in_notify = lfsck_layout_slave_in_notify,
5528 .lfsck_query = lfsck_layout_query,
5529 .lfsck_join = lfsck_layout_slave_join,
5532 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
5533 struct lfsck_component *com,
5534 struct lfsck_position *pos)
5536 struct lfsck_assistant_data *lad = com->lc_data;
5537 struct lfsck_layout_req *llr;
5539 if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
5543 if (list_empty(&lad->lad_req_list))
5546 llr = list_entry(lad->lad_req_list.next,
5547 struct lfsck_layout_req,
5549 pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
5552 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
5553 .la_handler_p1 = lfsck_layout_assistant_handler_p1,
5554 .la_handler_p2 = lfsck_layout_assistant_handler_p2,
5555 .la_fill_pos = lfsck_layout_assistant_fill_pos,
5556 .la_double_scan_result = lfsck_layout_double_scan_result,
5557 .la_req_fini = lfsck_layout_assistant_req_fini,
5558 .la_sync_failures = lfsck_layout_assistant_sync_failures,
5561 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
5563 struct lfsck_component *com;
5564 struct lfsck_layout *lo;
5565 struct dt_object *root = NULL;
5566 struct dt_object *obj;
5574 INIT_LIST_HEAD(&com->lc_link);
5575 INIT_LIST_HEAD(&com->lc_link_dir);
5576 init_rwsem(&com->lc_sem);
5577 atomic_set(&com->lc_ref, 1);
5578 com->lc_lfsck = lfsck;
5579 com->lc_type = LFSCK_TYPE_LAYOUT;
5580 if (lfsck->li_master) {
5581 com->lc_ops = &lfsck_layout_master_ops;
5582 com->lc_data = lfsck_assistant_data_init(
5583 &lfsck_layout_assistant_ops,
5585 if (com->lc_data == NULL)
5586 GOTO(out, rc = -ENOMEM);
5588 struct lfsck_layout_slave_data *llsd;
5590 com->lc_ops = &lfsck_layout_slave_ops;
5591 OBD_ALLOC_PTR(llsd);
5593 GOTO(out, rc = -ENOMEM);
5595 INIT_LIST_HEAD(&llsd->llsd_seq_list);
5596 INIT_LIST_HEAD(&llsd->llsd_master_list);
5597 spin_lock_init(&llsd->llsd_lock);
5598 llsd->llsd_rb_root = RB_ROOT;
5599 rwlock_init(&llsd->llsd_rb_lock);
5600 com->lc_data = llsd;
5602 com->lc_file_size = sizeof(*lo);
5603 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
5604 if (com->lc_file_ram == NULL)
5605 GOTO(out, rc = -ENOMEM);
5607 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
5608 if (com->lc_file_disk == NULL)
5609 GOTO(out, rc = -ENOMEM);
5611 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
5613 GOTO(out, rc = PTR_ERR(root));
5615 if (unlikely(!dt_try_as_dir(env, root)))
5616 GOTO(out, rc = -ENOTDIR);
5618 obj = local_file_find_or_create(env, lfsck->li_los, root,
5620 S_IFREG | S_IRUGO | S_IWUSR);
5622 GOTO(out, rc = PTR_ERR(obj));
5625 rc = lfsck_layout_load(env, com);
5627 rc = lfsck_layout_reset(env, com, true);
5628 else if (rc == -ENOENT)
5629 rc = lfsck_layout_init(env, com);
5634 lo = com->lc_file_ram;
5635 switch (lo->ll_status) {
5641 spin_lock(&lfsck->li_lock);
5642 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
5643 spin_unlock(&lfsck->li_lock);
5646 CERROR("%s: unknown lfsck_layout status %d\n",
5647 lfsck_lfsck2name(lfsck), lo->ll_status);
5649 case LS_SCANNING_PHASE1:
5650 case LS_SCANNING_PHASE2:
5651 /* No need to store the status to disk right now.
5652 * If the system crashed before the status stored,
5653 * it will be loaded back when next time. */
5654 lo->ll_status = LS_CRASHED;
5655 if (!lfsck->li_master)
5656 lo->ll_flags |= LF_INCOMPLETE;
5663 spin_lock(&lfsck->li_lock);
5664 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
5665 spin_unlock(&lfsck->li_lock);
5669 if (lo->ll_flags & LF_CRASHED_LASTID) {
5670 LASSERT(lfsck->li_out_notify != NULL);
5672 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5673 LE_LASTID_REBUILDING);
5679 if (root != NULL && !IS_ERR(root))
5680 lfsck_object_put(env, root);
5683 lfsck_component_cleanup(env, com);
5684 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
5685 lfsck_lfsck2name(lfsck), rc);
5691 struct lfsck_orphan_it {
5692 struct lfsck_component *loi_com;
5693 struct lfsck_rbtree_node *loi_lrn;
5694 struct lfsck_layout_slave_target *loi_llst;
5695 struct lu_fid loi_key;
5696 struct lu_orphan_rec loi_rec;
5698 unsigned int loi_over:1;
5701 static int lfsck_fid_match_idx(const struct lu_env *env,
5702 struct lfsck_instance *lfsck,
5703 const struct lu_fid *fid, int idx)
5705 struct seq_server_site *ss;
5706 struct lu_server_fld *sf;
5707 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
5710 /* All abnormal cases will be returned to MDT0. */
5711 if (!fid_is_norm(fid)) {
5718 ss = lfsck_dev_site(lfsck);
5719 if (unlikely(ss == NULL))
5722 sf = ss->ss_server_fld;
5723 LASSERT(sf != NULL);
5725 fld_range_set_any(range);
5726 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
5730 if (!fld_range_is_mdt(range))
5733 if (range->lsr_index == idx)
5739 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
5740 struct dt_object *obj)
5742 struct dt_device *dev = lfsck_obj2dev(obj);
5743 struct thandle *handle;
5747 handle = dt_trans_create(env, dev);
5751 rc = dt_declare_ref_del(env, obj, handle);
5755 rc = dt_declare_destroy(env, obj, handle);
5759 rc = dt_trans_start_local(env, dev, handle);
5763 dt_write_lock(env, obj, 0);
5764 rc = dt_ref_del(env, obj, handle);
5766 rc = dt_destroy(env, obj, handle);
5767 dt_write_unlock(env, obj);
5772 dt_trans_stop(env, dev, handle);
5774 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
5775 PFID(lfsck_dto2fid(obj)), rc);
5780 static int lfsck_orphan_index_lookup(const struct lu_env *env,
5781 struct dt_object *dt,
5783 const struct dt_key *key)
5788 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
5789 struct dt_object *dt,
5790 const struct dt_rec *rec,
5791 const struct dt_key *key,
5792 struct thandle *handle)
5797 static int lfsck_orphan_index_insert(const struct lu_env *env,
5798 struct dt_object *dt,
5799 const struct dt_rec *rec,
5800 const struct dt_key *key,
5801 struct thandle *handle,
5807 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
5808 struct dt_object *dt,
5809 const struct dt_key *key,
5810 struct thandle *handle)
5815 static int lfsck_orphan_index_delete(const struct lu_env *env,
5816 struct dt_object *dt,
5817 const struct dt_key *key,
5818 struct thandle *handle)
5823 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
5824 struct dt_object *dt,
5827 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
5828 struct lfsck_instance *lfsck;
5829 struct lfsck_component *com = NULL;
5830 struct lfsck_layout_slave_data *llsd;
5831 struct lfsck_orphan_it *it = NULL;
5832 struct lfsck_layout *lo;
5836 lfsck = lfsck_instance_find(dev, true, false);
5837 if (unlikely(lfsck == NULL))
5838 RETURN(ERR_PTR(-ENXIO));
5840 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
5841 if (unlikely(com == NULL))
5842 GOTO(out, rc = -ENOENT);
5844 lo = com->lc_file_ram;
5845 if (lo->ll_flags & LF_INCOMPLETE)
5846 GOTO(out, rc = -ESRCH);
5848 llsd = com->lc_data;
5849 if (!llsd->llsd_rbtree_valid)
5850 GOTO(out, rc = -ESRCH);
5854 GOTO(out, rc = -ENOMEM);
5856 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
5857 if (it->loi_llst == NULL)
5858 GOTO(out, rc = -ENXIO);
5860 if (dev->dd_record_fid_accessed) {
5861 /* The first iteration against the rbtree, scan the whole rbtree
5862 * to remove the nodes which do NOT need to be handled. */
5863 write_lock(&llsd->llsd_rb_lock);
5864 if (dev->dd_record_fid_accessed) {
5865 struct rb_node *node;
5866 struct rb_node *next;
5867 struct lfsck_rbtree_node *lrn;
5869 /* No need to record the fid accessing anymore. */
5870 dev->dd_record_fid_accessed = 0;
5872 node = rb_first(&llsd->llsd_rb_root);
5873 while (node != NULL) {
5874 next = rb_next(node);
5875 lrn = rb_entry(node, struct lfsck_rbtree_node,
5877 if (atomic_read(&lrn->lrn_known_count) <=
5878 atomic_read(&lrn->lrn_accessed_count)) {
5879 rb_erase(node, &llsd->llsd_rb_root);
5880 lfsck_rbtree_free(lrn);
5885 write_unlock(&llsd->llsd_rb_lock);
5888 /* read lock the rbtree when init, and unlock when fini */
5889 read_lock(&llsd->llsd_rb_lock);
5897 lfsck_component_put(env, com);
5899 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
5900 lfsck_lfsck2name(lfsck), rc);
5902 lfsck_instance_put(env, lfsck);
5907 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
5910 return (struct dt_it *)it;
5913 static void lfsck_orphan_it_fini(const struct lu_env *env,
5916 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
5917 struct lfsck_component *com = it->loi_com;
5918 struct lfsck_layout_slave_data *llsd;
5919 struct lfsck_layout_slave_target *llst;
5922 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
5923 lfsck_lfsck2name(com->lc_lfsck));
5925 llsd = com->lc_data;
5926 read_unlock(&llsd->llsd_rb_lock);
5927 llst = it->loi_llst;
5928 LASSERT(llst != NULL);
5930 /* Save the key and hash for iterate next. */
5931 llst->llst_fid = it->loi_key;
5932 llst->llst_hash = it->loi_hash;
5933 lfsck_layout_llst_put(llst);
5934 lfsck_component_put(env, com);
5940 * \retval +1: the iteration finished
5941 * \retval 0: on success, not finished
5942 * \retval -ve: on error
5944 static int lfsck_orphan_it_next(const struct lu_env *env,
5947 struct lfsck_thread_info *info = lfsck_env_info(env);
5948 struct filter_fid_old *pfid = &info->lti_old_pfid;
5949 struct lu_attr *la = &info->lti_la;
5950 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
5951 struct lu_fid *key = &it->loi_key;
5952 struct lu_orphan_rec *rec = &it->loi_rec;
5953 struct lfsck_component *com = it->loi_com;
5954 struct lfsck_instance *lfsck = com->lc_lfsck;
5955 struct lfsck_layout_slave_data *llsd = com->lc_data;
5956 struct dt_object *obj;
5957 struct lfsck_rbtree_node *lrn;
5961 __u32 idx = it->loi_llst->llst_index;
5971 lrn = lfsck_rbtree_search(llsd, key, &exact);
5979 key->f_seq = lrn->lrn_seq;
5980 key->f_oid = lrn->lrn_first_oid;
5985 if (unlikely(key->f_oid == 0)) {
5992 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
5998 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
5999 atomic_read(&lrn->lrn_accessed_count))) {
6000 struct rb_node *next = rb_next(&lrn->lrn_node);
6002 while (next != NULL) {
6003 lrn = rb_entry(next, struct lfsck_rbtree_node,
6005 if (atomic_read(&lrn->lrn_known_count) >
6006 atomic_read(&lrn->lrn_accessed_count))
6008 next = rb_next(next);
6017 key->f_seq = lrn->lrn_seq;
6018 key->f_oid = lrn->lrn_first_oid;
6022 pos = key->f_oid - lrn->lrn_first_oid;
6025 pos = find_next_bit(lrn->lrn_known_bitmap,
6026 LFSCK_RBTREE_BITMAP_WIDTH, pos);
6027 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
6028 key->f_oid = lrn->lrn_first_oid + pos;
6029 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
6037 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
6042 key->f_oid = lrn->lrn_first_oid + pos;
6043 obj = lfsck_object_find_bottom(env, lfsck, key);
6046 if (rc == -ENOENT) {
6053 dt_read_lock(env, obj, 0);
6054 if (dt_object_exists(obj) == 0 ||
6055 lfsck_is_dead_obj(obj)) {
6056 dt_read_unlock(env, obj);
6057 lfsck_object_put(env, obj);
6062 rc = dt_attr_get(env, obj, la);
6066 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
6068 if (rc == -ENODATA) {
6069 /* For the pre-created OST-object, update the bitmap to avoid
6070 * others LFSCK (second phase) iteration to touch it again. */
6071 if (la->la_ctime == 0) {
6072 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
6073 atomic_inc(&lrn->lrn_accessed_count);
6075 /* For the race between repairing dangling referenced
6076 * MDT-object and unlink the file, it may left orphan
6077 * OST-object there. Destroy it now! */
6078 if (unlikely(!(la->la_mode & S_ISUID))) {
6079 dt_read_unlock(env, obj);
6080 lfsck_layout_destroy_orphan(env, obj);
6081 lfsck_object_put(env, obj);
6085 } else if (idx == 0) {
6086 /* If the orphan OST-object has no parent information,
6087 * regard it as referenced by the MDT-object on MDT0. */
6088 fid_zero(&rec->lor_fid);
6089 rec->lor_uid = la->la_uid;
6090 rec->lor_gid = la->la_gid;
6094 dt_read_unlock(env, obj);
6095 lfsck_object_put(env, obj);
6103 if (rc != sizeof(struct filter_fid) &&
6104 rc != sizeof(struct filter_fid_old))
6105 GOTO(out, rc = -EINVAL);
6107 fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
6108 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
6109 * MDT-object's FID::f_ver, instead it is the OST-object index in its
6110 * parent MDT-object's layout EA. */
6111 save = rec->lor_fid.f_stripe_idx;
6112 rec->lor_fid.f_ver = 0;
6113 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
6114 /* If the orphan OST-object does not claim the MDT, then next.
6116 * If we do not know whether it matches or not, then return it
6117 * to the MDT for further check. */
6119 dt_read_unlock(env, obj);
6120 lfsck_object_put(env, obj);
6125 rec->lor_fid.f_stripe_idx = save;
6126 rec->lor_uid = la->la_uid;
6127 rec->lor_gid = la->la_gid;
6129 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
6130 lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
6131 rec->lor_uid, rec->lor_gid);
6136 dt_read_unlock(env, obj);
6137 lfsck_object_put(env, obj);
6145 * \retval +1: locate to the exactly position
6146 * \retval 0: cannot locate to the exactly position,
6147 * call next() to move to a valid position.
6148 * \retval -ve: on error
6150 static int lfsck_orphan_it_get(const struct lu_env *env,
6152 const struct dt_key *key)
6154 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6157 it->loi_key = *(struct lu_fid *)key;
6158 rc = lfsck_orphan_it_next(env, di);
6168 static void lfsck_orphan_it_put(const struct lu_env *env,
6173 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
6174 const struct dt_it *di)
6176 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6178 return (struct dt_key *)&it->loi_key;
6181 static int lfsck_orphan_it_key_size(const struct lu_env *env,
6182 const struct dt_it *di)
6184 return sizeof(struct lu_fid);
6187 static int lfsck_orphan_it_rec(const struct lu_env *env,
6188 const struct dt_it *di,
6192 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6194 *(struct lu_orphan_rec *)rec = it->loi_rec;
6199 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
6200 const struct dt_it *di)
6202 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6204 return it->loi_hash;
6208 * \retval +1: locate to the exactly position
6209 * \retval 0: cannot locate to the exactly position,
6210 * call next() to move to a valid position.
6211 * \retval -ve: on error
6213 static int lfsck_orphan_it_load(const struct lu_env *env,
6214 const struct dt_it *di,
6217 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6218 struct lfsck_layout_slave_target *llst = it->loi_llst;
6221 LASSERT(llst != NULL);
6223 if (hash != llst->llst_hash) {
6224 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
6225 "iteration does not match the one when fini "
6226 "%llu, to be reset.\n",
6227 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
6229 fid_zero(&llst->llst_fid);
6230 llst->llst_hash = 0;
6233 it->loi_key = llst->llst_fid;
6234 it->loi_hash = llst->llst_hash;
6235 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
6245 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
6246 const struct dt_it *di,
6252 const struct dt_index_operations lfsck_orphan_index_ops = {
6253 .dio_lookup = lfsck_orphan_index_lookup,
6254 .dio_declare_insert = lfsck_orphan_index_declare_insert,
6255 .dio_insert = lfsck_orphan_index_insert,
6256 .dio_declare_delete = lfsck_orphan_index_declare_delete,
6257 .dio_delete = lfsck_orphan_index_delete,
6259 .init = lfsck_orphan_it_init,
6260 .fini = lfsck_orphan_it_fini,
6261 .get = lfsck_orphan_it_get,
6262 .put = lfsck_orphan_it_put,
6263 .next = lfsck_orphan_it_next,
6264 .key = lfsck_orphan_it_key,
6265 .key_size = lfsck_orphan_it_key_size,
6266 .rec = lfsck_orphan_it_rec,
6267 .store = lfsck_orphan_it_store,
6268 .load = lfsck_orphan_it_load,
6269 .key_rec = lfsck_orphan_it_key_rec,