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);
719 fid_le_to_cpu(&des->ll_fid_latest_scanned_phase2,
720 &src->ll_fid_latest_scanned_phase2);
723 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
724 const struct lfsck_layout *src)
728 des->ll_magic = cpu_to_le32(src->ll_magic);
729 des->ll_status = cpu_to_le32(src->ll_status);
730 des->ll_flags = cpu_to_le32(src->ll_flags);
731 des->ll_success_count = cpu_to_le32(src->ll_success_count);
732 des->ll_run_time_phase1 = cpu_to_le32(src->ll_run_time_phase1);
733 des->ll_run_time_phase2 = cpu_to_le32(src->ll_run_time_phase2);
734 des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
735 des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
736 des->ll_time_last_checkpoint =
737 cpu_to_le64(src->ll_time_last_checkpoint);
738 des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
739 des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
740 des->ll_pos_first_inconsistent =
741 cpu_to_le64(src->ll_pos_first_inconsistent);
742 des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
743 des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
744 des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
745 des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
746 for (i = 0; i < LLIT_MAX; i++)
747 des->ll_objs_repaired[i] =
748 cpu_to_le64(src->ll_objs_repaired[i]);
749 des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
750 des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
751 fid_cpu_to_le(&des->ll_fid_latest_scanned_phase2,
752 &src->ll_fid_latest_scanned_phase2);
756 * Load the OST bitmap from the lfsck_layout trace file.
758 * \param[in] env pointer to the thread context
759 * \param[in] com pointer to the lfsck component
761 * \retval 0 for success
762 * \retval negative error number on failure or data corruption
764 static int lfsck_layout_load_bitmap(const struct lu_env *env,
765 struct lfsck_component *com)
767 struct dt_object *obj = com->lc_obj;
768 struct lfsck_assistant_data *lad = com->lc_data;
769 struct lfsck_layout *lo = com->lc_file_ram;
770 struct cfs_bitmap *bitmap = lad->lad_bitmap;
771 loff_t pos = com->lc_file_size;
777 if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
779 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
781 nbits = lo->ll_bitmap_size;
783 if (unlikely(nbits < BITS_PER_LONG))
784 nbits = BITS_PER_LONG;
786 if (nbits > bitmap->size) {
787 __u32 new_bits = bitmap->size;
788 struct cfs_bitmap *new_bitmap;
790 while (new_bits < nbits)
793 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
794 if (new_bitmap == NULL)
797 lad->lad_bitmap = new_bitmap;
798 CFS_FREE_BITMAP(bitmap);
802 if (lo->ll_bitmap_size == 0) {
803 lad->lad_incomplete = 0;
804 CFS_RESET_BITMAP(bitmap);
809 size = (lo->ll_bitmap_size + 7) >> 3;
810 rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
812 RETURN(rc >= 0 ? -EINVAL : rc);
814 if (cfs_bitmap_check_empty(bitmap))
815 lad->lad_incomplete = 0;
817 lad->lad_incomplete = 1;
823 * Load the layout LFSCK trace file from disk.
825 * The layout LFSCK trace file records the layout LFSCK status information
826 * and other statistics, such as how many objects have been scanned, and how
827 * many objects have been repaired, and etc. It also contains the bitmap for
828 * failed OSTs during the layout LFSCK. All these information will be loaded
829 * from disk to RAM when the layout LFSCK component setup.
831 * \param[in] env pointer to the thread context
832 * \param[in] com pointer to the lfsck component
834 * \retval positive number for file data corruption, the caller
835 * should reset the layout LFSCK trace file
836 * \retval 0 for success
837 * \retval negative error number on failure
839 static int lfsck_layout_load(const struct lu_env *env,
840 struct lfsck_component *com)
842 struct lfsck_layout *lo = com->lc_file_ram;
843 ssize_t size = com->lc_file_size;
847 rc = dt_read(env, com->lc_obj,
848 lfsck_buf_get(env, com->lc_file_disk, size), &pos);
852 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
853 lfsck_lfsck2name(com->lc_lfsck), rc);
855 } else if (rc != size) {
856 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
857 lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
861 lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
862 if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
863 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
864 "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
865 lo->ll_magic, LFSCK_LAYOUT_MAGIC);
873 * Store the layout LFSCK trace file on disk.
875 * The layout LFSCK trace file records the layout LFSCK status information
876 * and other statistics, such as how many objects have been scanned, and how
877 * many objects have been repaired, and etc. It also contains the bitmap for
878 * failed OSTs during the layout LFSCK. All these information will be synced
879 * from RAM to disk periodically.
881 * \param[in] env pointer to the thread context
882 * \param[in] com pointer to the lfsck component
884 * \retval 0 for success
885 * \retval negative error number on failure
887 static int lfsck_layout_store(const struct lu_env *env,
888 struct lfsck_component *com)
890 struct dt_object *obj = com->lc_obj;
891 struct lfsck_instance *lfsck = com->lc_lfsck;
892 struct lfsck_layout *lo_ram = com->lc_file_ram;
893 struct lfsck_layout *lo = com->lc_file_disk;
895 struct dt_device *dev = lfsck_obj2dev(obj);
896 struct cfs_bitmap *bitmap = NULL;
898 ssize_t size = com->lc_file_size;
903 if (lfsck->li_master) {
904 struct lfsck_assistant_data *lad = com->lc_data;
906 bitmap = lad->lad_bitmap;
907 nbits = bitmap->size;
910 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
913 lo_ram->ll_bitmap_size = nbits;
914 lfsck_layout_cpu_to_le(lo, lo_ram);
915 th = dt_trans_create(env, dev);
917 GOTO(log, rc = PTR_ERR(th));
919 rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
924 if (bitmap != NULL) {
925 rc = dt_declare_record_write(env, obj,
926 lfsck_buf_get(env, bitmap->data, nbits >> 3),
932 rc = dt_trans_start_local(env, dev, th);
937 rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
941 if (bitmap != NULL) {
943 rc = dt_record_write(env, obj,
944 lfsck_buf_get(env, bitmap->data, nbits >> 3),
951 dt_trans_stop(env, dev, th);
955 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
956 lfsck_lfsck2name(lfsck), rc);
961 static int lfsck_layout_init(const struct lu_env *env,
962 struct lfsck_component *com)
964 struct lfsck_layout *lo = com->lc_file_ram;
967 memset(lo, 0, com->lc_file_size);
968 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
969 lo->ll_status = LS_INIT;
970 down_write(&com->lc_sem);
971 rc = lfsck_layout_store(env, com);
972 if (rc == 0 && com->lc_lfsck->li_master)
973 rc = lfsck_load_sub_trace_files(env, com,
974 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
975 up_write(&com->lc_sem);
980 static int fid_is_for_ostobj(const struct lu_env *env,
981 struct lfsck_instance *lfsck,
982 struct dt_object *obj, const struct lu_fid *fid)
984 struct seq_server_site *ss = lfsck_dev_site(lfsck);
985 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
986 struct lustre_mdt_attrs *lma;
989 fld_range_set_any(range);
990 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
992 if (fld_range_is_ost(range))
998 lma = &lfsck_env_info(env)->lti_lma;
999 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, lma, sizeof(*lma)),
1001 if (rc == sizeof(*lma)) {
1002 lustre_lma_swab(lma);
1004 return lma->lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1007 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1012 static struct lfsck_layout_seq *
1013 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1015 struct lfsck_layout_seq *lls;
1017 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1018 if (lls->lls_seq == seq)
1021 if (lls->lls_seq > seq)
1029 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1030 struct lfsck_layout_seq *lls)
1032 struct lfsck_layout_seq *tmp;
1033 struct list_head *pos = &llsd->llsd_seq_list;
1035 list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1036 if (lls->lls_seq < tmp->lls_seq) {
1037 pos = &tmp->lls_list;
1041 list_add_tail(&lls->lls_list, pos);
1045 lfsck_layout_lastid_create(const struct lu_env *env,
1046 struct lfsck_instance *lfsck,
1047 struct dt_object *obj)
1049 struct lfsck_thread_info *info = lfsck_env_info(env);
1050 struct lu_attr *la = &info->lti_la;
1051 struct dt_object_format *dof = &info->lti_dof;
1052 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1053 struct dt_device *dt = lfsck_obj2dev(obj);
1060 if (bk->lb_param & LPF_DRYRUN)
1063 memset(la, 0, sizeof(*la));
1064 la->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1065 la->la_valid = LA_MODE | LA_UID | LA_GID;
1066 memset(dof, 0, sizeof(*dof));
1067 dof->dof_type = dt_mode_to_dft(S_IFREG);
1069 th = dt_trans_create(env, dt);
1071 GOTO(log, rc = PTR_ERR(th));
1073 rc = dt_declare_create(env, obj, la, NULL, dof, th);
1077 rc = dt_declare_record_write(env, obj,
1078 lfsck_buf_get(env, &lastid,
1084 rc = dt_trans_start_local(env, dt, th);
1088 dt_write_lock(env, obj, 0);
1089 if (likely(dt_object_exists(obj) == 0)) {
1090 rc = dt_create(env, obj, la, NULL, dof, th);
1092 rc = dt_record_write(env, obj,
1093 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1096 dt_write_unlock(env, obj);
1101 dt_trans_stop(env, dt, th);
1104 CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1106 lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1112 lfsck_layout_lastid_reload(const struct lu_env *env,
1113 struct lfsck_component *com,
1114 struct lfsck_layout_seq *lls)
1120 dt_read_lock(env, lls->lls_lastid_obj, 0);
1121 rc = dt_record_read(env, lls->lls_lastid_obj,
1122 lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1123 dt_read_unlock(env, lls->lls_lastid_obj);
1124 if (unlikely(rc != 0))
1127 lastid = le64_to_cpu(lastid);
1128 if (lastid < lls->lls_lastid_known) {
1129 struct lfsck_instance *lfsck = com->lc_lfsck;
1130 struct lfsck_layout *lo = com->lc_file_ram;
1132 lls->lls_lastid = lls->lls_lastid_known;
1134 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1135 LASSERT(lfsck->li_out_notify != NULL);
1137 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1138 LE_LASTID_REBUILDING);
1139 lo->ll_flags |= LF_CRASHED_LASTID;
1141 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1142 "LAST_ID file (1) for the sequence %#llx"
1143 ", old value %llu, known value %llu\n",
1144 lfsck_lfsck2name(lfsck), lls->lls_seq,
1145 lastid, lls->lls_lastid);
1147 } else if (lastid >= lls->lls_lastid) {
1148 lls->lls_lastid = lastid;
1156 lfsck_layout_lastid_store(const struct lu_env *env,
1157 struct lfsck_component *com)
1159 struct lfsck_instance *lfsck = com->lc_lfsck;
1160 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1161 struct dt_device *dt = lfsck->li_bottom;
1162 struct lfsck_layout_slave_data *llsd = com->lc_data;
1163 struct lfsck_layout_seq *lls;
1169 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1172 if (!lls->lls_dirty)
1175 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1176 "<seq> %#llx as <oid> %llu\n",
1177 lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1179 if (bk->lb_param & LPF_DRYRUN) {
1184 th = dt_trans_create(env, dt);
1187 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1188 "the LAST_ID for <seq> %#llx(1): rc = %d\n",
1189 lfsck_lfsck2name(com->lc_lfsck),
1194 lastid = cpu_to_le64(lls->lls_lastid);
1195 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1196 lfsck_buf_get(env, &lastid,
1202 rc = dt_trans_start_local(env, dt, th);
1206 dt_write_lock(env, lls->lls_lastid_obj, 0);
1207 rc = dt_record_write(env, lls->lls_lastid_obj,
1208 lfsck_buf_get(env, &lastid,
1209 sizeof(lastid)), &pos, th);
1210 dt_write_unlock(env, lls->lls_lastid_obj);
1215 dt_trans_stop(env, dt, th);
1218 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1219 "the LAST_ID for <seq> %#llx(2): rc = %d\n",
1220 lfsck_lfsck2name(com->lc_lfsck),
1229 lfsck_layout_lastid_load(const struct lu_env *env,
1230 struct lfsck_component *com,
1231 struct lfsck_layout_seq *lls)
1233 struct lfsck_instance *lfsck = com->lc_lfsck;
1234 struct lfsck_layout *lo = com->lc_file_ram;
1235 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
1236 struct dt_object *obj;
1241 lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1242 obj = dt_locate(env, lfsck->li_bottom, fid);
1244 RETURN(PTR_ERR(obj));
1246 /* LAST_ID crashed, to be rebuilt */
1247 if (dt_object_exists(obj) == 0) {
1248 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1249 LASSERT(lfsck->li_out_notify != NULL);
1251 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1252 LE_LASTID_REBUILDING);
1253 lo->ll_flags |= LF_CRASHED_LASTID;
1255 CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1256 "LAST_ID file for sequence %#llx\n",
1257 lfsck_lfsck2name(lfsck), lls->lls_seq);
1259 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1261 struct l_wait_info lwi = LWI_TIMEOUT(
1262 cfs_time_seconds(cfs_fail_val),
1265 /* Some others may changed the cfs_fail_val
1266 * as zero after above check, re-check it for
1267 * sure to avoid falling into wait for ever. */
1268 if (likely(lwi.lwi_timeout > 0)) {
1269 struct ptlrpc_thread *thread =
1272 up_write(&com->lc_sem);
1273 l_wait_event(thread->t_ctl_waitq,
1274 !thread_is_running(thread),
1276 down_write(&com->lc_sem);
1281 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1283 dt_read_lock(env, obj, 0);
1284 rc = dt_read(env, obj,
1285 lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1287 dt_read_unlock(env, obj);
1288 if (rc != 0 && rc != sizeof(__u64))
1289 GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1291 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1292 LASSERT(lfsck->li_out_notify != NULL);
1294 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1295 LE_LASTID_REBUILDING);
1296 lo->ll_flags |= LF_CRASHED_LASTID;
1298 CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1299 "LAST_ID file for the sequence %#llx"
1301 lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1304 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1312 lfsck_object_put(env, obj);
1314 lls->lls_lastid_obj = obj;
1319 static void lfsck_layout_record_failure(const struct lu_env *env,
1320 struct lfsck_instance *lfsck,
1321 struct lfsck_layout *lo)
1325 lo->ll_objs_failed_phase1++;
1326 cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1328 if (lo->ll_pos_first_inconsistent == 0 ||
1329 lo->ll_pos_first_inconsistent < cookie) {
1330 lo->ll_pos_first_inconsistent = cookie;
1332 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1333 "inconsistency at the pos [%llu]\n",
1334 lfsck_lfsck2name(lfsck),
1335 lo->ll_pos_first_inconsistent);
1339 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1340 struct lfsck_component *com,
1343 struct lfsck_instance *lfsck = com->lc_lfsck;
1344 struct lfsck_layout *lo = com->lc_file_ram;
1346 down_write(&com->lc_sem);
1347 lo->ll_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
1348 HALF_SEC - com->lc_time_last_checkpoint);
1349 lo->ll_time_last_checkpoint = cfs_time_current_sec();
1350 lo->ll_objs_checked_phase2 += com->lc_new_checked;
1353 if (lo->ll_flags & LF_INCOMPLETE) {
1354 lo->ll_status = LS_PARTIAL;
1356 if (lfsck->li_master) {
1357 struct lfsck_assistant_data *lad = com->lc_data;
1359 if (lad->lad_incomplete)
1360 lo->ll_status = LS_PARTIAL;
1362 lo->ll_status = LS_COMPLETED;
1364 lo->ll_status = LS_COMPLETED;
1367 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1368 lo->ll_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
1369 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1370 lo->ll_success_count++;
1371 } else if (rc == 0) {
1372 if (lfsck->li_status != 0)
1373 lo->ll_status = lfsck->li_status;
1375 lo->ll_status = LS_STOPPED;
1377 lo->ll_status = LS_FAILED;
1380 rc = lfsck_layout_store(env, com);
1381 up_write(&com->lc_sem);
1386 static int lfsck_layout_trans_stop(const struct lu_env *env,
1387 struct dt_device *dev,
1388 struct thandle *handle, int result)
1392 /* XXX: If there is something worng or it needs to repair nothing,
1393 * then notify the lower to stop the modification. Currently,
1394 * we use th_result for such purpose, that may be replaced by
1395 * some rollback mechanism in the future. */
1396 handle->th_result = result;
1397 rc = dt_trans_stop(env, dev, handle);
1399 return result > 0 ? 0 : result;
1401 return rc == 0 ? 1 : rc;
1404 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1405 struct lfsck_component *com,
1406 const struct lu_fid *pfid,
1407 const struct lu_fid *cfid,
1408 __u32 ea_off, __u32 ost_idx)
1410 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
1411 struct lu_fid *rec = &lfsck_env_info(env)->lti_fid4;
1412 struct dt_device *dev;
1413 struct dt_object *obj;
1414 struct thandle *th = NULL;
1419 idx = lfsck_sub_trace_file_fid2idx(pfid);
1420 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1421 dev = lfsck_obj2dev(obj);
1422 fid_cpu_to_be(key, pfid);
1423 key->f_ver = cpu_to_be32(ea_off);
1424 fid_cpu_to_be(rec, cfid);
1425 rec->f_ver = cpu_to_be32(ost_idx);
1427 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1429 th = dt_trans_create(env, dev);
1431 GOTO(unlock, rc = PTR_ERR(th));
1433 rc = dt_declare_insert(env, obj,
1434 (const struct dt_rec *)rec,
1435 (const struct dt_key *)key, th);
1439 rc = dt_trans_start_local(env, dev, th);
1443 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1444 (const struct dt_key *)key, th, 1);
1449 if (th != NULL && !IS_ERR(th))
1450 dt_trans_stop(env, dev, th);
1452 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1454 CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", ea_off = %u, "
1455 "ost_idx = %u, into the trace file for further dangling check: "
1456 "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1457 PFID(pfid), PFID(cfid), ea_off, ost_idx, rc);
1462 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1463 struct lfsck_component *com,
1464 const struct lu_fid *fid,
1467 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
1468 struct dt_device *dev;
1469 struct dt_object *obj;
1470 struct thandle *th = NULL;
1475 idx = lfsck_sub_trace_file_fid2idx(fid);
1476 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1477 dev = lfsck_obj2dev(obj);
1478 fid_cpu_to_be(key, fid);
1479 key->f_ver = cpu_to_be32(ea_off);
1481 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1483 th = dt_trans_create(env, dev);
1485 GOTO(unlock, rc = PTR_ERR(th));
1487 rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1491 rc = dt_trans_start_local(env, dev, th);
1495 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1500 if (th != NULL && !IS_ERR(th))
1501 dt_trans_stop(env, dev, th);
1503 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1505 CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1506 ", ea_off = %u from the trace file: rc = %d\n",
1507 lfsck_lfsck2name(com->lc_lfsck), PFID(fid), ea_off, rc);
1513 * Get the system default stripe size.
1515 * \param[in] env pointer to the thread context
1516 * \param[in] lfsck pointer to the lfsck instance
1517 * \param[out] size pointer to the default stripe size
1519 * \retval 0 for success
1520 * \retval negative error number on failure
1522 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1523 struct lfsck_instance *lfsck,
1526 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1527 struct dt_object *root;
1530 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1532 return PTR_ERR(root);
1534 /* Get the default stripe size via xattr_get on the backend root. */
1535 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1538 /* The lum->lmm_stripe_size is LE mode. The *size also
1539 * should be LE mode. So it is unnecessary to convert. */
1540 *size = lum->lmm_stripe_size;
1542 } else if (unlikely(rc == 0)) {
1546 lfsck_object_put(env, root);
1552 * \retval +1: repaired
1553 * \retval 0: did nothing
1554 * \retval -ve: on error
1556 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1557 struct thandle *handle,
1558 struct dt_object *parent,
1559 struct lu_fid *cfid,
1561 struct lov_ost_data_v1 *slot,
1562 int fl, __u32 ost_idx)
1564 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1565 struct lov_mds_md_v1 *lmm = buf->lb_buf;
1566 struct lu_buf ea_buf;
1572 magic = le32_to_cpu(lmm->lmm_magic);
1573 count = le16_to_cpu(lmm->lmm_stripe_count);
1575 fid_to_ostid(cfid, oi);
1576 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1577 slot->l_ost_gen = cpu_to_le32(0);
1578 slot->l_ost_idx = cpu_to_le32(ost_idx);
1580 if (le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE) {
1581 struct lov_ost_data_v1 *objs;
1584 if (magic == LOV_MAGIC_V1)
1585 objs = &lmm->lmm_objects[0];
1587 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1588 for (i = 0; i < count; i++, objs++) {
1589 if (objs != slot && lovea_slot_is_dummy(objs))
1593 /* If the @slot is the last dummy slot to be refilled,
1594 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1596 lmm->lmm_pattern &= ~cpu_to_le32(LOV_PATTERN_F_HOLE);
1599 lfsck_buf_init(&ea_buf, lmm, lov_mds_md_size(count, magic));
1600 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1608 * \retval +1: repaired
1609 * \retval 0: did nothing
1610 * \retval -ve: on error
1612 static int lfsck_layout_extend_lovea(const struct lu_env *env,
1613 struct lfsck_instance *lfsck,
1614 struct thandle *handle,
1615 struct dt_object *parent,
1616 struct lu_fid *cfid,
1617 struct lu_buf *buf, int fl,
1618 __u32 ost_idx, __u32 ea_off, bool reset)
1620 struct lov_mds_md_v1 *lmm = buf->lb_buf;
1621 struct lov_ost_data_v1 *objs;
1627 if (fl == LU_XATTR_CREATE || reset) {
1628 __u32 pattern = LOV_PATTERN_RAID0;
1631 LASSERT(buf->lb_len >= lov_mds_md_size(count, LOV_MAGIC_V1));
1633 if (ea_off != 0 || reset) {
1634 pattern |= LOV_PATTERN_F_HOLE;
1638 memset(lmm, 0, buf->lb_len);
1639 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1640 lmm->lmm_pattern = cpu_to_le32(pattern);
1641 fid_to_lmm_oi(lfsck_dto2fid(parent), &lmm->lmm_oi);
1642 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1644 rc = lfsck_layout_get_def_stripesize(env, lfsck,
1645 &lmm->lmm_stripe_size);
1649 objs = &lmm->lmm_objects[ea_off];
1651 __u32 magic = le32_to_cpu(lmm->lmm_magic);
1654 count = le16_to_cpu(lmm->lmm_stripe_count);
1655 if (magic == LOV_MAGIC_V1)
1656 objs = &lmm->lmm_objects[count];
1658 objs = &((struct lov_mds_md_v3 *)lmm)->
1661 gap = ea_off - count;
1664 LASSERT(buf->lb_len >= lov_mds_md_size(count, magic));
1667 memset(objs, 0, gap * sizeof(*objs));
1668 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
1672 lmm->lmm_layout_gen =
1673 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
1677 lmm->lmm_stripe_count = cpu_to_le16(count);
1678 rc = lfsck_layout_refill_lovea(env, handle, parent, cfid, buf, objs,
1681 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
1682 DFID": parent "DFID", OST-index %u, stripe-index %u, fl %d, "
1683 "reset %s, %s LOV EA hole: rc = %d\n",
1684 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
1685 ost_idx, ea_off, fl, reset ? "yes" : "no",
1686 hole ? "with" : "without", rc);
1691 static int __lfsck_layout_update_pfid(const struct lu_env *env,
1692 struct dt_object *child,
1693 const struct lu_fid *pfid, __u32 offset)
1695 struct dt_device *dev = lfsck_obj2dev(child);
1696 struct filter_fid *ff = &lfsck_env_info(env)->lti_new_pfid;
1697 struct thandle *handle;
1698 struct lu_buf buf = { NULL };
1701 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
1702 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
1703 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
1704 * MDT-object's FID::f_ver, instead it is the OST-object index in its
1705 * parent MDT-object's layout EA. */
1706 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
1707 lfsck_buf_init(&buf, ff, sizeof(struct filter_fid));
1709 handle = dt_trans_create(env, dev);
1711 RETURN(PTR_ERR(handle));
1713 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1717 rc = dt_trans_start_local(env, dev, handle);
1721 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
1726 dt_trans_stop(env, dev, handle);
1732 * \retval +1: repaired
1733 * \retval 0: did nothing
1734 * \retval -ve: on error
1736 static int lfsck_layout_update_pfid(const struct lu_env *env,
1737 struct lfsck_component *com,
1738 struct dt_object *parent,
1739 struct lu_fid *cfid,
1740 struct dt_device *cdev, __u32 ea_off)
1742 struct dt_object *child;
1746 child = lfsck_object_find_by_dev(env, cdev, cfid);
1748 RETURN(PTR_ERR(child));
1750 rc = __lfsck_layout_update_pfid(env, child,
1751 lu_object_fid(&parent->do_lu), ea_off);
1752 lfsck_object_put(env, child);
1754 RETURN(rc == 0 ? 1 : rc);
1758 * This function will create the MDT-object with the given (partial) LOV EA.
1760 * Under some data corruption cases, the MDT-object of the file may be lost,
1761 * but its OST-objects, or some of them are there. The layout LFSCK needs to
1762 * re-create the MDT-object with the orphan OST-object(s) information.
1764 * On the other hand, the LFSCK may has created some OST-object for repairing
1765 * dangling LOV EA reference, but as the LFSCK processing, it may find that
1766 * the old OST-object is there and should replace the former new created OST
1767 * object. Unfortunately, some others have modified such newly created object.
1768 * To keep the data (both new and old), the LFSCK will create MDT-object with
1769 * new FID to reference the original OST-object.
1771 * \param[in] env pointer to the thread context
1772 * \param[in] com pointer to the lfsck component
1773 * \param[in] ltd pointer to target device descriptor
1774 * \param[in] rec pointer to the record for the orphan OST-object
1775 * \param[in] cfid pointer to FID for the orphan OST-object
1776 * \param[in] infix additional information, such as the FID for original
1777 * MDT-object and the stripe offset in the LOV EA
1778 * \param[in] type the type for describing why the orphan MDT-object is
1779 * created. The rules are as following:
1781 * type "C": Multiple OST-objects claim the same MDT-object and the
1782 * same slot in the layout EA. Then the LFSCK will create
1783 * new MDT-object(s) to hold the conflict OST-object(s).
1785 * type "N": The orphan OST-object does not know which one was the
1786 * real parent MDT-object, so the LFSCK uses new FID for
1787 * its parent MDT-object.
1789 * type "R": The orphan OST-object knows its parent MDT-object FID,
1790 * but does not know the position (the file name) in the
1793 * type "D": The MDT-object is a directory, it may knows its parent
1794 * but because there is no valid linkEA, the LFSCK cannot
1795 * know where to put it back to the namespace.
1796 * type "O": The MDT-object has no linkEA, and there is no name
1797 * entry that references the MDT-object.
1799 * type "P": The orphan object to be created was a parent directory
1800 * of some MDT-object which linkEA shows that the @orphan
1801 * object is missing.
1803 * The orphan name will be like:
1804 * ${FID}-${infix}-${type}-${conflict_version}
1806 * \param[in] ea_off the stripe offset in the LOV EA
1808 * \retval positive on repaired something
1809 * \retval 0 if needs to repair nothing
1810 * \retval negative error number on failure
1812 static int lfsck_layout_recreate_parent(const struct lu_env *env,
1813 struct lfsck_component *com,
1814 struct lfsck_tgt_desc *ltd,
1815 struct lu_orphan_rec *rec,
1816 struct lu_fid *cfid,
1821 struct lfsck_thread_info *info = lfsck_env_info(env);
1822 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
1823 char *name = info->lti_key;
1824 struct lu_attr *la = &info->lti_la2;
1825 struct dt_object_format *dof = &info->lti_dof;
1826 struct lfsck_instance *lfsck = com->lc_lfsck;
1827 struct lu_fid *pfid = &rec->lor_fid;
1828 struct lu_fid *tfid = &info->lti_fid3;
1829 struct dt_device *dev = lfsck->li_bottom;
1830 struct dt_object *lpf = lfsck->li_lpf_obj;
1831 struct dt_object *pobj = NULL;
1832 struct dt_object *cobj = NULL;
1833 struct thandle *th = NULL;
1834 struct lu_buf *ea_buf = &info->lti_big_buf;
1835 struct lu_buf lov_buf;
1836 struct lfsck_lock_handle *llh = &info->lti_llh;
1837 struct linkea_data ldata = { NULL };
1838 struct lu_buf linkea_buf;
1839 const struct lu_name *pname;
1845 if (unlikely(lpf == NULL))
1846 GOTO(log, rc = -ENXIO);
1848 /* We use two separated transactions to repair the inconsistency.
1850 * 1) create the MDT-object locally.
1851 * 2) update the OST-object's PFID EA if necessary.
1853 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
1854 * updated when the layout LFSCK run next time.
1856 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
1857 * when the layout LFSCK run next time. */
1859 if (fid_is_zero(pfid)) {
1860 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
1864 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
1866 GOTO(log, rc = PTR_ERR(cobj));
1869 pobj = lfsck_object_find_by_dev(env, dev, pfid);
1871 GOTO(log, rc = PTR_ERR(pobj));
1873 LASSERT(infix != NULL);
1874 LASSERT(type != NULL);
1876 memset(la, 0, sizeof(*la));
1877 la->la_uid = rec->lor_uid;
1878 la->la_gid = rec->lor_gid;
1879 la->la_mode = S_IFREG | S_IRUSR;
1880 la->la_valid = LA_MODE | LA_UID | LA_GID;
1882 memset(dof, 0, sizeof(*dof));
1883 dof->dof_type = dt_mode_to_dft(S_IFREG);
1884 /* Because the dof->dof_reg.striped = 0, the LOD will not create
1885 * the stripe(s). The LFSCK will specify the LOV EA via
1886 * lfsck_layout_extend_lovea(). */
1888 size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
1889 if (ea_buf->lb_len < size) {
1890 lu_buf_realloc(ea_buf, size);
1891 if (ea_buf->lb_buf == NULL)
1892 GOTO(log, rc = -ENOMEM);
1897 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
1899 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1900 (const struct dt_key *)name);
1901 if (rc != 0 && rc != -ENOENT)
1905 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
1906 MDS_INODELOCK_UPDATE, LCK_PW);
1910 /* Re-check whether the name conflict with othrs after taken
1912 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
1913 (const struct dt_key *)name);
1914 if (unlikely(rc == 0)) {
1922 pname = lfsck_name_get_const(env, name, strlen(name));
1923 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
1924 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
1928 /* The 1st transaction. */
1929 th = dt_trans_create(env, dev);
1931 GOTO(unlock, rc = PTR_ERR(th));
1933 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
1937 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
1938 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
1939 LU_XATTR_REPLACE, th);
1943 dtrec->rec_fid = pfid;
1944 dtrec->rec_type = S_IFREG;
1945 rc = dt_declare_insert(env, lpf,
1946 (const struct dt_rec *)dtrec,
1947 (const struct dt_key *)name, th);
1951 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1952 ldata.ld_leh->leh_len);
1953 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
1954 XATTR_NAME_LINK, 0, th);
1958 rc = dt_trans_start_local(env, dev, th);
1962 dt_write_lock(env, pobj, 0);
1963 rc = dt_create(env, pobj, la, NULL, dof, th);
1965 rc = lfsck_layout_extend_lovea(env, lfsck, th, pobj, cfid,
1966 &lov_buf, 0, ltd->ltd_index, ea_off, true);
1967 dt_write_unlock(env, pobj);
1971 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
1972 (const struct dt_key *)name, th, 1);
1976 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
1977 if (rc == 0 && cobj != NULL) {
1978 dt_trans_stop(env, dev, th);
1981 /* The 2nd transaction. */
1982 rc = __lfsck_layout_update_pfid(env, cobj, pfid, ea_off);
1989 dt_trans_stop(env, dev, th);
1995 if (cobj != NULL && !IS_ERR(cobj))
1996 lfsck_object_put(env, cobj);
1997 if (pobj != NULL && !IS_ERR(pobj))
1998 lfsck_object_put(env, pobj);
2001 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2002 "recreate the lost MDT-object: parent "DFID
2003 ", child "DFID", OST-index %u, stripe-index %u, "
2004 "infix %s, type %s: rc = %d\n",
2005 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2006 ltd->ltd_index, ea_off, infix, type, rc);
2008 return rc >= 0 ? 1 : rc;
2011 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2012 struct lfsck_component *com,
2013 const struct lu_fid *fid,
2016 struct lfsck_thread_info *info = lfsck_env_info(env);
2017 struct lfsck_request *lr = &info->lti_lr;
2018 struct lfsck_instance *lfsck = com->lc_lfsck;
2019 struct lfsck_tgt_desc *ltd;
2020 struct ptlrpc_request *req;
2021 struct lfsck_request *tmp;
2022 struct obd_export *exp;
2026 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2027 if (unlikely(ltd == NULL))
2031 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2032 GOTO(put, rc = -EOPNOTSUPP);
2034 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2036 GOTO(put, rc = -ENOMEM);
2038 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2040 ptlrpc_request_free(req);
2045 memset(lr, 0, sizeof(*lr));
2046 lr->lr_event = LE_CONDITIONAL_DESTROY;
2047 lr->lr_active = LFSCK_TYPE_LAYOUT;
2050 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2052 ptlrpc_request_set_replen(req);
2054 rc = ptlrpc_queue_wait(req);
2055 ptlrpc_req_finished(req);
2065 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2066 struct lfsck_component *com,
2067 struct lfsck_request *lr)
2069 struct lfsck_thread_info *info = lfsck_env_info(env);
2070 struct lu_attr *la = &info->lti_la;
2071 union ldlm_policy_data *policy = &info->lti_policy;
2072 struct ldlm_res_id *resid = &info->lti_resid;
2073 struct lfsck_instance *lfsck = com->lc_lfsck;
2074 struct dt_device *dev = lfsck->li_bottom;
2075 struct lu_fid *fid = &lr->lr_fid;
2076 struct dt_object *obj;
2077 struct thandle *th = NULL;
2078 struct lustre_handle lh = { 0 };
2083 obj = lfsck_object_find_by_dev(env, dev, fid);
2085 RETURN(PTR_ERR(obj));
2087 dt_read_lock(env, obj, 0);
2088 if (dt_object_exists(obj) == 0 ||
2089 lfsck_is_dead_obj(obj)) {
2090 dt_read_unlock(env, obj);
2092 GOTO(put, rc = -ENOENT);
2095 /* Get obj's attr without lock firstly. */
2096 rc = dt_attr_get(env, obj, la);
2097 dt_read_unlock(env, obj);
2101 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2102 GOTO(put, rc = -ETXTBSY);
2104 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2105 LASSERT(lfsck->li_namespace != NULL);
2107 memset(policy, 0, sizeof(*policy));
2108 policy->l_extent.end = OBD_OBJECT_EOF;
2109 ost_fid_build_resid(fid, resid);
2110 rc = ldlm_cli_enqueue_local(lfsck->li_namespace, resid, LDLM_EXTENT,
2111 policy, LCK_EX, &flags, ldlm_blocking_ast,
2112 ldlm_completion_ast, NULL, NULL, 0,
2113 LVB_T_NONE, NULL, &lh);
2115 GOTO(put, rc = -EIO);
2117 dt_write_lock(env, obj, 0);
2118 /* Get obj's attr within lock again. */
2119 rc = dt_attr_get(env, obj, la);
2123 if (la->la_ctime != 0)
2124 GOTO(unlock, rc = -ETXTBSY);
2126 th = dt_trans_create(env, dev);
2128 GOTO(unlock, rc = PTR_ERR(th));
2130 rc = dt_declare_ref_del(env, obj, th);
2134 rc = dt_declare_destroy(env, obj, th);
2138 rc = dt_trans_start_local(env, dev, th);
2142 rc = dt_ref_del(env, obj, th);
2146 rc = dt_destroy(env, obj, th);
2148 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2149 "OST-object "DFID" that was created for reparing "
2150 "dangling referenced case. But the original missing "
2151 "OST-object is found now.\n",
2152 lfsck_lfsck2name(lfsck), PFID(fid));
2157 dt_trans_stop(env, dev, th);
2160 dt_write_unlock(env, obj);
2161 ldlm_lock_decref(&lh, LCK_EX);
2164 lfsck_object_put(env, obj);
2170 * Some OST-object has occupied the specified layout EA slot.
2171 * Such OST-object may be generated by the LFSCK when repair
2172 * dangling referenced MDT-object, which can be indicated by
2173 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2174 * is true and such OST-object has not been modified yet, we
2175 * will replace it with the orphan OST-object; otherwise the
2176 * LFSCK will create new MDT-object to reference the orphan.
2178 * \retval +1: repaired
2179 * \retval 0: did nothing
2180 * \retval -ve: on error
2182 static int lfsck_layout_conflict_create(const struct lu_env *env,
2183 struct lfsck_component *com,
2184 struct lfsck_tgt_desc *ltd,
2185 struct lu_orphan_rec *rec,
2186 struct dt_object *parent,
2187 struct lu_fid *cfid,
2188 struct lu_buf *ea_buf,
2189 struct lov_ost_data_v1 *slot,
2192 struct lfsck_thread_info *info = lfsck_env_info(env);
2193 struct lu_fid *cfid2 = &info->lti_fid2;
2194 struct ost_id *oi = &info->lti_oi;
2195 struct lov_mds_md_v1 *lmm = ea_buf->lb_buf;
2196 struct dt_device *dev = lfsck_obj2dev(parent);
2197 struct thandle *th = NULL;
2198 struct lustre_handle lh = { 0 };
2199 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2203 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2204 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2208 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2209 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2213 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2214 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2219 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2221 /* If the conflict OST-obejct is not created for fixing dangling
2222 * referenced MDT-object in former LFSCK check/repair, or it has
2223 * been modified by others, then we cannot destroy it. Re-create
2224 * a new MDT-object for the orphan OST-object. */
2225 if (rc == -ETXTBSY) {
2226 /* No need the layout lock on the original parent. */
2227 lfsck_ibits_unlock(&lh, LCK_EX);
2229 fid_zero(&rec->lor_fid);
2230 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2231 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2233 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2234 info->lti_tmpbuf, "C", ea_off);
2239 if (rc != 0 && rc != -ENOENT)
2242 th = dt_trans_create(env, dev);
2244 GOTO(unlock, rc = PTR_ERR(th));
2246 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2247 LU_XATTR_REPLACE, th);
2251 rc = dt_trans_start_local(env, dev, th);
2255 dt_write_lock(env, parent, 0);
2256 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2257 rc = lfsck_layout_refill_lovea(env, th, parent, cfid, ea_buf, slot,
2258 LU_XATTR_REPLACE, ltd->ltd_index);
2259 dt_write_unlock(env, parent);
2264 dt_trans_stop(env, dev, th);
2267 lfsck_ibits_unlock(&lh, LCK_EX);
2270 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2271 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2272 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2273 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2274 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2277 return rc >= 0 ? 1 : rc;
2281 * \retval +1: repaired
2282 * \retval 0: did nothing
2283 * \retval -ve: on error
2285 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2286 struct lfsck_component *com,
2287 struct lfsck_tgt_desc *ltd,
2288 struct lu_orphan_rec *rec,
2289 struct dt_object *parent,
2290 struct lu_fid *cfid,
2291 __u32 ost_idx, __u32 ea_off)
2293 struct lfsck_thread_info *info = lfsck_env_info(env);
2294 struct lu_buf *buf = &info->lti_big_buf;
2295 struct lu_fid *fid = &info->lti_fid2;
2296 struct ost_id *oi = &info->lti_oi;
2297 struct lfsck_instance *lfsck = com->lc_lfsck;
2298 struct dt_device *dt = lfsck_obj2dev(parent);
2299 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2300 struct thandle *handle = NULL;
2302 struct lov_mds_md_v1 *lmm;
2303 struct lov_ost_data_v1 *objs;
2304 struct lustre_handle lh = { 0 };
2311 bool locked = false;
2314 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2315 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2318 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2319 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2320 "stripe-index %u: rc = %d\n",
2321 lfsck_lfsck2name(lfsck), PFID(cfid),
2322 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off, rc);
2329 dt_write_unlock(env, parent);
2333 if (handle != NULL) {
2334 dt_trans_stop(env, dt, handle);
2339 GOTO(unlock_layout, rc);
2342 if (buf->lb_len < lovea_size) {
2343 lu_buf_realloc(buf, lovea_size);
2344 if (buf->lb_buf == NULL)
2345 GOTO(unlock_layout, rc = -ENOMEM);
2348 if (!(bk->lb_param & LPF_DRYRUN)) {
2349 handle = dt_trans_create(env, dt);
2351 GOTO(unlock_layout, rc = PTR_ERR(handle));
2353 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2358 rc = dt_trans_start_local(env, dt, handle);
2363 dt_write_lock(env, parent, 0);
2365 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2366 if (rc == -ERANGE) {
2367 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2370 } else if (rc == -ENODATA || rc == 0) {
2371 lovea_size = lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2372 /* If the declared is not big enough, re-try. */
2373 if (buf->lb_len < lovea_size) {
2377 fl = LU_XATTR_CREATE;
2378 } else if (rc < 0) {
2379 GOTO(unlock_parent, rc);
2380 } else if (unlikely(buf->lb_len == 0)) {
2383 fl = LU_XATTR_REPLACE;
2387 if (fl == LU_XATTR_CREATE) {
2388 if (bk->lb_param & LPF_DRYRUN)
2389 GOTO(unlock_parent, rc = 1);
2391 LASSERT(buf->lb_len >= lovea_size);
2393 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2394 buf, fl, ost_idx, ea_off, false);
2396 GOTO(unlock_parent, rc);
2400 rc1 = lfsck_layout_verify_header(lmm);
2402 /* If the LOV EA crashed, the rebuild it. */
2403 if (rc1 == -EINVAL) {
2404 if (bk->lb_param & LPF_DRYRUN)
2405 GOTO(unlock_parent, rc = 1);
2407 LASSERT(buf->lb_len >= lovea_size);
2409 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2410 buf, fl, ost_idx, ea_off, true);
2412 GOTO(unlock_parent, rc);
2415 /* For other unknown magic/pattern, keep the current LOV EA. */
2417 GOTO(unlock_parent, rc = rc1);
2419 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
2420 * been verified in lfsck_layout_verify_header() already. If some
2421 * new magic introduced in the future, then layout LFSCK needs to
2422 * be updated also. */
2423 magic = le32_to_cpu(lmm->lmm_magic);
2424 if (magic == LOV_MAGIC_V1) {
2425 objs = &lmm->lmm_objects[0];
2427 LASSERT(magic == LOV_MAGIC_V3);
2428 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
2431 count = le16_to_cpu(lmm->lmm_stripe_count);
2433 GOTO(unlock_parent, rc = -EINVAL);
2436 /* Exceed the current end of MDT-object layout EA. Then extend it. */
2437 if (count <= ea_off) {
2438 if (bk->lb_param & LPF_DRYRUN)
2439 GOTO(unlock_parent, rc = 1);
2441 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2442 /* If the declared is not big enough, re-try. */
2443 if (buf->lb_len < lovea_size) {
2448 rc = lfsck_layout_extend_lovea(env, lfsck, handle, parent, cfid,
2449 buf, fl, ost_idx, ea_off, false);
2451 GOTO(unlock_parent, rc);
2454 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
2456 for (i = 0; i < count; i++, objs++) {
2457 /* The MDT-object was created via lfsck_layout_recover_create()
2458 * by others before, and we fill the dummy layout EA. */
2459 if (lovea_slot_is_dummy(objs)) {
2463 if (bk->lb_param & LPF_DRYRUN)
2464 GOTO(unlock_parent, rc = 1);
2466 lmm->lmm_layout_gen =
2467 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2468 rc = lfsck_layout_refill_lovea(env, handle, parent,
2469 cfid, buf, objs, fl,
2472 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
2473 "dummy layout slot for "DFID": parent "DFID
2474 ", OST-index %u, stripe-index %u: rc = %d\n",
2475 lfsck_lfsck2name(lfsck), PFID(cfid),
2476 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
2478 GOTO(unlock_parent, rc);
2481 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2482 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
2484 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
2485 "invalid layout EA at the slot %d, index %u\n",
2486 lfsck_lfsck2name(lfsck),
2487 PFID(lfsck_dto2fid(parent)), i,
2488 le32_to_cpu(objs->l_ost_idx));
2490 GOTO(unlock_parent, rc);
2493 /* It should be rare case, the slot is there, but the LFSCK
2494 * does not handle it during the first-phase cycle scanning. */
2495 if (unlikely(lu_fid_eq(fid, cfid))) {
2497 GOTO(unlock_parent, rc = 0);
2499 /* Rare case that the OST-object index
2500 * does not match the parent MDT-object
2501 * layout EA. We trust the later one. */
2502 if (bk->lb_param & LPF_DRYRUN)
2503 GOTO(unlock_parent, rc = 1);
2505 dt_write_unlock(env, parent);
2507 dt_trans_stop(env, dt, handle);
2508 lfsck_ibits_unlock(&lh, LCK_EX);
2509 rc = lfsck_layout_update_pfid(env, com, parent,
2510 cfid, ltd->ltd_tgt, i);
2512 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
2513 "updated OST-object's pfid for "DFID
2514 ": parent "DFID", OST-index %u, "
2515 "stripe-index %u: rc = %d\n",
2516 lfsck_lfsck2name(lfsck), PFID(cfid),
2517 PFID(lfsck_dto2fid(parent)),
2518 ltd->ltd_index, i, rc);
2525 /* The MDT-object exists, but related layout EA slot is occupied
2527 if (bk->lb_param & LPF_DRYRUN)
2528 GOTO(unlock_parent, rc = 1);
2530 dt_write_unlock(env, parent);
2532 dt_trans_stop(env, dt, handle);
2533 lfsck_ibits_unlock(&lh, LCK_EX);
2534 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1)
2535 objs = &lmm->lmm_objects[ea_off];
2537 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
2538 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
2545 dt_write_unlock(env, parent);
2549 dt_trans_stop(env, dt, handle);
2552 lfsck_ibits_unlock(&lh, LCK_EX);
2557 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
2558 struct lfsck_component *com,
2559 struct lfsck_tgt_desc *ltd,
2560 struct lu_orphan_rec *rec,
2561 struct lu_fid *cfid)
2563 struct lfsck_layout *lo = com->lc_file_ram;
2564 struct lu_fid *pfid = &rec->lor_fid;
2565 struct dt_object *parent = NULL;
2566 __u32 ea_off = pfid->f_stripe_idx;
2570 if (!fid_is_sane(cfid))
2571 GOTO(out, rc = -EINVAL);
2573 if (fid_is_zero(pfid)) {
2574 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2580 if (!fid_is_sane(pfid))
2581 GOTO(out, rc = -EINVAL);
2583 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
2585 GOTO(out, rc = PTR_ERR(parent));
2587 if (unlikely(dt_object_remote(parent) != 0))
2588 GOTO(put, rc = -EXDEV);
2590 if (dt_object_exists(parent) == 0) {
2591 lfsck_object_put(env, parent);
2592 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2597 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
2598 GOTO(put, rc = -EISDIR);
2600 /* The orphan OST-object claims to be the parent's stripe, then
2601 * related dangling record in the trace file is meaningless. */
2602 rc = lfsck_layout_del_dangling_rec(env, com, pfid, ea_off);
2603 if (rc != 0 && rc != -ENOENT)
2606 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
2607 ltd->ltd_index, ea_off);
2613 lfsck_object_put(env, parent);
2615 /* The layout EA is changed, need to be reloaded next time. */
2616 lu_object_put_nocache(env, &parent->do_lu);
2619 down_write(&com->lc_sem);
2620 com->lc_new_scanned++;
2621 com->lc_new_checked++;
2623 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
2625 } else if (rc < 0) {
2626 lo->ll_objs_failed_phase2++;
2628 up_write(&com->lc_sem);
2633 static int lfsck_layout_scan_orphan(const struct lu_env *env,
2634 struct lfsck_component *com,
2635 struct lfsck_tgt_desc *ltd)
2637 struct lfsck_assistant_data *lad = com->lc_data;
2638 struct lfsck_instance *lfsck = com->lc_lfsck;
2639 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2640 struct lfsck_thread_info *info = lfsck_env_info(env);
2641 struct ost_id *oi = &info->lti_oi;
2642 struct lu_fid *fid = &info->lti_fid;
2643 struct dt_object *obj;
2644 const struct dt_it_ops *iops;
2649 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
2650 "scanning for OST%04x\n",
2651 lfsck_lfsck2name(lfsck), ltd->ltd_index);
2653 if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
2654 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
2655 "scanning for OST%04x\n",
2656 lfsck_lfsck2name(lfsck), ltd->ltd_index);
2661 ostid_set_seq(oi, FID_SEQ_IDIF);
2662 ostid_set_id(oi, 0);
2663 rc = ostid_to_fid(fid, oi, ltd->ltd_index);
2667 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
2668 if (unlikely(IS_ERR(obj)))
2669 GOTO(log, rc = PTR_ERR(obj));
2671 rc = obj->do_ops->do_index_try(env, obj,
2672 &dt_lfsck_layout_orphan_features);
2676 iops = &obj->do_index_ops->dio_it;
2677 di = iops->init(env, obj, 0);
2679 GOTO(put, rc = PTR_ERR(di));
2681 rc = iops->load(env, di, 0);
2683 /* -ESRCH means that the orphan OST-objects rbtree has been
2684 * cleanup because of the OSS server restart or other errors. */
2685 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
2690 rc = iops->next(env, di);
2702 struct lu_orphan_rec *rec = &info->lti_rec;
2704 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
2705 unlikely(!thread_is_running(&lfsck->li_thread)))
2708 key = iops->key(env, di);
2709 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
2710 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
2712 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
2713 &com->lc_fid_latest_scanned_phase2);
2714 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
2717 lfsck_control_speed_by_self(com);
2719 rc = iops->next(env, di);
2720 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
2727 iops->fini(env, di);
2729 lfsck_object_put(env, obj);
2732 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
2733 "scanning for OST%04x: rc = %d\n",
2734 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
2736 return rc > 0 ? 0 : rc;
2740 * Repair the MDT-object with dangling LOV EA reference.
2742 * we need to repair the inconsistency according to the users' requirement:
2744 * 1) Keep the inconsistency there and report the inconsistency case,
2745 * then give the chance to the application to find related issues,
2746 * and the users can make the decision about how to handle it with
2747 * more human knownledge. (by default)
2749 * 2) Re-create the missing OST-object with the FID/owner information.
2751 * \param[in] env pointer to the thread context
2752 * \param[in] com the layout LFSCK component
2753 * \param[in] parent the MDT-object with dangling LOV EA reference
2754 * \param[in] child the OST-object to be created
2755 * \param[in] ea_off the offset of the OST-object in the LOV EA
2756 * \param[in] ost_idx the index of OST on which the OST-object resides
2758 * \retval +1 for repair successfully
2759 * \retval 0 for did nothing
2760 * \retval negative error number on failure
2762 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
2763 struct lfsck_component *com,
2764 struct dt_object *parent,
2765 struct dt_object *child,
2766 __u32 ea_off, __u32 ost_idx, bool log)
2768 struct lfsck_thread_info *info = lfsck_env_info(env);
2769 struct filter_fid *ff = &info->lti_new_pfid;
2770 struct dt_object_format *dof = &info->lti_dof;
2771 struct lu_attr *la = &info->lti_la;
2772 struct lfsck_instance *lfsck = com->lc_lfsck;
2773 struct dt_device *dev = lfsck_obj2dev(child);
2774 const struct lu_fid *pfid = lfsck_dto2fid(parent);
2775 const struct lu_fid *cfid = lfsck_dto2fid(child);
2776 struct thandle *handle;
2778 struct lustre_handle lh = { 0 };
2782 if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
2785 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2786 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2791 rc = dt_attr_get(env, parent, la);
2795 la->la_mode = S_IFREG | 0666;
2796 la->la_atime = la->la_mtime = la->la_ctime = 0;
2797 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
2798 LA_ATIME | LA_MTIME | LA_CTIME;
2799 memset(dof, 0, sizeof(*dof));
2800 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2801 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2802 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2803 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2804 * parent MDT-object's layout EA. */
2805 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
2806 buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
2808 handle = dt_trans_create(env, dev);
2810 GOTO(unlock1, rc = PTR_ERR(handle));
2812 rc = dt_declare_create(env, child, la, NULL, dof, handle);
2816 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
2817 LU_XATTR_CREATE, handle);
2821 rc = dt_trans_start_local(env, dev, handle);
2825 dt_read_lock(env, parent, 0);
2826 if (unlikely(lfsck_is_dead_obj(parent)))
2827 GOTO(unlock2, rc = 0);
2829 if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
2830 struct ost_id *oi = &info->lti_oi;
2831 struct lu_fid *tfid = &info->lti_fid2;
2832 struct lu_buf *lovea = &info->lti_big_buf;
2833 struct lov_mds_md_v1 *lmm;
2834 struct lov_ost_data_v1 *objs;
2839 rc = lfsck_layout_get_lovea(env, parent, lovea);
2843 lmm = lovea->lb_buf;
2844 rc = lfsck_layout_verify_header(lmm);
2845 if (unlikely(rc != 0))
2848 count = le16_to_cpu(lmm->lmm_stripe_count);
2849 /* Someone changed the LOV EA, do nothing. */
2850 if (count <= ea_off)
2851 GOTO(unlock2, rc = 0);
2853 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which
2854 * has been verified in lfsck_layout_verify_header() already.
2855 * If some new magic introduced in the future, then the layout
2856 * LFSCK needs to be updated also. */
2857 magic = le32_to_cpu(lmm->lmm_magic);
2858 if (magic == LOV_MAGIC_V1) {
2859 objs = &lmm->lmm_objects[ea_off];
2861 LASSERT(magic == LOV_MAGIC_V3);
2863 objs = &((struct lov_mds_md_v3 *)lmm)->\
2864 lmm_objects[ea_off];
2867 ostid_le_to_cpu(&objs->l_ost_oi, oi);
2868 idx2 = le32_to_cpu(objs->l_ost_idx);
2869 rc = ostid_to_fid(tfid, oi, idx2);
2870 /* Someone changed the LOV EA, do nothing. */
2871 if (rc != 0 || !lu_fid_eq(tfid, cfid))
2875 rc = dt_create(env, child, la, NULL, dof, handle);
2879 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
2885 dt_read_unlock(env, parent);
2888 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
2891 lfsck_ibits_unlock(&lh, LCK_EX);
2895 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2896 "dangling reference for: parent "DFID", child "
2897 DFID", ea_off %u, ost_idx %u, %s: rc = %d\n",
2898 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2900 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
2901 "Create the lost OST-object as required" :
2902 "Keep the MDT-object there by default", rc);
2908 * Repair the MDT-object with dangling LOV EA reference.
2910 * Prepare parameters and call __lfsck_layout_repair_dangling()
2911 * to repair the dangling LOV EA reference.
2913 * \param[in] env pointer to the thread context
2914 * \param[in] com the layout LFSCK component
2915 * \param[in] pfid the MDT-object's FID
2916 * \param[in] cfid the FID for the OST-object to be created
2917 * \param[in] ea_off the offset of the OST-object in the LOV EA
2918 * \param[in] ost_idx the index of OST on which the OST-object resides
2920 * \retval +1 for repair successfully
2921 * \retval 0 for did nothing
2922 * \retval negative error number on failure
2924 static int lfsck_layout_repair_dangling(const struct lu_env *env,
2925 struct lfsck_component *com,
2926 const struct lu_fid *pfid,
2927 const struct lu_fid *cfid,
2928 __u32 ea_off, __u32 ost_idx)
2930 struct lfsck_instance *lfsck = com->lc_lfsck;
2931 struct dt_object *parent = NULL;
2932 struct dt_object *child = NULL;
2933 struct lfsck_tgt_desc *ltd;
2937 parent = lfsck_object_find_bottom(env, lfsck, pfid);
2939 GOTO(log, rc = PTR_ERR(parent));
2941 /* The MDT-object has been removed. */
2942 if (dt_object_exists(parent) == 0)
2945 ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
2946 if (unlikely(ltd == NULL))
2947 GOTO(log, rc = -ENODEV);
2949 child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2951 GOTO(log, rc = PTR_ERR(child));
2953 /* The OST-object has been created. */
2954 if (unlikely(dt_object_exists(child) != 0))
2957 rc = __lfsck_layout_repair_dangling(env, com, parent, child,
2958 ea_off, ost_idx, false);
2963 if (child != NULL && !IS_ERR(child))
2964 lfsck_object_put(env, child);
2966 if (parent != NULL && !IS_ERR(parent))
2967 lfsck_object_put(env, parent);
2970 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
2971 "dangling reference for: parent "DFID", child "
2972 DFID", ea_off %u, ost_idx %u, %s: rc = %d\n",
2973 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2975 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
2976 "Create the lost OST-object as required" :
2977 "Keep the MDT-object there by default", rc);
2982 /* If the OST-object does not recognize the MDT-object as its parent, and
2983 * there is no other MDT-object claims as its parent, then just trust the
2984 * given MDT-object as its parent. So update the OST-object filter_fid. */
2985 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
2986 struct lfsck_component *com,
2987 struct dt_object *parent,
2988 struct lfsck_layout_req *llr,
2991 struct lfsck_thread_info *info = lfsck_env_info(env);
2992 struct filter_fid *pfid = &info->lti_new_pfid;
2993 struct dt_object *child = llr->llr_child;
2994 struct dt_device *dev = lfsck_obj2dev(child);
2995 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
2996 struct thandle *handle;
2998 struct lustre_handle lh = { 0 };
3002 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3003 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3008 pfid->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3009 pfid->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3010 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3011 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3012 * parent MDT-object's layout EA. */
3013 pfid->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3014 buf = lfsck_buf_get(env, pfid, sizeof(struct filter_fid));
3016 handle = dt_trans_create(env, dev);
3018 GOTO(unlock1, rc = PTR_ERR(handle));
3020 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3024 rc = dt_attr_get(env, parent, la);
3028 la->la_valid = LA_UID | LA_GID;
3029 rc = dt_declare_attr_set(env, child, la, handle);
3033 rc = dt_trans_start_local(env, dev, handle);
3037 dt_write_lock(env, parent, 0);
3038 if (unlikely(lfsck_is_dead_obj(parent)))
3039 GOTO(unlock2, rc = 1);
3041 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3045 /* Get the latest parent's owner. */
3046 rc = dt_attr_get(env, parent, la);
3050 la->la_valid = LA_UID | LA_GID;
3051 rc = dt_attr_set(env, child, la, handle);
3056 dt_write_unlock(env, parent);
3059 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3062 lfsck_ibits_unlock(&lh, LCK_EX);
3066 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3067 "unmatched MDT-OST pair for: parent "DFID
3068 ", child "DFID", OST-index %u, stripe-index %u, "
3069 "owner %u/%u: rc = %d\n",
3070 lfsck_lfsck2name(com->lc_lfsck),
3071 PFID(lfsck_dto2fid(parent)),
3072 PFID(lfsck_dto2fid(child)),
3073 llr->llr_ost_idx, llr->llr_lov_idx,
3074 la->la_uid, la->la_gid, rc);
3079 /* If there are more than one MDT-objects claim as the OST-object's parent,
3080 * and the OST-object only recognizes one of them, then we need to generate
3081 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3082 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3083 struct lfsck_component *com,
3084 struct dt_object *parent,
3085 struct lfsck_layout_req *llr,
3089 struct lfsck_thread_info *info = lfsck_env_info(env);
3090 struct dt_allocation_hint *hint = &info->lti_hint;
3091 struct dt_object_format *dof = &info->lti_dof;
3092 struct ost_id *oi = &info->lti_oi;
3093 struct lfsck_instance *lfsck = com->lc_lfsck;
3094 struct dt_device *dev;
3095 struct lu_device *d =
3096 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3097 struct lu_object *o;
3098 struct lu_object *n;
3099 struct dt_object *child = NULL;
3100 struct thandle *handle = NULL;
3101 struct lov_mds_md_v1 *lmm;
3102 struct lov_ost_data_v1 *objs;
3103 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3105 struct lustre_handle lh = { 0 };
3106 struct lu_buf ea_buf;
3112 /* We use two separated transactions to repair the inconsistency.
3114 * 1) create the child (OST-object).
3115 * 2) update the parent LOV EA according to the child's FID.
3117 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3118 * handled as orphan when the layout LFSCK run next time.
3120 * If 1) failed, but 2) succeed, then such OST-object will be re-created
3121 * as dangling referened case when the layout LFSCK run next time. */
3123 /* The 1st transaction. */
3124 o = lu_object_anon(env, d, NULL);
3126 GOTO(log, rc = PTR_ERR(o));
3128 n = lu_object_locate(o->lo_header, d->ld_type);
3129 if (unlikely(n == NULL)) {
3130 lu_object_put_nocache(env, o);
3132 GOTO(log, rc = -EINVAL);
3135 child = container_of(n, struct dt_object, do_lu);
3136 memset(hint, 0, sizeof(*hint));
3137 rc = dt_attr_get(env, parent, la);
3141 la->la_valid = LA_UID | LA_GID;
3142 memset(dof, 0, sizeof(*dof));
3144 dev = lfsck_obj2dev(child);
3145 handle = dt_trans_create(env, dev);
3147 GOTO(log, rc = PTR_ERR(handle));
3149 rc = dt_declare_create(env, child, la, hint, dof, handle);
3153 rc = dt_trans_start_local(env, dev, handle);
3157 rc = dt_create(env, child, la, hint, dof, handle);
3158 dt_trans_stop(env, dev, handle);
3163 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3164 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3169 /* The 2nd transaction. */
3171 /* XXX: Generally, we should use bottom device (OSD) to update parent
3172 * LOV EA. But because the LOD-object still references the wrong
3173 * OSP-object that should be detached after the parent's LOV EA
3174 * refreshed. Unfortunately, there is no suitable API for that.
3175 * So we have to make the LOD to re-load the OSP-object(s) via
3176 * replacing the LOV EA against the LOD-object.
3178 * Once the DNE2 patches have been landed, we can replace the
3179 * LOD device with the OSD device. LU-6230. */
3181 dev = lfsck->li_next;
3182 parent = lfsck_object_locate(dev, parent);
3184 GOTO(log, rc = PTR_ERR(parent));
3186 handle = dt_trans_create(env, dev);
3188 GOTO(log, rc = PTR_ERR(handle));
3190 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3191 LU_XATTR_REPLACE, handle);
3195 rc = dt_trans_start_local(env, dev, handle);
3199 dt_write_lock(env, parent, 0);
3200 if (unlikely(lfsck_is_dead_obj(parent)))
3201 GOTO(unlock, rc = 0);
3203 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
3204 if (unlikely(rc == 0 || rc == -ENODATA || rc == -ERANGE))
3205 GOTO(unlock, rc = 0);
3208 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
3209 * been verified in lfsck_layout_verify_header() already. If some
3210 * new magic introduced in the future, then layout LFSCK needs to
3211 * be updated also. */
3212 magic = le32_to_cpu(lmm->lmm_magic);
3213 if (magic == LOV_MAGIC_V1) {
3214 objs = &lmm->lmm_objects[llr->llr_lov_idx];
3216 LASSERT(magic == LOV_MAGIC_V3);
3218 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
3221 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3222 index = le32_to_cpu(objs->l_ost_idx);
3223 rc = ostid_to_fid(&tfid, oi, index);
3224 /* Someone changed layout during the LFSCK, no need to repair then. */
3225 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
3226 GOTO(unlock, rc = 0);
3228 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3229 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3230 ostid_cpu_to_le(oi, &objs->l_ost_oi);
3231 objs->l_ost_gen = cpu_to_le32(0);
3232 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3233 lfsck_buf_init(&ea_buf, lmm,
3234 lov_mds_md_size(le16_to_cpu(lmm->lmm_stripe_count),
3236 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV,
3237 LU_XATTR_REPLACE, handle);
3239 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3242 dt_write_unlock(env, parent);
3246 dt_trans_stop(env, dev, handle);
3249 lfsck_ibits_unlock(&lh, LCK_EX);
3251 lfsck_object_put(env, child);
3254 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3255 "multiple references for: parent "DFID", OST-index %u, "
3256 "stripe-index %u, owner %u/%u: rc = %d\n",
3257 lfsck_lfsck2name(lfsck), PFID(pfid), llr->llr_ost_idx,
3258 llr->llr_lov_idx, la->la_uid, la->la_gid, rc);
3263 /* If the MDT-object and the OST-object have different owner information,
3264 * then trust the MDT-object, because the normal chown/chgrp handle order
3265 * is from MDT to OST, and it is possible that some chown/chgrp operation
3266 * is partly done. */
3267 static int lfsck_layout_repair_owner(const struct lu_env *env,
3268 struct lfsck_component *com,
3269 struct dt_object *parent,
3270 struct lfsck_layout_req *llr,
3271 struct lu_attr *pla,
3272 const struct lu_attr *cla)
3274 struct lfsck_thread_info *info = lfsck_env_info(env);
3275 struct lu_attr *tla = &info->lti_la2;
3276 struct dt_object *child = llr->llr_child;
3277 struct dt_device *dev = lfsck_obj2dev(child);
3278 struct thandle *handle;
3282 tla->la_uid = pla->la_uid;
3283 tla->la_gid = pla->la_gid;
3284 tla->la_valid = LA_UID | LA_GID;
3285 handle = dt_trans_create(env, dev);
3287 GOTO(log, rc = PTR_ERR(handle));
3289 rc = dt_declare_attr_set(env, child, tla, handle);
3293 rc = dt_trans_start_local(env, dev, handle);
3297 /* Use the dt_object lock to serialize with destroy and attr_set. */
3298 dt_read_lock(env, parent, 0);
3299 if (unlikely(lfsck_is_dead_obj(parent)))
3300 GOTO(unlock, rc = 1);
3302 /* Get the latest parent's owner. */
3303 rc = dt_attr_get(env, parent, pla);
3307 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
3308 if (unlikely(tla->la_uid != pla->la_uid ||
3309 tla->la_gid != pla->la_gid))
3312 rc = dt_attr_set(env, child, tla, handle);
3317 dt_read_unlock(env, parent);
3320 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3324 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3325 "inconsistent file owner for: parent "DFID", child "DFID
3326 ", OST-index %u, stripe-index %u, old owner %u/%u, "
3327 "new owner %u/%u: rc = %d\n",
3328 lfsck_lfsck2name(com->lc_lfsck),
3329 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
3330 llr->llr_ost_idx, llr->llr_lov_idx,
3331 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
3336 /* Check whether the OST-object correctly back points to the
3337 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
3338 static int lfsck_layout_check_parent(const struct lu_env *env,
3339 struct lfsck_component *com,
3340 struct lfsck_assistant_object *lso,
3341 const struct lu_fid *pfid,
3342 const struct lu_fid *cfid,
3343 const struct lu_attr *cla,
3344 struct lfsck_layout_req *llr,
3345 struct lu_buf *lov_ea, __u32 idx)
3347 struct lfsck_thread_info *info = lfsck_env_info(env);
3348 struct lu_buf *buf = &info->lti_big_buf;
3349 struct dt_object *tobj;
3350 struct lov_mds_md_v1 *lmm;
3351 struct lov_ost_data_v1 *objs;
3352 struct lustre_handle lh = { 0 };
3359 if (unlikely(!fid_is_sane(pfid)))
3360 RETURN(LLIT_UNMATCHED_PAIR);
3362 if (lu_fid_eq(pfid, &lso->lso_fid)) {
3363 if (likely(llr->llr_lov_idx == idx))
3366 RETURN(LLIT_UNMATCHED_PAIR);
3369 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
3371 RETURN(PTR_ERR(tobj));
3373 if (dt_object_exists(tobj) == 0 || lfsck_is_dead_obj(tobj) ||
3374 !S_ISREG(lfsck_object_type(tobj)))
3375 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3377 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
3378 * remote one on another MDT. Then check whether the given OST-object
3379 * is in such layout. If yes, it is multiple referenced, otherwise it
3380 * is unmatched referenced case. */
3381 rc = lfsck_layout_get_lovea(env, tobj, buf);
3382 if (rc == 0 || rc == -ENOENT)
3383 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3389 magic = le32_to_cpu(lmm->lmm_magic);
3390 if (magic == LOV_MAGIC_V1) {
3391 objs = &lmm->lmm_objects[0];
3393 LASSERT(magic == LOV_MAGIC_V3);
3394 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3397 count = le16_to_cpu(lmm->lmm_stripe_count);
3398 for (i = 0; i < count; i++, objs++) {
3399 struct lu_fid *tfid = &info->lti_fid2;
3400 struct ost_id *oi = &info->lti_oi;
3403 if (lovea_slot_is_dummy(objs))
3406 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3407 idx2 = le32_to_cpu(objs->l_ost_idx);
3408 rc = ostid_to_fid(tfid, oi, idx2);
3410 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3411 "invalid layout EA at the slot %d, index %u\n",
3412 lfsck_lfsck2name(com->lc_lfsck),
3413 PFID(pfid), i, idx2);
3415 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3418 if (lu_fid_eq(cfid, tfid)) {
3419 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
3420 MDS_INODELOCK_UPDATE |
3421 MDS_INODELOCK_LAYOUT |
3422 MDS_INODELOCK_XATTR,
3427 dt_read_lock(env, tobj, 0);
3429 /* For local MDT-object, re-check existence
3430 * after taken the lock. */
3431 if (!dt_object_remote(tobj)) {
3432 if (dt_object_exists(tobj) == 0 ||
3433 lfsck_is_dead_obj(tobj)) {
3434 rc = LLIT_UNMATCHED_PAIR;
3437 rc = LLIT_MULTIPLE_REFERENCED;
3443 /* For migration case, the new MDT-object and old
3444 * MDT-object may reference the same OST-object at
3445 * some migration internal time.
3447 * For remote MDT-object, the local MDT may not know
3448 * whether it has been removed or not. Try checking
3449 * for a non-existent xattr to check if this object
3450 * has been been removed or not. */
3451 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
3453 if (unlikely(rc == -ENOENT || rc >= 0)) {
3454 rc = LLIT_UNMATCHED_PAIR;
3455 } else if (rc == -ENODATA) {
3457 rc = LLIT_MULTIPLE_REFERENCED;
3464 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
3467 if (lustre_handle_is_used(&lh)) {
3468 dt_read_unlock(env, tobj);
3469 lfsck_ibits_unlock(&lh, LCK_EX);
3473 lfsck_object_put(env, tobj);
3478 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
3479 struct lfsck_component *com,
3480 struct lfsck_assistant_req *lar)
3482 struct lfsck_layout_req *llr =
3483 container_of0(lar, struct lfsck_layout_req, llr_lar);
3484 struct lfsck_assistant_object *lso = lar->lar_parent;
3485 struct lfsck_layout *lo = com->lc_file_ram;
3486 struct lfsck_thread_info *info = lfsck_env_info(env);
3487 struct filter_fid_old *pea = &info->lti_old_pfid;
3488 struct lu_fid *pfid = &info->lti_fid;
3489 struct lu_buf buf = { NULL };
3490 struct dt_object *parent = NULL;
3491 struct dt_object *child = llr->llr_child;
3492 struct lu_attr *pla = &lso->lso_attr;
3493 struct lu_attr *cla = &info->lti_la;
3494 struct lfsck_instance *lfsck = com->lc_lfsck;
3495 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3496 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
3504 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ASSISTANT_DIRECT, cfs_fail_val);
3506 rc = dt_attr_get(env, child, cla);
3507 if (rc == -ENOENT) {
3508 parent = lfsck_assistant_object_load(env, lfsck, lso);
3509 if (IS_ERR(parent)) {
3510 rc = PTR_ERR(parent);
3512 RETURN(rc == -ENOENT ? 0 : rc);
3515 type = LLIT_DANGLING;
3522 lfsck_buf_init(&buf, pea, sizeof(struct filter_fid_old));
3523 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
3524 if (unlikely(rc > 0 && rc != sizeof(struct filter_fid_old) &&
3525 rc != sizeof(struct filter_fid))) {
3526 type = LLIT_UNMATCHED_PAIR;
3530 if (rc < 0 && rc != -ENODATA)
3533 if (rc == 0 || rc == -ENODATA)
3534 GOTO(check_owner, rc = 0);
3536 fid_le_to_cpu(pfid, &pea->ff_parent);
3537 /* Currently, the filter_fid::ff_parent::f_ver is not the
3538 * real parent MDT-object's FID::f_ver, instead it is the
3539 * OST-object index in its parent MDT-object's layout EA. */
3540 idx = pfid->f_stripe_idx;
3542 rc = lfsck_layout_check_parent(env, com, lso, pfid,
3543 lu_object_fid(&child->do_lu),
3544 cla, llr, &buf, idx);
3554 /* Someone may has changed the owner after the parent attr pre-loaded.
3555 * It can be handled later inside the lfsck_layout_repair_owner(). */
3556 if (unlikely(cla->la_uid != pla->la_uid ||
3557 cla->la_gid != pla->la_gid)) {
3558 type = LLIT_INCONSISTENT_OWNER;
3563 if (type == LLIT_NONE)
3566 if (bk->lb_param & LPF_DRYRUN)
3569 if (parent == NULL) {
3570 parent = lfsck_assistant_object_load(env, lfsck, lso);
3571 if (IS_ERR(parent)) {
3572 rc = PTR_ERR(parent);
3583 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
3584 rc = lfsck_layout_ins_dangling_rec(env, com,
3585 lfsck_dto2fid(parent), lfsck_dto2fid(child),
3586 llr->llr_lov_idx, llr->llr_ost_idx);
3588 rc = __lfsck_layout_repair_dangling(env, com, parent,
3589 llr->llr_child, llr->llr_lov_idx,
3590 llr->llr_ost_idx, true);
3592 case LLIT_UNMATCHED_PAIR:
3593 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
3596 case LLIT_MULTIPLE_REFERENCED:
3597 rc = lfsck_layout_repair_multiple_references(env, com, parent,
3600 case LLIT_INCONSISTENT_OWNER:
3601 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
3611 down_write(&com->lc_sem);
3613 struct lfsck_assistant_data *lad = com->lc_data;
3615 if (unlikely(lad->lad_exit)) {
3617 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
3618 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
3619 rc == -EHOSTUNREACH) {
3620 /* If cannot touch the target server,
3621 * mark the LFSCK as INCOMPLETE. */
3622 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
3623 "talk with OST %x: rc = %d\n",
3624 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
3625 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
3626 lo->ll_objs_skipped++;
3629 lfsck_layout_record_failure(env, lfsck, lo);
3631 } else if (rc > 0 && (type != LLIT_DANGLING ||
3632 !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
3633 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
3634 "unknown type = %d\n", type);
3636 lo->ll_objs_repaired[type - 1]++;
3637 if (bk->lb_param & LPF_DRYRUN &&
3638 unlikely(lo->ll_pos_first_inconsistent == 0))
3639 lo->ll_pos_first_inconsistent =
3640 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
3643 up_write(&com->lc_sem);
3645 if (parent != NULL && !IS_ERR(parent))
3646 lfsck_object_put(env, parent);
3652 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
3653 struct lfsck_component *com,
3654 struct dt_object *obj, bool first)
3656 struct lfsck_instance *lfsck = com->lc_lfsck;
3657 struct ptlrpc_thread *thread = &lfsck->li_thread;
3658 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3659 struct lfsck_layout *lo = com->lc_file_ram;
3660 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
3663 struct lu_fid *pfid = &lfsck_env_info(env)->lti_fid3;
3664 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid4;
3670 di = iops->init(env, obj, 0);
3672 RETURN(PTR_ERR(di));
3675 fid_cpu_to_be(pfid, &lo->ll_fid_latest_scanned_phase2);
3678 rc = iops->get(env, di, (const struct dt_key *)pfid);
3683 /* The start one either has been processed or does not exist,
3685 rc = iops->next(env, di);
3691 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3692 unlikely(!thread_is_running(thread)))
3695 key = iops->key(env, di);
3704 fid_be_to_cpu(pfid, (const struct lu_fid *)key);
3705 ea_off = pfid->f_ver;
3707 if (!fid_is_sane(pfid)) {
3712 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
3714 fid_be_to_cpu(cfid, cfid);
3715 ost_idx = cfid->f_ver;
3717 if (!fid_is_sane(cfid)) {
3722 rc = lfsck_layout_repair_dangling(env, com, pfid, cfid,
3727 down_write(&com->lc_sem);
3728 com->lc_new_checked++;
3729 com->lc_new_scanned++;
3731 lo->ll_fid_latest_scanned_phase2 = *pfid;
3734 lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
3736 lo->ll_objs_failed_phase2++;
3737 up_write(&com->lc_sem);
3739 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
3742 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
3743 cfs_time_current())) &&
3744 com->lc_new_checked != 0) {
3745 down_write(&com->lc_sem);
3746 lo->ll_run_time_phase2 +=
3747 cfs_duration_sec(cfs_time_current() +
3748 HALF_SEC - com->lc_time_last_checkpoint);
3749 lo->ll_time_last_checkpoint = cfs_time_current_sec();
3750 lo->ll_objs_checked_phase2 += com->lc_new_checked;
3751 com->lc_new_checked = 0;
3752 lfsck_layout_store(env, com);
3753 up_write(&com->lc_sem);
3755 com->lc_time_last_checkpoint = cfs_time_current();
3756 com->lc_time_next_checkpoint =
3757 com->lc_time_last_checkpoint +
3758 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
3761 lfsck_control_speed_by_self(com);
3762 if (unlikely(!thread_is_running(thread)))
3765 rc = iops->next(env, di);
3774 iops->fini(env, di);
3779 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
3780 struct lfsck_component *com)
3782 struct lfsck_assistant_data *lad = com->lc_data;
3783 struct lfsck_instance *lfsck = com->lc_lfsck;
3784 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3785 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
3786 struct lfsck_tgt_desc *ltd;
3790 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
3791 lfsck_lfsck2name(lfsck));
3793 spin_lock(<ds->ltd_lock);
3794 while (!list_empty(&lad->lad_ost_phase2_list)) {
3795 ltd = list_entry(lad->lad_ost_phase2_list.next,
3796 struct lfsck_tgt_desc,
3797 ltd_layout_phase_list);
3798 list_del_init(<d->ltd_layout_phase_list);
3799 if (bk->lb_param & LPF_OST_ORPHAN) {
3800 spin_unlock(<ds->ltd_lock);
3801 rc = lfsck_layout_scan_orphan(env, com, ltd);
3802 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3805 if (unlikely(lad->lad_exit ||
3806 !thread_is_running(&lfsck->li_thread)))
3808 spin_lock(<ds->ltd_lock);
3812 if (list_empty(&lad->lad_ost_phase1_list))
3816 spin_unlock(<ds->ltd_lock);
3818 if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
3819 struct lfsck_layout *lo = com->lc_file_ram;
3822 com->lc_new_checked = 0;
3823 com->lc_new_scanned = 0;
3824 com->lc_time_last_checkpoint = cfs_time_current();
3825 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
3826 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
3828 i = lfsck_sub_trace_file_fid2idx(
3829 &lo->ll_fid_latest_scanned_phase2);
3830 rc = lfsck_layout_double_scan_one_trace_file(env, com,
3831 com->lc_sub_trace_objs[i].lsto_obj, true);
3832 while (rc > 0 && ++i < LFSCK_STF_COUNT)
3833 rc = lfsck_layout_double_scan_one_trace_file(env, com,
3834 com->lc_sub_trace_objs[i].lsto_obj, false);
3836 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
3837 "at the No. %d trace file: rc = %d\n",
3838 lfsck_lfsck2name(lfsck), i, rc);
3841 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
3842 lfsck_lfsck2name(lfsck), rc);
3848 lfsck_layout_slave_async_interpret(const struct lu_env *env,
3849 struct ptlrpc_request *req,
3852 struct lfsck_layout_slave_async_args *llsaa = args;
3853 struct obd_export *exp = llsaa->llsaa_exp;
3854 struct lfsck_component *com = llsaa->llsaa_com;
3855 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
3856 struct lfsck_layout_slave_data *llsd = com->lc_data;
3857 struct lfsck_reply *lr = NULL;
3861 /* It is probably caused by network trouble, or target crash,
3862 * it will try several times (depends on the obd_timeout, and
3863 * will not less than 3 times). But to make the LFSCK can go
3864 * ahead, we should not try for ever. After some try but still
3865 * hit failure, it will assume that the target exit the LFSCK
3866 * prcoessing and stop try. */
3867 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
3868 int max_try = max_t(int, obd_timeout / 30, 3);
3870 if (++(llst->llst_failures) > max_try)
3876 llst->llst_failures = 0;
3877 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
3878 if (lr->lr_status != LS_SCANNING_PHASE1 &&
3879 lr->lr_status != LS_SCANNING_PHASE2)
3884 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
3885 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
3886 llst->llst_index, lr != NULL ? lr->lr_status : rc,
3887 llst->llst_failures);
3889 lfsck_layout_llst_del(llsd, llst);
3892 lfsck_layout_llst_put(llst);
3893 lfsck_component_put(env, com);
3894 class_export_put(exp);
3899 static int lfsck_layout_async_query(const struct lu_env *env,
3900 struct lfsck_component *com,
3901 struct obd_export *exp,
3902 struct lfsck_layout_slave_target *llst,
3903 struct lfsck_request *lr,
3904 struct ptlrpc_request_set *set)
3906 struct lfsck_layout_slave_async_args *llsaa;
3907 struct ptlrpc_request *req;
3908 struct lfsck_request *tmp;
3912 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
3916 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
3918 ptlrpc_request_free(req);
3922 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3924 ptlrpc_request_set_replen(req);
3926 llsaa = ptlrpc_req_async_args(req);
3927 llsaa->llsaa_exp = exp;
3928 llsaa->llsaa_com = lfsck_component_get(com);
3929 llsaa->llsaa_llst = llst;
3930 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
3931 req->rq_allow_intr = 1;
3932 ptlrpc_set_add_req(set, req);
3937 static int lfsck_layout_async_notify(const struct lu_env *env,
3938 struct obd_export *exp,
3939 struct lfsck_request *lr,
3940 struct ptlrpc_request_set *set)
3942 struct ptlrpc_request *req;
3943 struct lfsck_request *tmp;
3947 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
3951 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
3953 ptlrpc_request_free(req);
3957 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
3959 ptlrpc_request_set_replen(req);
3960 req->rq_allow_intr = 1;
3961 ptlrpc_set_add_req(set, req);
3967 lfsck_layout_slave_query_master(const struct lu_env *env,
3968 struct lfsck_component *com)
3970 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
3971 struct lfsck_instance *lfsck = com->lc_lfsck;
3972 struct lfsck_layout_slave_data *llsd = com->lc_data;
3973 struct lfsck_layout_slave_target *llst;
3974 struct obd_export *exp;
3975 struct ptlrpc_request_set *set;
3980 set = ptlrpc_prep_set();
3982 GOTO(log, rc = -ENOMEM);
3984 memset(lr, 0, sizeof(*lr));
3985 lr->lr_event = LE_QUERY;
3986 lr->lr_active = LFSCK_TYPE_LAYOUT;
3988 llsd->llsd_touch_gen++;
3989 spin_lock(&llsd->llsd_lock);
3990 while (!list_empty(&llsd->llsd_master_list)) {
3991 llst = list_entry(llsd->llsd_master_list.next,
3992 struct lfsck_layout_slave_target,
3994 if (llst->llst_gen == llsd->llsd_touch_gen)
3997 llst->llst_gen = llsd->llsd_touch_gen;
3998 list_move_tail(&llst->llst_list,
3999 &llsd->llsd_master_list);
4000 atomic_inc(&llst->llst_ref);
4001 spin_unlock(&llsd->llsd_lock);
4003 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4006 lfsck_layout_llst_del(llsd, llst);
4007 lfsck_layout_llst_put(llst);
4008 spin_lock(&llsd->llsd_lock);
4012 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
4014 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4015 "query %s for layout: rc = %d\n",
4016 lfsck_lfsck2name(lfsck),
4017 exp->exp_obd->obd_name, rc);
4020 lfsck_layout_llst_put(llst);
4021 class_export_put(exp);
4023 spin_lock(&llsd->llsd_lock);
4025 spin_unlock(&llsd->llsd_lock);
4027 rc = ptlrpc_set_wait(set);
4028 ptlrpc_set_destroy(set);
4030 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4033 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4034 lfsck_lfsck2name(com->lc_lfsck), rc);
4040 lfsck_layout_slave_notify_master(const struct lu_env *env,
4041 struct lfsck_component *com,
4042 enum lfsck_events event, int result)
4044 struct lfsck_layout *lo = com->lc_file_ram;
4045 struct lfsck_instance *lfsck = com->lc_lfsck;
4046 struct lfsck_layout_slave_data *llsd = com->lc_data;
4047 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4048 struct lfsck_layout_slave_target *llst;
4049 struct obd_export *exp;
4050 struct ptlrpc_request_set *set;
4054 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4055 lfsck_lfsck2name(com->lc_lfsck));
4057 set = ptlrpc_prep_set();
4061 memset(lr, 0, sizeof(*lr));
4062 lr->lr_event = event;
4063 lr->lr_flags = LEF_FROM_OST;
4064 lr->lr_status = result;
4065 lr->lr_index = lfsck_dev_idx(lfsck);
4066 lr->lr_active = LFSCK_TYPE_LAYOUT;
4067 lr->lr_flags2 = lo->ll_flags;
4068 llsd->llsd_touch_gen++;
4069 spin_lock(&llsd->llsd_lock);
4070 while (!list_empty(&llsd->llsd_master_list)) {
4071 llst = list_entry(llsd->llsd_master_list.next,
4072 struct lfsck_layout_slave_target,
4074 if (llst->llst_gen == llsd->llsd_touch_gen)
4077 llst->llst_gen = llsd->llsd_touch_gen;
4078 list_move_tail(&llst->llst_list,
4079 &llsd->llsd_master_list);
4080 atomic_inc(&llst->llst_ref);
4081 spin_unlock(&llsd->llsd_lock);
4083 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4086 lfsck_layout_llst_del(llsd, llst);
4087 lfsck_layout_llst_put(llst);
4088 spin_lock(&llsd->llsd_lock);
4092 rc = lfsck_layout_async_notify(env, exp, lr, set);
4094 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4095 "notify %s for layout: rc = %d\n",
4096 lfsck_lfsck2name(lfsck),
4097 exp->exp_obd->obd_name, rc);
4099 lfsck_layout_llst_put(llst);
4100 class_export_put(exp);
4101 spin_lock(&llsd->llsd_lock);
4103 spin_unlock(&llsd->llsd_lock);
4105 ptlrpc_set_wait(set);
4106 ptlrpc_set_destroy(set);
4112 * \ret -ENODATA: unrecognized stripe
4113 * \ret = 0 : recognized stripe
4114 * \ret < 0 : other failures
4116 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4117 struct lfsck_component *com,
4118 struct lu_fid *cfid,
4119 struct lu_fid *pfid)
4121 struct lfsck_thread_info *info = lfsck_env_info(env);
4122 struct lu_buf *buf = &info->lti_big_buf;
4123 struct ost_id *oi = &info->lti_oi;
4124 struct dt_object *obj;
4125 struct lov_mds_md_v1 *lmm;
4126 struct lov_ost_data_v1 *objs;
4127 __u32 idx = pfid->f_stripe_idx;
4135 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4137 RETURN(PTR_ERR(obj));
4139 dt_read_lock(env, obj, 0);
4140 if (unlikely(dt_object_exists(obj) == 0 ||
4141 lfsck_is_dead_obj(obj)))
4142 GOTO(unlock, rc = -ENOENT);
4144 if (!S_ISREG(lfsck_object_type(obj)))
4145 GOTO(unlock, rc = -ENODATA);
4147 rc = lfsck_layout_get_lovea(env, obj, buf);
4152 GOTO(unlock, rc = -ENODATA);
4155 rc = lfsck_layout_verify_header(lmm);
4159 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4160 * been verified in lfsck_layout_verify_header() already. If some
4161 * new magic introduced in the future, then layout LFSCK needs to
4162 * be updated also. */
4163 magic = le32_to_cpu(lmm->lmm_magic);
4164 if (magic == LOV_MAGIC_V1) {
4165 objs = &lmm->lmm_objects[0];
4167 LASSERT(magic == LOV_MAGIC_V3);
4168 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4171 fid_to_ostid(cfid, oi);
4172 count = le16_to_cpu(lmm->lmm_stripe_count);
4173 for (i = 0; i < count; i++, objs++) {
4176 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
4177 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
4178 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
4181 GOTO(unlock, rc = -ENODATA);
4184 dt_read_unlock(env, obj);
4185 lfsck_object_put(env, obj);
4191 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
4192 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
4193 * layout EA from MDT to OST. On one hand, the OST no need to understand
4194 * the layout EA structure; on the other hand, it may cause trouble when
4195 * transfer large layout EA from MDT to OST via normal OUT RPC.
4197 * \ret > 0: unrecognized stripe
4198 * \ret = 0: recognized stripe
4199 * \ret < 0: other failures
4201 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
4202 struct lfsck_component *com,
4203 struct lu_fid *cfid,
4204 struct lu_fid *pfid)
4206 struct lfsck_instance *lfsck = com->lc_lfsck;
4207 struct obd_device *obd = lfsck->li_obd;
4208 struct seq_server_site *ss = lfsck_dev_site(lfsck);
4209 struct obd_export *exp = NULL;
4210 struct ptlrpc_request *req = NULL;
4211 struct lfsck_request *lr;
4212 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
4216 if (unlikely(fid_is_idif(pfid)))
4219 fld_range_set_any(range);
4220 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
4222 RETURN(rc == -ENOENT ? 1 : rc);
4224 if (unlikely(!fld_range_is_mdt(range)))
4227 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
4228 if (unlikely(exp == NULL))
4231 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
4232 GOTO(out, rc = -EOPNOTSUPP);
4234 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4236 GOTO(out, rc = -ENOMEM);
4238 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4240 ptlrpc_request_free(req);
4245 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4246 memset(lr, 0, sizeof(*lr));
4247 lr->lr_event = LE_PAIRS_VERIFY;
4248 lr->lr_active = LFSCK_TYPE_LAYOUT;
4249 lr->lr_fid = *cfid; /* OST-object itself FID. */
4250 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
4252 ptlrpc_request_set_replen(req);
4253 rc = ptlrpc_queue_wait(req);
4254 ptlrpc_req_finished(req);
4256 if (rc == -ENOENT || rc == -ENODATA)
4263 class_export_put(exp);
4268 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
4269 struct lfsck_component *com,
4270 struct lfsck_request *lr)
4272 struct dt_object *obj;
4276 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lr->lr_fid);
4278 GOTO(log, rc = PTR_ERR(obj));
4280 dt_write_lock(env, obj, 0);
4281 if (unlikely(dt_object_exists(obj) == 0 ||
4282 lfsck_is_dead_obj(obj)))
4283 GOTO(unlock, rc = 0);
4285 rc = __lfsck_layout_update_pfid(env, obj, &lr->lr_fid2,
4291 dt_write_unlock(env, obj);
4292 lfsck_object_put(env, obj);
4295 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
4296 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
4297 PFID(&lr->lr_fid), PFID(&lr->lr_fid2), rc);
4304 static void lfsck_layout_slave_quit(const struct lu_env *env,
4305 struct lfsck_component *com);
4307 static int lfsck_layout_reset(const struct lu_env *env,
4308 struct lfsck_component *com, bool init)
4310 struct lfsck_layout *lo = com->lc_file_ram;
4313 down_write(&com->lc_sem);
4315 memset(lo, 0, com->lc_file_size);
4317 __u32 count = lo->ll_success_count;
4318 __u64 last_time = lo->ll_time_last_complete;
4320 memset(lo, 0, com->lc_file_size);
4321 lo->ll_success_count = count;
4322 lo->ll_time_last_complete = last_time;
4325 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
4326 lo->ll_status = LS_INIT;
4328 if (com->lc_lfsck->li_master) {
4329 struct lfsck_assistant_data *lad = com->lc_data;
4331 lad->lad_incomplete = 0;
4332 CFS_RESET_BITMAP(lad->lad_bitmap);
4335 rc = lfsck_layout_store(env, com);
4336 if (rc == 0 && com->lc_lfsck->li_master)
4337 rc = lfsck_load_sub_trace_files(env, com,
4338 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
4339 up_write(&com->lc_sem);
4341 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
4342 lfsck_lfsck2name(com->lc_lfsck), rc);
4347 static void lfsck_layout_fail(const struct lu_env *env,
4348 struct lfsck_component *com, bool new_checked)
4350 struct lfsck_layout *lo = com->lc_file_ram;
4352 down_write(&com->lc_sem);
4354 com->lc_new_checked++;
4355 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
4356 up_write(&com->lc_sem);
4359 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
4360 struct lfsck_component *com, bool init)
4362 struct lfsck_instance *lfsck = com->lc_lfsck;
4363 struct lfsck_layout *lo = com->lc_file_ram;
4367 rc = lfsck_checkpoint_generic(env, com);
4369 return rc > 0 ? 0 : rc;
4372 down_write(&com->lc_sem);
4374 lo->ll_pos_latest_start =
4375 lfsck->li_pos_checkpoint.lp_oit_cookie;
4377 lo->ll_pos_last_checkpoint =
4378 lfsck->li_pos_checkpoint.lp_oit_cookie;
4379 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4380 HALF_SEC - lfsck->li_time_last_checkpoint);
4381 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4382 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4383 com->lc_new_checked = 0;
4386 rc = lfsck_layout_store(env, com);
4387 up_write(&com->lc_sem);
4389 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
4390 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
4391 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
4396 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
4397 struct lfsck_component *com, bool init)
4399 struct lfsck_instance *lfsck = com->lc_lfsck;
4400 struct lfsck_layout *lo = com->lc_file_ram;
4403 if (com->lc_new_checked == 0 && !init)
4406 down_write(&com->lc_sem);
4408 lo->ll_pos_latest_start =
4409 lfsck->li_pos_checkpoint.lp_oit_cookie;
4411 lo->ll_pos_last_checkpoint =
4412 lfsck->li_pos_checkpoint.lp_oit_cookie;
4413 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4414 HALF_SEC - lfsck->li_time_last_checkpoint);
4415 lo->ll_time_last_checkpoint = cfs_time_current_sec();
4416 lo->ll_objs_checked_phase1 += com->lc_new_checked;
4417 com->lc_new_checked = 0;
4420 rc = lfsck_layout_store(env, com);
4421 up_write(&com->lc_sem);
4423 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
4424 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
4425 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
4430 static int lfsck_layout_prep(const struct lu_env *env,
4431 struct lfsck_component *com,
4432 struct lfsck_start *start)
4434 struct lfsck_instance *lfsck = com->lc_lfsck;
4435 struct lfsck_layout *lo = com->lc_file_ram;
4436 struct lfsck_position *pos = &com->lc_pos_start;
4438 fid_zero(&pos->lp_dir_parent);
4439 pos->lp_dir_cookie = 0;
4440 if (lo->ll_status == LS_COMPLETED ||
4441 lo->ll_status == LS_PARTIAL ||
4442 /* To handle orphan, must scan from the beginning. */
4443 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
4446 rc = lfsck_layout_reset(env, com, false);
4448 rc = lfsck_set_param(env, lfsck, start, true);
4451 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
4452 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
4458 down_write(&com->lc_sem);
4459 lo->ll_time_latest_start = cfs_time_current_sec();
4460 spin_lock(&lfsck->li_lock);
4461 if (lo->ll_flags & LF_SCANNED_ONCE) {
4462 if (!lfsck->li_drop_dryrun ||
4463 lo->ll_pos_first_inconsistent == 0) {
4464 lo->ll_status = LS_SCANNING_PHASE2;
4465 list_move_tail(&com->lc_link,
4466 &lfsck->li_list_double_scan);
4467 pos->lp_oit_cookie = 0;
4471 lo->ll_status = LS_SCANNING_PHASE1;
4472 lo->ll_run_time_phase1 = 0;
4473 lo->ll_run_time_phase2 = 0;
4474 lo->ll_objs_checked_phase1 = 0;
4475 lo->ll_objs_checked_phase2 = 0;
4476 lo->ll_objs_failed_phase1 = 0;
4477 lo->ll_objs_failed_phase2 = 0;
4478 for (i = 0; i < LLIT_MAX; i++)
4479 lo->ll_objs_repaired[i] = 0;
4481 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4482 fid_zero(&com->lc_fid_latest_scanned_phase2);
4485 lo->ll_status = LS_SCANNING_PHASE1;
4486 if (!lfsck->li_drop_dryrun ||
4487 lo->ll_pos_first_inconsistent == 0)
4488 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
4490 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
4492 spin_unlock(&lfsck->li_lock);
4493 up_write(&com->lc_sem);
4498 static int lfsck_layout_slave_prep(const struct lu_env *env,
4499 struct lfsck_component *com,
4500 struct lfsck_start_param *lsp)
4502 struct lfsck_layout_slave_data *llsd = com->lc_data;
4503 struct lfsck_instance *lfsck = com->lc_lfsck;
4504 struct lfsck_layout *lo = com->lc_file_ram;
4505 struct lfsck_start *start = lsp->lsp_start;
4508 rc = lfsck_layout_prep(env, com, start);
4512 if (lo->ll_flags & LF_CRASHED_LASTID &&
4513 list_empty(&llsd->llsd_master_list)) {
4514 LASSERT(lfsck->li_out_notify != NULL);
4516 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
4517 LE_LASTID_REBUILDING);
4520 if (!lsp->lsp_index_valid)
4523 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
4524 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
4525 LASSERT(!llsd->llsd_rbtree_valid);
4527 write_lock(&llsd->llsd_rb_lock);
4528 rc = lfsck_rbtree_setup(env, com);
4529 write_unlock(&llsd->llsd_rb_lock);
4532 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
4533 "%llu]\n", lfsck_lfsck2name(lfsck),
4534 com->lc_pos_start.lp_oit_cookie);
4539 static int lfsck_layout_master_prep(const struct lu_env *env,
4540 struct lfsck_component *com,
4541 struct lfsck_start_param *lsp)
4546 rc = lfsck_layout_load_bitmap(env, com);
4548 rc = lfsck_layout_reset(env, com, false);
4550 rc = lfsck_set_param(env, com->lc_lfsck,
4551 lsp->lsp_start, true);
4557 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
4561 rc = lfsck_start_assistant(env, com, lsp);
4566 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
4567 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
4568 com->lc_pos_start.lp_oit_cookie);
4573 /* Pre-fetch the attribute for each stripe in the given layout EA. */
4574 static int lfsck_layout_scan_stripes(const struct lu_env *env,
4575 struct lfsck_component *com,
4576 struct dt_object *parent,
4577 struct lov_mds_md_v1 *lmm)
4579 struct lfsck_thread_info *info = lfsck_env_info(env);
4580 struct lfsck_instance *lfsck = com->lc_lfsck;
4581 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4582 struct lfsck_layout *lo = com->lc_file_ram;
4583 struct lfsck_assistant_data *lad = com->lc_data;
4584 struct lfsck_assistant_object *lso = NULL;
4585 struct lov_ost_data_v1 *objs;
4586 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4587 struct ptlrpc_thread *mthread = &lfsck->li_thread;
4588 struct ptlrpc_thread *athread = &lad->lad_thread;
4589 struct l_wait_info lwi = { 0 };
4597 lfsck_buf_init(&buf, &info->lti_old_pfid,
4598 sizeof(struct filter_fid_old));
4599 count = le16_to_cpu(lmm->lmm_stripe_count);
4600 /* Currently, we only support LOV_MAGIC_V1/LOV_MAGIC_V3 which has
4601 * been verified in lfsck_layout_verify_header() already. If some
4602 * new magic introduced in the future, then layout LFSCK needs to
4603 * be updated also. */
4604 magic = le32_to_cpu(lmm->lmm_magic);
4605 if (magic == LOV_MAGIC_V1) {
4606 objs = &lmm->lmm_objects[0];
4608 LASSERT(magic == LOV_MAGIC_V3);
4609 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4612 for (i = 0; i < count; i++, objs++) {
4613 struct lu_fid *fid = &info->lti_fid;
4614 struct ost_id *oi = &info->lti_oi;
4615 struct lfsck_layout_req *llr;
4616 struct lfsck_tgt_desc *tgt = NULL;
4617 struct dt_object *cobj = NULL;
4619 bool wakeup = false;
4621 if (unlikely(lovea_slot_is_dummy(objs)))
4624 l_wait_event(mthread->t_ctl_waitq,
4625 lad->lad_prefetched < bk->lb_async_windows ||
4626 !thread_is_running(mthread) ||
4627 thread_is_stopped(athread),
4630 if (unlikely(!thread_is_running(mthread)) ||
4631 thread_is_stopped(athread))
4634 if (unlikely(lfsck_is_dead_obj(parent)))
4637 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4638 index = le32_to_cpu(objs->l_ost_idx);
4639 rc = ostid_to_fid(fid, oi, index);
4641 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
4642 ": "DOSTID", idx:%u\n", lfsck_lfsck2name(lfsck),
4643 PFID(lfsck_dto2fid(parent)), POSTID(oi), index);
4647 tgt = lfsck_tgt_get(ltds, index);
4648 if (unlikely(tgt == NULL)) {
4649 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
4650 "did not join the layout LFSCK\n",
4651 lfsck_lfsck2name(lfsck), index);
4652 lfsck_lad_set_bitmap(env, com, index);
4656 /* There is potential deadlock race condition between object
4657 * destroy and layout LFSCK. Consider the following scenario:
4659 * 1) The LFSCK thread obtained the parent object firstly, at
4660 * that time, the parent object has not been destroyed yet.
4662 * 2) One RPC service thread destroyed the parent and all its
4663 * children objects. Because the LFSCK is referencing the
4664 * parent object, then the parent object will be marked as
4665 * dying in RAM. On the other hand, the parent object is
4666 * referencing all its children objects, then all children
4667 * objects will be marked as dying in RAM also.
4669 * 3) The LFSCK thread tries to find some child object with
4670 * the parent object referenced. Then it will find that the
4671 * child object is dying. According to the object visibility
4672 * rules: the object with dying flag cannot be returned to
4673 * others. So the LFSCK thread has to wait until the dying
4674 * object has been purged from RAM, then it can allocate a
4675 * new object (with the same FID) in RAM. Unfortunately, the
4676 * LFSCK thread itself is referencing the parent object, and
4677 * cause the parent object cannot be purged, then cause the
4678 * child object cannot be purged also. So the LFSCK thread
4679 * will fall into deadlock.
4681 * We introduce non-blocked version lu_object_find() to allow
4682 * the LFSCK thread to return failure immediately (instead of
4683 * wait) when it finds dying (child) object, then the LFSCK
4684 * thread can check whether the parent object is dying or not.
4685 * So avoid above deadlock. LU-5395 */
4686 cobj = lfsck_object_find_by_dev_nowait(env, tgt->ltd_tgt, fid);
4688 if (lfsck_is_dead_obj(parent)) {
4698 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_ASSISTANT_DIRECT)) {
4699 rc = dt_declare_attr_get(env, cobj);
4703 rc = dt_declare_xattr_get(env, cobj, &buf,
4710 struct lu_attr *attr = &info->lti_la;
4712 rc = dt_attr_get(env, parent, attr);
4716 lso = lfsck_assistant_object_init(env,
4717 lfsck_dto2fid(parent), attr,
4718 lfsck->li_pos_current.lp_oit_cookie, false);
4727 llr = lfsck_layout_assistant_req_init(lso, cobj, index, i);
4734 spin_lock(&lad->lad_lock);
4735 if (lad->lad_assistant_status < 0) {
4736 spin_unlock(&lad->lad_lock);
4737 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
4739 RETURN(lad->lad_assistant_status);
4742 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
4743 if (lad->lad_prefetched == 0)
4746 lad->lad_prefetched++;
4747 spin_unlock(&lad->lad_lock);
4749 wake_up_all(&athread->t_ctl_waitq);
4752 down_write(&com->lc_sem);
4753 com->lc_new_checked++;
4755 lfsck_layout_record_failure(env, lfsck, lo);
4756 up_write(&com->lc_sem);
4758 if (cobj != NULL && !IS_ERR(cobj))
4759 lfsck_object_put(env, cobj);
4761 if (likely(tgt != NULL))
4764 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4772 lfsck_assistant_object_put(env, lso);
4777 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
4778 * the OST-object's attribute and generate an structure lfsck_layout_req on the
4779 * list ::lad_req_list.
4781 * For each request on above list, the lfsck_layout_assistant thread compares
4782 * the OST side attribute with local attribute, if inconsistent, then repair it.
4784 * All above processing is async mode with pipeline. */
4785 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
4786 struct lfsck_component *com,
4787 struct dt_object *obj)
4789 struct lfsck_thread_info *info = lfsck_env_info(env);
4790 struct ost_id *oi = &info->lti_oi;
4791 struct lfsck_layout *lo = com->lc_file_ram;
4792 struct lfsck_assistant_data *lad = com->lc_data;
4793 struct lfsck_instance *lfsck = com->lc_lfsck;
4794 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4795 struct thandle *handle = NULL;
4796 struct lu_buf *buf = &info->lti_big_buf;
4797 struct lov_mds_md_v1 *lmm = NULL;
4798 struct dt_device *dev = lfsck_obj2dev(obj);
4799 struct lustre_handle lh = { 0 };
4800 struct lu_buf ea_buf = { NULL };
4803 bool locked = false;
4804 bool stripe = false;
4805 bool bad_oi = false;
4808 if (!S_ISREG(lfsck_object_type(obj)))
4811 if (lad->lad_assistant_status < 0)
4812 GOTO(out, rc = -ESRCH);
4814 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
4815 lmm_oi_cpu_to_le(oi, oi);
4816 dt_read_lock(env, obj, 0);
4820 if (dt_object_exists(obj) == 0 ||
4821 lfsck_is_dead_obj(obj))
4824 rc = lfsck_layout_get_lovea(env, obj, buf);
4830 rc = lfsck_layout_verify_header(lmm);
4831 /* If the LOV EA crashed, then it is possible to be rebuilt later
4832 * when handle orphan OST-objects. */
4836 if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0)
4837 GOTO(out, stripe = true);
4839 /* Inconsistent lmm_oi, should be repaired. */
4843 if (bk->lb_param & LPF_DRYRUN) {
4844 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4846 GOTO(out, stripe = true);
4849 if (!lustre_handle_is_used(&lh)) {
4850 dt_read_unlock(env, obj);
4852 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4853 MDS_INODELOCK_LAYOUT |
4854 MDS_INODELOCK_XATTR, LCK_EX);
4858 handle = dt_trans_create(env, dev);
4860 GOTO(out, rc = PTR_ERR(handle));
4862 lfsck_buf_init(&ea_buf, lmm, size);
4863 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4864 LU_XATTR_REPLACE, handle);
4868 rc = dt_trans_start_local(env, dev, handle);
4872 dt_write_lock(env, obj, 0);
4878 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
4879 LU_XATTR_REPLACE, handle);
4883 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4885 GOTO(out, stripe = true);
4889 if (lustre_handle_is_used(&lh))
4890 dt_write_unlock(env, obj);
4892 dt_read_unlock(env, obj);
4895 if (handle != NULL && !IS_ERR(handle))
4896 dt_trans_stop(env, dev, handle);
4898 lfsck_ibits_unlock(&lh, LCK_EX);
4901 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
4902 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
4903 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
4904 PFID(lfsck_dto2fid(obj)), rc);
4907 rc = lfsck_layout_scan_stripes(env, com, obj, lmm);
4909 down_write(&com->lc_sem);
4910 com->lc_new_checked++;
4912 lfsck_layout_record_failure(env, lfsck, lo);
4913 up_write(&com->lc_sem);
4919 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
4920 struct lfsck_component *com,
4921 struct dt_object *obj)
4923 struct lfsck_instance *lfsck = com->lc_lfsck;
4924 struct lfsck_layout *lo = com->lc_file_ram;
4925 const struct lu_fid *fid = lfsck_dto2fid(obj);
4926 struct lfsck_layout_slave_data *llsd = com->lc_data;
4927 struct lfsck_layout_seq *lls;
4933 LASSERT(llsd != NULL);
4935 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
4936 cfs_fail_val == lfsck_dev_idx(lfsck)) {
4937 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(1),
4939 struct ptlrpc_thread *thread = &lfsck->li_thread;
4941 l_wait_event(thread->t_ctl_waitq,
4942 !thread_is_running(thread),
4946 lfsck_rbtree_update_bitmap(env, com, fid, false);
4948 down_write(&com->lc_sem);
4949 if (fid_is_idif(fid))
4951 else if (!fid_is_norm(fid) ||
4952 !fid_is_for_ostobj(env, lfsck, obj, fid))
4953 GOTO(unlock, rc = 0);
4956 com->lc_new_checked++;
4958 lls = lfsck_layout_seq_lookup(llsd, seq);
4961 if (unlikely(lls == NULL))
4962 GOTO(unlock, rc = -ENOMEM);
4964 INIT_LIST_HEAD(&lls->lls_list);
4966 rc = lfsck_layout_lastid_load(env, com, lls);
4968 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4969 "load LAST_ID for %#llx: rc = %d\n",
4970 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
4971 lo->ll_objs_failed_phase1++;
4976 lfsck_layout_seq_insert(llsd, lls);
4979 if (unlikely(fid_is_last_id(fid)))
4980 GOTO(unlock, rc = 0);
4982 if (fid_is_idif(fid))
4983 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
4987 if (oid > lls->lls_lastid_known)
4988 lls->lls_lastid_known = oid;
4990 if (oid > lls->lls_lastid) {
4991 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
4992 /* OFD may create new objects during LFSCK scanning. */
4993 rc = lfsck_layout_lastid_reload(env, com, lls);
4994 if (unlikely(rc != 0)) {
4995 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
4996 "reload LAST_ID for %#llx: rc = %d\n",
4997 lfsck_lfsck2name(com->lc_lfsck),
5003 if (oid <= lls->lls_lastid ||
5004 lo->ll_flags & LF_CRASHED_LASTID)
5005 GOTO(unlock, rc = 0);
5007 LASSERT(lfsck->li_out_notify != NULL);
5009 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5010 LE_LASTID_REBUILDING);
5011 lo->ll_flags |= LF_CRASHED_LASTID;
5013 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
5014 "LAST_ID file (2) for the sequence %#llx"
5015 ", old value %llu, known value %llu\n",
5016 lfsck_lfsck2name(lfsck), lls->lls_seq,
5017 lls->lls_lastid, oid);
5020 lls->lls_lastid = oid;
5024 GOTO(unlock, rc = 0);
5027 up_write(&com->lc_sem);
5032 static int lfsck_layout_exec_dir(const struct lu_env *env,
5033 struct lfsck_component *com,
5034 struct lfsck_assistant_object *lso,
5035 struct lu_dirent *ent, __u16 type)
5040 static int lfsck_layout_master_post(const struct lu_env *env,
5041 struct lfsck_component *com,
5042 int result, bool init)
5044 struct lfsck_instance *lfsck = com->lc_lfsck;
5045 struct lfsck_layout *lo = com->lc_file_ram;
5049 lfsck_post_generic(env, com, &result);
5051 down_write(&com->lc_sem);
5052 spin_lock(&lfsck->li_lock);
5054 lo->ll_pos_last_checkpoint =
5055 lfsck->li_pos_checkpoint.lp_oit_cookie;
5058 if (lo->ll_flags & LF_INCOMPLETE)
5059 lo->ll_status = LS_PARTIAL;
5061 lo->ll_status = LS_SCANNING_PHASE2;
5062 lo->ll_flags |= LF_SCANNED_ONCE;
5063 lo->ll_flags &= ~LF_UPGRADE;
5064 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5065 } else if (result == 0) {
5066 if (lfsck->li_status != 0)
5067 lo->ll_status = lfsck->li_status;
5069 lo->ll_status = LS_STOPPED;
5070 if (lo->ll_status != LS_PAUSED)
5071 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5073 lo->ll_status = LS_FAILED;
5074 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5076 spin_unlock(&lfsck->li_lock);
5079 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5080 HALF_SEC - lfsck->li_time_last_checkpoint);
5081 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5082 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5083 com->lc_new_checked = 0;
5086 rc = lfsck_layout_store(env, com);
5087 up_write(&com->lc_sem);
5089 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
5090 lfsck_lfsck2name(lfsck), rc);
5095 static int lfsck_layout_slave_post(const struct lu_env *env,
5096 struct lfsck_component *com,
5097 int result, bool init)
5099 struct lfsck_instance *lfsck = com->lc_lfsck;
5100 struct lfsck_layout *lo = com->lc_file_ram;
5104 down_write(&com->lc_sem);
5105 rc = lfsck_layout_lastid_store(env, com);
5109 LASSERT(lfsck->li_out_notify != NULL);
5111 spin_lock(&lfsck->li_lock);
5113 lo->ll_pos_last_checkpoint =
5114 lfsck->li_pos_checkpoint.lp_oit_cookie;
5117 lo->ll_status = LS_SCANNING_PHASE2;
5118 lo->ll_flags |= LF_SCANNED_ONCE;
5119 if (lo->ll_flags & LF_CRASHED_LASTID) {
5121 lo->ll_flags &= ~LF_CRASHED_LASTID;
5123 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
5124 "crashed LAST_ID files successfully\n",
5125 lfsck_lfsck2name(lfsck));
5127 lo->ll_flags &= ~LF_UPGRADE;
5128 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5129 } else if (result == 0) {
5130 if (lfsck->li_status != 0)
5131 lo->ll_status = lfsck->li_status;
5133 lo->ll_status = LS_STOPPED;
5134 if (lo->ll_status != LS_PAUSED)
5135 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5137 lo->ll_status = LS_FAILED;
5138 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5140 spin_unlock(&lfsck->li_lock);
5143 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5147 lo->ll_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
5148 HALF_SEC - lfsck->li_time_last_checkpoint);
5149 lo->ll_time_last_checkpoint = cfs_time_current_sec();
5150 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5151 com->lc_new_checked = 0;
5154 rc = lfsck_layout_store(env, com);
5155 up_write(&com->lc_sem);
5157 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
5159 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
5160 lfsck_lfsck2name(lfsck), rc);
5165 static void lfsck_layout_dump(const struct lu_env *env,
5166 struct lfsck_component *com, struct seq_file *m)
5168 struct lfsck_instance *lfsck = com->lc_lfsck;
5169 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5170 struct lfsck_layout *lo = com->lc_file_ram;
5172 down_read(&com->lc_sem);
5173 seq_printf(m, "name: lfsck_layout\n"
5179 lfsck_status2name(lo->ll_status));
5181 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
5183 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
5185 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
5187 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
5189 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
5191 seq_printf(m, "latest_start_position: %llu\n"
5192 "last_checkpoint_position: %llu\n"
5193 "first_failure_position: %llu\n",
5194 lo->ll_pos_latest_start,
5195 lo->ll_pos_last_checkpoint,
5196 lo->ll_pos_first_inconsistent);
5198 seq_printf(m, "success_count: %u\n"
5199 "repaired_dangling: %llu\n"
5200 "repaired_unmatched_pair: %llu\n"
5201 "repaired_multiple_referenced: %llu\n"
5202 "repaired_orphan: %llu\n"
5203 "repaired_inconsistent_owner: %llu\n"
5204 "repaired_others: %llu\n"
5206 "failed_phase1: %llu\n"
5207 "failed_phase2: %llu\n",
5208 lo->ll_success_count,
5209 lo->ll_objs_repaired[LLIT_DANGLING - 1],
5210 lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
5211 lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
5212 lo->ll_objs_repaired[LLIT_ORPHAN - 1],
5213 lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
5214 lo->ll_objs_repaired[LLIT_OTHERS - 1],
5215 lo->ll_objs_skipped,
5216 lo->ll_objs_failed_phase1,
5217 lo->ll_objs_failed_phase2);
5219 if (lo->ll_status == LS_SCANNING_PHASE1) {
5221 const struct dt_it_ops *iops;
5222 cfs_duration_t duration = cfs_time_current() -
5223 lfsck->li_time_last_checkpoint;
5224 __u64 checked = lo->ll_objs_checked_phase1 +
5225 com->lc_new_checked;
5226 __u64 speed = checked;
5227 __u64 new_checked = com->lc_new_checked *
5228 msecs_to_jiffies(MSEC_PER_SEC);
5229 __u32 rtime = lo->ll_run_time_phase1 +
5230 cfs_duration_sec(duration + HALF_SEC);
5233 do_div(new_checked, duration);
5235 do_div(speed, rtime);
5236 seq_printf(m, "checked_phase1: %llu\n"
5237 "checked_phase2: %llu\n"
5238 "run_time_phase1: %u seconds\n"
5239 "run_time_phase2: %u seconds\n"
5240 "average_speed_phase1: %llu items/sec\n"
5241 "average_speed_phase2: N/A\n"
5242 "real-time_speed_phase1: %llu items/sec\n"
5243 "real-time_speed_phase2: N/A\n",
5245 lo->ll_objs_checked_phase2,
5247 lo->ll_run_time_phase2,
5251 LASSERT(lfsck->li_di_oit != NULL);
5253 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
5255 /* The low layer otable-based iteration position may NOT
5256 * exactly match the layout-based directory traversal
5257 * cookie. Generally, it is not a serious issue. But the
5258 * caller should NOT make assumption on that. */
5259 pos = iops->store(env, lfsck->li_di_oit);
5260 if (!lfsck->li_current_oit_processed)
5262 seq_printf(m, "current_position: %llu\n", pos);
5264 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
5265 cfs_duration_t duration = cfs_time_current() -
5266 com->lc_time_last_checkpoint;
5267 __u64 checked = lo->ll_objs_checked_phase2 +
5268 com->lc_new_checked;
5269 __u64 speed1 = lo->ll_objs_checked_phase1;
5270 __u64 speed2 = checked;
5271 __u64 new_checked = com->lc_new_checked *
5272 msecs_to_jiffies(MSEC_PER_SEC);
5273 __u32 rtime = lo->ll_run_time_phase2 +
5274 cfs_duration_sec(duration + HALF_SEC);
5277 do_div(new_checked, duration);
5278 if (lo->ll_run_time_phase1 != 0)
5279 do_div(speed1, lo->ll_run_time_phase1);
5281 do_div(speed2, rtime);
5282 seq_printf(m, "checked_phase1: %llu\n"
5283 "checked_phase2: %llu\n"
5284 "run_time_phase1: %u seconds\n"
5285 "run_time_phase2: %u seconds\n"
5286 "average_speed_phase1: %llu items/sec\n"
5287 "average_speed_phase2: %llu items/sec\n"
5288 "real-time_speed_phase1: N/A\n"
5289 "real-time_speed_phase2: %llu items/sec\n"
5290 "current_position: "DFID"\n",
5291 lo->ll_objs_checked_phase1,
5293 lo->ll_run_time_phase1,
5298 PFID(&com->lc_fid_latest_scanned_phase2));
5300 __u64 speed1 = lo->ll_objs_checked_phase1;
5301 __u64 speed2 = lo->ll_objs_checked_phase2;
5303 if (lo->ll_run_time_phase1 != 0)
5304 do_div(speed1, lo->ll_run_time_phase1);
5305 if (lo->ll_run_time_phase2 != 0)
5306 do_div(speed2, lo->ll_run_time_phase2);
5307 seq_printf(m, "checked_phase1: %llu\n"
5308 "checked_phase2: %llu\n"
5309 "run_time_phase1: %u seconds\n"
5310 "run_time_phase2: %u seconds\n"
5311 "average_speed_phase1: %llu items/sec\n"
5312 "average_speed_phase2: %llu objs/sec\n"
5313 "real-time_speed_phase1: N/A\n"
5314 "real-time_speed_phase2: N/A\n"
5315 "current_position: N/A\n",
5316 lo->ll_objs_checked_phase1,
5317 lo->ll_objs_checked_phase2,
5318 lo->ll_run_time_phase1,
5319 lo->ll_run_time_phase2,
5324 up_read(&com->lc_sem);
5327 static int lfsck_layout_master_double_scan(const struct lu_env *env,
5328 struct lfsck_component *com)
5330 struct lfsck_layout *lo = com->lc_file_ram;
5331 struct lfsck_assistant_data *lad = com->lc_data;
5332 struct lfsck_instance *lfsck = com->lc_lfsck;
5333 struct lfsck_tgt_descs *ltds;
5334 struct lfsck_tgt_desc *ltd;
5335 struct lfsck_tgt_desc *next;
5338 rc = lfsck_double_scan_generic(env, com, lo->ll_status);
5340 if (thread_is_stopped(&lad->lad_thread)) {
5341 LASSERT(list_empty(&lad->lad_req_list));
5342 LASSERT(list_empty(&lad->lad_ost_phase1_list));
5343 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
5345 ltds = &lfsck->li_ost_descs;
5346 spin_lock(<ds->ltd_lock);
5347 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5348 ltd_layout_phase_list) {
5349 list_del_init(<d->ltd_layout_phase_list);
5351 spin_unlock(<ds->ltd_lock);
5353 ltds = &lfsck->li_mdt_descs;
5354 spin_lock(<ds->ltd_lock);
5355 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5356 ltd_layout_phase_list) {
5357 list_del_init(<d->ltd_layout_phase_list);
5359 spin_unlock(<ds->ltd_lock);
5365 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
5366 struct lfsck_component *com)
5368 struct lfsck_instance *lfsck = com->lc_lfsck;
5369 struct lfsck_layout_slave_data *llsd = com->lc_data;
5370 struct lfsck_layout *lo = com->lc_file_ram;
5371 struct ptlrpc_thread *thread = &lfsck->li_thread;
5375 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
5376 lfsck_lfsck2name(lfsck));
5378 atomic_inc(&lfsck->li_double_scan_count);
5380 if (lo->ll_flags & LF_INCOMPLETE)
5383 com->lc_new_checked = 0;
5384 com->lc_new_scanned = 0;
5385 com->lc_time_last_checkpoint = cfs_time_current();
5386 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
5387 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5390 struct l_wait_info lwi = LWI_TIMEOUT(cfs_time_seconds(30),
5393 rc = lfsck_layout_slave_query_master(env, com);
5394 if (list_empty(&llsd->llsd_master_list)) {
5395 if (unlikely(!thread_is_running(thread)))
5406 rc = l_wait_event(thread->t_ctl_waitq,
5407 !thread_is_running(thread) ||
5408 lo->ll_flags & LF_INCOMPLETE ||
5409 list_empty(&llsd->llsd_master_list),
5411 if (unlikely(!thread_is_running(thread)))
5414 if (lo->ll_flags & LF_INCOMPLETE)
5417 if (rc == -ETIMEDOUT)
5420 GOTO(done, rc = (rc < 0 ? rc : 1));
5424 rc = lfsck_layout_double_scan_result(env, com, rc);
5425 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
5426 (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
5427 lfsck_layout_slave_quit(env, com);
5428 if (atomic_dec_and_test(&lfsck->li_double_scan_count))
5429 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5431 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
5432 "status %d: rc = %d\n",
5433 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
5438 static void lfsck_layout_master_data_release(const struct lu_env *env,
5439 struct lfsck_component *com)
5441 struct lfsck_assistant_data *lad = com->lc_data;
5442 struct lfsck_instance *lfsck = com->lc_lfsck;
5443 struct lfsck_tgt_descs *ltds;
5444 struct lfsck_tgt_desc *ltd;
5445 struct lfsck_tgt_desc *next;
5447 LASSERT(lad != NULL);
5448 LASSERT(thread_is_init(&lad->lad_thread) ||
5449 thread_is_stopped(&lad->lad_thread));
5450 LASSERT(list_empty(&lad->lad_req_list));
5452 com->lc_data = NULL;
5454 ltds = &lfsck->li_ost_descs;
5455 spin_lock(<ds->ltd_lock);
5456 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5457 ltd_layout_phase_list) {
5458 list_del_init(<d->ltd_layout_phase_list);
5460 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5461 ltd_layout_phase_list) {
5462 list_del_init(<d->ltd_layout_phase_list);
5464 list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
5466 list_del_init(<d->ltd_layout_list);
5468 spin_unlock(<ds->ltd_lock);
5470 ltds = &lfsck->li_mdt_descs;
5471 spin_lock(<ds->ltd_lock);
5472 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5473 ltd_layout_phase_list) {
5474 list_del_init(<d->ltd_layout_phase_list);
5476 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5477 ltd_layout_phase_list) {
5478 list_del_init(<d->ltd_layout_phase_list);
5480 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
5482 list_del_init(<d->ltd_layout_list);
5484 spin_unlock(<ds->ltd_lock);
5486 if (likely(lad->lad_bitmap != NULL))
5487 CFS_FREE_BITMAP(lad->lad_bitmap);
5492 static void lfsck_layout_slave_data_release(const struct lu_env *env,
5493 struct lfsck_component *com)
5495 struct lfsck_layout_slave_data *llsd = com->lc_data;
5497 lfsck_layout_slave_quit(env, com);
5498 com->lc_data = NULL;
5502 static void lfsck_layout_master_quit(const struct lu_env *env,
5503 struct lfsck_component *com)
5505 struct lfsck_assistant_data *lad = com->lc_data;
5506 struct lfsck_instance *lfsck = com->lc_lfsck;
5507 struct lfsck_tgt_descs *ltds;
5508 struct lfsck_tgt_desc *ltd;
5509 struct lfsck_tgt_desc *next;
5511 LASSERT(lad != NULL);
5513 lfsck_quit_generic(env, com);
5515 LASSERT(thread_is_init(&lad->lad_thread) ||
5516 thread_is_stopped(&lad->lad_thread));
5517 LASSERT(list_empty(&lad->lad_req_list));
5519 ltds = &lfsck->li_ost_descs;
5520 spin_lock(<ds->ltd_lock);
5521 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
5522 ltd_layout_phase_list) {
5523 list_del_init(<d->ltd_layout_phase_list);
5525 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
5526 ltd_layout_phase_list) {
5527 list_del_init(<d->ltd_layout_phase_list);
5529 spin_unlock(<ds->ltd_lock);
5531 ltds = &lfsck->li_mdt_descs;
5532 spin_lock(<ds->ltd_lock);
5533 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
5534 ltd_layout_phase_list) {
5535 list_del_init(<d->ltd_layout_phase_list);
5537 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
5538 ltd_layout_phase_list) {
5539 list_del_init(<d->ltd_layout_phase_list);
5541 spin_unlock(<ds->ltd_lock);
5544 static void lfsck_layout_slave_quit(const struct lu_env *env,
5545 struct lfsck_component *com)
5547 struct lfsck_layout_slave_data *llsd = com->lc_data;
5548 struct lfsck_layout_seq *lls;
5549 struct lfsck_layout_seq *next;
5550 struct lfsck_layout_slave_target *llst;
5552 LASSERT(llsd != NULL);
5554 down_write(&com->lc_sem);
5555 list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
5557 list_del_init(&lls->lls_list);
5558 lfsck_object_put(env, lls->lls_lastid_obj);
5561 up_write(&com->lc_sem);
5563 spin_lock(&llsd->llsd_lock);
5564 while (!list_empty(&llsd->llsd_master_list)) {
5565 llst = list_entry(llsd->llsd_master_list.next,
5566 struct lfsck_layout_slave_target, llst_list);
5567 list_del_init(&llst->llst_list);
5568 spin_unlock(&llsd->llsd_lock);
5569 lfsck_layout_llst_put(llst);
5570 spin_lock(&llsd->llsd_lock);
5572 spin_unlock(&llsd->llsd_lock);
5574 lfsck_rbtree_cleanup(env, com);
5577 static int lfsck_layout_master_in_notify(const struct lu_env *env,
5578 struct lfsck_component *com,
5579 struct lfsck_request *lr,
5582 struct lfsck_instance *lfsck = com->lc_lfsck;
5583 struct lfsck_layout *lo = com->lc_file_ram;
5584 struct lfsck_assistant_data *lad = com->lc_data;
5585 struct lfsck_tgt_descs *ltds;
5586 struct lfsck_tgt_desc *ltd;
5590 if (lr->lr_event == LE_PAIRS_VERIFY) {
5593 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
5599 CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
5600 "from %s %x, status %d, flags %x, flags2 %x\n",
5601 lfsck_lfsck2name(lfsck), lr->lr_event,
5602 (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
5603 lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
5605 if (lr->lr_event != LE_PHASE1_DONE &&
5606 lr->lr_event != LE_PHASE2_DONE &&
5607 lr->lr_event != LE_PEER_EXIT)
5610 if (lr->lr_flags & LEF_FROM_OST)
5611 ltds = &lfsck->li_ost_descs;
5613 ltds = &lfsck->li_mdt_descs;
5614 spin_lock(<ds->ltd_lock);
5615 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
5617 spin_unlock(<ds->ltd_lock);
5622 list_del_init(<d->ltd_layout_phase_list);
5623 switch (lr->lr_event) {
5624 case LE_PHASE1_DONE:
5625 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
5626 if (lr->lr_flags2 & LF_INCOMPLETE) {
5627 if (lr->lr_flags & LEF_FROM_OST)
5628 lfsck_lad_set_bitmap(env, com,
5631 lo->ll_flags |= LF_INCOMPLETE;
5633 ltd->ltd_layout_done = 1;
5634 list_del_init(<d->ltd_layout_list);
5639 if (lr->lr_flags & LEF_FROM_OST) {
5640 if (list_empty(<d->ltd_layout_list))
5641 list_add_tail(<d->ltd_layout_list,
5642 &lad->lad_ost_list);
5643 list_add_tail(<d->ltd_layout_phase_list,
5644 &lad->lad_ost_phase2_list);
5646 if (list_empty(<d->ltd_layout_list))
5647 list_add_tail(<d->ltd_layout_list,
5648 &lad->lad_mdt_list);
5649 list_add_tail(<d->ltd_layout_phase_list,
5650 &lad->lad_mdt_phase2_list);
5653 case LE_PHASE2_DONE:
5654 ltd->ltd_layout_done = 1;
5655 if (!list_empty(<d->ltd_layout_list)) {
5656 list_del_init(<d->ltd_layout_list);
5657 if (lr->lr_flags2 & LF_INCOMPLETE) {
5658 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
5666 ltd->ltd_layout_done = 1;
5667 list_del_init(<d->ltd_layout_list);
5668 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
5669 !(lr->lr_flags & LEF_FROM_OST))
5670 lo->ll_flags |= LF_INCOMPLETE;
5675 spin_unlock(<ds->ltd_lock);
5677 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
5678 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5680 memset(stop, 0, sizeof(*stop));
5681 stop->ls_status = lr->lr_status;
5682 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5683 lfsck_stop(env, lfsck->li_bottom, stop);
5684 } else if (lfsck_phase2_next_ready(lad)) {
5685 wake_up_all(&lad->lad_thread.t_ctl_waitq);
5691 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
5692 struct lfsck_component *com,
5693 struct lfsck_request *lr,
5696 struct lfsck_instance *lfsck = com->lc_lfsck;
5697 struct lfsck_layout_slave_data *llsd = com->lc_data;
5698 struct lfsck_layout_slave_target *llst;
5702 switch (lr->lr_event) {
5703 case LE_FID_ACCESSED:
5704 lfsck_rbtree_update_bitmap(env, com, &lr->lr_fid, true);
5706 case LE_CONDITIONAL_DESTROY:
5707 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
5709 case LE_PAIRS_VERIFY: {
5710 lr->lr_status = LPVS_INIT;
5711 /* Firstly, if the MDT-object which is claimed via OST-object
5712 * local stored PFID xattr recognizes the OST-object, then it
5713 * must be that the client given PFID is wrong. */
5714 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5719 lr->lr_status = LPVS_INCONSISTENT;
5720 /* The OST-object local stored PFID xattr is stale. We need to
5721 * check whether the MDT-object that is claimed via the client
5722 * given PFID information recognizes the OST-object or not. If
5723 * matches, then need to update the OST-object's PFID xattr. */
5724 rc = lfsck_layout_slave_check_pairs(env, com, &lr->lr_fid,
5727 * We are not sure whether the client given PFID information
5728 * is correct or not, do nothing to avoid improper fixing.
5731 * The client given PFID information is also invalid, we can
5732 * NOT fix the OST-object inconsistency.
5737 lr->lr_status = LPVS_INCONSISTENT_TOFIX;
5738 rc = lfsck_layout_slave_repair_pfid(env, com, lr);
5742 case LE_PHASE1_DONE: {
5743 if (lr->lr_flags2 & LF_INCOMPLETE) {
5744 struct lfsck_layout *lo = com->lc_file_ram;
5746 lo->ll_flags |= LF_INCOMPLETE;
5747 llst = lfsck_layout_llst_find_and_del(llsd,
5751 lfsck_layout_llst_put(llst);
5752 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5758 case LE_PHASE2_DONE:
5760 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
5761 "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
5762 lr->lr_event, lr->lr_index, lr->lr_status);
5768 llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
5772 lfsck_layout_llst_put(llst);
5773 if (list_empty(&llsd->llsd_master_list))
5774 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
5776 if (lr->lr_event == LE_PEER_EXIT &&
5777 (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
5778 (list_empty(&llsd->llsd_master_list) &&
5779 (lr->lr_status == LS_STOPPED ||
5780 lr->lr_status == LS_CO_STOPPED)))) {
5781 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
5783 memset(stop, 0, sizeof(*stop));
5784 stop->ls_status = lr->lr_status;
5785 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
5786 lfsck_stop(env, lfsck->li_bottom, stop);
5792 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
5796 for (i = 0; i < LLIT_MAX; i++)
5797 *count += lo->ll_objs_repaired[i];
5800 static int lfsck_layout_query_all(const struct lu_env *env,
5801 struct lfsck_component *com,
5802 __u32 *mdts_count, __u32 *osts_count,
5805 struct lfsck_layout *lo = com->lc_file_ram;
5806 struct lfsck_tgt_descs *ltds;
5807 struct lfsck_tgt_desc *ltd;
5812 rc = lfsck_query_all(env, com);
5816 ltds = &com->lc_lfsck->li_mdt_descs;
5817 down_read(<ds->ltd_rw_sem);
5818 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5819 ltd = lfsck_ltd2tgt(ltds, idx);
5820 LASSERT(ltd != NULL);
5822 mdts_count[ltd->ltd_layout_status]++;
5823 *repaired += ltd->ltd_layout_repaired;
5825 up_read(<ds->ltd_rw_sem);
5827 ltds = &com->lc_lfsck->li_ost_descs;
5828 down_read(<ds->ltd_rw_sem);
5829 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
5830 ltd = lfsck_ltd2tgt(ltds, idx);
5831 LASSERT(ltd != NULL);
5833 osts_count[ltd->ltd_layout_status]++;
5834 *repaired += ltd->ltd_layout_repaired;
5836 up_read(<ds->ltd_rw_sem);
5838 down_read(&com->lc_sem);
5839 mdts_count[lo->ll_status]++;
5840 lfsck_layout_repaired(lo, repaired);
5841 up_read(&com->lc_sem);
5846 static int lfsck_layout_query(const struct lu_env *env,
5847 struct lfsck_component *com,
5848 struct lfsck_request *req,
5849 struct lfsck_reply *rep,
5850 struct lfsck_query *que, int idx)
5852 struct lfsck_layout *lo = com->lc_file_ram;
5856 LASSERT(com->lc_lfsck->li_master);
5858 rc = lfsck_layout_query_all(env, com,
5859 que->lu_mdts_count[idx],
5860 que->lu_osts_count[idx],
5861 &que->lu_repaired[idx]);
5863 down_read(&com->lc_sem);
5864 rep->lr_status = lo->ll_status;
5865 if (req->lr_flags & LEF_QUERY_ALL)
5866 lfsck_layout_repaired(lo, &rep->lr_repaired);
5867 up_read(&com->lc_sem);
5873 /* with lfsck::li_lock held */
5874 static int lfsck_layout_slave_join(const struct lu_env *env,
5875 struct lfsck_component *com,
5876 struct lfsck_start_param *lsp)
5878 struct lfsck_instance *lfsck = com->lc_lfsck;
5879 struct lfsck_layout_slave_data *llsd = com->lc_data;
5880 struct lfsck_layout_slave_target *llst;
5881 struct lfsck_start *start = lsp->lsp_start;
5885 if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
5888 if (!lsp->lsp_index_valid)
5891 /* If someone is running the LFSCK without orphan handling,
5892 * it will not maintain the object accessing rbtree. So we
5893 * cannot join it for orphan handling. */
5894 if (!llsd->llsd_rbtree_valid)
5897 spin_unlock(&lfsck->li_lock);
5898 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5899 spin_lock(&lfsck->li_lock);
5900 if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
5901 spin_unlock(&lfsck->li_lock);
5902 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
5905 lfsck_layout_llst_put(llst);
5906 spin_lock(&lfsck->li_lock);
5913 static struct lfsck_operations lfsck_layout_master_ops = {
5914 .lfsck_reset = lfsck_layout_reset,
5915 .lfsck_fail = lfsck_layout_fail,
5916 .lfsck_checkpoint = lfsck_layout_master_checkpoint,
5917 .lfsck_prep = lfsck_layout_master_prep,
5918 .lfsck_exec_oit = lfsck_layout_master_exec_oit,
5919 .lfsck_exec_dir = lfsck_layout_exec_dir,
5920 .lfsck_post = lfsck_layout_master_post,
5921 .lfsck_dump = lfsck_layout_dump,
5922 .lfsck_double_scan = lfsck_layout_master_double_scan,
5923 .lfsck_data_release = lfsck_layout_master_data_release,
5924 .lfsck_quit = lfsck_layout_master_quit,
5925 .lfsck_in_notify = lfsck_layout_master_in_notify,
5926 .lfsck_query = lfsck_layout_query,
5929 static struct lfsck_operations lfsck_layout_slave_ops = {
5930 .lfsck_reset = lfsck_layout_reset,
5931 .lfsck_fail = lfsck_layout_fail,
5932 .lfsck_checkpoint = lfsck_layout_slave_checkpoint,
5933 .lfsck_prep = lfsck_layout_slave_prep,
5934 .lfsck_exec_oit = lfsck_layout_slave_exec_oit,
5935 .lfsck_exec_dir = lfsck_layout_exec_dir,
5936 .lfsck_post = lfsck_layout_slave_post,
5937 .lfsck_dump = lfsck_layout_dump,
5938 .lfsck_double_scan = lfsck_layout_slave_double_scan,
5939 .lfsck_data_release = lfsck_layout_slave_data_release,
5940 .lfsck_quit = lfsck_layout_slave_quit,
5941 .lfsck_in_notify = lfsck_layout_slave_in_notify,
5942 .lfsck_query = lfsck_layout_query,
5943 .lfsck_join = lfsck_layout_slave_join,
5946 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
5947 struct lfsck_component *com,
5948 struct lfsck_position *pos)
5950 struct lfsck_assistant_data *lad = com->lc_data;
5951 struct lfsck_layout_req *llr;
5953 if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
5957 if (list_empty(&lad->lad_req_list))
5960 llr = list_entry(lad->lad_req_list.next,
5961 struct lfsck_layout_req,
5963 pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
5966 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
5967 .la_handler_p1 = lfsck_layout_assistant_handler_p1,
5968 .la_handler_p2 = lfsck_layout_assistant_handler_p2,
5969 .la_fill_pos = lfsck_layout_assistant_fill_pos,
5970 .la_double_scan_result = lfsck_layout_double_scan_result,
5971 .la_req_fini = lfsck_layout_assistant_req_fini,
5972 .la_sync_failures = lfsck_layout_assistant_sync_failures,
5975 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
5977 struct lfsck_component *com;
5978 struct lfsck_layout *lo;
5979 struct dt_object *root = NULL;
5980 struct dt_object *obj;
5989 INIT_LIST_HEAD(&com->lc_link);
5990 INIT_LIST_HEAD(&com->lc_link_dir);
5991 init_rwsem(&com->lc_sem);
5992 atomic_set(&com->lc_ref, 1);
5993 com->lc_lfsck = lfsck;
5994 com->lc_type = LFSCK_TYPE_LAYOUT;
5995 if (lfsck->li_master) {
5996 com->lc_ops = &lfsck_layout_master_ops;
5997 com->lc_data = lfsck_assistant_data_init(
5998 &lfsck_layout_assistant_ops,
6000 if (com->lc_data == NULL)
6001 GOTO(out, rc = -ENOMEM);
6003 for (i = 0; i < LFSCK_STF_COUNT; i++)
6004 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6006 struct lfsck_layout_slave_data *llsd;
6008 com->lc_ops = &lfsck_layout_slave_ops;
6009 OBD_ALLOC_PTR(llsd);
6011 GOTO(out, rc = -ENOMEM);
6013 INIT_LIST_HEAD(&llsd->llsd_seq_list);
6014 INIT_LIST_HEAD(&llsd->llsd_master_list);
6015 spin_lock_init(&llsd->llsd_lock);
6016 llsd->llsd_rb_root = RB_ROOT;
6017 rwlock_init(&llsd->llsd_rb_lock);
6018 com->lc_data = llsd;
6020 com->lc_file_size = sizeof(*lo);
6021 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6022 if (com->lc_file_ram == NULL)
6023 GOTO(out, rc = -ENOMEM);
6025 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6026 if (com->lc_file_disk == NULL)
6027 GOTO(out, rc = -ENOMEM);
6029 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6031 GOTO(out, rc = PTR_ERR(root));
6033 if (unlikely(!dt_try_as_dir(env, root)))
6034 GOTO(out, rc = -ENOTDIR);
6036 obj = local_file_find_or_create(env, lfsck->li_los, root,
6038 S_IFREG | S_IRUGO | S_IWUSR);
6040 GOTO(out, rc = PTR_ERR(obj));
6043 rc = lfsck_layout_load(env, com);
6045 rc = lfsck_layout_reset(env, com, true);
6046 else if (rc == -ENOENT)
6047 rc = lfsck_layout_init(env, com);
6048 else if (lfsck->li_master)
6049 rc = lfsck_load_sub_trace_files(env, com,
6050 &dt_lfsck_layout_dangling_features,
6051 LFSCK_LAYOUT, false);
6056 lo = com->lc_file_ram;
6057 switch (lo->ll_status) {
6063 spin_lock(&lfsck->li_lock);
6064 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6065 spin_unlock(&lfsck->li_lock);
6068 CERROR("%s: unknown lfsck_layout status %d\n",
6069 lfsck_lfsck2name(lfsck), lo->ll_status);
6071 case LS_SCANNING_PHASE1:
6072 case LS_SCANNING_PHASE2:
6073 /* No need to store the status to disk right now.
6074 * If the system crashed before the status stored,
6075 * it will be loaded back when next time. */
6076 lo->ll_status = LS_CRASHED;
6077 if (!lfsck->li_master)
6078 lo->ll_flags |= LF_INCOMPLETE;
6085 spin_lock(&lfsck->li_lock);
6086 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6087 spin_unlock(&lfsck->li_lock);
6091 if (lo->ll_flags & LF_CRASHED_LASTID) {
6092 LASSERT(lfsck->li_out_notify != NULL);
6094 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6095 LE_LASTID_REBUILDING);
6101 if (root != NULL && !IS_ERR(root))
6102 lfsck_object_put(env, root);
6105 lfsck_component_cleanup(env, com);
6106 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
6107 lfsck_lfsck2name(lfsck), rc);
6113 struct lfsck_orphan_it {
6114 struct lfsck_component *loi_com;
6115 struct lfsck_rbtree_node *loi_lrn;
6116 struct lfsck_layout_slave_target *loi_llst;
6117 struct lu_fid loi_key;
6118 struct lu_orphan_rec loi_rec;
6120 unsigned int loi_over:1;
6123 static int lfsck_fid_match_idx(const struct lu_env *env,
6124 struct lfsck_instance *lfsck,
6125 const struct lu_fid *fid, int idx)
6127 struct seq_server_site *ss;
6128 struct lu_server_fld *sf;
6129 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
6132 /* All abnormal cases will be returned to MDT0. */
6133 if (!fid_is_norm(fid)) {
6140 ss = lfsck_dev_site(lfsck);
6141 if (unlikely(ss == NULL))
6144 sf = ss->ss_server_fld;
6145 LASSERT(sf != NULL);
6147 fld_range_set_any(range);
6148 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
6152 if (!fld_range_is_mdt(range))
6155 if (range->lsr_index == idx)
6161 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
6162 struct dt_object *obj)
6164 struct dt_device *dev = lfsck_obj2dev(obj);
6165 struct thandle *handle;
6169 handle = dt_trans_create(env, dev);
6173 rc = dt_declare_ref_del(env, obj, handle);
6177 rc = dt_declare_destroy(env, obj, handle);
6181 rc = dt_trans_start_local(env, dev, handle);
6185 dt_write_lock(env, obj, 0);
6186 rc = dt_ref_del(env, obj, handle);
6188 rc = dt_destroy(env, obj, handle);
6189 dt_write_unlock(env, obj);
6194 dt_trans_stop(env, dev, handle);
6196 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
6197 PFID(lfsck_dto2fid(obj)), rc);
6202 static int lfsck_orphan_index_lookup(const struct lu_env *env,
6203 struct dt_object *dt,
6205 const struct dt_key *key)
6210 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
6211 struct dt_object *dt,
6212 const struct dt_rec *rec,
6213 const struct dt_key *key,
6214 struct thandle *handle)
6219 static int lfsck_orphan_index_insert(const struct lu_env *env,
6220 struct dt_object *dt,
6221 const struct dt_rec *rec,
6222 const struct dt_key *key,
6223 struct thandle *handle,
6229 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
6230 struct dt_object *dt,
6231 const struct dt_key *key,
6232 struct thandle *handle)
6237 static int lfsck_orphan_index_delete(const struct lu_env *env,
6238 struct dt_object *dt,
6239 const struct dt_key *key,
6240 struct thandle *handle)
6245 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
6246 struct dt_object *dt,
6249 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
6250 struct lfsck_instance *lfsck;
6251 struct lfsck_component *com = NULL;
6252 struct lfsck_layout_slave_data *llsd;
6253 struct lfsck_orphan_it *it = NULL;
6254 struct lfsck_layout *lo;
6258 lfsck = lfsck_instance_find(dev, true, false);
6259 if (unlikely(lfsck == NULL))
6260 RETURN(ERR_PTR(-ENXIO));
6262 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
6263 if (unlikely(com == NULL))
6264 GOTO(out, rc = -ENOENT);
6266 lo = com->lc_file_ram;
6267 if (lo->ll_flags & LF_INCOMPLETE)
6268 GOTO(out, rc = -ESRCH);
6270 llsd = com->lc_data;
6271 if (!llsd->llsd_rbtree_valid)
6272 GOTO(out, rc = -ESRCH);
6276 GOTO(out, rc = -ENOMEM);
6278 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
6279 if (it->loi_llst == NULL)
6280 GOTO(out, rc = -ENXIO);
6282 if (dev->dd_record_fid_accessed) {
6283 /* The first iteration against the rbtree, scan the whole rbtree
6284 * to remove the nodes which do NOT need to be handled. */
6285 write_lock(&llsd->llsd_rb_lock);
6286 if (dev->dd_record_fid_accessed) {
6287 struct rb_node *node;
6288 struct rb_node *next;
6289 struct lfsck_rbtree_node *lrn;
6291 /* No need to record the fid accessing anymore. */
6292 dev->dd_record_fid_accessed = 0;
6294 node = rb_first(&llsd->llsd_rb_root);
6295 while (node != NULL) {
6296 next = rb_next(node);
6297 lrn = rb_entry(node, struct lfsck_rbtree_node,
6299 if (atomic_read(&lrn->lrn_known_count) <=
6300 atomic_read(&lrn->lrn_accessed_count)) {
6301 rb_erase(node, &llsd->llsd_rb_root);
6302 lfsck_rbtree_free(lrn);
6307 write_unlock(&llsd->llsd_rb_lock);
6310 /* read lock the rbtree when init, and unlock when fini */
6311 read_lock(&llsd->llsd_rb_lock);
6319 lfsck_component_put(env, com);
6321 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
6322 lfsck_lfsck2name(lfsck), rc);
6324 lfsck_instance_put(env, lfsck);
6329 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
6332 return (struct dt_it *)it;
6335 static void lfsck_orphan_it_fini(const struct lu_env *env,
6338 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6339 struct lfsck_component *com = it->loi_com;
6340 struct lfsck_layout_slave_data *llsd;
6341 struct lfsck_layout_slave_target *llst;
6344 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
6345 lfsck_lfsck2name(com->lc_lfsck));
6347 llsd = com->lc_data;
6348 read_unlock(&llsd->llsd_rb_lock);
6349 llst = it->loi_llst;
6350 LASSERT(llst != NULL);
6352 /* Save the key and hash for iterate next. */
6353 llst->llst_fid = it->loi_key;
6354 llst->llst_hash = it->loi_hash;
6355 lfsck_layout_llst_put(llst);
6356 lfsck_component_put(env, com);
6362 * \retval +1: the iteration finished
6363 * \retval 0: on success, not finished
6364 * \retval -ve: on error
6366 static int lfsck_orphan_it_next(const struct lu_env *env,
6369 struct lfsck_thread_info *info = lfsck_env_info(env);
6370 struct filter_fid_old *pfid = &info->lti_old_pfid;
6371 struct lu_attr *la = &info->lti_la;
6372 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6373 struct lu_fid *key = &it->loi_key;
6374 struct lu_orphan_rec *rec = &it->loi_rec;
6375 struct lfsck_component *com = it->loi_com;
6376 struct lfsck_instance *lfsck = com->lc_lfsck;
6377 struct lfsck_layout_slave_data *llsd = com->lc_data;
6378 struct dt_object *obj;
6379 struct lfsck_rbtree_node *lrn;
6383 __u32 idx = it->loi_llst->llst_index;
6393 lrn = lfsck_rbtree_search(llsd, key, &exact);
6401 key->f_seq = lrn->lrn_seq;
6402 key->f_oid = lrn->lrn_first_oid;
6407 if (unlikely(key->f_oid == 0)) {
6414 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
6420 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
6421 atomic_read(&lrn->lrn_accessed_count))) {
6422 struct rb_node *next = rb_next(&lrn->lrn_node);
6424 while (next != NULL) {
6425 lrn = rb_entry(next, struct lfsck_rbtree_node,
6427 if (atomic_read(&lrn->lrn_known_count) >
6428 atomic_read(&lrn->lrn_accessed_count))
6430 next = rb_next(next);
6439 key->f_seq = lrn->lrn_seq;
6440 key->f_oid = lrn->lrn_first_oid;
6444 pos = key->f_oid - lrn->lrn_first_oid;
6447 pos = find_next_bit(lrn->lrn_known_bitmap,
6448 LFSCK_RBTREE_BITMAP_WIDTH, pos);
6449 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
6450 key->f_oid = lrn->lrn_first_oid + pos;
6451 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
6459 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
6464 key->f_oid = lrn->lrn_first_oid + pos;
6465 obj = lfsck_object_find_bottom(env, lfsck, key);
6468 if (rc == -ENOENT) {
6475 dt_read_lock(env, obj, 0);
6476 if (dt_object_exists(obj) == 0 ||
6477 lfsck_is_dead_obj(obj)) {
6478 dt_read_unlock(env, obj);
6479 lfsck_object_put(env, obj);
6484 rc = dt_attr_get(env, obj, la);
6488 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, pfid, sizeof(*pfid)),
6490 if (rc == -ENODATA) {
6491 /* For the pre-created OST-object, update the bitmap to avoid
6492 * others LFSCK (second phase) iteration to touch it again. */
6493 if (la->la_ctime == 0) {
6494 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
6495 atomic_inc(&lrn->lrn_accessed_count);
6497 /* For the race between repairing dangling referenced
6498 * MDT-object and unlink the file, it may left orphan
6499 * OST-object there. Destroy it now! */
6500 if (unlikely(!(la->la_mode & S_ISUID))) {
6501 dt_read_unlock(env, obj);
6502 lfsck_layout_destroy_orphan(env, obj);
6503 lfsck_object_put(env, obj);
6507 } else if (idx == 0) {
6508 /* If the orphan OST-object has no parent information,
6509 * regard it as referenced by the MDT-object on MDT0. */
6510 fid_zero(&rec->lor_fid);
6511 rec->lor_uid = la->la_uid;
6512 rec->lor_gid = la->la_gid;
6516 dt_read_unlock(env, obj);
6517 lfsck_object_put(env, obj);
6525 if (rc != sizeof(struct filter_fid) &&
6526 rc != sizeof(struct filter_fid_old))
6527 GOTO(out, rc = -EINVAL);
6529 fid_le_to_cpu(&rec->lor_fid, &pfid->ff_parent);
6530 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
6531 * MDT-object's FID::f_ver, instead it is the OST-object index in its
6532 * parent MDT-object's layout EA. */
6533 save = rec->lor_fid.f_stripe_idx;
6534 rec->lor_fid.f_ver = 0;
6535 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_fid, idx);
6536 /* If the orphan OST-object does not claim the MDT, then next.
6538 * If we do not know whether it matches or not, then return it
6539 * to the MDT for further check. */
6541 dt_read_unlock(env, obj);
6542 lfsck_object_put(env, obj);
6547 rec->lor_fid.f_stripe_idx = save;
6548 rec->lor_uid = la->la_uid;
6549 rec->lor_gid = la->la_gid;
6551 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u\n",
6552 lfsck_lfsck2name(com->lc_lfsck), PFID(key), PFID(&rec->lor_fid),
6553 rec->lor_uid, rec->lor_gid);
6558 dt_read_unlock(env, obj);
6559 lfsck_object_put(env, obj);
6567 * \retval +1: locate to the exactly position
6568 * \retval 0: cannot locate to the exactly position,
6569 * call next() to move to a valid position.
6570 * \retval -ve: on error
6572 static int lfsck_orphan_it_get(const struct lu_env *env,
6574 const struct dt_key *key)
6576 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6579 it->loi_key = *(struct lu_fid *)key;
6580 rc = lfsck_orphan_it_next(env, di);
6590 static void lfsck_orphan_it_put(const struct lu_env *env,
6595 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
6596 const struct dt_it *di)
6598 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6600 return (struct dt_key *)&it->loi_key;
6603 static int lfsck_orphan_it_key_size(const struct lu_env *env,
6604 const struct dt_it *di)
6606 return sizeof(struct lu_fid);
6609 static int lfsck_orphan_it_rec(const struct lu_env *env,
6610 const struct dt_it *di,
6614 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6616 *(struct lu_orphan_rec *)rec = it->loi_rec;
6621 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
6622 const struct dt_it *di)
6624 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6626 return it->loi_hash;
6630 * \retval +1: locate to the exactly position
6631 * \retval 0: cannot locate to the exactly position,
6632 * call next() to move to a valid position.
6633 * \retval -ve: on error
6635 static int lfsck_orphan_it_load(const struct lu_env *env,
6636 const struct dt_it *di,
6639 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
6640 struct lfsck_layout_slave_target *llst = it->loi_llst;
6643 LASSERT(llst != NULL);
6645 if (hash != llst->llst_hash) {
6646 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
6647 "iteration does not match the one when fini "
6648 "%llu, to be reset.\n",
6649 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
6651 fid_zero(&llst->llst_fid);
6652 llst->llst_hash = 0;
6655 it->loi_key = llst->llst_fid;
6656 it->loi_hash = llst->llst_hash;
6657 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
6667 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
6668 const struct dt_it *di,
6674 const struct dt_index_operations lfsck_orphan_index_ops = {
6675 .dio_lookup = lfsck_orphan_index_lookup,
6676 .dio_declare_insert = lfsck_orphan_index_declare_insert,
6677 .dio_insert = lfsck_orphan_index_insert,
6678 .dio_declare_delete = lfsck_orphan_index_declare_delete,
6679 .dio_delete = lfsck_orphan_index_delete,
6681 .init = lfsck_orphan_it_init,
6682 .fini = lfsck_orphan_it_fini,
6683 .get = lfsck_orphan_it_get,
6684 .put = lfsck_orphan_it_put,
6685 .next = lfsck_orphan_it_next,
6686 .key = lfsck_orphan_it_key,
6687 .key_size = lfsck_orphan_it_key_size,
6688 .rec = lfsck_orphan_it_rec,
6689 .store = lfsck_orphan_it_store,
6690 .load = lfsck_orphan_it_load,
6691 .key_rec = lfsck_orphan_it_key_rec,