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, 2017, Intel Corporation.
26 * lustre/lfsck/lfsck_layout.c
28 * Author: Fan, Yong <fan.yong@intel.com>
32 # define EXPORT_SYMTAB
34 #define DEBUG_SUBSYSTEM S_LFSCK
36 #include <linux/bitops.h>
37 #include <linux/rbtree.h>
39 #include <lu_object.h>
40 #include <dt_object.h>
41 #include <lustre_fid.h>
42 #include <lustre_lib.h>
43 #include <lustre_net.h>
44 #include <md_object.h>
45 #include <obd_class.h>
47 #include "lfsck_internal.h"
49 #define LFSCK_LAYOUT_MAGIC_V1 0xB173AE14
50 #define LFSCK_LAYOUT_MAGIC_V2 0xB1734D76
51 #define LFSCK_LAYOUT_MAGIC_V3 0xB17371B9
52 #define LFSCK_LAYOUT_MAGIC_V4 0xB1732FED
54 #define LFSCK_LAYOUT_MAGIC LFSCK_LAYOUT_MAGIC_V4
56 struct lfsck_layout_seq {
57 struct list_head lls_list;
60 __u64 lls_lastid_known;
61 struct dt_object *lls_lastid_obj;
62 unsigned int lls_dirty:1;
65 struct lfsck_layout_slave_target {
66 /* link into lfsck_layout_slave_data::llsd_master_list. */
67 struct list_head llst_list;
68 /* The position for next record in the rbtree for iteration. */
69 struct lu_fid llst_fid;
70 /* Dummy hash for iteration against the rbtree. */
75 /* How many times we have failed to get the master status. */
79 struct lfsck_layout_slave_data {
80 /* list for lfsck_layout_seq */
81 struct list_head llsd_seq_list;
83 /* list for the masters involve layout verification. */
84 struct list_head llsd_master_list;
87 struct dt_object *llsd_rb_obj;
88 struct rb_root llsd_rb_root;
89 struct rw_semaphore llsd_rb_rwsem;
90 unsigned int llsd_rbtree_valid:1;
93 struct lfsck_layout_slave_async_args {
94 struct obd_export *llsaa_exp;
95 struct lfsck_component *llsaa_com;
96 struct lfsck_layout_slave_target *llsaa_llst;
99 static inline bool lfsck_comp_extent_aligned(__u64 size)
101 return (size & (LOV_MIN_STRIPE_SIZE - 1)) == 0;
105 lfsck_layout_llst_put(struct lfsck_layout_slave_target *llst)
107 if (atomic_dec_and_test(&llst->llst_ref)) {
108 LASSERT(list_empty(&llst->llst_list));
115 lfsck_layout_llst_add(struct lfsck_layout_slave_data *llsd, __u32 index)
117 struct lfsck_layout_slave_target *llst;
118 struct lfsck_layout_slave_target *tmp;
125 INIT_LIST_HEAD(&llst->llst_list);
127 llst->llst_index = index;
128 atomic_set(&llst->llst_ref, 1);
130 spin_lock(&llsd->llsd_lock);
131 list_for_each_entry(tmp, &llsd->llsd_master_list, llst_list) {
132 if (tmp->llst_index == index) {
138 list_add_tail(&llst->llst_list, &llsd->llsd_master_list);
139 spin_unlock(&llsd->llsd_lock);
148 lfsck_layout_llst_del(struct lfsck_layout_slave_data *llsd,
149 struct lfsck_layout_slave_target *llst)
153 spin_lock(&llsd->llsd_lock);
154 if (!list_empty(&llst->llst_list)) {
155 list_del_init(&llst->llst_list);
158 spin_unlock(&llsd->llsd_lock);
161 lfsck_layout_llst_put(llst);
164 static inline struct lfsck_layout_slave_target *
165 lfsck_layout_llst_find_and_del(struct lfsck_layout_slave_data *llsd,
166 __u32 index, bool unlink)
168 struct lfsck_layout_slave_target *llst;
170 spin_lock(&llsd->llsd_lock);
171 list_for_each_entry(llst, &llsd->llsd_master_list, llst_list) {
172 if (llst->llst_index == index) {
174 list_del_init(&llst->llst_list);
176 atomic_inc(&llst->llst_ref);
177 spin_unlock(&llsd->llsd_lock);
182 spin_unlock(&llsd->llsd_lock);
187 static struct lfsck_layout_req *
188 lfsck_layout_assistant_req_init(struct lfsck_assistant_object *lso,
189 struct dt_object *child, __u32 comp_id,
190 __u32 ost_idx, __u32 lov_idx)
192 struct lfsck_layout_req *llr;
196 return ERR_PTR(-ENOMEM);
198 INIT_LIST_HEAD(&llr->llr_lar.lar_list);
199 llr->llr_lar.lar_parent = lfsck_assistant_object_get(lso);
200 llr->llr_child = child;
201 llr->llr_comp_id = comp_id;
202 llr->llr_ost_idx = ost_idx;
203 llr->llr_lov_idx = lov_idx;
208 static void lfsck_layout_assistant_req_fini(const struct lu_env *env,
209 struct lfsck_assistant_req *lar)
211 struct lfsck_layout_req *llr =
212 container_of0(lar, struct lfsck_layout_req, llr_lar);
214 lfsck_object_put(env, llr->llr_child);
215 lfsck_assistant_object_put(env, lar->lar_parent);
220 lfsck_layout_assistant_sync_failures_interpret(const struct lu_env *env,
221 struct ptlrpc_request *req,
225 struct lfsck_async_interpret_args *laia = args;
226 struct lfsck_tgt_desc *ltd = laia->laia_ltd;
228 ltd->ltd_synced_failures = 1;
229 atomic_dec(laia->laia_count);
236 * Notify remote LFSCK instances about former failures.
238 * The local LFSCK instance has recorded which OSTs have ever failed to respond
239 * some LFSCK verification requests (maybe because of network issues or the OST
240 * itself trouble). During the respond gap, the OST may missed some OST-objects
241 * verification, then the OST cannot know whether related OST-objects have been
242 * referenced by related MDT-objects or not, then in the second-stage scanning,
243 * these OST-objects will be regarded as orphan, if the OST-object contains bad
244 * parent FID for back reference, then it will misguide the LFSCK to make wrong
245 * fixing for the fake orphan.
247 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
248 * it will scan the bitmap for the ever failed OSTs, and notify them that they
249 * have ever missed some OST-object verification and should skip the handling
250 * for orphan OST-objects on all MDTs that are in the layout LFSCK.
252 * \param[in] env pointer to the thread context
253 * \param[in] com pointer to the lfsck component
254 * \param[in] lr pointer to the lfsck request
256 static void lfsck_layout_assistant_sync_failures(const struct lu_env *env,
257 struct lfsck_component *com,
258 struct lfsck_request *lr)
260 struct lfsck_async_interpret_args *laia =
261 &lfsck_env_info(env)->lti_laia2;
262 struct lfsck_assistant_data *lad = com->lc_data;
263 struct lfsck_layout *lo = com->lc_file_ram;
264 struct lfsck_instance *lfsck = com->lc_lfsck;
265 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
266 struct lfsck_tgt_desc *ltd;
267 struct ptlrpc_request_set *set;
273 if (!test_bit(LAD_INCOMPLETE, &lad->lad_flags))
276 /* If the MDT has ever failed to verfiy some OST-objects,
277 * then sync failures with them firstly. */
278 lr->lr_flags2 = lo->ll_flags | LF_INCOMPLETE;
280 atomic_set(&count, 0);
281 memset(laia, 0, sizeof(*laia));
282 laia->laia_count = &count;
283 set = ptlrpc_prep_set();
285 GOTO(out, rc = -ENOMEM);
287 down_read(<ds->ltd_rw_sem);
288 cfs_foreach_bit(lad->lad_bitmap, idx) {
289 ltd = lfsck_ltd2tgt(ltds, idx);
293 laia->laia_ltd = ltd;
294 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
295 lfsck_layout_assistant_sync_failures_interpret,
298 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
299 "notify target %x for %s phase1 done: "
300 "rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
301 ltd->ltd_index, lad->lad_name, rc);
308 up_read(<ds->ltd_rw_sem);
310 if (rc == 0 && atomic_read(&count) > 0)
311 rc = ptlrpc_set_wait(env, set);
313 ptlrpc_set_destroy(set);
315 if (rc == 0 && atomic_read(&count) > 0)
322 /* If failed to sync failures with the OSTs, then have to
323 * mark the whole LFSCK as LF_INCOMPLETE to skip the whole
324 * subsequent orphan OST-object handling. */
325 lo->ll_flags |= LF_INCOMPLETE;
327 lr->lr_flags2 = lo->ll_flags;
330 static int lfsck_layout_verify_header_v1v3(struct dt_object *obj,
331 struct lov_mds_md_v1 *lmm,
332 __u64 start, __u32 comp_id)
337 magic = le32_to_cpu(lmm->lmm_magic);
338 /* If magic crashed, keep it there. Sometime later, during OST-object
339 * orphan handling, if some OST-object(s) back-point to it, it can be
340 * verified and repaired. */
341 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
344 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
349 CDEBUG(D_LFSCK, "%s LOV EA magic %u for the file "DFID"\n",
350 rc == -EINVAL ? "Unknown" : "Unsupported",
351 magic, PFID(lfsck_dto2fid(obj)));
356 pattern = le32_to_cpu(lmm->lmm_pattern);
359 /* XXX: DoM file verification will be supportted via LU-11081. */
360 if (lov_pattern(pattern) == LOV_PATTERN_MDT) {
362 CDEBUG(D_LFSCK, "The DoM entry for "DFID" is not "
363 "the first component in the mirror %x/%llu\n",
364 PFID(lfsck_dto2fid(obj)), comp_id, start);
371 if (!lov_pattern_supported_normal_comp(lov_pattern(pattern))) {
372 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u for the file "
373 DFID" in the component %x\n",
374 pattern, PFID(lfsck_dto2fid(obj)), comp_id);
382 static int lfsck_layout_verify_header_foreign(struct dt_object *obj,
383 struct lov_foreign_md *lfm,
386 /* magic has been verified already */
387 __u32 value_len = le32_to_cpu(lfm->lfm_length);
388 /* type and flags are not checked for instance */
390 CDEBUG(D_INFO, "foreign LOV EA, magic %x, len %u, type %x, flags %x, for file "DFID"\n",
391 le32_to_cpu(lfm->lfm_magic), value_len,
392 le32_to_cpu(lfm->lfm_type), le32_to_cpu(lfm->lfm_flags),
393 PFID(lfsck_dto2fid(obj)));
395 if (len != value_len + offsetof(typeof(*lfm), lfm_value))
396 CDEBUG(D_LFSCK, "foreign LOV EA internal size %u does not match EA full size %zu for file "DFID"\n",
397 value_len, len, PFID(lfsck_dto2fid(obj)));
399 /* nothing to repair */
403 static int lfsck_layout_verify_header(struct dt_object *obj,
404 struct lov_mds_md_v1 *lmm, size_t len)
408 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1) {
409 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
411 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
413 if (unlikely(count == 0)) {
414 CDEBUG(D_LFSCK, "the PFL file "DFID" contains invalid "
415 "components count 0\n",
416 PFID(lfsck_dto2fid(obj)));
421 for (i = 0; i < count && !rc; i++) {
422 struct lov_comp_md_entry_v1 *lcme =
423 &lcm->lcm_entries[i];
424 __u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
425 __u64 end = le64_to_cpu(lcme->lcme_extent.e_end);
426 __u32 comp_id = le32_to_cpu(lcme->lcme_id);
428 if (unlikely(comp_id == LCME_ID_INVAL ||
429 comp_id > LCME_ID_MAX)) {
430 CDEBUG(D_LFSCK, "found invalid FPL ID %u "
431 "for the file "DFID" at idx %d\n",
432 comp_id, PFID(lfsck_dto2fid(obj)), i);
437 if (unlikely(start >= end ||
438 !lfsck_comp_extent_aligned(start) ||
439 (!lfsck_comp_extent_aligned(end) &&
440 end != LUSTRE_EOF))) {
441 CDEBUG(D_LFSCK, "found invalid FPL extent "
442 "range [%llu - %llu) for the file "
444 start, end, PFID(lfsck_dto2fid(obj)), i);
449 rc = lfsck_layout_verify_header_v1v3(obj,
450 (struct lov_mds_md_v1 *)((char *)lmm +
451 le32_to_cpu(lcme->lcme_offset)), start,
454 } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_FOREIGN) {
455 rc = lfsck_layout_verify_header_foreign(obj,
456 (struct lov_foreign_md *)lmm,
459 rc = lfsck_layout_verify_header_v1v3(obj, lmm, 1, 0);
465 static int lfsck_layout_get_lovea(const struct lu_env *env,
466 struct dt_object *obj, struct lu_buf *buf)
472 rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
474 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
476 return !rc ? -ENODATA : rc;
478 lu_buf_realloc(buf, rc);
479 if (buf->lb_buf == NULL)
486 return !rc ? -ENODATA : rc;
488 if (unlikely(buf->lb_buf == NULL)) {
489 lu_buf_alloc(buf, rc);
490 if (buf->lb_buf == NULL)
496 rc1 = lfsck_layout_verify_header(obj, buf->lb_buf, rc);
498 return rc1 ? rc1 : rc;
501 #define LFSCK_RBTREE_BITMAP_SIZE PAGE_SIZE
502 #define LFSCK_RBTREE_BITMAP_WIDTH (LFSCK_RBTREE_BITMAP_SIZE << 3)
503 #define LFSCK_RBTREE_BITMAP_MASK (LFSCK_RBTREE_BITMAP_WIDTH - 1)
505 struct lfsck_rbtree_node {
506 struct rb_node lrn_node;
509 atomic_t lrn_known_count;
510 atomic_t lrn_accessed_count;
511 void *lrn_known_bitmap;
512 void *lrn_accessed_bitmap;
515 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
516 __u64 seq, __u32 oid)
518 if (seq < lrn->lrn_seq)
521 if (seq > lrn->lrn_seq)
524 if (oid < lrn->lrn_first_oid)
527 if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
533 /* The caller should hold llsd->llsd_rb_lock. */
534 static struct lfsck_rbtree_node *
535 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
536 const struct lu_fid *fid, bool *exact)
538 struct rb_node *node = llsd->llsd_rb_root.rb_node;
539 struct rb_node *prev = NULL;
540 struct lfsck_rbtree_node *lrn = NULL;
546 while (node != NULL) {
548 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
549 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
551 node = node->rb_left;
553 node = node->rb_right;
561 /* If there is no exactly matched one, then to the next valid one. */
564 /* The rbtree is empty. */
571 node = rb_next(prev);
573 /* The end of the rbtree. */
577 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
582 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
583 const struct lu_fid *fid)
585 struct lfsck_rbtree_node *lrn;
589 return ERR_PTR(-ENOMEM);
591 OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
592 if (lrn->lrn_known_bitmap == NULL) {
595 return ERR_PTR(-ENOMEM);
598 OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
599 if (lrn->lrn_accessed_bitmap == NULL) {
600 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
603 return ERR_PTR(-ENOMEM);
606 RB_CLEAR_NODE(&lrn->lrn_node);
607 lrn->lrn_seq = fid_seq(fid);
608 lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
609 atomic_set(&lrn->lrn_known_count, 0);
610 atomic_set(&lrn->lrn_accessed_count, 0);
615 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
617 OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
618 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
622 /* The caller should hold lock. */
623 static struct lfsck_rbtree_node *
624 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
625 struct lfsck_rbtree_node *lrn)
627 struct rb_node **pos = &llsd->llsd_rb_root.rb_node;
628 struct rb_node *parent = NULL;
629 struct lfsck_rbtree_node *tmp;
632 while (*pos != NULL) {
634 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
635 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
637 pos = &(*pos)->rb_left;
639 pos = &(*pos)->rb_right;
644 rb_link_node(&lrn->lrn_node, parent, pos);
645 rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
650 static const struct dt_index_operations lfsck_orphan_index_ops;
652 static int lfsck_rbtree_setup(const struct lu_env *env,
653 struct lfsck_component *com)
655 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
656 struct lfsck_instance *lfsck = com->lc_lfsck;
657 struct dt_device *dev = lfsck->li_bottom;
658 struct lfsck_layout_slave_data *llsd = com->lc_data;
659 struct dt_object *obj;
661 fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
662 fid->f_oid = lfsck_dev_idx(lfsck);
664 obj = dt_locate(env, dev, fid);
666 RETURN(PTR_ERR(obj));
668 /* Generate an in-RAM object to stand for the layout rbtree.
669 * Scanning the layout rbtree will be via the iteration over
670 * the object. In the future, the rbtree may be written onto
671 * disk with the object.
673 * Mark the object to be as exist. */
674 obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
675 obj->do_index_ops = &lfsck_orphan_index_ops;
676 llsd->llsd_rb_obj = obj;
677 llsd->llsd_rbtree_valid = 1;
678 dev->dd_record_fid_accessed = 1;
680 CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
681 lfsck_lfsck2name(lfsck));
686 static void lfsck_rbtree_cleanup(const struct lu_env *env,
687 struct lfsck_component *com)
689 struct lfsck_instance *lfsck = com->lc_lfsck;
690 struct lfsck_layout_slave_data *llsd = com->lc_data;
691 struct rb_node *node = rb_first(&llsd->llsd_rb_root);
692 struct rb_node *next;
693 struct lfsck_rbtree_node *lrn;
695 lfsck->li_bottom->dd_record_fid_accessed = 0;
696 /* Invalid the rbtree, then no others will use it. */
697 down_write(&llsd->llsd_rb_rwsem);
698 llsd->llsd_rbtree_valid = 0;
699 up_write(&llsd->llsd_rb_rwsem);
701 while (node != NULL) {
702 next = rb_next(node);
703 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
704 rb_erase(node, &llsd->llsd_rb_root);
705 lfsck_rbtree_free(lrn);
709 if (llsd->llsd_rb_obj != NULL) {
710 lfsck_object_put(env, llsd->llsd_rb_obj);
711 llsd->llsd_rb_obj = NULL;
714 CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
715 lfsck_lfsck2name(lfsck));
718 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
719 struct lfsck_component *com,
720 const struct lu_fid *fid,
723 struct lfsck_layout_slave_data *llsd = com->lc_data;
724 struct lfsck_rbtree_node *lrn;
730 if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
733 if (!fid_is_idif(fid) && !fid_is_norm(fid))
736 down_read(&llsd->llsd_rb_rwsem);
737 if (!llsd->llsd_rbtree_valid)
738 GOTO(unlock, rc = 0);
740 lrn = lfsck_rbtree_search(llsd, fid, NULL);
742 struct lfsck_rbtree_node *tmp;
746 up_read(&llsd->llsd_rb_rwsem);
747 tmp = lfsck_rbtree_new(env, fid);
749 GOTO(out, rc = PTR_ERR(tmp));
752 down_write(&llsd->llsd_rb_rwsem);
753 if (!llsd->llsd_rbtree_valid) {
754 lfsck_rbtree_free(tmp);
755 GOTO(unlock, rc = 0);
758 lrn = lfsck_rbtree_insert(llsd, tmp);
760 lfsck_rbtree_free(tmp);
763 idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
764 /* Any accessed object must be a known object. */
765 if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
766 atomic_inc(&lrn->lrn_known_count);
767 if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
768 atomic_inc(&lrn->lrn_accessed_count);
770 GOTO(unlock, rc = 0);
774 up_write(&llsd->llsd_rb_rwsem);
776 up_read(&llsd->llsd_rb_rwsem);
778 if (rc != 0 && accessed) {
779 struct lfsck_layout *lo = com->lc_file_ram;
781 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
782 "bitmap, and will cause incorrect LFSCK OST-object "
783 "handling, so disable it to cancel orphan handling "
784 "for related device. rc = %d\n",
785 lfsck_lfsck2name(com->lc_lfsck), rc);
787 lo->ll_flags |= LF_INCOMPLETE;
788 lfsck_rbtree_cleanup(env, com);
792 static inline void lldk_le_to_cpu(struct lfsck_layout_dangling_key *des,
793 const struct lfsck_layout_dangling_key *src)
795 fid_le_to_cpu(&des->lldk_fid, &src->lldk_fid);
796 des->lldk_comp_id = le32_to_cpu(src->lldk_comp_id);
797 des->lldk_ea_off = le32_to_cpu(src->lldk_ea_off);
800 static inline void lldk_cpu_to_le(struct lfsck_layout_dangling_key *des,
801 const struct lfsck_layout_dangling_key *src)
803 fid_cpu_to_le(&des->lldk_fid, &src->lldk_fid);
804 des->lldk_comp_id = cpu_to_le32(src->lldk_comp_id);
805 des->lldk_ea_off = cpu_to_le32(src->lldk_ea_off);
808 static inline void lldk_be_to_cpu(struct lfsck_layout_dangling_key *des,
809 const struct lfsck_layout_dangling_key *src)
811 fid_be_to_cpu(&des->lldk_fid, &src->lldk_fid);
812 des->lldk_comp_id = be32_to_cpu(src->lldk_comp_id);
813 des->lldk_ea_off = be32_to_cpu(src->lldk_ea_off);
816 static inline void lldk_cpu_to_be(struct lfsck_layout_dangling_key *des,
817 const struct lfsck_layout_dangling_key *src)
819 fid_cpu_to_be(&des->lldk_fid, &src->lldk_fid);
820 des->lldk_comp_id = cpu_to_be32(src->lldk_comp_id);
821 des->lldk_ea_off = cpu_to_be32(src->lldk_ea_off);
824 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
825 const struct lfsck_layout *src)
829 des->ll_magic = le32_to_cpu(src->ll_magic);
830 des->ll_status = le32_to_cpu(src->ll_status);
831 des->ll_flags = le32_to_cpu(src->ll_flags);
832 des->ll_success_count = le32_to_cpu(src->ll_success_count);
833 des->ll_run_time_phase1 = le64_to_cpu(src->ll_run_time_phase1);
834 des->ll_run_time_phase2 = le64_to_cpu(src->ll_run_time_phase2);
835 des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
836 des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
837 des->ll_time_last_checkpoint =
838 le64_to_cpu(src->ll_time_last_checkpoint);
839 des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
840 des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
841 des->ll_pos_first_inconsistent =
842 le64_to_cpu(src->ll_pos_first_inconsistent);
843 des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
844 des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
845 des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
846 des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
847 for (i = 0; i < LLIT_MAX; i++)
848 des->ll_objs_repaired[i] =
849 le64_to_cpu(src->ll_objs_repaired[i]);
850 des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
851 des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
852 lldk_le_to_cpu(&des->ll_lldk_latest_scanned_phase2,
853 &src->ll_lldk_latest_scanned_phase2);
856 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
857 const struct lfsck_layout *src)
861 des->ll_magic = cpu_to_le32(src->ll_magic);
862 des->ll_status = cpu_to_le32(src->ll_status);
863 des->ll_flags = cpu_to_le32(src->ll_flags);
864 des->ll_success_count = cpu_to_le32(src->ll_success_count);
865 des->ll_run_time_phase1 = cpu_to_le64(src->ll_run_time_phase1);
866 des->ll_run_time_phase2 = cpu_to_le64(src->ll_run_time_phase2);
867 des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
868 des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
869 des->ll_time_last_checkpoint =
870 cpu_to_le64(src->ll_time_last_checkpoint);
871 des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
872 des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
873 des->ll_pos_first_inconsistent =
874 cpu_to_le64(src->ll_pos_first_inconsistent);
875 des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
876 des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
877 des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
878 des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
879 for (i = 0; i < LLIT_MAX; i++)
880 des->ll_objs_repaired[i] =
881 cpu_to_le64(src->ll_objs_repaired[i]);
882 des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
883 des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
884 lldk_cpu_to_le(&des->ll_lldk_latest_scanned_phase2,
885 &src->ll_lldk_latest_scanned_phase2);
889 * Load the OST bitmap from the lfsck_layout trace file.
891 * \param[in] env pointer to the thread context
892 * \param[in] com pointer to the lfsck component
894 * \retval 0 for success
895 * \retval negative error number on failure or data corruption
897 static int lfsck_layout_load_bitmap(const struct lu_env *env,
898 struct lfsck_component *com)
900 struct dt_object *obj = com->lc_obj;
901 struct lfsck_assistant_data *lad = com->lc_data;
902 struct lfsck_layout *lo = com->lc_file_ram;
903 struct cfs_bitmap *bitmap = lad->lad_bitmap;
904 loff_t pos = com->lc_file_size;
910 if (com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size >
912 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_bitmap->size;
914 nbits = lo->ll_bitmap_size;
916 if (unlikely(nbits < BITS_PER_LONG))
917 nbits = BITS_PER_LONG;
919 if (nbits > bitmap->size) {
920 __u32 new_bits = bitmap->size;
921 struct cfs_bitmap *new_bitmap;
923 while (new_bits < nbits)
926 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
927 if (new_bitmap == NULL)
930 lad->lad_bitmap = new_bitmap;
931 CFS_FREE_BITMAP(bitmap);
935 if (lo->ll_bitmap_size == 0) {
936 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
937 CFS_RESET_BITMAP(bitmap);
942 size = (lo->ll_bitmap_size + 7) >> 3;
943 rc = dt_read(env, obj, lfsck_buf_get(env, bitmap->data, size), &pos);
945 RETURN(rc >= 0 ? -EINVAL : rc);
947 if (cfs_bitmap_check_empty(bitmap))
948 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
950 set_bit(LAD_INCOMPLETE, &lad->lad_flags);
956 * Load the layout LFSCK trace file from disk.
958 * The layout LFSCK trace file records the layout LFSCK status information
959 * and other statistics, such as how many objects have been scanned, and how
960 * many objects have been repaired, and etc. It also contains the bitmap for
961 * failed OSTs during the layout LFSCK. All these information will be loaded
962 * from disk to RAM when the layout LFSCK component setup.
964 * \param[in] env pointer to the thread context
965 * \param[in] com pointer to the lfsck component
967 * \retval positive number for file data corruption, the caller
968 * should reset the layout LFSCK trace file
969 * \retval 0 for success
970 * \retval negative error number on failure
972 static int lfsck_layout_load(const struct lu_env *env,
973 struct lfsck_component *com)
975 struct lfsck_layout *lo = com->lc_file_ram;
976 ssize_t size = com->lc_file_size;
980 rc = dt_read(env, com->lc_obj,
981 lfsck_buf_get(env, com->lc_file_disk, size), &pos);
985 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
986 lfsck_lfsck2name(com->lc_lfsck), rc);
988 } else if (rc != size) {
989 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
990 lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
994 lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
995 if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
996 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
997 "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
998 lo->ll_magic, LFSCK_LAYOUT_MAGIC);
1006 * Store the layout LFSCK trace file on disk.
1008 * The layout LFSCK trace file records the layout LFSCK status information
1009 * and other statistics, such as how many objects have been scanned, and how
1010 * many objects have been repaired, and etc. It also contains the bitmap for
1011 * failed OSTs during the layout LFSCK. All these information will be synced
1012 * from RAM to disk periodically.
1014 * \param[in] env pointer to the thread context
1015 * \param[in] com pointer to the lfsck component
1017 * \retval 0 for success
1018 * \retval negative error number on failure
1020 static int lfsck_layout_store(const struct lu_env *env,
1021 struct lfsck_component *com)
1023 struct dt_object *obj = com->lc_obj;
1024 struct lfsck_instance *lfsck = com->lc_lfsck;
1025 struct lfsck_layout *lo_ram = com->lc_file_ram;
1026 struct lfsck_layout *lo = com->lc_file_disk;
1028 struct dt_device *dev = lfsck_obj2dev(obj);
1029 struct cfs_bitmap *bitmap = NULL;
1031 ssize_t size = com->lc_file_size;
1036 if (lfsck->li_master) {
1037 struct lfsck_assistant_data *lad = com->lc_data;
1039 bitmap = lad->lad_bitmap;
1040 nbits = bitmap->size;
1043 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
1046 lo_ram->ll_bitmap_size = nbits;
1047 lfsck_layout_cpu_to_le(lo, lo_ram);
1048 th = dt_trans_create(env, dev);
1050 GOTO(log, rc = PTR_ERR(th));
1052 rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
1057 if (bitmap != NULL) {
1058 rc = dt_declare_record_write(env, obj,
1059 lfsck_buf_get(env, bitmap->data, nbits >> 3),
1065 rc = dt_trans_start_local(env, dev, th);
1070 rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
1074 if (bitmap != NULL) {
1076 rc = dt_record_write(env, obj,
1077 lfsck_buf_get(env, bitmap->data, nbits >> 3),
1084 dt_trans_stop(env, dev, th);
1088 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
1089 lfsck_lfsck2name(lfsck), rc);
1094 static int lfsck_layout_init(const struct lu_env *env,
1095 struct lfsck_component *com)
1097 struct lfsck_layout *lo = com->lc_file_ram;
1100 memset(lo, 0, com->lc_file_size);
1101 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
1102 lo->ll_status = LS_INIT;
1103 down_write(&com->lc_sem);
1104 rc = lfsck_layout_store(env, com);
1105 if (rc == 0 && com->lc_lfsck->li_master)
1106 rc = lfsck_load_sub_trace_files(env, com,
1107 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
1108 up_write(&com->lc_sem);
1113 static int fid_is_for_ostobj(const struct lu_env *env,
1114 struct lfsck_instance *lfsck,
1115 struct dt_object *obj, const struct lu_fid *fid)
1117 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1118 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
1119 struct lustre_ost_attrs *loa;
1122 fld_range_set_any(range);
1123 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1125 if (fld_range_is_ost(range))
1131 loa = &lfsck_env_info(env)->lti_loa;
1132 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, loa, sizeof(*loa)),
1134 if (rc >= (int)sizeof(struct lustre_mdt_attrs)) {
1135 lustre_lma_swab(&loa->loa_lma);
1137 return loa->loa_lma.lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1140 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1145 static struct lfsck_layout_seq *
1146 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1148 struct lfsck_layout_seq *lls;
1150 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1151 if (lls->lls_seq == seq)
1154 if (lls->lls_seq > seq)
1162 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1163 struct lfsck_layout_seq *lls)
1165 struct lfsck_layout_seq *tmp;
1166 struct list_head *pos = &llsd->llsd_seq_list;
1168 list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1169 if (lls->lls_seq < tmp->lls_seq) {
1170 pos = &tmp->lls_list;
1174 list_add_tail(&lls->lls_list, pos);
1178 lfsck_layout_lastid_create(const struct lu_env *env,
1179 struct lfsck_instance *lfsck,
1180 struct dt_object *obj)
1182 struct lfsck_thread_info *info = lfsck_env_info(env);
1183 struct lu_attr *la = &info->lti_la;
1184 struct dt_object_format *dof = &info->lti_dof;
1185 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1186 struct dt_device *dt = lfsck_obj2dev(obj);
1193 if (bk->lb_param & LPF_DRYRUN)
1196 memset(la, 0, sizeof(*la));
1197 la->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1198 la->la_valid = LA_MODE | LA_UID | LA_GID;
1199 memset(dof, 0, sizeof(*dof));
1200 dof->dof_type = dt_mode_to_dft(S_IFREG);
1202 th = dt_trans_create(env, dt);
1204 GOTO(log, rc = PTR_ERR(th));
1206 rc = dt_declare_create(env, obj, la, NULL, dof, th);
1210 rc = dt_declare_record_write(env, obj,
1211 lfsck_buf_get(env, &lastid,
1217 rc = dt_trans_start_local(env, dt, th);
1221 dt_write_lock(env, obj, 0);
1222 if (likely(dt_object_exists(obj) == 0)) {
1223 rc = dt_create(env, obj, la, NULL, dof, th);
1225 rc = dt_record_write(env, obj,
1226 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1229 dt_write_unlock(env, obj);
1234 dt_trans_stop(env, dt, th);
1237 CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1239 lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1245 lfsck_layout_lastid_reload(const struct lu_env *env,
1246 struct lfsck_component *com,
1247 struct lfsck_layout_seq *lls)
1253 dt_read_lock(env, lls->lls_lastid_obj, 0);
1254 rc = dt_record_read(env, lls->lls_lastid_obj,
1255 lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1256 dt_read_unlock(env, lls->lls_lastid_obj);
1257 if (unlikely(rc != 0))
1260 lastid = le64_to_cpu(lastid);
1261 if (lastid < lls->lls_lastid_known) {
1262 struct lfsck_instance *lfsck = com->lc_lfsck;
1263 struct lfsck_layout *lo = com->lc_file_ram;
1265 lls->lls_lastid = lls->lls_lastid_known;
1267 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1268 LASSERT(lfsck->li_out_notify != NULL);
1270 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1271 LE_LASTID_REBUILDING);
1272 lo->ll_flags |= LF_CRASHED_LASTID;
1274 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1275 "LAST_ID file (1) for the sequence %#llx"
1276 ", old value %llu, known value %llu\n",
1277 lfsck_lfsck2name(lfsck), lls->lls_seq,
1278 lastid, lls->lls_lastid);
1280 } else if (lastid >= lls->lls_lastid) {
1281 lls->lls_lastid = lastid;
1289 lfsck_layout_lastid_store(const struct lu_env *env,
1290 struct lfsck_component *com)
1292 struct lfsck_instance *lfsck = com->lc_lfsck;
1293 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1294 struct dt_device *dt = lfsck->li_bottom;
1295 struct lfsck_layout_slave_data *llsd = com->lc_data;
1296 struct lfsck_layout_seq *lls;
1302 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1305 if (!lls->lls_dirty)
1308 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1309 "<seq> %#llx as <oid> %llu\n",
1310 lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1312 if (bk->lb_param & LPF_DRYRUN) {
1317 th = dt_trans_create(env, dt);
1320 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1321 "the LAST_ID for <seq> %#llx(1): rc = %d\n",
1322 lfsck_lfsck2name(com->lc_lfsck),
1327 lastid = cpu_to_le64(lls->lls_lastid);
1328 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1329 lfsck_buf_get(env, &lastid,
1335 rc = dt_trans_start_local(env, dt, th);
1339 dt_write_lock(env, lls->lls_lastid_obj, 0);
1340 rc = dt_record_write(env, lls->lls_lastid_obj,
1341 lfsck_buf_get(env, &lastid,
1342 sizeof(lastid)), &pos, th);
1343 dt_write_unlock(env, lls->lls_lastid_obj);
1348 dt_trans_stop(env, dt, th);
1351 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1352 "the LAST_ID for <seq> %#llx(2): rc = %d\n",
1353 lfsck_lfsck2name(com->lc_lfsck),
1362 lfsck_layout_lastid_load(const struct lu_env *env,
1363 struct lfsck_component *com,
1364 struct lfsck_layout_seq *lls)
1366 struct lfsck_instance *lfsck = com->lc_lfsck;
1367 struct lfsck_layout *lo = com->lc_file_ram;
1368 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
1369 struct dt_object *obj;
1374 lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1375 obj = dt_locate(env, lfsck->li_bottom, fid);
1377 RETURN(PTR_ERR(obj));
1379 /* LAST_ID crashed, to be rebuilt */
1380 if (dt_object_exists(obj) == 0) {
1381 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1382 LASSERT(lfsck->li_out_notify != NULL);
1384 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1385 LE_LASTID_REBUILDING);
1386 lo->ll_flags |= LF_CRASHED_LASTID;
1388 CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1389 "LAST_ID file for sequence %#llx\n",
1390 lfsck_lfsck2name(lfsck), lls->lls_seq);
1392 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1394 struct ptlrpc_thread *thread =
1397 up_write(&com->lc_sem);
1398 wait_event_idle_timeout(
1399 thread->t_ctl_waitq,
1400 !thread_is_running(thread),
1401 cfs_time_seconds(cfs_fail_val));
1402 down_write(&com->lc_sem);
1406 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1408 dt_read_lock(env, obj, 0);
1409 rc = dt_read(env, obj,
1410 lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1412 dt_read_unlock(env, obj);
1413 if (rc != 0 && rc != sizeof(__u64))
1414 GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1416 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1417 LASSERT(lfsck->li_out_notify != NULL);
1419 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1420 LE_LASTID_REBUILDING);
1421 lo->ll_flags |= LF_CRASHED_LASTID;
1423 CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1424 "LAST_ID file for the sequence %#llx"
1426 lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1429 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1437 lfsck_object_put(env, obj);
1439 lls->lls_lastid_obj = obj;
1444 static void lfsck_layout_record_failure(const struct lu_env *env,
1445 struct lfsck_instance *lfsck,
1446 struct lfsck_layout *lo)
1450 lo->ll_objs_failed_phase1++;
1451 cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1453 if (lo->ll_pos_first_inconsistent == 0 ||
1454 lo->ll_pos_first_inconsistent < cookie) {
1455 lo->ll_pos_first_inconsistent = cookie;
1457 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1458 "inconsistency at the pos [%llu]\n",
1459 lfsck_lfsck2name(lfsck),
1460 lo->ll_pos_first_inconsistent);
1464 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1465 struct lfsck_component *com,
1468 struct lfsck_instance *lfsck = com->lc_lfsck;
1469 struct lfsck_layout *lo = com->lc_file_ram;
1471 CDEBUG(D_LFSCK, "%s: layout LFSCK double scan: rc = %d\n",
1472 lfsck_lfsck2name(lfsck), rc);
1474 down_write(&com->lc_sem);
1475 lo->ll_run_time_phase2 += ktime_get_seconds() -
1476 com->lc_time_last_checkpoint;
1477 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
1478 lo->ll_objs_checked_phase2 += com->lc_new_checked;
1481 if (lo->ll_flags & LF_INCOMPLETE) {
1482 lo->ll_status = LS_PARTIAL;
1484 if (lfsck->li_master) {
1485 struct lfsck_assistant_data *lad = com->lc_data;
1487 if (test_bit(LAD_INCOMPLETE, &lad->lad_flags))
1488 lo->ll_status = LS_PARTIAL;
1490 lo->ll_status = LS_COMPLETED;
1492 lo->ll_status = LS_COMPLETED;
1495 lo->ll_flags &= ~LF_SCANNED_ONCE;
1496 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1497 lo->ll_flags &= ~LF_INCONSISTENT;
1498 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1499 lo->ll_success_count++;
1500 } else if (rc == 0) {
1501 if (lfsck->li_status != 0)
1502 lo->ll_status = lfsck->li_status;
1504 lo->ll_status = LS_STOPPED;
1506 lo->ll_status = LS_FAILED;
1509 rc = lfsck_layout_store(env, com);
1510 up_write(&com->lc_sem);
1512 CDEBUG(D_LFSCK, "%s: layout LFSCK double scan result %u: rc = %d\n",
1513 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
1518 static int lfsck_layout_trans_stop(const struct lu_env *env,
1519 struct dt_device *dev,
1520 struct thandle *handle, int result)
1524 /* XXX: If there is something worng or it needs to repair nothing,
1525 * then notify the lower to stop the modification. Currently,
1526 * we use th_result for such purpose, that may be replaced by
1527 * some rollback mechanism in the future. */
1528 handle->th_result = result;
1529 rc = dt_trans_stop(env, dev, handle);
1531 return result > 0 ? 0 : result;
1533 return rc == 0 ? 1 : rc;
1536 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1537 struct lfsck_component *com,
1538 const struct lu_fid *pfid,
1539 const struct lu_fid *cfid,
1540 __u32 comp_id, __u32 ea_off,
1543 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1544 struct lu_fid *rec = &lfsck_env_info(env)->lti_fid3;
1545 struct dt_device *dev;
1546 struct dt_object *obj;
1547 struct thandle *th = NULL;
1552 idx = lfsck_sub_trace_file_fid2idx(pfid);
1553 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1554 dev = lfsck_obj2dev(obj);
1556 fid_cpu_to_be(&key->lldk_fid, pfid);
1557 key->lldk_comp_id = cpu_to_be32(comp_id);
1558 key->lldk_ea_off = cpu_to_be32(ea_off);
1560 fid_cpu_to_be(rec, cfid);
1561 rec->f_ver = cpu_to_be32(ost_idx);
1563 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1565 th = dt_trans_create(env, dev);
1567 GOTO(unlock, rc = PTR_ERR(th));
1569 rc = dt_declare_insert(env, obj,
1570 (const struct dt_rec *)rec,
1571 (const struct dt_key *)key, th);
1575 rc = dt_trans_start_local(env, dev, th);
1579 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1580 (const struct dt_key *)key, th);
1585 if (th && !IS_ERR(th))
1586 dt_trans_stop(env, dev, th);
1588 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1590 CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", comp_id = %u, "
1591 "ea_off = %u, ost_idx = %u, into the trace file for further "
1592 "dangling check: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1593 PFID(pfid), PFID(cfid), comp_id, ea_off, ost_idx, rc);
1598 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1599 struct lfsck_component *com,
1600 const struct lu_fid *fid,
1601 __u32 comp_id, __u32 ea_off)
1603 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1604 struct dt_device *dev;
1605 struct dt_object *obj;
1606 struct thandle *th = NULL;
1611 idx = lfsck_sub_trace_file_fid2idx(fid);
1612 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1613 dev = lfsck_obj2dev(obj);
1615 fid_cpu_to_be(&key->lldk_fid, fid);
1616 key->lldk_comp_id = cpu_to_be32(comp_id);
1617 key->lldk_ea_off = cpu_to_be32(ea_off);
1619 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1621 th = dt_trans_create(env, dev);
1623 GOTO(unlock, rc = PTR_ERR(th));
1625 rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1629 rc = dt_trans_start_local(env, dev, th);
1633 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1638 if (th && !IS_ERR(th))
1639 dt_trans_stop(env, dev, th);
1641 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1643 CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1644 ", comp_id = %u, ea_off = %u from the trace file: rc = %d\n",
1645 lfsck_lfsck2name(com->lc_lfsck), PFID(fid), comp_id, ea_off, rc);
1651 * Get the system default stripe size.
1653 * \param[in] env pointer to the thread context
1654 * \param[in] lfsck pointer to the lfsck instance
1655 * \param[out] size pointer to the default stripe size
1657 * \retval 0 for success
1658 * \retval negative error number on failure
1660 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1661 struct lfsck_instance *lfsck,
1664 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1665 struct dt_object *root;
1668 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1670 return PTR_ERR(root);
1672 /* Get the default stripe size via xattr_get on the backend root. */
1673 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1676 /* The lum->lmm_stripe_size is LE mode. The *size also
1677 * should be LE mode. So it is unnecessary to convert. */
1678 *size = lum->lmm_stripe_size;
1680 } else if (unlikely(rc == 0)) {
1684 lfsck_object_put(env, root);
1690 * \retval +1: repaired
1691 * \retval 0: did nothing
1692 * \retval -ve: on error
1694 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1695 struct lfsck_instance *lfsck,
1696 struct thandle *handle,
1697 struct dt_object *parent,
1698 const struct lu_fid *cfid,
1700 struct lov_mds_md_v1 *lmm,
1701 struct lov_ost_data_v1 *slot,
1702 int fl, __u32 ost_idx, int size)
1704 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1705 struct lu_buf ea_buf;
1712 magic = le32_to_cpu(lmm->lmm_magic);
1713 pattern = le32_to_cpu(lmm->lmm_pattern);
1714 count = le16_to_cpu(lmm->lmm_stripe_count);
1716 fid_to_ostid(cfid, oi);
1717 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1718 slot->l_ost_gen = cpu_to_le32(0);
1719 slot->l_ost_idx = cpu_to_le32(ost_idx);
1721 if (pattern & LOV_PATTERN_F_HOLE) {
1722 struct lov_ost_data_v1 *objs;
1725 if (magic == LOV_MAGIC_V1)
1726 objs = &lmm->lmm_objects[0];
1728 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1729 for (i = 0; i < count; i++, objs++) {
1730 if (lovea_slot_is_dummy(objs))
1734 /* If the @slot is the last dummy slot to be refilled,
1735 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1738 cpu_to_le32(pattern & ~LOV_PATTERN_F_HOLE);
1740 CDEBUG(D_LFSCK, "%s: remove layout HOLE for "DFID
1741 ": parent "DFID"\n", lfsck_lfsck2name(lfsck),
1742 PFID(cfid), PFID(lfsck_dto2fid(parent)));
1746 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
1747 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1754 static struct lov_ost_data_v1 *
1755 __lfsck_layout_new_v1_lovea(struct lov_mds_md_v1 *lmm,
1756 const struct lu_fid *pfid,
1757 __u32 stripe_size, __u32 ea_off,
1758 __u32 pattern, __u16 count)
1760 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1761 lmm->lmm_pattern = cpu_to_le32(pattern);
1762 fid_to_lmm_oi(pfid, &lmm->lmm_oi);
1763 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1764 lmm->lmm_stripe_size = cpu_to_le32(stripe_size);
1765 lmm->lmm_stripe_count = cpu_to_le16(count);
1766 lmm->lmm_layout_gen = cpu_to_le16(1);
1767 memset(&lmm->lmm_objects[0], 0,
1768 sizeof(struct lov_ost_data_v1) * count);
1770 return &lmm->lmm_objects[ea_off];
1773 static int lfsck_layout_new_v1_lovea(const struct lu_env *env,
1774 struct lfsck_instance *lfsck,
1775 struct ost_layout *ol,
1776 struct dt_object *parent,
1777 struct lu_buf *buf, __u32 ea_off,
1778 struct lov_mds_md_v1 **lmm,
1779 struct lov_ost_data_v1 **objs)
1782 __u32 stripe_size = ol->ol_stripe_size;
1783 __u32 pattern = LOV_PATTERN_RAID0;
1786 if (ol->ol_stripe_count != 0)
1787 count = ol->ol_stripe_count;
1791 size = lov_mds_md_size(count, LOV_MAGIC_V1);
1792 LASSERTF(buf->lb_len >= size,
1793 "buffer len %d is less than real size %d\n",
1794 (int)buf->lb_len, size);
1796 if (stripe_size == 0) {
1799 rc = lfsck_layout_get_def_stripesize(env, lfsck, &stripe_size);
1805 if (ol->ol_stripe_count > 1 ||
1806 (ol->ol_stripe_count == 0 && ea_off != 0)) {
1807 pattern |= LOV_PATTERN_F_HOLE;
1808 memset(&(*lmm)->lmm_objects[0], 0,
1809 count * sizeof(struct lov_ost_data_v1));
1812 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1813 stripe_size, ea_off, pattern, count);
1818 static int lfsck_layout_new_comp_lovea(const struct lu_env *env,
1819 struct lu_orphan_rec_v3 *rec,
1820 struct dt_object *parent,
1821 struct lu_buf *buf, __u32 ea_off,
1822 struct lov_mds_md_v1 **lmm,
1823 struct lov_ost_data_v1 **objs)
1825 struct ost_layout *ol = &rec->lor_layout;
1826 struct lov_comp_md_v1 *lcm;
1827 struct lov_comp_md_entry_v1 *lcme;
1828 __u32 pattern = LOV_PATTERN_RAID0;
1829 __u32 offset = sizeof(*lcm) + sizeof(*lcme);
1830 int lcme_size = lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1831 int size = offset + lcme_size;
1833 LASSERTF(buf->lb_len >= size,
1834 "buffer len %d is less than real size %d\n",
1835 (int)buf->lb_len, size);
1838 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1839 lcm->lcm_size = cpu_to_le32(size);
1840 if (rec->lor_range) {
1841 lcm->lcm_layout_gen = cpu_to_le32(rec->lor_layout_version +
1843 lcm->lcm_flags = cpu_to_le16(LCM_FL_WRITE_PENDING);
1844 } else if (rec->lor_layout_version) {
1845 lcm->lcm_layout_gen = cpu_to_le32(rec->lor_layout_version +
1847 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
1849 lcm->lcm_layout_gen = cpu_to_le32(1);
1850 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
1852 lcm->lcm_entry_count = cpu_to_le16(1);
1853 /* Currently, we do not know how many mirrors will be, set it as zero
1854 * at the beginning. It will be updated when more mirrors are found. */
1855 lcm->lcm_mirror_count = 0;
1857 lcme = &lcm->lcm_entries[0];
1858 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1859 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1860 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1861 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1862 lcme->lcme_offset = cpu_to_le32(offset);
1863 lcme->lcme_size = cpu_to_le32(lcme_size);
1864 lcme->lcme_layout_gen = lcm->lcm_layout_gen;
1865 if (ol->ol_stripe_count > 1)
1866 pattern |= LOV_PATTERN_F_HOLE;
1868 *lmm = buf->lb_buf + offset;
1869 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1870 ol->ol_stripe_size, ea_off,
1871 pattern, ol->ol_stripe_count);
1876 static void lfsck_layout_update_lcm(struct lov_comp_md_v1 *lcm,
1877 struct lov_comp_md_entry_v1 *lcme,
1878 __u32 version, __u32 range)
1880 struct lov_comp_md_entry_v1 *tmp;
1881 __u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
1882 __u64 end = le64_to_cpu(lcme->lcme_extent.e_end);
1883 __u32 gen = version + range;
1886 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1887 __u16 flags = le16_to_cpu(lcm->lcm_flags);
1891 lcme->lcme_layout_gen = cpu_to_le32(gen);
1892 if (le32_to_cpu(lcm->lcm_layout_gen) < gen)
1893 lcm->lcm_layout_gen = cpu_to_le32(gen);
1896 lcm->lcm_flags = cpu_to_le16(LCM_FL_WRITE_PENDING);
1897 else if (flags == LCM_FL_NONE && le16_to_cpu(lcm->lcm_mirror_count) > 0)
1898 lcm->lcm_flags = cpu_to_le16(LCM_FL_RDONLY);
1900 for (i = 0; i < count; i++) {
1901 tmp = &lcm->lcm_entries[i];
1902 if (le64_to_cpu(tmp->lcme_extent.e_end) <= start)
1905 if (le64_to_cpu(tmp->lcme_extent.e_start) >= end)
1908 if (le32_to_cpu(tmp->lcme_flags) & LCME_FL_STALE)
1911 tmp_gen = le32_to_cpu(tmp->lcme_layout_gen);
1912 /* "lcme_layout_gen == 0" but without LCME_FL_STALE flag,
1913 * then it should be the latest version of all mirrors. */
1914 if (tmp_gen == 0 || tmp_gen > gen) {
1915 lcme->lcme_flags = cpu_to_le32(
1916 le32_to_cpu(lcme->lcme_flags) | LCME_FL_STALE);
1921 tmp->lcme_flags = cpu_to_le32(
1922 le32_to_cpu(tmp->lcme_flags) | LCME_FL_STALE);
1926 static int lfsck_layout_add_comp(const struct lu_env *env,
1927 struct lfsck_instance *lfsck,
1928 struct thandle *handle,
1929 struct lu_orphan_rec_v3 *rec,
1930 struct dt_object *parent,
1931 const struct lu_fid *cfid,
1932 struct lu_buf *buf, __u32 ost_idx,
1933 __u32 ea_off, int pos, bool new_mirror)
1935 struct ost_layout *ol = &rec->lor_layout;
1936 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1937 struct lov_comp_md_entry_v1 *lcme;
1938 struct lov_mds_md_v1 *lmm;
1939 struct lov_ost_data_v1 *objs;
1940 int added = sizeof(*lcme) +
1941 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1942 int size = le32_to_cpu(lcm->lcm_size) + added;
1946 __u32 pattern = LOV_PATTERN_RAID0;
1947 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1950 lu_buf_check_and_grow(buf, size);
1951 /* set the lcm again because lu_buf_check_and_grow() may
1952 * have reallocated the buf. */
1954 lcm->lcm_size = cpu_to_le32(size);
1955 lcm->lcm_entry_count = cpu_to_le16(count + 1);
1957 le16_add_cpu(&lcm->lcm_mirror_count, 1);
1959 /* 1. Move the component bodies from [pos, count-1] to [pos+1, count]
1960 * with distance of 'added'. */
1963 for (i = pos; i < count; i++) {
1964 lcme = &lcm->lcm_entries[i];
1965 size += le32_to_cpu(lcme->lcme_size);
1968 offset = le32_to_cpu(lcm->lcm_entries[pos].lcme_offset);
1969 memmove(buf->lb_buf + offset + added,
1970 buf->lb_buf + offset, size);
1974 /* 2. Move the component header [0, pos-1] to [0, pos-1] with distance
1975 * of 'sizeof(struct lov_comp_md_entry_v1)' */
1977 for (i = 0; i < pos; i++) {
1978 lcme = &lcm->lcm_entries[i];
1979 size += le32_to_cpu(lcme->lcme_size);
1982 offset = le32_to_cpu(lcm->lcm_entries[0].lcme_offset);
1983 memmove(buf->lb_buf + offset + sizeof(*lcme),
1984 buf->lb_buf + offset, size);
1987 /* 3. Recalculate the enter offset for the component [pos, count-1] */
1988 for (i = count - 1; i >= pos; i--) {
1989 lcm->lcm_entries[i + 1] = lcm->lcm_entries[i];
1990 lcm->lcm_entries[i + 1].lcme_offset =
1991 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i + 1].
1992 lcme_offset) + added);
1995 /* 4. Recalculate the enter offset for the component [0, pos) */
1996 for (i = 0; i < pos; i++) {
1997 lcm->lcm_entries[i].lcme_offset =
1998 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i].
1999 lcme_offset) + sizeof(*lcme));
2002 offset = sizeof(*lcm) + sizeof(*lcme) * (count + 1) + size;
2003 /* 4. Insert the new component header (entry) at the slot 'pos'. */
2004 lcme = &lcm->lcm_entries[pos];
2005 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
2006 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
2007 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
2008 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
2009 lcme->lcme_offset = cpu_to_le32(offset);
2010 lcme->lcme_size = cpu_to_le32(lov_mds_md_size(ol->ol_stripe_count,
2013 if (ol->ol_stripe_count > 1)
2014 pattern |= LOV_PATTERN_F_HOLE;
2016 lmm = buf->lb_buf + offset;
2017 /* 5. Insert teh new component body at the 'offset'. */
2018 objs = __lfsck_layout_new_v1_lovea(lmm, lfsck_dto2fid(parent),
2019 ol->ol_stripe_size, ea_off,
2020 pattern, ol->ol_stripe_count);
2022 /* 6. Update mirror related flags and version. */
2023 lfsck_layout_update_lcm(lcm, lcme, rec->lor_layout_version,
2026 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
2027 lmm, objs, LU_XATTR_REPLACE, ost_idx,
2028 le32_to_cpu(lcm->lcm_size));
2030 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant add new COMP for "
2031 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2032 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2033 "comp_end %llu, layout version %u, range %u, "
2034 "%s LOV EA hole: rc = %d\n",
2035 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2036 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2037 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2038 rec->lor_layout_version, rec->lor_range,
2039 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2040 "with" : "without", rc);
2045 static int lfsck_layout_extend_v1v3_lovea(const struct lu_env *env,
2046 struct lfsck_instance *lfsck,
2047 struct thandle *handle,
2048 struct ost_layout *ol,
2049 struct dt_object *parent,
2050 const struct lu_fid *cfid,
2051 struct lu_buf *buf, __u32 ost_idx,
2054 struct lov_mds_md_v1 *lmm = buf->lb_buf;
2055 struct lov_ost_data_v1 *objs;
2056 __u16 count = le16_to_cpu(lmm->lmm_stripe_count);
2057 __u32 magic = le32_to_cpu(lmm->lmm_magic);
2063 /* The original LOVEA maybe re-generated via old filter_fid, at
2064 * that time, we do not know the stripe count and stripe size. */
2065 if (ol->ol_stripe_count > count)
2066 count = ol->ol_stripe_count;
2067 if (ol->ol_stripe_size != 0 &&
2068 ol->ol_stripe_size != le32_to_cpu(lmm->lmm_stripe_size))
2069 lmm->lmm_stripe_size = cpu_to_le32(ol->ol_stripe_size);
2071 if (magic == LOV_MAGIC_V1)
2072 objs = &lmm->lmm_objects[count];
2074 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[count];
2076 gap = ea_off - count;
2080 size = lov_mds_md_size(count, magic);
2081 LASSERTF(buf->lb_len >= size,
2082 "buffer len %d is less than real size %d\n",
2083 (int)buf->lb_len, size);
2086 memset(objs, 0, gap * sizeof(*objs));
2087 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
2090 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2091 lmm->lmm_stripe_count = cpu_to_le16(count);
2094 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
2095 lmm, objs, LU_XATTR_REPLACE, ost_idx, size);
2097 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
2098 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2099 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2100 "comp_end %llu, %s LOV EA hole: rc = %d\n",
2101 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2102 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2103 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2104 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2105 "with" : "without", rc);
2111 * \retval +1: repaired
2112 * \retval 0: did nothing
2113 * \retval -ve: on error
2115 static int lfsck_layout_update_lovea(const struct lu_env *env,
2116 struct lfsck_instance *lfsck,
2117 struct thandle *handle,
2118 struct lu_orphan_rec_v3 *rec,
2119 struct dt_object *parent,
2120 const struct lu_fid *cfid,
2121 struct lu_buf *buf, int fl,
2122 __u32 ost_idx, __u32 ea_off)
2124 struct ost_layout *ol = &rec->lor_layout;
2125 struct lov_mds_md_v1 *lmm = NULL;
2126 struct lov_ost_data_v1 *objs = NULL;
2130 if (ol->ol_comp_id != 0)
2131 rc = lfsck_layout_new_comp_lovea(env, rec, parent, buf, ea_off,
2134 rc = lfsck_layout_new_v1_lovea(env, lfsck, &rec->lor_layout,
2135 parent, buf, ea_off, &lmm,
2138 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid,
2139 buf, lmm, objs, fl, ost_idx, rc);
2141 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant created layout EA for "
2142 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2143 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2144 "comp_end %llu, layout version %u, range %u, fl %d, "
2145 "%s LOV EA hole: rc = %d\n",
2146 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2147 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2148 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2149 rec->lor_layout_version, rec->lor_range, fl,
2150 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2151 "with" : "without", rc);
2156 static int __lfsck_layout_update_pfid(const struct lu_env *env,
2157 struct dt_object *child,
2158 const struct lu_fid *pfid,
2159 const struct ost_layout *ol, __u32 offset,
2160 __u32 version, __u32 range)
2162 struct dt_device *dev = lfsck_obj2dev(child);
2163 struct filter_fid *ff = &lfsck_env_info(env)->lti_ff;
2164 struct thandle *handle;
2165 struct lu_buf buf = { NULL };
2168 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2169 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2170 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2171 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2172 * parent MDT-object's layout EA. */
2173 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
2174 ost_layout_cpu_to_le(&ff->ff_layout, ol);
2175 ff->ff_layout_version = cpu_to_le32(version);
2176 ff->ff_range = cpu_to_le32(range);
2177 lfsck_buf_init(&buf, ff, sizeof(*ff));
2179 handle = dt_trans_create(env, dev);
2181 RETURN(PTR_ERR(handle));
2183 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2187 rc = dt_trans_start_local(env, dev, handle);
2191 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2196 dt_trans_stop(env, dev, handle);
2202 * \retval +1: repaired
2203 * \retval 0: did nothing
2204 * \retval -ve: on error
2206 static int lfsck_layout_update_pfid(const struct lu_env *env,
2207 struct lfsck_component *com,
2208 struct dt_object *parent,
2209 struct lu_fid *cfid,
2210 struct dt_device *cdev,
2211 struct lu_orphan_rec_v3 *rec, __u32 ea_off)
2213 struct dt_object *child;
2217 child = lfsck_object_find_by_dev(env, cdev, cfid);
2219 RETURN(PTR_ERR(child));
2221 rc = __lfsck_layout_update_pfid(env, child,
2222 lu_object_fid(&parent->do_lu),
2223 &rec->lor_layout, ea_off,
2224 rec->lor_layout_version,
2226 lfsck_object_put(env, child);
2228 RETURN(rc == 0 ? 1 : rc);
2231 static int lfsck_lovea_size(struct ost_layout *ol, __u32 ea_off)
2233 if (ol->ol_comp_id != 0)
2234 return sizeof(struct lov_comp_md_v1) +
2235 sizeof(struct lov_comp_md_entry_v1) +
2236 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2238 if (ol->ol_stripe_count != 0)
2239 return lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2241 return lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2245 * This function will create the MDT-object with the given (partial) LOV EA.
2247 * Under some data corruption cases, the MDT-object of the file may be lost,
2248 * but its OST-objects, or some of them are there. The layout LFSCK needs to
2249 * re-create the MDT-object with the orphan OST-object(s) information.
2251 * On the other hand, the LFSCK may has created some OST-object for repairing
2252 * dangling LOV EA reference, but as the LFSCK processing, it may find that
2253 * the old OST-object is there and should replace the former new created OST
2254 * object. Unfortunately, some others have modified such newly created object.
2255 * To keep the data (both new and old), the LFSCK will create MDT-object with
2256 * new FID to reference the original OST-object.
2258 * \param[in] env pointer to the thread context
2259 * \param[in] com pointer to the lfsck component
2260 * \param[in] ltd pointer to target device descriptor
2261 * \param[in] rec pointer to the record for the orphan OST-object
2262 * \param[in] cfid pointer to FID for the orphan OST-object
2263 * \param[in] infix additional information, such as the FID for original
2264 * MDT-object and the stripe offset in the LOV EA
2265 * \param[in] type the type for describing why the orphan MDT-object is
2266 * created. The rules are as following:
2268 * type "C": Multiple OST-objects claim the same MDT-object and the
2269 * same slot in the layout EA. Then the LFSCK will create
2270 * new MDT-object(s) to hold the conflict OST-object(s).
2272 * type "N": The orphan OST-object does not know which one was the
2273 * real parent MDT-object, so the LFSCK uses new FID for
2274 * its parent MDT-object.
2276 * type "R": The orphan OST-object knows its parent MDT-object FID,
2277 * but does not know the position (the file name) in the
2280 * type "D": The MDT-object is a directory, it may knows its parent
2281 * but because there is no valid linkEA, the LFSCK cannot
2282 * know where to put it back to the namespace.
2283 * type "O": The MDT-object has no linkEA, and there is no name
2284 * entry that references the MDT-object.
2286 * type "P": The orphan object to be created was a parent directory
2287 * of some MDT-object which linkEA shows that the @orphan
2288 * object is missing.
2290 * The orphan name will be like:
2291 * ${FID}-${infix}-${type}-${conflict_version}
2293 * \param[in] ea_off the stripe offset in the LOV EA
2295 * \retval positive on repaired something
2296 * \retval 0 if needs to repair nothing
2297 * \retval negative error number on failure
2299 static int lfsck_layout_recreate_parent(const struct lu_env *env,
2300 struct lfsck_component *com,
2301 struct lfsck_tgt_desc *ltd,
2302 struct lu_orphan_rec_v3 *rec,
2303 struct lu_fid *cfid,
2308 struct lfsck_thread_info *info = lfsck_env_info(env);
2309 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
2310 char *name = info->lti_key;
2311 struct lu_attr *la = &info->lti_la2;
2312 struct dt_object_format *dof = &info->lti_dof;
2313 struct lfsck_instance *lfsck = com->lc_lfsck;
2314 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
2315 struct lu_fid *tfid = &info->lti_fid3;
2316 struct dt_device *dev = lfsck->li_bottom;
2317 struct dt_object *lpf = lfsck->li_lpf_obj;
2318 struct dt_object *pobj = NULL;
2319 struct dt_object *cobj = NULL;
2320 struct thandle *th = NULL;
2321 struct lu_buf *ea_buf = &info->lti_big_buf;
2322 struct lu_buf lov_buf;
2323 struct lfsck_lock_handle *llh = &info->lti_llh;
2324 struct linkea_data ldata = { NULL };
2325 struct lu_buf linkea_buf;
2326 const struct lu_name *pname;
2332 if (lfsck_is_dryrun(lfsck))
2335 if (unlikely(lpf == NULL))
2336 GOTO(log, rc = -ENXIO);
2338 /* We use two separated transactions to repair the inconsistency.
2340 * 1) create the MDT-object locally.
2341 * 2) update the OST-object's PFID EA if necessary.
2343 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
2344 * updated when the layout LFSCK run next time.
2346 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
2347 * when the layout LFSCK run next time. */
2349 if (fid_is_zero(pfid)) {
2350 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
2354 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2356 GOTO(log, rc = PTR_ERR(cobj));
2359 pobj = lfsck_object_find_by_dev(env, dev, pfid);
2361 GOTO(log, rc = PTR_ERR(pobj));
2363 LASSERT(infix != NULL);
2364 LASSERT(type != NULL);
2366 memset(la, 0, sizeof(*la));
2367 la->la_uid = rec->lor_rec.lor_uid;
2368 la->la_gid = rec->lor_rec.lor_gid;
2369 la->la_mode = S_IFREG | S_IRUSR;
2370 la->la_valid = LA_MODE | LA_UID | LA_GID;
2372 memset(dof, 0, sizeof(*dof));
2373 dof->dof_type = dt_mode_to_dft(S_IFREG);
2374 /* Because the dof->dof_reg.striped = 0, the LOD will not create
2375 * the stripe(s). The LFSCK will specify the LOV EA via
2376 * lfsck_layout_update_lovea(). */
2378 size = lfsck_lovea_size(&rec->lor_layout, ea_off);
2379 if (ea_buf->lb_len < size) {
2380 lu_buf_realloc(ea_buf, size);
2381 if (ea_buf->lb_buf == NULL)
2382 GOTO(log, rc = -ENOMEM);
2387 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
2389 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2390 (const struct dt_key *)name);
2391 if (rc != 0 && rc != -ENOENT)
2395 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
2396 MDS_INODELOCK_UPDATE, LCK_PW);
2400 /* Re-check whether the name conflict with othrs after taken
2402 rc = dt_lookup(env, lfsck->li_lpf_obj, (struct dt_rec *)tfid,
2403 (const struct dt_key *)name);
2404 if (unlikely(rc == 0)) {
2412 pname = lfsck_name_get_const(env, name, strlen(name));
2413 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
2414 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
2418 /* The 1st transaction. */
2419 th = dt_trans_create(env, dev);
2421 GOTO(unlock, rc = PTR_ERR(th));
2423 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
2427 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
2428 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
2429 LU_XATTR_CREATE, th);
2433 dtrec->rec_fid = pfid;
2434 dtrec->rec_type = S_IFREG;
2435 rc = dt_declare_insert(env, lpf,
2436 (const struct dt_rec *)dtrec,
2437 (const struct dt_key *)name, th);
2441 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2442 ldata.ld_leh->leh_len);
2443 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
2444 XATTR_NAME_LINK, 0, th);
2448 rc = dt_trans_start_local(env, dev, th);
2452 dt_write_lock(env, pobj, 0);
2453 rc = dt_create(env, pobj, la, NULL, dof, th);
2455 rc = lfsck_layout_update_lovea(env, lfsck, th, rec, pobj, cfid,
2456 &lov_buf, LU_XATTR_CREATE, ltd->ltd_index, ea_off);
2457 dt_write_unlock(env, pobj);
2461 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
2462 (const struct dt_key *)name, th);
2466 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
2467 if (rc == 0 && cobj != NULL) {
2468 dt_trans_stop(env, dev, th);
2471 /* The 2nd transaction. */
2472 rc = __lfsck_layout_update_pfid(env, cobj, pfid,
2473 &rec->lor_layout, ea_off,
2474 rec->lor_layout_version,
2482 dt_trans_stop(env, dev, th);
2488 if (cobj != NULL && !IS_ERR(cobj))
2489 lfsck_object_put(env, cobj);
2490 if (pobj != NULL && !IS_ERR(pobj))
2491 lfsck_object_put(env, pobj);
2494 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2495 "recreate the lost MDT-object: parent "DFID
2496 ", child "DFID", OST-index %u, stripe-index %u, "
2497 "infix %s, type %s: rc = %d\n",
2498 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2499 ltd->ltd_index, ea_off, infix, type, rc);
2501 return rc >= 0 ? 1 : rc;
2504 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2505 struct lfsck_component *com,
2506 const struct lu_fid *fid,
2509 struct lfsck_thread_info *info = lfsck_env_info(env);
2510 struct lfsck_request *lr = &info->lti_lr;
2511 struct lfsck_instance *lfsck = com->lc_lfsck;
2512 struct lfsck_tgt_desc *ltd;
2513 struct ptlrpc_request *req;
2514 struct lfsck_request *tmp;
2515 struct obd_export *exp;
2519 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2520 if (unlikely(ltd == NULL))
2524 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2525 GOTO(put, rc = -EOPNOTSUPP);
2527 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2529 GOTO(put, rc = -ENOMEM);
2531 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2533 ptlrpc_request_free(req);
2538 memset(lr, 0, sizeof(*lr));
2539 lr->lr_event = LE_CONDITIONAL_DESTROY;
2540 lr->lr_active = LFSCK_TYPE_LAYOUT;
2543 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2545 ptlrpc_request_set_replen(req);
2547 rc = ptlrpc_queue_wait(req);
2548 ptlrpc_req_finished(req);
2558 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2559 struct lfsck_component *com,
2560 struct lfsck_request *lr)
2562 struct lfsck_thread_info *info = lfsck_env_info(env);
2563 struct lu_attr *la = &info->lti_la;
2564 union ldlm_policy_data *policy = &info->lti_policy;
2565 struct ldlm_res_id *resid = &info->lti_resid;
2566 struct lfsck_instance *lfsck = com->lc_lfsck;
2567 struct dt_device *dev = lfsck->li_bottom;
2568 struct lu_fid *fid = &lr->lr_fid;
2569 struct dt_object *obj;
2570 struct thandle *th = NULL;
2571 struct lustre_handle lh = { 0 };
2576 obj = lfsck_object_find_by_dev(env, dev, fid);
2578 RETURN(PTR_ERR(obj));
2580 dt_read_lock(env, obj, 0);
2581 if (dt_object_exists(obj) == 0 ||
2582 lfsck_is_dead_obj(obj)) {
2583 dt_read_unlock(env, obj);
2585 GOTO(put, rc = -ENOENT);
2588 /* Get obj's attr without lock firstly. */
2589 rc = dt_attr_get(env, obj, la);
2590 dt_read_unlock(env, obj);
2594 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2595 GOTO(put, rc = -ETXTBSY);
2597 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2598 LASSERT(lfsck->li_namespace != NULL);
2600 memset(policy, 0, sizeof(*policy));
2601 policy->l_extent.end = OBD_OBJECT_EOF;
2602 ost_fid_build_resid(fid, resid);
2603 rc = ldlm_cli_enqueue_local(env, lfsck->li_namespace, resid,
2604 LDLM_EXTENT, policy, LCK_EX, &flags,
2605 ldlm_blocking_ast, ldlm_completion_ast,
2606 NULL, NULL, 0, LVB_T_NONE, NULL, &lh);
2608 GOTO(put, rc = -EIO);
2610 dt_write_lock(env, obj, 0);
2611 /* Get obj's attr within lock again. */
2612 rc = dt_attr_get(env, obj, la);
2616 if (la->la_ctime != 0)
2617 GOTO(unlock, rc = -ETXTBSY);
2619 th = dt_trans_create(env, dev);
2621 GOTO(unlock, rc = PTR_ERR(th));
2623 rc = dt_declare_ref_del(env, obj, th);
2627 rc = dt_declare_destroy(env, obj, th);
2631 rc = dt_trans_start_local(env, dev, th);
2635 rc = dt_ref_del(env, obj, th);
2639 rc = dt_destroy(env, obj, th);
2641 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2642 "OST-object "DFID" that was created for reparing "
2643 "dangling referenced case. But the original missing "
2644 "OST-object is found now.\n",
2645 lfsck_lfsck2name(lfsck), PFID(fid));
2650 dt_trans_stop(env, dev, th);
2653 dt_write_unlock(env, obj);
2654 ldlm_lock_decref(&lh, LCK_EX);
2657 lfsck_object_put(env, obj);
2663 * Some OST-object has occupied the specified layout EA slot.
2664 * Such OST-object may be generated by the LFSCK when repair
2665 * dangling referenced MDT-object, which can be indicated by
2666 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2667 * is true and such OST-object has not been modified yet, we
2668 * will replace it with the orphan OST-object; otherwise the
2669 * LFSCK will create new MDT-object to reference the orphan.
2671 * \retval +1: repaired
2672 * \retval 0: did nothing
2673 * \retval -ve: on error
2675 static int lfsck_layout_conflict_create(const struct lu_env *env,
2676 struct lfsck_component *com,
2677 struct lfsck_tgt_desc *ltd,
2678 struct lu_orphan_rec_v3 *rec,
2679 struct dt_object *parent,
2680 struct lu_fid *cfid,
2681 struct lu_buf *ea_buf,
2682 struct lov_mds_md_v1 *lmm,
2683 struct lov_ost_data_v1 *slot,
2684 __u32 ea_off, int lovea_size)
2686 struct lfsck_thread_info *info = lfsck_env_info(env);
2687 struct lu_fid *cfid2 = &info->lti_fid2;
2688 struct ost_id *oi = &info->lti_oi;
2689 struct dt_device *dev = lfsck_obj2dev(parent);
2690 struct thandle *th = NULL;
2691 struct lustre_handle lh = { 0 };
2692 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2696 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2697 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2701 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2702 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2706 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2707 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2712 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2714 /* If the conflict OST-obejct is not created for fixing dangling
2715 * referenced MDT-object in former LFSCK check/repair, or it has
2716 * been modified by others, then we cannot destroy it. Re-create
2717 * a new MDT-object for the orphan OST-object. */
2718 if (rc == -ETXTBSY) {
2719 /* No need the layout lock on the original parent. */
2720 lfsck_ibits_unlock(&lh, LCK_EX);
2722 fid_zero(&rec->lor_rec.lor_fid);
2723 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2724 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2726 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2727 info->lti_tmpbuf, "C", ea_off);
2732 if (rc != 0 && rc != -ENOENT)
2735 if (lfsck_is_dryrun(com->lc_lfsck))
2736 GOTO(unlock, rc = 0);
2738 th = dt_trans_create(env, dev);
2740 GOTO(unlock, rc = PTR_ERR(th));
2742 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2743 LU_XATTR_REPLACE, th);
2747 rc = dt_trans_start_local(env, dev, th);
2751 dt_write_lock(env, parent, 0);
2752 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2753 rc = lfsck_layout_refill_lovea(env, com->lc_lfsck, th, parent, cfid,
2754 ea_buf, lmm, slot, LU_XATTR_REPLACE,
2755 ltd->ltd_index, lovea_size);
2756 dt_write_unlock(env, parent);
2761 dt_trans_stop(env, dev, th);
2764 lfsck_ibits_unlock(&lh, LCK_EX);
2767 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2768 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2769 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2770 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2771 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2774 return rc >= 0 ? 1 : rc;
2778 * \retval +1: repaired
2779 * \retval 0: did nothing
2780 * \retval -ve: on error
2782 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2783 struct lfsck_component *com,
2784 struct lfsck_tgt_desc *ltd,
2785 struct lu_orphan_rec_v3 *rec,
2786 struct dt_object *parent,
2787 struct lu_fid *cfid,
2788 __u32 ost_idx, __u32 ea_off)
2790 struct lfsck_thread_info *info = lfsck_env_info(env);
2791 struct lu_buf *buf = &info->lti_big_buf;
2792 struct lu_fid *fid = &info->lti_fid2;
2793 struct ost_id *oi = &info->lti_oi;
2794 struct lfsck_instance *lfsck = com->lc_lfsck;
2795 struct dt_device *dt = lfsck_obj2dev(parent);
2796 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2797 struct ost_layout *ol = &rec->lor_layout;
2798 struct lov_comp_md_v1 *lcm = NULL;
2799 struct lov_comp_md_entry_v1 *lcme = NULL;
2800 struct thandle *handle = NULL;
2802 struct lov_mds_md_v1 *lmm;
2803 struct lov_ost_data_v1 *objs;
2804 struct lustre_handle lh = { 0 };
2813 bool locked = false;
2814 bool new_mirror = true;
2817 if (lfsck_is_dryrun(lfsck))
2820 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2821 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2824 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2825 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2826 "stripe-index %u, comp_id %u, comp_start %llu, "
2827 "comp_end %llu, layout version %u, range %u: rc = %d\n",
2828 lfsck_lfsck2name(lfsck), PFID(cfid),
2829 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off,
2830 ol->ol_comp_id, ol->ol_comp_start,
2831 ol->ol_comp_end, rec->lor_layout_version,
2832 rec->lor_range, rc);
2839 dt_write_unlock(env, parent);
2843 if (handle != NULL) {
2844 dt_trans_stop(env, dt, handle);
2849 GOTO(unlock_layout, rc);
2852 if (buf->lb_len < lovea_size) {
2853 lu_buf_realloc(buf, lovea_size);
2854 if (buf->lb_buf == NULL)
2855 GOTO(unlock_layout, rc = -ENOMEM);
2858 if (!(bk->lb_param & LPF_DRYRUN)) {
2859 handle = dt_trans_create(env, dt);
2861 GOTO(unlock_layout, rc = PTR_ERR(handle));
2863 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2868 rc = dt_trans_start_local(env, dt, handle);
2873 dt_write_lock(env, parent, 0);
2875 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2876 if (rc == -ERANGE) {
2877 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2880 } else if (rc == -ENODATA || rc == 0) {
2881 lovea_size = lfsck_lovea_size(ol, ea_off);
2882 /* If the declared is not big enough, re-try. */
2883 if (buf->lb_len < lovea_size) {
2887 fl = LU_XATTR_CREATE;
2888 } else if (rc < 0) {
2889 GOTO(unlock_parent, rc);
2890 } else if (unlikely(buf->lb_len == 0)) {
2893 fl = LU_XATTR_REPLACE;
2897 if (fl == LU_XATTR_CREATE) {
2898 if (bk->lb_param & LPF_DRYRUN)
2899 GOTO(unlock_parent, rc = 1);
2901 LASSERT(buf->lb_len >= lovea_size);
2903 rc = lfsck_layout_update_lovea(env, lfsck, handle, rec, parent,
2904 cfid, buf, fl, ost_idx, ea_off);
2906 GOTO(unlock_parent, rc);
2910 rc1 = lfsck_layout_verify_header(parent, lmm, lovea_size);
2912 /* If the LOV EA crashed, the rebuild it. */
2913 if (rc1 == -EINVAL) {
2914 if (bk->lb_param & LPF_DRYRUN)
2915 GOTO(unlock_parent, rc = 1);
2917 LASSERT(buf->lb_len >= lovea_size);
2919 rc = lfsck_layout_update_lovea(env, lfsck, handle, rec, parent,
2920 cfid, buf, fl, ost_idx, ea_off);
2922 GOTO(unlock_parent, rc);
2925 /* For other unknown magic/pattern, keep the current LOV EA. */
2926 if (rc1 == -EOPNOTSUPP)
2927 GOTO(unlock_parent, rc1 = 0);
2930 GOTO(unlock_parent, rc = rc1);
2932 magic = le32_to_cpu(lmm->lmm_magic);
2933 if (magic == LOV_MAGIC_COMP_V1) {
2936 __u16 mirror_id0 = mirror_id_of(ol->ol_comp_id);
2939 if (bk->lb_param & LPF_DRYRUN)
2940 GOTO(unlock_parent, rc = 1);
2943 count = le16_to_cpu(lcm->lcm_entry_count);
2944 for (i = 0; i < count; pos = ++i) {
2945 lcme = &lcm->lcm_entries[i];
2946 start = le64_to_cpu(lcme->lcme_extent.e_start);
2947 end = le64_to_cpu(lcme->lcme_extent.e_end);
2948 mirror_id1 = mirror_id_of(le32_to_cpu(lcme->lcme_id));
2950 if (mirror_id0 > mirror_id1)
2953 if (mirror_id0 < mirror_id1)
2957 if (end <= ol->ol_comp_start)
2960 if (start >= ol->ol_comp_end)
2963 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
2964 magic = le32_to_cpu(lmm->lmm_magic);
2965 flags = le32_to_cpu(lcme->lcme_flags);
2969 rc = lfsck_layout_add_comp(env, lfsck, handle, rec, parent,
2970 cfid, buf, ost_idx, ea_off, pos, new_mirror);
2972 GOTO(unlock_parent, rc);
2976 count = le16_to_cpu(lmm->lmm_stripe_count);
2978 GOTO(unlock_parent, rc = -EINVAL);
2981 /* Exceed the current end of MDT-object layout EA. Then extend it. */
2982 if (count <= ea_off) {
2983 if (bk->lb_param & LPF_DRYRUN)
2984 GOTO(unlock_parent, rc = 1);
2986 lovea_size = lov_mds_md_size(ea_off + 1, magic);
2987 /* If the declared is not big enough, re-try. */
2988 if (buf->lb_len < lovea_size) {
2996 lcme->lcme_flags = cpu_to_le32(flags | LCME_FL_INIT);
2997 lfsck_layout_update_lcm(lcm, lcme,
2998 rec->lor_layout_version,
3002 rc = lfsck_layout_extend_v1v3_lovea(env, lfsck, handle, ol,
3003 parent, cfid, buf, ost_idx, ea_off);
3005 GOTO(unlock_parent, rc);
3008 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
3010 if (magic == LOV_MAGIC_V1) {
3011 objs = &lmm->lmm_objects[0];
3013 LASSERT(magic == LOV_MAGIC_V3);
3014 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3017 for (i = 0; i < count; i++, objs++) {
3018 /* The MDT-object was created via lfsck_layout_recover_create()
3019 * by others before, and we fill the dummy layout EA. */
3020 if ((lcme && !(flags & LCME_FL_INIT)) ||
3021 lovea_slot_is_dummy(objs)) {
3025 if (bk->lb_param & LPF_DRYRUN)
3026 GOTO(unlock_parent, rc = 1);
3028 lmm->lmm_layout_gen =
3029 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3033 if (le32_to_cpu(lmm->lmm_stripe_size) !=
3034 ol->ol_stripe_size ||
3035 le16_to_cpu(lmm->lmm_stripe_count) !=
3036 ol->ol_stripe_count ||
3037 le64_to_cpu(lcme->lcme_extent.e_start) !=
3038 ol->ol_comp_start ||
3039 le64_to_cpu(lcme->lcme_extent.e_end) !=
3041 CDEBUG(D_LFSCK, "%s: found invalid "
3042 "component for "DFID ": parent "DFID
3043 ", stripe-index %u, stripe_size %u, "
3044 "stripe_count %u, comp_id %u, "
3045 "comp_start %llu, comp_end %llu, "
3046 "cur_stripe_size %u, "
3047 "cur_stripe_count %u, "
3048 "cur_comp_start %llu, "
3049 "cur_comp_end %llu\n",
3050 lfsck_lfsck2name(lfsck), PFID(cfid),
3051 PFID(lfsck_dto2fid(parent)), ea_off,
3053 ol->ol_stripe_count, ol->ol_comp_id,
3054 ol->ol_comp_start, ol->ol_comp_end,
3055 le32_to_cpu(lmm->lmm_stripe_size),
3056 le16_to_cpu(lmm->lmm_stripe_count),
3057 le64_to_cpu(lcme->lcme_extent.e_start),
3058 le64_to_cpu(lcme->lcme_extent.e_end));
3060 GOTO(unlock_parent, rc = -EINVAL);
3063 lovea_size = le32_to_cpu(lcm->lcm_size);
3064 lcme->lcme_flags = cpu_to_le32(flags |
3066 lfsck_layout_update_lcm(lcm, lcme,
3067 rec->lor_layout_version,
3071 LASSERTF(buf->lb_len >= lovea_size,
3072 "buffer len %d is less than real size %d\n",
3073 (int)buf->lb_len, (int)lovea_size);
3075 rc = lfsck_layout_refill_lovea(env, lfsck, handle,
3076 parent, cfid, buf, lmm, objs,
3077 fl, ost_idx, lovea_size);
3079 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
3080 "dummy layout slot for "DFID": parent "DFID
3081 ", OST-index %u, stripe-index %u: rc = %d\n",
3082 lfsck_lfsck2name(lfsck), PFID(cfid),
3083 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
3085 GOTO(unlock_parent, rc);
3088 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3089 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
3091 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3092 "invalid layout EA at the slot %d, index %u\n",
3093 lfsck_lfsck2name(lfsck),
3094 PFID(lfsck_dto2fid(parent)), i,
3095 le32_to_cpu(objs->l_ost_idx));
3097 GOTO(unlock_parent, rc);
3100 /* It should be rare case, the slot is there, but the LFSCK
3101 * does not handle it during the first-phase cycle scanning. */
3102 if (unlikely(lu_fid_eq(fid, cfid))) {
3104 GOTO(unlock_parent, rc = 0);
3106 /* Rare case that the OST-object index
3107 * does not match the parent MDT-object
3108 * layout EA. We trust the later one. */
3109 if (bk->lb_param & LPF_DRYRUN)
3110 GOTO(unlock_parent, rc = 1);
3112 dt_write_unlock(env, parent);
3114 dt_trans_stop(env, dt, handle);
3115 lfsck_ibits_unlock(&lh, LCK_EX);
3116 rc = lfsck_layout_update_pfid(env, com, parent,
3120 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
3121 "updated OST-object's pfid for "DFID
3122 ": parent "DFID", OST-index %u, "
3123 "stripe-index %u: rc = %d\n",
3124 lfsck_lfsck2name(lfsck), PFID(cfid),
3125 PFID(lfsck_dto2fid(parent)),
3126 ltd->ltd_index, i, rc);
3133 /* The MDT-object exists, but related layout EA slot is occupied
3135 if (bk->lb_param & LPF_DRYRUN)
3136 GOTO(unlock_parent, rc = 1);
3138 dt_write_unlock(env, parent);
3140 dt_trans_stop(env, dt, handle);
3141 lfsck_ibits_unlock(&lh, LCK_EX);
3142 if (magic == LOV_MAGIC_V1)
3143 objs = &lmm->lmm_objects[ea_off];
3145 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
3146 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
3147 buf, lmm, objs, ea_off, lovea_size);
3153 dt_write_unlock(env, parent);
3157 dt_trans_stop(env, dt, handle);
3160 lfsck_ibits_unlock(&lh, LCK_EX);
3165 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
3166 struct lfsck_component *com,
3167 struct lfsck_tgt_desc *ltd,
3168 struct lu_orphan_rec_v3 *rec,
3169 struct lu_fid *cfid)
3171 struct lfsck_layout *lo = com->lc_file_ram;
3172 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
3173 struct dt_object *parent = NULL;
3174 __u32 ea_off = pfid->f_stripe_idx;
3178 if (!fid_is_sane(cfid))
3179 GOTO(out, rc = -EINVAL);
3182 if (fid_is_zero(pfid)) {
3183 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3188 if (!fid_is_sane(pfid))
3189 GOTO(out, rc = -EINVAL);
3191 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
3193 GOTO(out, rc = PTR_ERR(parent));
3195 if (unlikely(dt_object_remote(parent) != 0))
3196 GOTO(put, rc = -EXDEV);
3198 if (dt_object_exists(parent) == 0) {
3199 lfsck_object_put(env, parent);
3200 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3205 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
3206 GOTO(put, rc = -EISDIR);
3208 /* The orphan OST-object claims to be the parent's stripe, then
3209 * related dangling record in the trace file is meaningless. */
3210 rc = lfsck_layout_del_dangling_rec(env, com, pfid,
3211 rec->lor_layout.ol_comp_id, ea_off);
3212 if (rc && rc != -ENOENT)
3215 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
3216 ltd->ltd_index, ea_off);
3222 lfsck_object_put(env, parent);
3224 /* The layout EA is changed, need to be reloaded next time. */
3225 dt_object_put_nocache(env, parent);
3228 down_write(&com->lc_sem);
3229 com->lc_new_scanned++;
3230 com->lc_new_checked++;
3232 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
3234 } else if (rc < 0) {
3235 lo->ll_objs_failed_phase2++;
3237 up_write(&com->lc_sem);
3242 static int lfsck_layout_scan_orphan(const struct lu_env *env,
3243 struct lfsck_component *com,
3244 struct lfsck_tgt_desc *ltd)
3246 struct lfsck_assistant_data *lad = com->lc_data;
3247 struct lfsck_instance *lfsck = com->lc_lfsck;
3248 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3249 struct lfsck_thread_info *info = lfsck_env_info(env);
3250 struct lu_fid *fid = &info->lti_fid;
3251 struct dt_object *obj;
3252 const struct dt_it_ops *iops;
3257 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
3258 "scanning for OST%04x\n",
3259 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3261 if (cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
3262 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
3263 "scanning for OST%04x\n",
3264 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3269 fid->f_seq = fid_idif_seq(0, ltd->ltd_index);
3270 fid->f_oid = fid->f_ver = 0;
3272 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
3273 if (unlikely(IS_ERR(obj)))
3274 GOTO(log, rc = PTR_ERR(obj));
3276 rc = obj->do_ops->do_index_try(env, obj,
3277 &dt_lfsck_layout_orphan_features);
3281 iops = &obj->do_index_ops->dio_it;
3282 di = iops->init(env, obj, 0);
3284 GOTO(put, rc = PTR_ERR(di));
3286 rc = iops->load(env, di, 0);
3288 /* -ESRCH means that the orphan OST-objects rbtree has been
3289 * cleanup because of the OSS server restart or other errors. */
3290 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
3295 rc = iops->next(env, di);
3307 struct lu_orphan_rec_v3 *rec = &info->lti_rec;
3309 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3310 unlikely(!thread_is_running(&lfsck->li_thread)))
3313 key = iops->key(env, di);
3314 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
3315 /* Remote target OST may be runnning old LFSCK */
3316 memset(rec, 0, sizeof(*rec));
3317 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
3319 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
3320 &com->lc_fid_latest_scanned_phase2);
3321 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3324 lfsck_control_speed_by_self(com);
3326 rc = iops->next(env, di);
3327 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
3334 iops->fini(env, di);
3336 lfsck_object_put(env, obj);
3339 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
3340 "scanning for OST%04x: rc = %d\n",
3341 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
3343 return rc > 0 ? 0 : rc;
3346 static int lfsck_lov2layout(struct lov_mds_md_v1 *lmm, struct filter_fid *ff,
3349 struct ost_layout *ol = &ff->ff_layout;
3350 __u32 magic = le32_to_cpu(lmm->lmm_magic);
3354 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3) {
3355 ol->ol_stripe_size = lmm->lmm_stripe_size;
3356 ol->ol_stripe_count = lmm->lmm_stripe_count;
3357 ol->ol_comp_start = 0;
3358 ol->ol_comp_end = 0;
3360 ff->ff_layout_version = 0;
3362 } else if (magic == LOV_MAGIC_COMP_V1) {
3363 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
3364 struct lov_comp_md_entry_v1 *lcme = NULL;
3365 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3368 for (i = 0; i < count; i++) {
3369 lcme = &lcm->lcm_entries[i];
3370 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3371 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3378 /* The comp has been removed, do nothing. */
3382 lmm = (void *)lmm + le32_to_cpu(lcme->lcme_offset);
3383 ol->ol_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
3384 ol->ol_stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
3385 ol->ol_comp_start = le64_to_cpu(lcme->lcme_extent.e_start);
3386 ol->ol_comp_end = le64_to_cpu(lcme->lcme_extent.e_end);
3387 ol->ol_comp_id = le32_to_cpu(lcme->lcme_id);
3388 ff->ff_layout_version = le32_to_cpu(lcme->lcme_layout_gen);
3391 GOTO(out, rc = -EINVAL);
3401 * Repair the MDT-object with dangling LOV EA reference.
3403 * we need to repair the inconsistency according to the users' requirement:
3405 * 1) Keep the inconsistency there and report the inconsistency case,
3406 * then give the chance to the application to find related issues,
3407 * and the users can make the decision about how to handle it with
3408 * more human knownledge. (by default)
3410 * 2) Re-create the missing OST-object with the FID/owner information.
3412 * \param[in] env pointer to the thread context
3413 * \param[in] com the layout LFSCK component
3414 * \param[in] parent the MDT-object with dangling LOV EA reference
3415 * \param[in] child the OST-object to be created
3416 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3417 * \param[in] ea_off the offset of the OST-object in the LOV EA
3418 * \param[in] ost_idx the index of OST on which the OST-object resides
3420 * \retval +1 for repair successfully
3421 * \retval 0 for did nothing
3422 * \retval negative error number on failure
3424 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
3425 struct lfsck_component *com,
3426 struct dt_object *parent,
3427 struct dt_object *child,
3428 __u32 comp_id, __u32 ea_off,
3429 __u32 ost_idx, bool log)
3431 struct lfsck_thread_info *info = lfsck_env_info(env);
3432 struct filter_fid *ff = &info->lti_ff;
3433 struct dt_object_format *dof = &info->lti_dof;
3434 struct lu_attr *la = &info->lti_la;
3435 struct lfsck_instance *lfsck = com->lc_lfsck;
3436 struct dt_device *dev = lfsck_obj2dev(child);
3437 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3438 const struct lu_fid *cfid = lfsck_dto2fid(child);
3439 struct lu_buf *tbuf = &info->lti_big_buf;
3440 struct thandle *handle;
3442 struct lustre_handle lh = { 0 };
3446 if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
3449 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3450 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3455 rc = dt_attr_get(env, parent, la);
3459 la->la_mode = S_IFREG | 0666;
3460 la->la_atime = la->la_mtime = la->la_ctime = 0;
3461 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
3462 LA_ATIME | LA_MTIME | LA_CTIME;
3463 memset(dof, 0, sizeof(*dof));
3464 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
3465 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
3466 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3467 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3468 * parent MDT-object's layout EA. */
3469 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
3471 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3472 if (unlikely(rc == -ENODATA))
3477 rc = lfsck_lov2layout(tbuf->lb_buf, ff, comp_id);
3481 buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
3482 handle = dt_trans_create(env, dev);
3484 GOTO(unlock1, rc = PTR_ERR(handle));
3486 rc = dt_declare_create(env, child, la, NULL, dof, handle);
3490 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
3491 LU_XATTR_CREATE, handle);
3495 rc = dt_trans_start_local(env, dev, handle);
3499 dt_read_lock(env, parent, 0);
3500 if (unlikely(lfsck_is_dead_obj(parent)))
3501 GOTO(unlock2, rc = 0);
3503 if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
3504 struct ost_id *oi = &info->lti_oi;
3505 struct lu_fid *tfid = &info->lti_fid2;
3506 struct lu_buf *lovea = &info->lti_big_buf;
3507 struct lov_mds_md_v1 *lmm;
3508 struct lov_ost_data_v1 *objs;
3513 rc = lfsck_layout_get_lovea(env, parent, lovea);
3514 if (unlikely(rc == -ENODATA))
3519 lmm = lovea->lb_buf;
3520 magic = le32_to_cpu(lmm->lmm_magic);
3521 if (magic == LOV_MAGIC_COMP_V1) {
3522 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3523 struct lov_comp_md_entry_v1 *lcme;
3524 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3527 for (i = 0; i < count; i++) {
3528 lcme = &lcm->lcm_entries[i];
3529 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3530 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3533 lmm = lovea->lb_buf +
3534 le32_to_cpu(lcme->lcme_offset);
3535 magic = le32_to_cpu(lmm->lmm_magic);
3540 /* Someone removed the component, do nothing. */
3541 GOTO(unlock2, rc = 0);
3545 count = le16_to_cpu(lmm->lmm_stripe_count);
3546 /* Someone changed the LOV EA, do nothing. */
3547 if (count <= ea_off)
3548 GOTO(unlock2, rc = 0);
3550 if (magic == LOV_MAGIC_V1) {
3551 objs = &lmm->lmm_objects[ea_off];
3553 LASSERT(magic == LOV_MAGIC_V3);
3555 objs = &((struct lov_mds_md_v3 *)lmm)->\
3556 lmm_objects[ea_off];
3559 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3560 idx2 = le32_to_cpu(objs->l_ost_idx);
3561 rc = ostid_to_fid(tfid, oi, idx2);
3562 /* Someone changed the LOV EA, do nothing. */
3563 if (rc != 0 || !lu_fid_eq(tfid, cfid))
3567 rc = dt_create(env, child, la, NULL, dof, handle);
3571 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
3577 dt_read_unlock(env, parent);
3580 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3583 lfsck_ibits_unlock(&lh, LCK_EX);
3587 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3588 "dangling reference for: parent "DFID", child "
3589 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: "
3591 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3592 comp_id, ea_off, ost_idx,
3593 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3594 "Create the lost OST-object as required" :
3595 "Keep the MDT-object there by default", rc);
3601 * Repair the MDT-object with dangling LOV EA reference.
3603 * Prepare parameters and call __lfsck_layout_repair_dangling()
3604 * to repair the dangling LOV EA reference.
3606 * \param[in] env pointer to the thread context
3607 * \param[in] com the layout LFSCK component
3608 * \param[in] pfid the MDT-object's FID
3609 * \param[in] cfid the FID for the OST-object to be created
3610 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3611 * \param[in] ea_off the offset of the OST-object in the LOV EA
3612 * \param[in] ost_idx the index of OST on which the OST-object resides
3614 * \retval +1 for repair successfully
3615 * \retval 0 for did nothing
3616 * \retval negative error number on failure
3618 static int lfsck_layout_repair_dangling(const struct lu_env *env,
3619 struct lfsck_component *com,
3620 const struct lu_fid *pfid,
3621 const struct lu_fid *cfid,
3622 __u32 comp_id, __u32 ea_off,
3625 struct lfsck_instance *lfsck = com->lc_lfsck;
3626 struct dt_object *parent = NULL;
3627 struct dt_object *child = NULL;
3628 struct lfsck_tgt_desc *ltd;
3632 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3634 GOTO(log, rc = PTR_ERR(parent));
3636 /* The MDT-object has been removed. */
3637 if (dt_object_exists(parent) == 0)
3640 ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
3641 if (unlikely(ltd == NULL))
3642 GOTO(log, rc = -ENODEV);
3644 child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
3646 GOTO(log, rc = PTR_ERR(child));
3648 /* The OST-object has been created. */
3649 if (unlikely(dt_object_exists(child) != 0))
3652 rc = __lfsck_layout_repair_dangling(env, com, parent, child,
3653 comp_id, ea_off, ost_idx, false);
3658 if (child != NULL && !IS_ERR(child))
3659 lfsck_object_put(env, child);
3661 if (parent != NULL && !IS_ERR(parent))
3662 lfsck_object_put(env, parent);
3665 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3666 "dangling reference for: parent "DFID", child "
3667 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: rc = %d\n",
3668 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3669 comp_id, ea_off, ost_idx,
3670 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3671 "Create the lost OST-object as required" :
3672 "Keep the MDT-object there by default", rc);
3677 /* If the OST-object does not recognize the MDT-object as its parent, and
3678 * there is no other MDT-object claims as its parent, then just trust the
3679 * given MDT-object as its parent. So update the OST-object filter_fid. */
3680 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
3681 struct lfsck_component *com,
3682 struct dt_object *parent,
3683 struct lfsck_layout_req *llr,
3686 struct lfsck_thread_info *info = lfsck_env_info(env);
3687 struct filter_fid *ff = &info->lti_ff;
3688 struct dt_object *child = llr->llr_child;
3689 struct dt_device *dev = lfsck_obj2dev(child);
3690 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
3691 struct lu_buf *tbuf = &info->lti_big_buf;
3692 struct thandle *handle;
3694 struct lustre_handle lh = { 0 };
3698 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3699 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3704 ff->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3705 ff->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3706 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3707 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3708 * parent MDT-object's layout EA. */
3709 ff->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3711 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3712 if (unlikely(rc == -ENODATA))
3717 rc = lfsck_lov2layout(tbuf->lb_buf, ff, llr->llr_comp_id);
3721 buf = lfsck_buf_get(env, ff, sizeof(*ff));
3723 handle = dt_trans_create(env, dev);
3725 GOTO(unlock1, rc = PTR_ERR(handle));
3727 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3731 rc = dt_attr_get(env, parent, la);
3735 la->la_valid = LA_UID | LA_GID;
3736 rc = dt_declare_attr_set(env, child, la, handle);
3740 rc = dt_trans_start_local(env, dev, handle);
3744 dt_write_lock(env, parent, 0);
3745 if (unlikely(lfsck_is_dead_obj(parent)))
3746 GOTO(unlock2, rc = 1);
3748 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3752 /* Get the latest parent's owner. */
3753 rc = dt_attr_get(env, parent, la);
3757 la->la_valid = LA_UID | LA_GID;
3758 rc = dt_attr_set(env, child, la, handle);
3763 dt_write_unlock(env, parent);
3766 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3769 lfsck_ibits_unlock(&lh, LCK_EX);
3773 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3774 "unmatched MDT-OST pair for: parent "DFID
3775 ", child "DFID", comp_id %u, OST-index %u, "
3776 "stripe-index %u, owner %u/%u: rc = %d\n",
3777 lfsck_lfsck2name(com->lc_lfsck),
3778 PFID(lfsck_dto2fid(parent)),
3779 PFID(lfsck_dto2fid(child)),
3780 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3781 la->la_uid, la->la_gid, rc);
3786 /* If there are more than one MDT-objects claim as the OST-object's parent,
3787 * and the OST-object only recognizes one of them, then we need to generate
3788 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3789 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3790 struct lfsck_component *com,
3791 struct dt_object *parent,
3792 struct lfsck_layout_req *llr,
3795 struct lfsck_thread_info *info = lfsck_env_info(env);
3796 struct dt_allocation_hint *hint = &info->lti_hint;
3797 struct dt_object_format *dof = &info->lti_dof;
3798 struct ost_id *oi = &info->lti_oi;
3799 struct lu_buf *buf = &info->lti_big_buf;
3800 struct lfsck_instance *lfsck = com->lc_lfsck;
3801 struct dt_device *dev;
3802 struct lu_device *d =
3803 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3804 struct lu_object *o;
3805 struct lu_object *n;
3806 struct dt_object *child = NULL;
3807 struct thandle *handle = NULL;
3808 struct lov_mds_md_v1 *lmm;
3809 struct lov_ost_data_v1 *objs;
3810 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3812 struct lustre_handle lh = { 0 };
3818 /* We use two separated transactions to repair the inconsistency.
3820 * 1) create the child (OST-object).
3821 * 2) update the parent LOV EA according to the child's FID.
3823 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3824 * handled as orphan when the layout LFSCK run next time.
3826 * If 1) failed, but 2) succeed, then such OST-object will be re-created
3827 * as dangling referened case when the layout LFSCK run next time. */
3829 /* The 1st transaction. */
3830 o = lu_object_anon(env, d, NULL);
3832 GOTO(log, rc = PTR_ERR(o));
3834 n = lu_object_locate(o->lo_header, d->ld_type);
3835 if (unlikely(n == NULL)) {
3836 lu_object_put_nocache(env, o);
3838 GOTO(log, rc = -EINVAL);
3841 child = container_of(n, struct dt_object, do_lu);
3842 memset(hint, 0, sizeof(*hint));
3843 rc = dt_attr_get(env, parent, la);
3847 la->la_valid = LA_UID | LA_GID;
3848 memset(dof, 0, sizeof(*dof));
3850 dev = lfsck_obj2dev(child);
3851 handle = dt_trans_create(env, dev);
3853 GOTO(log, rc = PTR_ERR(handle));
3855 rc = dt_declare_create(env, child, la, hint, dof, handle);
3859 rc = dt_trans_start_local(env, dev, handle);
3863 rc = dt_create(env, child, la, hint, dof, handle);
3864 dt_trans_stop(env, dev, handle);
3869 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3870 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3875 /* The 2nd transaction. */
3877 /* XXX: Generally, we should use bottom device (OSD) to update parent
3878 * LOV EA. But because the LOD-object still references the wrong
3879 * OSP-object that should be detached after the parent's LOV EA
3880 * refreshed. Unfortunately, there is no suitable API for that.
3881 * So we have to make the LOD to re-load the OSP-object(s) via
3882 * replacing the LOV EA against the LOD-object.
3884 * Once the DNE2 patches have been landed, we can replace the
3885 * LOD device with the OSD device. LU-6230. */
3887 dev = lfsck->li_next;
3888 parent = lfsck_object_locate(dev, parent);
3890 GOTO(log, rc = PTR_ERR(parent));
3892 handle = dt_trans_create(env, dev);
3894 GOTO(log, rc = PTR_ERR(handle));
3896 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3897 LU_XATTR_REPLACE, handle);
3901 rc = dt_trans_start_local(env, dev, handle);
3905 dt_write_lock(env, parent, 0);
3906 if (unlikely(lfsck_is_dead_obj(parent)))
3907 GOTO(unlock, rc = 0);
3909 rc = lfsck_layout_get_lovea(env, parent, buf);
3910 if (unlikely(rc == -ENODATA))
3916 magic = le32_to_cpu(lmm->lmm_magic);
3917 if (magic == LOV_MAGIC_COMP_V1) {
3918 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3919 struct lov_comp_md_entry_v1 *lcme;
3920 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3923 LASSERT(llr->llr_comp_id != 0);
3925 for (i = 0; i < count; i++) {
3926 lcme = &lcm->lcm_entries[i];
3927 if (le32_to_cpu(lcme->lcme_id) == llr->llr_comp_id) {
3928 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3931 le32_add_cpu(&lcm->lcm_layout_gen, 1);
3933 le32_to_cpu(lcme->lcme_offset);
3934 magic = le32_to_cpu(lmm->lmm_magic);
3939 GOTO(unlock, rc = 0);
3943 if (magic == LOV_MAGIC_V1) {
3944 objs = &lmm->lmm_objects[llr->llr_lov_idx];
3946 LASSERT(magic == LOV_MAGIC_V3);
3948 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
3951 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3952 index = le32_to_cpu(objs->l_ost_idx);
3953 rc = ostid_to_fid(&tfid, oi, index);
3954 /* Someone changed layout during the LFSCK, no need to repair then. */
3955 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
3956 GOTO(unlock, rc = 0);
3958 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3959 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
3960 ostid_cpu_to_le(oi, &objs->l_ost_oi);
3961 objs->l_ost_gen = cpu_to_le32(0);
3962 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
3963 rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3964 LU_XATTR_REPLACE, handle);
3966 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
3969 dt_write_unlock(env, parent);
3973 dt_trans_stop(env, dev, handle);
3976 lfsck_ibits_unlock(&lh, LCK_EX);
3978 lfsck_object_put(env, child);
3981 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3982 "multiple references for: parent "DFID", comp_id %u, "
3983 "OST-index %u, stripe-index %u, owner %u/%u: rc = %d\n",
3984 lfsck_lfsck2name(lfsck), PFID(pfid),
3985 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3986 la->la_uid, la->la_gid, rc);
3991 /* If the MDT-object and the OST-object have different owner information,
3992 * then trust the MDT-object, because the normal chown/chgrp handle order
3993 * is from MDT to OST, and it is possible that some chown/chgrp operation
3994 * is partly done. */
3995 static int lfsck_layout_repair_owner(const struct lu_env *env,
3996 struct lfsck_component *com,
3997 struct dt_object *parent,
3998 struct lfsck_layout_req *llr,
3999 struct lu_attr *pla,
4000 const struct lu_attr *cla)
4002 struct lfsck_thread_info *info = lfsck_env_info(env);
4003 struct lu_attr *tla = &info->lti_la2;
4004 struct dt_object *child = llr->llr_child;
4005 struct dt_device *dev = lfsck_obj2dev(child);
4006 struct thandle *handle;
4008 dt_obj_version_t version;
4011 tla->la_uid = pla->la_uid;
4012 tla->la_gid = pla->la_gid;
4013 tla->la_valid = LA_UID | LA_GID;
4014 handle = dt_trans_create(env, dev);
4016 GOTO(log, rc = PTR_ERR(handle));
4018 rc = dt_declare_attr_set(env, child, tla, handle);
4022 rc = dt_trans_start_local(env, dev, handle);
4026 /* Use the dt_object lock to serialize with destroy and attr_set. */
4027 dt_read_lock(env, parent, 0);
4028 if (unlikely(lfsck_is_dead_obj(parent)))
4029 GOTO(unlock, rc = 1);
4031 version = dt_version_get(env, child);
4032 if (version == -EOPNOTSUPP)
4035 /* Get the latest parent's owner. */
4036 rc = dt_attr_get(env, parent, pla);
4040 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
4041 if (unlikely((!version && tla->la_ctime == 0) ||
4042 tla->la_uid != pla->la_uid || tla->la_gid != pla->la_gid))
4045 rc = dt_attr_set(env, child, tla, handle);
4050 dt_read_unlock(env, parent);
4053 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
4057 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
4058 "inconsistent file owner for: parent "DFID", child "DFID
4059 ", OST-index %u, stripe-index %u, old owner %u/%u, "
4060 "new owner %u/%u: rc = %d\n",
4061 lfsck_lfsck2name(com->lc_lfsck),
4062 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
4063 llr->llr_ost_idx, llr->llr_lov_idx,
4064 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
4069 #define CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid, msg) \
4070 CDEBUG(D_LFSCK, "%s:("DFID"|"DFID")/"DFID":XATTR %s: %s\n", \
4071 lfsck_lfsck2name(lfsck), PFID(&lso->lso_fid), PFID(pfid), \
4072 PFID(cfid), XATTR_NAME_FID, msg);
4074 /* Check whether the OST-object correctly back points to the
4075 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
4076 static int lfsck_layout_check_parent(const struct lu_env *env,
4077 struct lfsck_component *com,
4078 struct lfsck_assistant_object *lso,
4079 struct filter_fid *ff,
4080 const struct lu_fid *cfid,
4081 const struct lu_attr *cla,
4082 struct lfsck_layout_req *llr)
4084 struct lfsck_thread_info *info = lfsck_env_info(env);
4085 struct lu_buf *buf = &info->lti_big_buf;
4086 struct lu_fid *pfid = &info->lti_fid;
4087 struct dt_object *tobj;
4088 struct lov_mds_md_v1 *lmm;
4089 struct lov_ost_data_v1 *objs;
4090 struct lustre_handle lh = { 0 };
4091 struct lfsck_instance *lfsck = com->lc_lfsck;
4099 *pfid = ff->ff_parent;
4100 idx = pfid->f_stripe_idx;
4103 if (unlikely(!fid_is_sane(pfid))) {
4104 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4105 "the parent FID is invalid");
4107 RETURN(LLIT_UNMATCHED_PAIR);
4110 if (lu_fid_eq(pfid, &lso->lso_fid)) {
4111 if (likely(llr->llr_lov_idx == idx))
4114 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4115 "the stripe index is unmatched");
4117 RETURN(LLIT_UNMATCHED_PAIR);
4120 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4122 RETURN(PTR_ERR(tobj));
4124 if (dt_object_exists(tobj) == 0) {
4125 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4126 "the parent is nonexistent");
4128 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4131 if (lfsck_is_dead_obj(tobj)) {
4132 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4133 "the parent is dead object");
4135 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4138 if (!S_ISREG(lfsck_object_type(tobj))) {
4139 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4140 "the parent is not a regular file");
4142 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4145 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
4146 * remote one on another MDT. Then check whether the given OST-object
4147 * is in such layout. If yes, it is multiple referenced, otherwise it
4148 * is unmatched referenced case. */
4149 rc = lfsck_layout_get_lovea(env, tobj, buf);
4150 if (rc == 0 || rc == -ENODATA || rc == -ENOENT) {
4151 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4152 "the parent has no stripe data");
4154 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4157 if (unlikely(rc == -EOPNOTSUPP))
4158 GOTO(out, rc = LLIT_NONE);
4164 magic = le32_to_cpu(lmm->lmm_magic);
4165 if (magic == LOV_MAGIC_COMP_V1) {
4166 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4167 struct lov_comp_md_entry_v1 *lcme;
4169 if (ff->ff_layout.ol_comp_id == 0) {
4170 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4171 "the parent has incorrect comp_id");
4173 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4176 count = le16_to_cpu(lcm->lcm_entry_count);
4177 for (i = 0; i < count; i++) {
4178 lcme = &lcm->lcm_entries[i];
4179 if (le32_to_cpu(lcme->lcme_id) ==
4180 ff->ff_layout.ol_comp_id) {
4182 le32_to_cpu(lcme->lcme_offset);
4183 magic = le32_to_cpu(lmm->lmm_magic);
4184 if (!(le32_to_cpu(lcme->lcme_flags) &
4186 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid,
4188 "the parent has uninitialized component");
4190 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4197 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4198 "the parent has no matched comp_id");
4200 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4204 if (magic == LOV_MAGIC_V1) {
4205 objs = &lmm->lmm_objects[0];
4207 LASSERT(magic == LOV_MAGIC_V3);
4208 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4211 count = le16_to_cpu(lmm->lmm_stripe_count);
4212 for (i = 0; i < count; i++, objs++) {
4213 struct lu_fid *tfid = &info->lti_fid2;
4214 struct ost_id *oi = &info->lti_oi;
4217 if (lovea_slot_is_dummy(objs))
4220 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4221 idx2 = le32_to_cpu(objs->l_ost_idx);
4222 rc = ostid_to_fid(tfid, oi, idx2);
4224 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
4225 "invalid layout EA at the slot %d, index %u\n",
4226 lfsck_lfsck2name(com->lc_lfsck),
4227 PFID(pfid), i, idx2);
4229 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4232 if (lu_fid_eq(cfid, tfid)) {
4233 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
4234 MDS_INODELOCK_UPDATE |
4235 MDS_INODELOCK_LAYOUT |
4236 MDS_INODELOCK_XATTR,
4241 dt_read_lock(env, tobj, 0);
4243 /* For local MDT-object, re-check existence
4244 * after taken the lock. */
4245 if (!dt_object_remote(tobj)) {
4246 if (dt_object_exists(tobj) == 0 ||
4247 lfsck_is_dead_obj(tobj)) {
4248 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid,
4250 "the parent doesn't exist anymore after lock");
4252 rc = LLIT_UNMATCHED_PAIR;
4254 rc = LLIT_MULTIPLE_REFERENCED;
4260 /* For migration case, the new MDT-object and old
4261 * MDT-object may reference the same OST-object at
4262 * some migration internal time.
4264 * For remote MDT-object, the local MDT may not know
4265 * whether it has been removed or not. Try checking
4266 * for a non-existent xattr to check if this object
4267 * has been been removed or not. */
4268 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
4270 if (unlikely(rc == -ENOENT || rc >= 0)) {
4271 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4272 "the parent is remote object and nonexistent after lock");
4274 rc = LLIT_UNMATCHED_PAIR;
4275 } else if (rc == -ENODATA) {
4276 rc = LLIT_MULTIPLE_REFERENCED;
4283 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4284 "the parent has no matched stripe");
4286 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4289 if (lustre_handle_is_used(&lh)) {
4290 dt_read_unlock(env, tobj);
4291 lfsck_ibits_unlock(&lh, LCK_EX);
4295 lfsck_object_put(env, tobj);
4300 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
4301 struct lfsck_component *com,
4302 struct lfsck_assistant_req *lar)
4304 struct lfsck_layout_req *llr =
4305 container_of0(lar, struct lfsck_layout_req, llr_lar);
4306 struct lfsck_assistant_object *lso = lar->lar_parent;
4307 struct lfsck_layout *lo = com->lc_file_ram;
4308 struct lfsck_thread_info *info = lfsck_env_info(env);
4309 struct filter_fid *ff = &info->lti_ff;
4310 struct lu_buf buf = { .lb_buf = ff,
4311 .lb_len = sizeof(*ff) };
4312 struct dt_object *parent = NULL;
4313 struct dt_object *child = llr->llr_child;
4314 struct lu_attr *pla = &lso->lso_attr;
4315 struct lu_attr *cla = &info->lti_la;
4316 struct lfsck_instance *lfsck = com->lc_lfsck;
4317 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4318 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
4325 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ENGINE_DELAY, cfs_fail_val);
4327 rc = dt_attr_get(env, child, cla);
4328 if (rc == -ENOENT) {
4329 parent = lfsck_assistant_object_load(env, lfsck, lso);
4330 if (IS_ERR(parent)) {
4331 rc = PTR_ERR(parent);
4333 RETURN(rc == -ENOENT ? 0 : rc);
4336 type = LLIT_DANGLING;
4343 lfsck_buf_init(&buf, ff, sizeof(*ff));
4344 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
4345 if (unlikely(rc > 0 && rc < sizeof(struct lu_fid))) {
4346 CDEBUG(D_LFSCK, "%s:"DFID"/"DFID": "
4347 "the child object's %s is corrupted\n",
4348 lfsck_lfsck2name(lfsck), PFID(&lso->lso_fid),
4349 PFID(lu_object_fid(&child->do_lu)),
4352 type = LLIT_UNMATCHED_PAIR;
4356 if (rc < 0 && rc != -ENODATA)
4359 if (rc == 0 || rc == -ENODATA)
4360 GOTO(check_owner, rc = 0);
4362 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
4363 rc = lfsck_layout_check_parent(env, com, lso, ff,
4364 lu_object_fid(&child->do_lu), cla, llr);
4374 /* Someone may has changed the owner after the parent attr pre-loaded.
4375 * It can be handled later inside the lfsck_layout_repair_owner(). */
4376 if (unlikely(cla->la_uid != pla->la_uid ||
4377 cla->la_gid != pla->la_gid)) {
4378 type = LLIT_INCONSISTENT_OWNER;
4383 if (type == LLIT_NONE)
4386 if (bk->lb_param & LPF_DRYRUN)
4389 if (parent == NULL) {
4390 parent = lfsck_assistant_object_load(env, lfsck, lso);
4391 if (IS_ERR(parent)) {
4392 rc = PTR_ERR(parent);
4403 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
4404 rc = lfsck_layout_ins_dangling_rec(env, com,
4405 lfsck_dto2fid(parent), lfsck_dto2fid(child),
4406 llr->llr_comp_id, llr->llr_lov_idx,
4409 rc = __lfsck_layout_repair_dangling(env, com, parent,
4416 case LLIT_UNMATCHED_PAIR:
4417 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
4420 case LLIT_MULTIPLE_REFERENCED:
4421 rc = lfsck_layout_repair_multiple_references(env, com, parent,
4424 case LLIT_INCONSISTENT_OWNER:
4425 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
4435 down_write(&com->lc_sem);
4437 struct lfsck_assistant_data *lad = com->lc_data;
4439 if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags))) {
4441 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
4442 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
4443 rc == -EHOSTUNREACH) {
4444 /* If cannot touch the target server,
4445 * mark the LFSCK as INCOMPLETE. */
4446 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
4447 "talk with OST %x: rc = %d\n",
4448 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
4449 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
4450 lo->ll_objs_skipped++;
4453 lfsck_layout_record_failure(env, lfsck, lo);
4455 } else if (rc > 0 && (type != LLIT_DANGLING ||
4456 !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
4457 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
4458 "unknown type = %d\n", type);
4460 lo->ll_objs_repaired[type - 1]++;
4461 if (bk->lb_param & LPF_DRYRUN &&
4462 unlikely(lo->ll_pos_first_inconsistent == 0))
4463 lo->ll_pos_first_inconsistent =
4464 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
4467 up_write(&com->lc_sem);
4469 if (parent != NULL && !IS_ERR(parent))
4470 lfsck_object_put(env, parent);
4476 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
4477 struct lfsck_component *com,
4478 struct dt_object *obj, bool first)
4480 struct lfsck_instance *lfsck = com->lc_lfsck;
4481 struct ptlrpc_thread *thread = &lfsck->li_thread;
4482 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4483 struct lfsck_layout *lo = com->lc_file_ram;
4484 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
4487 struct lfsck_layout_dangling_key *parent =
4488 &lfsck_env_info(env)->lti_lldk;
4489 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
4494 di = iops->init(env, obj, 0);
4496 RETURN(PTR_ERR(di));
4499 lldk_cpu_to_be(parent, &lo->ll_lldk_latest_scanned_phase2);
4501 memset(parent, 0, sizeof(*parent));
4502 rc = iops->get(env, di, (const struct dt_key *)parent);
4507 /* The start one either has been processed or does not exist,
4509 rc = iops->next(env, di);
4515 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
4516 unlikely(!thread_is_running(thread)))
4519 key = iops->key(env, di);
4528 lldk_be_to_cpu(parent,
4529 (const struct lfsck_layout_dangling_key *)key);
4530 if (!fid_is_sane(&parent->lldk_fid)) {
4535 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
4537 fid_be_to_cpu(cfid, cfid);
4538 ost_idx = cfid->f_ver;
4540 if (!fid_is_sane(cfid)) {
4545 rc = lfsck_layout_repair_dangling(env, com,
4546 &parent->lldk_fid, cfid,
4547 parent->lldk_comp_id,
4548 parent->lldk_ea_off, ost_idx);
4552 down_write(&com->lc_sem);
4553 com->lc_new_checked++;
4554 com->lc_new_scanned++;
4556 lo->ll_lldk_latest_scanned_phase2 = *parent;
4559 lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
4561 lo->ll_objs_failed_phase2++;
4562 up_write(&com->lc_sem);
4564 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4567 if (unlikely(com->lc_time_next_checkpoint <=
4568 ktime_get_seconds()) &&
4569 com->lc_new_checked != 0) {
4570 down_write(&com->lc_sem);
4571 lo->ll_run_time_phase2 += ktime_get_seconds() -
4572 com->lc_time_last_checkpoint;
4573 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
4574 lo->ll_objs_checked_phase2 += com->lc_new_checked;
4575 com->lc_new_checked = 0;
4576 lfsck_layout_store(env, com);
4577 up_write(&com->lc_sem);
4579 com->lc_time_last_checkpoint = ktime_get_seconds();
4580 com->lc_time_next_checkpoint =
4581 com->lc_time_last_checkpoint +
4582 LFSCK_CHECKPOINT_INTERVAL;
4585 lfsck_control_speed_by_self(com);
4586 if (unlikely(!thread_is_running(thread)))
4589 rc = iops->next(env, di);
4598 iops->fini(env, di);
4603 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
4604 struct lfsck_component *com)
4606 struct lfsck_assistant_data *lad = com->lc_data;
4607 struct lfsck_instance *lfsck = com->lc_lfsck;
4608 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4609 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4610 struct lfsck_tgt_desc *ltd;
4614 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
4615 lfsck_lfsck2name(lfsck));
4617 spin_lock(<ds->ltd_lock);
4618 while (!list_empty(&lad->lad_ost_phase2_list)) {
4619 ltd = list_entry(lad->lad_ost_phase2_list.next,
4620 struct lfsck_tgt_desc,
4621 ltd_layout_phase_list);
4622 list_del_init(<d->ltd_layout_phase_list);
4623 if (bk->lb_param & LPF_OST_ORPHAN) {
4624 spin_unlock(<ds->ltd_lock);
4625 rc = lfsck_layout_scan_orphan(env, com, ltd);
4626 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
4629 if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags) ||
4630 !thread_is_running(&lfsck->li_thread)))
4632 spin_lock(<ds->ltd_lock);
4636 if (list_empty(&lad->lad_ost_phase1_list))
4640 spin_unlock(<ds->ltd_lock);
4642 if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
4643 struct lfsck_layout *lo = com->lc_file_ram;
4646 com->lc_new_checked = 0;
4647 com->lc_new_scanned = 0;
4648 com->lc_time_last_checkpoint = ktime_get_seconds();
4649 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4650 LFSCK_CHECKPOINT_INTERVAL;
4652 i = lfsck_sub_trace_file_fid2idx(
4653 &lo->ll_lldk_latest_scanned_phase2.lldk_fid);
4654 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4655 com->lc_sub_trace_objs[i].lsto_obj, true);
4656 while (rc > 0 && ++i < LFSCK_STF_COUNT)
4657 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4658 com->lc_sub_trace_objs[i].lsto_obj, false);
4660 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
4661 "at the No. %d trace file: rc = %d\n",
4662 lfsck_lfsck2name(lfsck), i, rc);
4665 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
4666 lfsck_lfsck2name(lfsck), rc);
4672 lfsck_layout_slave_async_interpret(const struct lu_env *env,
4673 struct ptlrpc_request *req,
4676 struct lfsck_layout_slave_async_args *llsaa = args;
4677 struct obd_export *exp = llsaa->llsaa_exp;
4678 struct lfsck_component *com = llsaa->llsaa_com;
4679 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
4680 struct lfsck_layout_slave_data *llsd = com->lc_data;
4681 struct lfsck_reply *lr = NULL;
4685 /* It is probably caused by network trouble, or target crash,
4686 * it will try several times (depends on the obd_timeout, and
4687 * will not less than 3 times). But to make the LFSCK can go
4688 * ahead, we should not try for ever. After some try but still
4689 * hit failure, it will assume that the target exit the LFSCK
4690 * prcoessing and stop try. */
4691 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
4692 int max_try = max_t(int, obd_timeout / 30, 3);
4694 if (++(llst->llst_failures) > max_try)
4700 llst->llst_failures = 0;
4701 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
4702 if (lr->lr_status != LS_SCANNING_PHASE1 &&
4703 lr->lr_status != LS_SCANNING_PHASE2)
4708 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
4709 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
4710 llst->llst_index, lr != NULL ? lr->lr_status : rc,
4711 llst->llst_failures);
4713 lfsck_layout_llst_del(llsd, llst);
4716 lfsck_layout_llst_put(llst);
4717 lfsck_component_put(env, com);
4718 class_export_put(exp);
4723 static int lfsck_layout_async_query(const struct lu_env *env,
4724 struct lfsck_component *com,
4725 struct obd_export *exp,
4726 struct lfsck_layout_slave_target *llst,
4727 struct lfsck_request *lr,
4728 struct ptlrpc_request_set *set)
4730 struct lfsck_layout_slave_async_args *llsaa;
4731 struct ptlrpc_request *req;
4732 struct lfsck_request *tmp;
4736 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
4740 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
4742 ptlrpc_request_free(req);
4746 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4748 ptlrpc_request_set_replen(req);
4750 llsaa = ptlrpc_req_async_args(llsaa, req);
4751 llsaa->llsaa_exp = exp;
4752 llsaa->llsaa_com = lfsck_component_get(com);
4753 llsaa->llsaa_llst = llst;
4754 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
4755 req->rq_allow_intr = 1;
4756 req->rq_no_delay = 1;
4757 ptlrpc_set_add_req(set, req);
4762 static int lfsck_layout_async_notify(const struct lu_env *env,
4763 struct obd_export *exp,
4764 struct lfsck_request *lr,
4765 struct ptlrpc_request_set *set)
4767 struct ptlrpc_request *req;
4768 struct lfsck_request *tmp;
4772 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4776 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4778 ptlrpc_request_free(req);
4782 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4784 ptlrpc_request_set_replen(req);
4785 req->rq_allow_intr = 1;
4786 req->rq_no_delay = 1;
4787 ptlrpc_set_add_req(set, req);
4793 lfsck_layout_slave_query_master(const struct lu_env *env,
4794 struct lfsck_component *com)
4796 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4797 struct lfsck_instance *lfsck = com->lc_lfsck;
4798 struct lfsck_layout_slave_data *llsd = com->lc_data;
4799 struct lfsck_layout_slave_target *llst;
4800 struct obd_export *exp;
4801 struct ptlrpc_request_set *set;
4806 set = ptlrpc_prep_set();
4808 GOTO(log, rc = -ENOMEM);
4810 memset(lr, 0, sizeof(*lr));
4811 lr->lr_event = LE_QUERY;
4812 lr->lr_active = LFSCK_TYPE_LAYOUT;
4814 llsd->llsd_touch_gen++;
4815 spin_lock(&llsd->llsd_lock);
4816 while (!list_empty(&llsd->llsd_master_list)) {
4817 llst = list_entry(llsd->llsd_master_list.next,
4818 struct lfsck_layout_slave_target,
4820 if (llst->llst_gen == llsd->llsd_touch_gen)
4823 llst->llst_gen = llsd->llsd_touch_gen;
4824 list_move_tail(&llst->llst_list,
4825 &llsd->llsd_master_list);
4826 atomic_inc(&llst->llst_ref);
4827 spin_unlock(&llsd->llsd_lock);
4829 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4832 lfsck_layout_llst_del(llsd, llst);
4833 lfsck_layout_llst_put(llst);
4834 spin_lock(&llsd->llsd_lock);
4838 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
4840 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4841 "query %s for layout: rc = %d\n",
4842 lfsck_lfsck2name(lfsck),
4843 exp->exp_obd->obd_name, rc);
4846 lfsck_layout_llst_put(llst);
4847 class_export_put(exp);
4849 spin_lock(&llsd->llsd_lock);
4851 spin_unlock(&llsd->llsd_lock);
4853 rc = ptlrpc_set_wait(env, set);
4854 ptlrpc_set_destroy(set);
4856 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4859 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4860 lfsck_lfsck2name(com->lc_lfsck), rc);
4866 lfsck_layout_slave_notify_master(const struct lu_env *env,
4867 struct lfsck_component *com,
4868 enum lfsck_events event, int result)
4870 struct lfsck_layout *lo = com->lc_file_ram;
4871 struct lfsck_instance *lfsck = com->lc_lfsck;
4872 struct lfsck_layout_slave_data *llsd = com->lc_data;
4873 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4874 struct lfsck_layout_slave_target *llst;
4875 struct obd_export *exp;
4876 struct ptlrpc_request_set *set;
4880 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4881 lfsck_lfsck2name(com->lc_lfsck));
4883 set = ptlrpc_prep_set();
4887 memset(lr, 0, sizeof(*lr));
4888 lr->lr_event = event;
4889 lr->lr_flags = LEF_FROM_OST;
4890 lr->lr_status = result;
4891 lr->lr_index = lfsck_dev_idx(lfsck);
4892 lr->lr_active = LFSCK_TYPE_LAYOUT;
4893 lr->lr_flags2 = lo->ll_flags;
4894 llsd->llsd_touch_gen++;
4895 spin_lock(&llsd->llsd_lock);
4896 while (!list_empty(&llsd->llsd_master_list)) {
4897 llst = list_entry(llsd->llsd_master_list.next,
4898 struct lfsck_layout_slave_target,
4900 if (llst->llst_gen == llsd->llsd_touch_gen)
4903 llst->llst_gen = llsd->llsd_touch_gen;
4904 list_move_tail(&llst->llst_list,
4905 &llsd->llsd_master_list);
4906 atomic_inc(&llst->llst_ref);
4907 spin_unlock(&llsd->llsd_lock);
4909 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4912 lfsck_layout_llst_del(llsd, llst);
4913 lfsck_layout_llst_put(llst);
4914 spin_lock(&llsd->llsd_lock);
4918 rc = lfsck_layout_async_notify(env, exp, lr, set);
4920 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4921 "notify %s for layout: rc = %d\n",
4922 lfsck_lfsck2name(lfsck),
4923 exp->exp_obd->obd_name, rc);
4925 lfsck_layout_llst_put(llst);
4926 class_export_put(exp);
4927 spin_lock(&llsd->llsd_lock);
4929 spin_unlock(&llsd->llsd_lock);
4931 ptlrpc_set_wait(env, set);
4932 ptlrpc_set_destroy(set);
4938 * \ret -ENODATA: unrecognized stripe
4939 * \ret = 0 : recognized stripe
4940 * \ret < 0 : other failures
4942 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4943 struct lfsck_component *com,
4944 struct lu_fid *cfid,
4945 struct lu_fid *pfid, __u32 comp_id)
4947 struct lfsck_thread_info *info = lfsck_env_info(env);
4948 struct lu_buf *buf = &info->lti_big_buf;
4949 struct ost_id *oi = &info->lti_oi;
4950 struct dt_object *obj;
4951 struct lov_mds_md_v1 *lmm;
4952 struct lov_ost_data_v1 *objs;
4953 __u32 idx = pfid->f_stripe_idx;
4961 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4963 RETURN(PTR_ERR(obj));
4965 dt_read_lock(env, obj, 0);
4966 if (unlikely(dt_object_exists(obj) == 0 ||
4967 lfsck_is_dead_obj(obj)))
4968 GOTO(unlock, rc = -ENOENT);
4970 if (!S_ISREG(lfsck_object_type(obj)))
4971 GOTO(unlock, rc = -ENODATA);
4973 rc = lfsck_layout_get_lovea(env, obj, buf);
4978 magic = le32_to_cpu(lmm->lmm_magic);
4979 if (magic == LOV_MAGIC_COMP_V1) {
4980 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4981 struct lov_comp_md_entry_v1 *lcme;
4984 GOTO(unlock, rc = -ENODATA);
4986 count = le16_to_cpu(lcm->lcm_entry_count);
4987 for (i = 0; i < count; i++) {
4988 lcme = &lcm->lcm_entries[i];
4989 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
4991 le32_to_cpu(lcme->lcme_offset);
4992 magic = le32_to_cpu(lmm->lmm_magic);
4993 if (!(le32_to_cpu(lcme->lcme_flags) &
4995 GOTO(unlock, rc = -ENODATA);
5001 GOTO(unlock, rc = -ENODATA);
5005 if (magic == LOV_MAGIC_V1) {
5006 objs = &lmm->lmm_objects[0];
5008 LASSERT(magic == LOV_MAGIC_V3);
5009 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5012 fid_to_ostid(cfid, oi);
5013 count = le16_to_cpu(lmm->lmm_stripe_count);
5014 for (i = 0; i < count; i++, objs++) {
5017 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
5018 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
5019 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
5022 GOTO(unlock, rc = -ENODATA);
5025 dt_read_unlock(env, obj);
5026 lfsck_object_put(env, obj);
5032 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
5033 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
5034 * layout EA from MDT to OST. On one hand, the OST no need to understand
5035 * the layout EA structure; on the other hand, it may cause trouble when
5036 * transfer large layout EA from MDT to OST via normal OUT RPC.
5038 * \ret > 0: unrecognized stripe
5039 * \ret = 0: recognized stripe
5040 * \ret < 0: other failures
5042 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
5043 struct lfsck_component *com,
5044 struct lu_fid *cfid,
5045 struct lu_fid *pfid, __u32 comp_id)
5047 struct lfsck_instance *lfsck = com->lc_lfsck;
5048 struct obd_device *obd = lfsck->li_obd;
5049 struct seq_server_site *ss = lfsck_dev_site(lfsck);
5050 struct obd_export *exp = NULL;
5051 struct ptlrpc_request *req = NULL;
5052 struct lfsck_request *lr;
5053 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
5057 if (unlikely(fid_is_idif(pfid)))
5060 fld_range_set_any(range);
5061 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
5063 RETURN(rc == -ENOENT ? 1 : rc);
5065 if (unlikely(!fld_range_is_mdt(range)))
5068 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
5069 if (unlikely(exp == NULL))
5072 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
5073 GOTO(out, rc = -EOPNOTSUPP);
5075 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
5077 GOTO(out, rc = -ENOMEM);
5079 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
5081 ptlrpc_request_free(req);
5086 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
5087 memset(lr, 0, sizeof(*lr));
5088 lr->lr_event = LE_PAIRS_VERIFY;
5089 lr->lr_active = LFSCK_TYPE_LAYOUT;
5090 lr->lr_fid = *cfid; /* OST-object itself FID. */
5091 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
5092 lr->lr_comp_id = comp_id;
5094 ptlrpc_request_set_replen(req);
5095 rc = ptlrpc_queue_wait(req);
5096 ptlrpc_req_finished(req);
5098 if (rc == -ENOENT || rc == -ENODATA)
5105 class_export_put(exp);
5110 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
5111 struct lfsck_component *com,
5112 struct lfsck_req_local *lrl)
5114 struct dt_object *obj;
5118 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lrl->lrl_fid);
5120 GOTO(log, rc = PTR_ERR(obj));
5122 dt_write_lock(env, obj, 0);
5123 if (unlikely(dt_object_exists(obj) == 0 ||
5124 lfsck_is_dead_obj(obj)))
5125 GOTO(unlock, rc = 0);
5127 rc = __lfsck_layout_update_pfid(env, obj, &lrl->lrl_ff_client.ff_parent,
5128 &lrl->lrl_ff_client.ff_layout,
5129 lrl->lrl_ff_client.ff_layout_version,
5130 lrl->lrl_ff_client.ff_range,
5131 lrl->lrl_ff_client.ff_parent.f_ver);
5136 dt_write_unlock(env, obj);
5137 lfsck_object_put(env, obj);
5140 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
5141 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
5142 PFID(&lrl->lrl_fid), PFID(&lrl->lrl_ff_client.ff_parent), rc);
5149 static void lfsck_layout_slave_quit(const struct lu_env *env,
5150 struct lfsck_component *com);
5152 static int lfsck_layout_reset(const struct lu_env *env,
5153 struct lfsck_component *com, bool init)
5155 struct lfsck_layout *lo = com->lc_file_ram;
5158 down_write(&com->lc_sem);
5160 memset(lo, 0, com->lc_file_size);
5162 __u32 count = lo->ll_success_count;
5163 time64_t last_time = lo->ll_time_last_complete;
5165 memset(lo, 0, com->lc_file_size);
5166 lo->ll_success_count = count;
5167 lo->ll_time_last_complete = last_time;
5170 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
5171 lo->ll_status = LS_INIT;
5173 if (com->lc_lfsck->li_master) {
5174 struct lfsck_assistant_data *lad = com->lc_data;
5176 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
5177 CFS_RESET_BITMAP(lad->lad_bitmap);
5180 rc = lfsck_layout_store(env, com);
5181 if (rc == 0 && com->lc_lfsck->li_master)
5182 rc = lfsck_load_sub_trace_files(env, com,
5183 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
5184 up_write(&com->lc_sem);
5186 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
5187 lfsck_lfsck2name(com->lc_lfsck), rc);
5192 static void lfsck_layout_fail(const struct lu_env *env,
5193 struct lfsck_component *com, bool new_checked)
5195 struct lfsck_layout *lo = com->lc_file_ram;
5197 down_write(&com->lc_sem);
5199 com->lc_new_checked++;
5200 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
5201 up_write(&com->lc_sem);
5204 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
5205 struct lfsck_component *com, bool init)
5207 struct lfsck_instance *lfsck = com->lc_lfsck;
5208 struct lfsck_layout *lo = com->lc_file_ram;
5212 rc = lfsck_checkpoint_generic(env, com);
5214 return rc > 0 ? 0 : rc;
5217 down_write(&com->lc_sem);
5219 lo->ll_pos_latest_start =
5220 lfsck->li_pos_checkpoint.lp_oit_cookie;
5222 lo->ll_pos_last_checkpoint =
5223 lfsck->li_pos_checkpoint.lp_oit_cookie;
5224 lo->ll_run_time_phase1 += ktime_get_seconds() -
5225 lfsck->li_time_last_checkpoint;
5226 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5227 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5228 com->lc_new_checked = 0;
5231 rc = lfsck_layout_store(env, com);
5232 up_write(&com->lc_sem);
5234 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
5235 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5236 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5241 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
5242 struct lfsck_component *com, bool init)
5244 struct lfsck_instance *lfsck = com->lc_lfsck;
5245 struct lfsck_layout *lo = com->lc_file_ram;
5248 if (com->lc_new_checked == 0 && !init)
5251 down_write(&com->lc_sem);
5253 lo->ll_pos_latest_start =
5254 lfsck->li_pos_checkpoint.lp_oit_cookie;
5256 lo->ll_pos_last_checkpoint =
5257 lfsck->li_pos_checkpoint.lp_oit_cookie;
5258 lo->ll_run_time_phase1 += ktime_get_seconds() -
5259 lfsck->li_time_last_checkpoint;
5260 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5261 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5262 com->lc_new_checked = 0;
5265 rc = lfsck_layout_store(env, com);
5266 up_write(&com->lc_sem);
5268 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
5269 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5270 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5275 static int lfsck_layout_prep(const struct lu_env *env,
5276 struct lfsck_component *com,
5277 struct lfsck_start *start)
5279 struct lfsck_instance *lfsck = com->lc_lfsck;
5280 struct lfsck_layout *lo = com->lc_file_ram;
5281 struct lfsck_position *pos = &com->lc_pos_start;
5283 fid_zero(&pos->lp_dir_parent);
5284 pos->lp_dir_cookie = 0;
5285 if (lo->ll_status == LS_COMPLETED ||
5286 lo->ll_status == LS_PARTIAL ||
5287 /* To handle orphan, must scan from the beginning. */
5288 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
5291 rc = lfsck_layout_reset(env, com, false);
5293 rc = lfsck_set_param(env, lfsck, start, true);
5296 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
5297 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
5303 down_write(&com->lc_sem);
5304 lo->ll_time_latest_start = ktime_get_real_seconds();
5305 spin_lock(&lfsck->li_lock);
5306 if (lo->ll_flags & LF_SCANNED_ONCE) {
5307 if (!lfsck->li_drop_dryrun ||
5308 lo->ll_pos_first_inconsistent == 0) {
5309 lo->ll_status = LS_SCANNING_PHASE2;
5310 list_move_tail(&com->lc_link,
5311 &lfsck->li_list_double_scan);
5312 pos->lp_oit_cookie = 0;
5316 lo->ll_status = LS_SCANNING_PHASE1;
5317 lo->ll_run_time_phase1 = 0;
5318 lo->ll_run_time_phase2 = 0;
5319 lo->ll_objs_checked_phase1 = 0;
5320 lo->ll_objs_checked_phase2 = 0;
5321 lo->ll_objs_failed_phase1 = 0;
5322 lo->ll_objs_failed_phase2 = 0;
5323 for (i = 0; i < LLIT_MAX; i++)
5324 lo->ll_objs_repaired[i] = 0;
5326 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5327 fid_zero(&com->lc_fid_latest_scanned_phase2);
5330 lo->ll_status = LS_SCANNING_PHASE1;
5331 if (!lfsck->li_drop_dryrun ||
5332 lo->ll_pos_first_inconsistent == 0)
5333 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
5335 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5337 spin_unlock(&lfsck->li_lock);
5338 up_write(&com->lc_sem);
5343 static int lfsck_layout_slave_prep(const struct lu_env *env,
5344 struct lfsck_component *com,
5345 struct lfsck_start_param *lsp)
5347 struct lfsck_layout_slave_data *llsd = com->lc_data;
5348 struct lfsck_instance *lfsck = com->lc_lfsck;
5349 struct lfsck_layout *lo = com->lc_file_ram;
5350 struct lfsck_start *start = lsp->lsp_start;
5353 rc = lfsck_layout_prep(env, com, start);
5357 if (lo->ll_flags & LF_CRASHED_LASTID &&
5358 list_empty(&llsd->llsd_master_list)) {
5359 LASSERT(lfsck->li_out_notify != NULL);
5361 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5362 LE_LASTID_REBUILDING);
5365 if (!lsp->lsp_index_valid)
5368 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5369 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
5370 LASSERT(!llsd->llsd_rbtree_valid);
5372 down_write(&llsd->llsd_rb_rwsem);
5373 rc = lfsck_rbtree_setup(env, com);
5374 up_write(&llsd->llsd_rb_rwsem);
5377 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
5378 "%llu]\n", lfsck_lfsck2name(lfsck),
5379 com->lc_pos_start.lp_oit_cookie);
5384 static int lfsck_layout_master_prep(const struct lu_env *env,
5385 struct lfsck_component *com,
5386 struct lfsck_start_param *lsp)
5391 rc = lfsck_layout_load_bitmap(env, com);
5393 rc = lfsck_layout_reset(env, com, false);
5395 rc = lfsck_set_param(env, com->lc_lfsck,
5396 lsp->lsp_start, true);
5402 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
5406 rc = lfsck_start_assistant(env, com, lsp);
5411 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
5412 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
5413 com->lc_pos_start.lp_oit_cookie);
5418 /* Pre-fetch the attribute for each stripe in the given layout EA. */
5419 static int lfsck_layout_scan_stripes(const struct lu_env *env,
5420 struct lfsck_component *com,
5421 struct dt_object *parent,
5422 struct lov_mds_md_v1 *lmm, __u32 comp_id)
5424 struct lfsck_thread_info *info = lfsck_env_info(env);
5425 struct lfsck_instance *lfsck = com->lc_lfsck;
5426 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5427 struct lfsck_layout *lo = com->lc_file_ram;
5428 struct lfsck_assistant_data *lad = com->lc_data;
5429 struct lfsck_assistant_object *lso = NULL;
5430 struct lov_ost_data_v1 *objs;
5431 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
5432 struct ptlrpc_thread *mthread = &lfsck->li_thread;
5433 struct ptlrpc_thread *athread = &lad->lad_thread;
5441 lfsck_buf_init(&buf, &info->lti_ff, sizeof(struct filter_fid));
5442 magic = le32_to_cpu(lmm->lmm_magic);
5443 if (magic == LOV_MAGIC_V1) {
5444 objs = &lmm->lmm_objects[0];
5446 LASSERT(magic == LOV_MAGIC_V3);
5447 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5450 count = le16_to_cpu(lmm->lmm_stripe_count);
5451 for (i = 0; i < count; i++, objs++) {
5452 struct lu_fid *fid = &info->lti_fid;
5453 struct ost_id *oi = &info->lti_oi;
5454 struct lfsck_layout_req *llr;
5455 struct lfsck_tgt_desc *tgt = NULL;
5456 struct dt_object *cobj = NULL;
5458 bool wakeup = false;
5460 if (unlikely(lovea_slot_is_dummy(objs)))
5463 wait_event_idle(mthread->t_ctl_waitq,
5464 lad->lad_prefetched < bk->lb_async_windows ||
5465 !thread_is_running(mthread) ||
5466 thread_is_stopped(athread));
5468 if (unlikely(!thread_is_running(mthread)) ||
5469 thread_is_stopped(athread))
5472 if (unlikely(lfsck_is_dead_obj(parent)))
5475 ostid_le_to_cpu(&objs->l_ost_oi, oi);
5476 index = le32_to_cpu(objs->l_ost_idx);
5477 rc = ostid_to_fid(fid, oi, index);
5479 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
5480 ": "DOSTID", idx %u, comp_id %u\n",
5481 lfsck_lfsck2name(lfsck),
5482 PFID(lfsck_dto2fid(parent)), POSTID(oi),
5487 tgt = lfsck_tgt_get(ltds, index);
5488 if (unlikely(tgt == NULL)) {
5489 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
5490 "did not join the layout LFSCK, comp_id %u\n",
5491 lfsck_lfsck2name(lfsck), index, comp_id);
5492 lfsck_lad_set_bitmap(env, com, index);
5496 /* There is potential deadlock race condition between object
5497 * destroy and layout LFSCK. Consider the following scenario:
5499 * 1) The LFSCK thread obtained the parent object firstly, at
5500 * that time, the parent object has not been destroyed yet.
5502 * 2) One RPC service thread destroyed the parent and all its
5503 * children objects. Because the LFSCK is referencing the
5504 * parent object, then the parent object will be marked as
5505 * dying in RAM. On the other hand, the parent object is
5506 * referencing all its children objects, then all children
5507 * objects will be marked as dying in RAM also.
5509 * 3) The LFSCK thread tries to find some child object with
5510 * the parent object referenced. Then it will find that the
5511 * child object is dying. According to the object visibility
5512 * rules: the object with dying flag cannot be returned to
5513 * others. So the LFSCK thread has to wait until the dying
5514 * object has been purged from RAM, then it can allocate a
5515 * new object (with the same FID) in RAM. Unfortunately, the
5516 * LFSCK thread itself is referencing the parent object, and
5517 * cause the parent object cannot be purged, then cause the
5518 * child object cannot be purged also. So the LFSCK thread
5519 * will fall into deadlock.
5521 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
5523 if (lfsck_is_dead_obj(parent)) {
5533 rc = dt_declare_attr_get(env, cobj);
5537 rc = dt_declare_xattr_get(env, cobj, &buf, XATTR_NAME_FID);
5542 struct lu_attr *attr = &info->lti_la;
5544 rc = dt_attr_get(env, parent, attr);
5548 lso = lfsck_assistant_object_init(env,
5549 lfsck_dto2fid(parent), attr,
5550 lfsck->li_pos_current.lp_oit_cookie, false);
5559 llr = lfsck_layout_assistant_req_init(lso, cobj, comp_id,
5567 spin_lock(&lad->lad_lock);
5568 if (lad->lad_assistant_status < 0) {
5569 spin_unlock(&lad->lad_lock);
5570 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
5572 RETURN(lad->lad_assistant_status);
5575 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
5576 if (lad->lad_prefetched == 0)
5579 lad->lad_prefetched++;
5580 spin_unlock(&lad->lad_lock);
5582 wake_up_all(&athread->t_ctl_waitq);
5585 down_write(&com->lc_sem);
5586 com->lc_new_checked++;
5588 lfsck_layout_record_failure(env, lfsck, lo);
5589 up_write(&com->lc_sem);
5591 if (cobj != NULL && !IS_ERR(cobj))
5592 lfsck_object_put(env, cobj);
5594 if (likely(tgt != NULL))
5597 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5605 lfsck_assistant_object_put(env, lso);
5610 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
5611 * the OST-object's attribute and generate an structure lfsck_layout_req on the
5612 * list ::lad_req_list.
5614 * For each request on above list, the lfsck_layout_assistant thread compares
5615 * the OST side attribute with local attribute, if inconsistent, then repair it.
5617 * All above processing is async mode with pipeline. */
5618 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
5619 struct lfsck_component *com,
5620 struct dt_object *obj)
5622 struct lfsck_thread_info *info = lfsck_env_info(env);
5623 struct ost_id *oi = &info->lti_oi;
5624 struct lfsck_layout *lo = com->lc_file_ram;
5625 struct lfsck_assistant_data *lad = com->lc_data;
5626 struct lfsck_instance *lfsck = com->lc_lfsck;
5627 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5628 struct thandle *handle = NULL;
5629 struct lu_buf *buf = &info->lti_big_buf;
5630 struct lov_mds_md_v1 *lmm = NULL;
5631 struct dt_device *dev = lfsck_obj2dev(obj);
5632 struct lustre_handle lh = { 0 };
5633 struct lu_buf ea_buf = { NULL };
5634 struct lov_comp_md_v1 *lcm = NULL;
5635 struct lov_comp_md_entry_v1 *lcme = NULL;
5640 bool locked = false;
5641 bool stripe = false;
5642 bool bad_oi = false;
5645 if (!S_ISREG(lfsck_object_type(obj)))
5648 if (lad->lad_assistant_status < 0)
5649 GOTO(out, rc = -ESRCH);
5651 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
5652 lmm_oi_cpu_to_le(oi, oi);
5653 dt_read_lock(env, obj, 0);
5658 if (dt_object_exists(obj) == 0 ||
5659 lfsck_is_dead_obj(obj))
5662 rc = lfsck_layout_get_lovea(env, obj, buf);
5663 if (rc == -EINVAL || rc == -ENODATA || rc == -EOPNOTSUPP)
5664 /* Skip bad lov EA during the 1st cycle scanning, and
5665 * try to recover it via orphan in the 2nd scanning. */
5672 magic = le32_to_cpu(lmm->lmm_magic);
5673 if (magic == LOV_MAGIC_COMP_V1) {
5677 count = le16_to_cpu(lcm->lcm_entry_count);
5678 for (i = 0; i < count; i++) {
5679 lcme = &lcm->lcm_entries[i];
5680 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5681 if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) != 0)
5685 GOTO(out, stripe = true);
5686 } else if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0) {
5687 GOTO(out, stripe = true);
5691 /* Inconsistent lmm_oi, should be repaired. */
5694 if (bk->lb_param & LPF_DRYRUN) {
5695 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5697 GOTO(out, stripe = true);
5700 if (!lustre_handle_is_used(&lh)) {
5701 dt_read_unlock(env, obj);
5703 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5704 MDS_INODELOCK_LAYOUT |
5705 MDS_INODELOCK_XATTR, LCK_EX);
5709 handle = dt_trans_create(env, dev);
5711 GOTO(out, rc = PTR_ERR(handle));
5713 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
5714 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5715 LU_XATTR_REPLACE, handle);
5719 rc = dt_trans_start_local(env, dev, handle);
5723 dt_write_lock(env, obj, 0);
5729 if (magic == LOV_MAGIC_COMP_V1) {
5732 for (i = 0; i < count; i++) {
5733 lcme = &lcm->lcm_entries[i];
5734 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5741 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5742 LU_XATTR_REPLACE, handle);
5746 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5748 GOTO(out, stripe = true);
5752 if (lustre_handle_is_used(&lh))
5753 dt_write_unlock(env, obj);
5755 dt_read_unlock(env, obj);
5758 if (handle != NULL && !IS_ERR(handle))
5759 dt_trans_stop(env, dev, handle);
5761 lfsck_ibits_unlock(&lh, LCK_EX);
5764 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
5765 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
5766 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
5767 PFID(lfsck_dto2fid(obj)), rc);
5770 if (magic == LOV_MAGIC_COMP_V1) {
5773 for (i = 0; i < count; i++) {
5774 lcme = &lcm->lcm_entries[i];
5775 if (!(le32_to_cpu(lcme->lcme_flags) &
5779 rc = lfsck_layout_scan_stripes(env, com, obj,
5780 (struct lov_mds_md_v1 *)(buf->lb_buf +
5781 le32_to_cpu(lcme->lcme_offset)),
5782 le32_to_cpu(lcme->lcme_id));
5785 rc = lfsck_layout_scan_stripes(env, com, obj, lmm, 0);
5788 down_write(&com->lc_sem);
5789 com->lc_new_checked++;
5791 lfsck_layout_record_failure(env, lfsck, lo);
5792 up_write(&com->lc_sem);
5798 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
5799 struct lfsck_component *com,
5800 struct dt_object *obj)
5802 struct lfsck_instance *lfsck = com->lc_lfsck;
5803 struct lfsck_layout *lo = com->lc_file_ram;
5804 const struct lu_fid *fid = lfsck_dto2fid(obj);
5805 struct lfsck_layout_slave_data *llsd = com->lc_data;
5806 struct lfsck_layout_seq *lls;
5812 LASSERT(llsd != NULL);
5814 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
5815 cfs_fail_val == lfsck_dev_idx(lfsck)) {
5816 struct ptlrpc_thread *thread = &lfsck->li_thread;
5818 wait_event_idle_timeout(thread->t_ctl_waitq,
5819 !thread_is_running(thread),
5820 cfs_time_seconds(1));
5823 lfsck_rbtree_update_bitmap(env, com, fid, false);
5825 down_write(&com->lc_sem);
5826 if (fid_is_idif(fid))
5828 else if (!fid_is_norm(fid) ||
5829 !fid_is_for_ostobj(env, lfsck, obj, fid))
5830 GOTO(unlock, rc = 0);
5833 com->lc_new_checked++;
5835 lls = lfsck_layout_seq_lookup(llsd, seq);
5838 if (unlikely(lls == NULL))
5839 GOTO(unlock, rc = -ENOMEM);
5841 INIT_LIST_HEAD(&lls->lls_list);
5843 rc = lfsck_layout_lastid_load(env, com, lls);
5845 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5846 "load LAST_ID for %#llx: rc = %d\n",
5847 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
5848 lo->ll_objs_failed_phase1++;
5853 lfsck_layout_seq_insert(llsd, lls);
5856 if (unlikely(fid_is_last_id(fid)))
5857 GOTO(unlock, rc = 0);
5859 if (fid_is_idif(fid))
5860 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
5864 if (oid > lls->lls_lastid_known)
5865 lls->lls_lastid_known = oid;
5867 if (oid > lls->lls_lastid) {
5868 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
5869 /* OFD may create new objects during LFSCK scanning. */
5870 rc = lfsck_layout_lastid_reload(env, com, lls);
5871 if (unlikely(rc != 0)) {
5872 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5873 "reload LAST_ID for %#llx: rc = %d\n",
5874 lfsck_lfsck2name(com->lc_lfsck),
5880 if (oid <= lls->lls_lastid ||
5881 lo->ll_flags & LF_CRASHED_LASTID)
5882 GOTO(unlock, rc = 0);
5884 LASSERT(lfsck->li_out_notify != NULL);
5886 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5887 LE_LASTID_REBUILDING);
5888 lo->ll_flags |= LF_CRASHED_LASTID;
5890 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
5891 "LAST_ID file (2) for the sequence %#llx"
5892 ", old value %llu, known value %llu\n",
5893 lfsck_lfsck2name(lfsck), lls->lls_seq,
5894 lls->lls_lastid, oid);
5897 lls->lls_lastid = oid;
5901 GOTO(unlock, rc = 0);
5904 up_write(&com->lc_sem);
5909 static int lfsck_layout_exec_dir(const struct lu_env *env,
5910 struct lfsck_component *com,
5911 struct lfsck_assistant_object *lso,
5912 struct lu_dirent *ent, __u16 type)
5917 static int lfsck_layout_master_post(const struct lu_env *env,
5918 struct lfsck_component *com,
5919 int result, bool init)
5921 struct lfsck_instance *lfsck = com->lc_lfsck;
5922 struct lfsck_layout *lo = com->lc_file_ram;
5926 lfsck_post_generic(env, com, &result);
5928 down_write(&com->lc_sem);
5929 spin_lock(&lfsck->li_lock);
5931 lo->ll_pos_last_checkpoint =
5932 lfsck->li_pos_checkpoint.lp_oit_cookie;
5935 if (lo->ll_flags & LF_INCOMPLETE)
5936 lo->ll_status = LS_PARTIAL;
5938 lo->ll_status = LS_SCANNING_PHASE2;
5939 lo->ll_flags |= LF_SCANNED_ONCE;
5940 lo->ll_flags &= ~LF_UPGRADE;
5941 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5942 } else if (result == 0) {
5943 if (lfsck->li_status != 0)
5944 lo->ll_status = lfsck->li_status;
5946 lo->ll_status = LS_STOPPED;
5947 if (lo->ll_status != LS_PAUSED)
5948 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5950 lo->ll_status = LS_FAILED;
5951 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
5953 spin_unlock(&lfsck->li_lock);
5956 lo->ll_run_time_phase1 += ktime_get_seconds() -
5957 lfsck->li_time_last_checkpoint;
5958 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5959 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5960 com->lc_new_checked = 0;
5963 rc = lfsck_layout_store(env, com);
5964 up_write(&com->lc_sem);
5966 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
5967 lfsck_lfsck2name(lfsck), rc);
5972 static int lfsck_layout_slave_post(const struct lu_env *env,
5973 struct lfsck_component *com,
5974 int result, bool init)
5976 struct lfsck_instance *lfsck = com->lc_lfsck;
5977 struct lfsck_layout *lo = com->lc_file_ram;
5981 down_write(&com->lc_sem);
5982 rc = lfsck_layout_lastid_store(env, com);
5986 LASSERT(lfsck->li_out_notify != NULL);
5988 spin_lock(&lfsck->li_lock);
5990 lo->ll_pos_last_checkpoint =
5991 lfsck->li_pos_checkpoint.lp_oit_cookie;
5994 lo->ll_status = LS_SCANNING_PHASE2;
5995 lo->ll_flags |= LF_SCANNED_ONCE;
5996 if (lo->ll_flags & LF_CRASHED_LASTID) {
5998 lo->ll_flags &= ~LF_CRASHED_LASTID;
6000 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
6001 "crashed LAST_ID files successfully\n",
6002 lfsck_lfsck2name(lfsck));
6004 lo->ll_flags &= ~LF_UPGRADE;
6005 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
6006 } else if (result == 0) {
6007 if (lfsck->li_status != 0)
6008 lo->ll_status = lfsck->li_status;
6010 lo->ll_status = LS_STOPPED;
6011 if (lo->ll_status != LS_PAUSED)
6012 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6014 lo->ll_status = LS_FAILED;
6015 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6017 spin_unlock(&lfsck->li_lock);
6020 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6024 lo->ll_run_time_phase1 += ktime_get_seconds() -
6025 lfsck->li_time_last_checkpoint;
6026 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
6027 lo->ll_objs_checked_phase1 += com->lc_new_checked;
6028 com->lc_new_checked = 0;
6031 rc = lfsck_layout_store(env, com);
6032 up_write(&com->lc_sem);
6034 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
6036 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
6037 lfsck_lfsck2name(lfsck), rc);
6042 static void lfsck_layout_dump(const struct lu_env *env,
6043 struct lfsck_component *com, struct seq_file *m)
6045 struct lfsck_instance *lfsck = com->lc_lfsck;
6046 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6047 struct lfsck_layout *lo = com->lc_file_ram;
6050 down_read(&com->lc_sem);
6051 if (bk->lb_param & LPF_DRYRUN)
6052 prefix = "inconsistent";
6054 prefix = "repaired";
6056 seq_printf(m, "name: lfsck_layout\n"
6062 lfsck_status2name(lo->ll_status));
6064 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
6066 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
6068 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
6070 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
6072 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
6074 seq_printf(m, "latest_start_position: %llu\n"
6075 "last_checkpoint_position: %llu\n"
6076 "first_failure_position: %llu\n",
6077 lo->ll_pos_latest_start,
6078 lo->ll_pos_last_checkpoint,
6079 lo->ll_pos_first_inconsistent);
6081 seq_printf(m, "success_count: %u\n"
6082 "%s_dangling: %llu\n"
6083 "%s_unmatched_pair: %llu\n"
6084 "%s_multiple_referenced: %llu\n"
6086 "%s_inconsistent_owner: %llu\n"
6089 "failed_phase1: %llu\n"
6090 "failed_phase2: %llu\n",
6091 lo->ll_success_count,
6092 prefix, lo->ll_objs_repaired[LLIT_DANGLING - 1],
6093 prefix, lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
6094 prefix, lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
6095 prefix, lo->ll_objs_repaired[LLIT_ORPHAN - 1],
6096 prefix, lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
6097 prefix, lo->ll_objs_repaired[LLIT_OTHERS - 1],
6098 lo->ll_objs_skipped,
6099 lo->ll_objs_failed_phase1,
6100 lo->ll_objs_failed_phase2);
6102 if (lo->ll_status == LS_SCANNING_PHASE1) {
6103 time64_t duration = ktime_get_seconds() -
6104 lfsck->li_time_last_checkpoint;
6105 u64 checked = lo->ll_objs_checked_phase1 +
6106 com->lc_new_checked;
6107 u64 speed = checked;
6108 u64 new_checked = com->lc_new_checked;
6109 time64_t rtime = lo->ll_run_time_phase1 + duration;
6113 new_checked = div64_s64(new_checked, duration);
6115 speed = div64_s64(speed, rtime);
6116 seq_printf(m, "checked_phase1: %llu\n"
6117 "checked_phase2: %llu\n"
6118 "run_time_phase1: %lld seconds\n"
6119 "run_time_phase2: %lld seconds\n"
6120 "average_speed_phase1: %llu items/sec\n"
6121 "average_speed_phase2: N/A\n"
6122 "real_time_speed_phase1: %llu items/sec\n"
6123 "real_time_speed_phase2: N/A\n",
6125 lo->ll_objs_checked_phase2,
6127 lo->ll_run_time_phase2,
6131 if (likely(lfsck->li_di_oit)) {
6132 const struct dt_it_ops *iops =
6133 &lfsck->li_obj_oit->do_index_ops->dio_it;
6135 /* The low layer otable-based iteration position may NOT
6136 * exactly match the layout-based directory traversal
6137 * cookie. Generally, it is not a serious issue. But the
6138 * caller should NOT make assumption on that. */
6139 pos = iops->store(env, lfsck->li_di_oit);
6140 if (!lfsck->li_current_oit_processed)
6143 pos = lo->ll_pos_last_checkpoint;
6146 seq_printf(m, "current_position: %llu\n", pos);
6147 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
6148 time64_t duration = ktime_get_seconds() -
6149 com->lc_time_last_checkpoint;
6150 u64 checked = lo->ll_objs_checked_phase2 +
6151 com->lc_new_checked;
6152 u64 speed1 = lo->ll_objs_checked_phase1;
6153 u64 speed2 = checked;
6154 u64 new_checked = com->lc_new_checked;
6155 time64_t rtime = lo->ll_run_time_phase2 + duration;
6158 new_checked = div64_s64(new_checked, duration);
6159 if (lo->ll_run_time_phase1 != 0)
6160 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
6162 speed2 = div64_s64(speed2, rtime);
6163 seq_printf(m, "checked_phase1: %llu\n"
6164 "checked_phase2: %llu\n"
6165 "run_time_phase1: %lld seconds\n"
6166 "run_time_phase2: %lld seconds\n"
6167 "average_speed_phase1: %llu items/sec\n"
6168 "average_speed_phase2: %llu items/sec\n"
6169 "real_time_speed_phase1: N/A\n"
6170 "real_time_speed_phase2: %llu items/sec\n"
6171 "current_position: "DFID"\n",
6172 lo->ll_objs_checked_phase1,
6174 lo->ll_run_time_phase1,
6179 PFID(&com->lc_fid_latest_scanned_phase2));
6181 __u64 speed1 = lo->ll_objs_checked_phase1;
6182 __u64 speed2 = lo->ll_objs_checked_phase2;
6184 if (lo->ll_run_time_phase1 != 0)
6185 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
6186 if (lo->ll_run_time_phase2 != 0)
6187 speed2 = div64_s64(speed2, lo->ll_run_time_phase2);
6188 seq_printf(m, "checked_phase1: %llu\n"
6189 "checked_phase2: %llu\n"
6190 "run_time_phase1: %lld seconds\n"
6191 "run_time_phase2: %lld seconds\n"
6192 "average_speed_phase1: %llu items/sec\n"
6193 "average_speed_phase2: %llu objs/sec\n"
6194 "real_time_speed_phase1: N/A\n"
6195 "real_time_speed_phase2: N/A\n"
6196 "current_position: N/A\n",
6197 lo->ll_objs_checked_phase1,
6198 lo->ll_objs_checked_phase2,
6199 lo->ll_run_time_phase1,
6200 lo->ll_run_time_phase2,
6205 up_read(&com->lc_sem);
6208 static int lfsck_layout_master_double_scan(const struct lu_env *env,
6209 struct lfsck_component *com)
6211 struct lfsck_layout *lo = com->lc_file_ram;
6212 struct lfsck_assistant_data *lad = com->lc_data;
6213 struct lfsck_instance *lfsck = com->lc_lfsck;
6214 struct lfsck_tgt_descs *ltds;
6215 struct lfsck_tgt_desc *ltd;
6216 struct lfsck_tgt_desc *next;
6219 rc = lfsck_double_scan_generic(env, com, lo->ll_status);
6221 if (thread_is_stopped(&lad->lad_thread)) {
6222 LASSERT(list_empty(&lad->lad_req_list));
6223 LASSERT(list_empty(&lad->lad_ost_phase1_list));
6224 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
6226 ltds = &lfsck->li_ost_descs;
6227 spin_lock(<ds->ltd_lock);
6228 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6229 ltd_layout_phase_list) {
6230 list_del_init(<d->ltd_layout_phase_list);
6232 spin_unlock(<ds->ltd_lock);
6234 ltds = &lfsck->li_mdt_descs;
6235 spin_lock(<ds->ltd_lock);
6236 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6237 ltd_layout_phase_list) {
6238 list_del_init(<d->ltd_layout_phase_list);
6240 spin_unlock(<ds->ltd_lock);
6246 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
6247 struct lfsck_component *com)
6249 struct lfsck_instance *lfsck = com->lc_lfsck;
6250 struct lfsck_layout_slave_data *llsd = com->lc_data;
6251 struct lfsck_layout *lo = com->lc_file_ram;
6252 struct ptlrpc_thread *thread = &lfsck->li_thread;
6256 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
6257 lfsck_lfsck2name(lfsck));
6259 atomic_inc(&lfsck->li_double_scan_count);
6261 if (lo->ll_flags & LF_INCOMPLETE)
6264 com->lc_new_checked = 0;
6265 com->lc_new_scanned = 0;
6266 com->lc_time_last_checkpoint = ktime_get_seconds();
6267 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6268 LFSCK_CHECKPOINT_INTERVAL;
6271 rc = lfsck_layout_slave_query_master(env, com);
6272 if (list_empty(&llsd->llsd_master_list)) {
6273 if (unlikely(!thread_is_running(thread)))
6284 rc = wait_event_idle_timeout(
6285 thread->t_ctl_waitq,
6286 !thread_is_running(thread) ||
6287 lo->ll_flags & LF_INCOMPLETE ||
6288 list_empty(&llsd->llsd_master_list),
6289 cfs_time_seconds(30));
6290 if (unlikely(!thread_is_running(thread)))
6293 if (lo->ll_flags & LF_INCOMPLETE)
6303 rc = lfsck_layout_double_scan_result(env, com, rc);
6304 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
6305 (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
6306 lfsck_layout_slave_quit(env, com);
6307 if (atomic_dec_and_test(&lfsck->li_double_scan_count))
6308 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6310 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
6311 "status %d: rc = %d\n",
6312 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
6317 static void lfsck_layout_master_data_release(const struct lu_env *env,
6318 struct lfsck_component *com)
6320 struct lfsck_assistant_data *lad = com->lc_data;
6321 struct lfsck_instance *lfsck = com->lc_lfsck;
6322 struct lfsck_tgt_descs *ltds;
6323 struct lfsck_tgt_desc *ltd;
6324 struct lfsck_tgt_desc *next;
6326 LASSERT(lad != NULL);
6327 LASSERT(thread_is_init(&lad->lad_thread) ||
6328 thread_is_stopped(&lad->lad_thread));
6329 LASSERT(list_empty(&lad->lad_req_list));
6331 com->lc_data = NULL;
6333 ltds = &lfsck->li_ost_descs;
6334 spin_lock(<ds->ltd_lock);
6335 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6336 ltd_layout_phase_list) {
6337 list_del_init(<d->ltd_layout_phase_list);
6339 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6340 ltd_layout_phase_list) {
6341 list_del_init(<d->ltd_layout_phase_list);
6343 list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
6345 list_del_init(<d->ltd_layout_list);
6347 spin_unlock(<ds->ltd_lock);
6349 ltds = &lfsck->li_mdt_descs;
6350 spin_lock(<ds->ltd_lock);
6351 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6352 ltd_layout_phase_list) {
6353 list_del_init(<d->ltd_layout_phase_list);
6355 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6356 ltd_layout_phase_list) {
6357 list_del_init(<d->ltd_layout_phase_list);
6359 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
6361 list_del_init(<d->ltd_layout_list);
6363 spin_unlock(<ds->ltd_lock);
6365 if (likely(lad->lad_bitmap != NULL))
6366 CFS_FREE_BITMAP(lad->lad_bitmap);
6371 static void lfsck_layout_slave_data_release(const struct lu_env *env,
6372 struct lfsck_component *com)
6374 struct lfsck_layout_slave_data *llsd = com->lc_data;
6376 lfsck_layout_slave_quit(env, com);
6377 com->lc_data = NULL;
6381 static void lfsck_layout_master_quit(const struct lu_env *env,
6382 struct lfsck_component *com)
6384 struct lfsck_assistant_data *lad = com->lc_data;
6385 struct lfsck_instance *lfsck = com->lc_lfsck;
6386 struct lfsck_tgt_descs *ltds;
6387 struct lfsck_tgt_desc *ltd;
6388 struct lfsck_tgt_desc *next;
6390 LASSERT(lad != NULL);
6392 lfsck_quit_generic(env, com);
6394 LASSERT(thread_is_init(&lad->lad_thread) ||
6395 thread_is_stopped(&lad->lad_thread));
6396 LASSERT(list_empty(&lad->lad_req_list));
6398 ltds = &lfsck->li_ost_descs;
6399 spin_lock(<ds->ltd_lock);
6400 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6401 ltd_layout_phase_list) {
6402 list_del_init(<d->ltd_layout_phase_list);
6404 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6405 ltd_layout_phase_list) {
6406 list_del_init(<d->ltd_layout_phase_list);
6408 spin_unlock(<ds->ltd_lock);
6410 ltds = &lfsck->li_mdt_descs;
6411 spin_lock(<ds->ltd_lock);
6412 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6413 ltd_layout_phase_list) {
6414 list_del_init(<d->ltd_layout_phase_list);
6416 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6417 ltd_layout_phase_list) {
6418 list_del_init(<d->ltd_layout_phase_list);
6420 spin_unlock(<ds->ltd_lock);
6423 static void lfsck_layout_slave_quit(const struct lu_env *env,
6424 struct lfsck_component *com)
6426 struct lfsck_layout_slave_data *llsd = com->lc_data;
6427 struct lfsck_layout_seq *lls;
6428 struct lfsck_layout_seq *next;
6429 struct lfsck_layout_slave_target *llst;
6431 LASSERT(llsd != NULL);
6433 down_write(&com->lc_sem);
6434 list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
6436 list_del_init(&lls->lls_list);
6437 lfsck_object_put(env, lls->lls_lastid_obj);
6440 up_write(&com->lc_sem);
6442 spin_lock(&llsd->llsd_lock);
6443 while (!list_empty(&llsd->llsd_master_list)) {
6444 llst = list_entry(llsd->llsd_master_list.next,
6445 struct lfsck_layout_slave_target, llst_list);
6446 list_del_init(&llst->llst_list);
6447 spin_unlock(&llsd->llsd_lock);
6448 lfsck_layout_llst_put(llst);
6449 spin_lock(&llsd->llsd_lock);
6451 spin_unlock(&llsd->llsd_lock);
6453 lfsck_rbtree_cleanup(env, com);
6456 static int lfsck_layout_master_in_notify(const struct lu_env *env,
6457 struct lfsck_component *com,
6458 struct lfsck_request *lr)
6460 struct lfsck_instance *lfsck = com->lc_lfsck;
6461 struct lfsck_layout *lo = com->lc_file_ram;
6462 struct lfsck_assistant_data *lad = com->lc_data;
6463 struct lfsck_tgt_descs *ltds;
6464 struct lfsck_tgt_desc *ltd;
6468 if (lr->lr_event == LE_PAIRS_VERIFY) {
6471 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
6478 CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
6479 "from %s %x, status %d, flags %x, flags2 %x\n",
6480 lfsck_lfsck2name(lfsck), lr->lr_event,
6481 (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
6482 lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
6484 if (lr->lr_event != LE_PHASE1_DONE &&
6485 lr->lr_event != LE_PHASE2_DONE &&
6486 lr->lr_event != LE_PEER_EXIT)
6489 if (lr->lr_flags & LEF_FROM_OST)
6490 ltds = &lfsck->li_ost_descs;
6492 ltds = &lfsck->li_mdt_descs;
6493 spin_lock(<ds->ltd_lock);
6494 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
6496 spin_unlock(<ds->ltd_lock);
6501 list_del_init(<d->ltd_layout_phase_list);
6502 switch (lr->lr_event) {
6503 case LE_PHASE1_DONE:
6504 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
6505 if (lr->lr_flags2 & LF_INCOMPLETE) {
6506 if (lr->lr_flags & LEF_FROM_OST)
6507 lfsck_lad_set_bitmap(env, com,
6510 lo->ll_flags |= LF_INCOMPLETE;
6512 ltd->ltd_layout_done = 1;
6513 list_del_init(<d->ltd_layout_list);
6518 if (lr->lr_flags & LEF_FROM_OST) {
6519 if (list_empty(<d->ltd_layout_list))
6520 list_add_tail(<d->ltd_layout_list,
6521 &lad->lad_ost_list);
6522 list_add_tail(<d->ltd_layout_phase_list,
6523 &lad->lad_ost_phase2_list);
6525 if (list_empty(<d->ltd_layout_list))
6526 list_add_tail(<d->ltd_layout_list,
6527 &lad->lad_mdt_list);
6528 list_add_tail(<d->ltd_layout_phase_list,
6529 &lad->lad_mdt_phase2_list);
6532 case LE_PHASE2_DONE:
6533 ltd->ltd_layout_done = 1;
6534 if (!list_empty(<d->ltd_layout_list))
6535 list_del_init(<d->ltd_layout_list);
6537 if (lr->lr_flags2 & LF_INCOMPLETE) {
6538 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
6545 ltd->ltd_layout_done = 1;
6546 list_del_init(<d->ltd_layout_list);
6547 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
6548 !(lr->lr_flags & LEF_FROM_OST))
6549 lo->ll_flags |= LF_INCOMPLETE;
6554 spin_unlock(<ds->ltd_lock);
6556 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
6557 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6559 memset(stop, 0, sizeof(*stop));
6560 stop->ls_status = lr->lr_status;
6561 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6562 lfsck_stop(env, lfsck->li_bottom, stop);
6563 } else if (lfsck_phase2_next_ready(lad)) {
6564 wake_up_all(&lad->lad_thread.t_ctl_waitq);
6570 static int lfsck_layout_slave_in_notify_local(const struct lu_env *env,
6571 struct lfsck_component *com,
6572 struct lfsck_req_local *lrl,
6577 switch (lrl->lrl_event) {
6578 case LEL_FID_ACCESSED:
6579 lfsck_rbtree_update_bitmap(env, com, &lrl->lrl_fid, true);
6581 case LEL_PAIRS_VERIFY_LOCAL: {
6584 lrl->lrl_status = LPVS_INIT;
6585 /* Firstly, if the MDT-object which is claimed via OST-object
6586 * local stored PFID xattr recognizes the OST-object, then it
6587 * must be that the client given PFID is wrong. */
6588 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6589 &lrl->lrl_ff_local.ff_parent,
6590 lrl->lrl_ff_local.ff_layout.ol_comp_id);
6594 lrl->lrl_status = LPVS_INCONSISTENT;
6595 /* The OST-object local stored PFID xattr is stale. We need to
6596 * check whether the MDT-object that is claimed via the client
6597 * given PFID information recognizes the OST-object or not. If
6598 * matches, then need to update the OST-object's PFID xattr. */
6599 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6600 &lrl->lrl_ff_client.ff_parent,
6601 lrl->lrl_ff_client.ff_layout.ol_comp_id);
6603 * We are not sure whether the client given PFID information
6604 * is correct or not, do nothing to avoid improper fixing.
6607 * The client given PFID information is also invalid, we can
6608 * NOT fix the OST-object inconsistency.
6611 lrl->lrl_status = LPVS_INCONSISTENT_TOFIX;
6612 rc = lfsck_layout_slave_repair_pfid(env, com, lrl);
6621 RETURN(-EOPNOTSUPP);
6624 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
6625 struct lfsck_component *com,
6626 struct lfsck_request *lr)
6628 struct lfsck_instance *lfsck = com->lc_lfsck;
6629 struct lfsck_layout_slave_data *llsd = com->lc_data;
6630 struct lfsck_layout_slave_target *llst;
6634 switch (lr->lr_event) {
6635 case LE_CONDITIONAL_DESTROY:
6636 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
6638 case LE_PHASE1_DONE: {
6639 if (lr->lr_flags2 & LF_INCOMPLETE) {
6640 struct lfsck_layout *lo = com->lc_file_ram;
6642 lo->ll_flags |= LF_INCOMPLETE;
6643 llst = lfsck_layout_llst_find_and_del(llsd,
6647 lfsck_layout_llst_put(llst);
6648 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6654 case LE_PHASE2_DONE:
6656 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
6657 "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
6658 lr->lr_event, lr->lr_index, lr->lr_status);
6664 llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
6668 lfsck_layout_llst_put(llst);
6669 if (list_empty(&llsd->llsd_master_list))
6670 wake_up_all(&lfsck->li_thread.t_ctl_waitq);
6672 if (lr->lr_event == LE_PEER_EXIT &&
6673 (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
6674 (list_empty(&llsd->llsd_master_list) &&
6675 (lr->lr_status == LS_STOPPED ||
6676 lr->lr_status == LS_CO_STOPPED)))) {
6677 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6679 memset(stop, 0, sizeof(*stop));
6680 stop->ls_status = lr->lr_status;
6681 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6682 lfsck_stop(env, lfsck->li_bottom, stop);
6688 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
6692 for (i = 0; i < LLIT_MAX; i++)
6693 *count += lo->ll_objs_repaired[i];
6696 static int lfsck_layout_query_all(const struct lu_env *env,
6697 struct lfsck_component *com,
6698 __u32 *mdts_count, __u32 *osts_count,
6701 struct lfsck_layout *lo = com->lc_file_ram;
6702 struct lfsck_tgt_descs *ltds;
6703 struct lfsck_tgt_desc *ltd;
6708 rc = lfsck_query_all(env, com);
6712 ltds = &com->lc_lfsck->li_mdt_descs;
6713 down_read(<ds->ltd_rw_sem);
6714 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
6715 ltd = lfsck_ltd2tgt(ltds, idx);
6716 LASSERT(ltd != NULL);
6718 mdts_count[ltd->ltd_layout_status]++;
6719 *repaired += ltd->ltd_layout_repaired;
6721 up_read(<ds->ltd_rw_sem);
6723 ltds = &com->lc_lfsck->li_ost_descs;
6724 down_read(<ds->ltd_rw_sem);
6725 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
6726 ltd = lfsck_ltd2tgt(ltds, idx);
6727 LASSERT(ltd != NULL);
6729 osts_count[ltd->ltd_layout_status]++;
6730 *repaired += ltd->ltd_layout_repaired;
6732 up_read(<ds->ltd_rw_sem);
6734 down_read(&com->lc_sem);
6735 mdts_count[lo->ll_status]++;
6736 lfsck_layout_repaired(lo, repaired);
6737 up_read(&com->lc_sem);
6742 static int lfsck_layout_query(const struct lu_env *env,
6743 struct lfsck_component *com,
6744 struct lfsck_request *req,
6745 struct lfsck_reply *rep,
6746 struct lfsck_query *que, int idx)
6748 struct lfsck_layout *lo = com->lc_file_ram;
6752 LASSERT(com->lc_lfsck->li_master);
6754 rc = lfsck_layout_query_all(env, com,
6755 que->lu_mdts_count[idx],
6756 que->lu_osts_count[idx],
6757 &que->lu_repaired[idx]);
6759 down_read(&com->lc_sem);
6760 rep->lr_status = lo->ll_status;
6761 if (req->lr_flags & LEF_QUERY_ALL)
6762 lfsck_layout_repaired(lo, &rep->lr_repaired);
6763 up_read(&com->lc_sem);
6769 /* with lfsck::li_lock held */
6770 static int lfsck_layout_slave_join(const struct lu_env *env,
6771 struct lfsck_component *com,
6772 struct lfsck_start_param *lsp)
6774 struct lfsck_instance *lfsck = com->lc_lfsck;
6775 struct lfsck_layout_slave_data *llsd = com->lc_data;
6776 struct lfsck_layout_slave_target *llst;
6777 struct lfsck_start *start = lsp->lsp_start;
6781 if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
6784 if (!lsp->lsp_index_valid)
6787 /* If someone is running the LFSCK without orphan handling,
6788 * it will not maintain the object accessing rbtree. So we
6789 * cannot join it for orphan handling. */
6790 if (!llsd->llsd_rbtree_valid)
6793 spin_unlock(&lfsck->li_lock);
6794 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
6795 spin_lock(&lfsck->li_lock);
6796 if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
6797 spin_unlock(&lfsck->li_lock);
6798 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
6801 lfsck_layout_llst_put(llst);
6802 spin_lock(&lfsck->li_lock);
6809 static struct lfsck_operations lfsck_layout_master_ops = {
6810 .lfsck_reset = lfsck_layout_reset,
6811 .lfsck_fail = lfsck_layout_fail,
6812 .lfsck_checkpoint = lfsck_layout_master_checkpoint,
6813 .lfsck_prep = lfsck_layout_master_prep,
6814 .lfsck_exec_oit = lfsck_layout_master_exec_oit,
6815 .lfsck_exec_dir = lfsck_layout_exec_dir,
6816 .lfsck_post = lfsck_layout_master_post,
6817 .lfsck_dump = lfsck_layout_dump,
6818 .lfsck_double_scan = lfsck_layout_master_double_scan,
6819 .lfsck_data_release = lfsck_layout_master_data_release,
6820 .lfsck_quit = lfsck_layout_master_quit,
6821 .lfsck_in_notify = lfsck_layout_master_in_notify,
6822 .lfsck_query = lfsck_layout_query,
6825 static struct lfsck_operations lfsck_layout_slave_ops = {
6826 .lfsck_reset = lfsck_layout_reset,
6827 .lfsck_fail = lfsck_layout_fail,
6828 .lfsck_checkpoint = lfsck_layout_slave_checkpoint,
6829 .lfsck_prep = lfsck_layout_slave_prep,
6830 .lfsck_exec_oit = lfsck_layout_slave_exec_oit,
6831 .lfsck_exec_dir = lfsck_layout_exec_dir,
6832 .lfsck_post = lfsck_layout_slave_post,
6833 .lfsck_dump = lfsck_layout_dump,
6834 .lfsck_double_scan = lfsck_layout_slave_double_scan,
6835 .lfsck_data_release = lfsck_layout_slave_data_release,
6836 .lfsck_quit = lfsck_layout_slave_quit,
6837 .lfsck_in_notify_local = lfsck_layout_slave_in_notify_local,
6838 .lfsck_in_notify = lfsck_layout_slave_in_notify,
6839 .lfsck_query = lfsck_layout_query,
6840 .lfsck_join = lfsck_layout_slave_join,
6843 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
6844 struct lfsck_component *com,
6845 struct lfsck_position *pos)
6847 struct lfsck_assistant_data *lad = com->lc_data;
6848 struct lfsck_layout_req *llr;
6850 if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
6854 if (list_empty(&lad->lad_req_list))
6857 llr = list_entry(lad->lad_req_list.next,
6858 struct lfsck_layout_req,
6860 pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
6863 struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
6864 .la_handler_p1 = lfsck_layout_assistant_handler_p1,
6865 .la_handler_p2 = lfsck_layout_assistant_handler_p2,
6866 .la_fill_pos = lfsck_layout_assistant_fill_pos,
6867 .la_double_scan_result = lfsck_layout_double_scan_result,
6868 .la_req_fini = lfsck_layout_assistant_req_fini,
6869 .la_sync_failures = lfsck_layout_assistant_sync_failures,
6872 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
6874 struct lfsck_component *com;
6875 struct lfsck_layout *lo;
6876 struct dt_object *root = NULL;
6877 struct dt_object *obj;
6886 INIT_LIST_HEAD(&com->lc_link);
6887 INIT_LIST_HEAD(&com->lc_link_dir);
6888 init_rwsem(&com->lc_sem);
6889 atomic_set(&com->lc_ref, 1);
6890 com->lc_lfsck = lfsck;
6891 com->lc_type = LFSCK_TYPE_LAYOUT;
6892 if (lfsck->li_master) {
6893 com->lc_ops = &lfsck_layout_master_ops;
6894 com->lc_data = lfsck_assistant_data_init(
6895 &lfsck_layout_assistant_ops,
6897 if (com->lc_data == NULL)
6898 GOTO(out, rc = -ENOMEM);
6900 for (i = 0; i < LFSCK_STF_COUNT; i++)
6901 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6903 struct lfsck_layout_slave_data *llsd;
6905 com->lc_ops = &lfsck_layout_slave_ops;
6906 OBD_ALLOC_PTR(llsd);
6908 GOTO(out, rc = -ENOMEM);
6910 INIT_LIST_HEAD(&llsd->llsd_seq_list);
6911 INIT_LIST_HEAD(&llsd->llsd_master_list);
6912 spin_lock_init(&llsd->llsd_lock);
6913 llsd->llsd_rb_root = RB_ROOT;
6914 init_rwsem(&llsd->llsd_rb_rwsem);
6915 com->lc_data = llsd;
6917 com->lc_file_size = sizeof(*lo);
6918 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6919 if (com->lc_file_ram == NULL)
6920 GOTO(out, rc = -ENOMEM);
6922 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6923 if (com->lc_file_disk == NULL)
6924 GOTO(out, rc = -ENOMEM);
6926 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6928 GOTO(out, rc = PTR_ERR(root));
6930 if (unlikely(!dt_try_as_dir(env, root)))
6931 GOTO(out, rc = -ENOTDIR);
6933 obj = local_file_find_or_create(env, lfsck->li_los, root,
6935 S_IFREG | S_IRUGO | S_IWUSR);
6937 GOTO(out, rc = PTR_ERR(obj));
6940 rc = lfsck_layout_load(env, com);
6942 rc = lfsck_layout_reset(env, com, true);
6943 } else if (rc == -ENOENT) {
6944 rc = lfsck_layout_init(env, com);
6945 } else if (lfsck->li_master) {
6946 rc = lfsck_load_sub_trace_files(env, com,
6947 &dt_lfsck_layout_dangling_features,
6948 LFSCK_LAYOUT, false);
6950 rc = lfsck_layout_reset(env, com, true);
6956 lo = com->lc_file_ram;
6957 switch (lo->ll_status) {
6963 spin_lock(&lfsck->li_lock);
6964 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6965 spin_unlock(&lfsck->li_lock);
6968 CERROR("%s: unknown lfsck_layout status %d\n",
6969 lfsck_lfsck2name(lfsck), lo->ll_status);
6971 case LS_SCANNING_PHASE1:
6972 case LS_SCANNING_PHASE2:
6973 /* No need to store the status to disk right now.
6974 * If the system crashed before the status stored,
6975 * it will be loaded back when next time. */
6976 lo->ll_status = LS_CRASHED;
6977 if (!lfsck->li_master)
6978 lo->ll_flags |= LF_INCOMPLETE;
6985 spin_lock(&lfsck->li_lock);
6986 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6987 spin_unlock(&lfsck->li_lock);
6991 if (lo->ll_flags & LF_CRASHED_LASTID) {
6992 LASSERT(lfsck->li_out_notify != NULL);
6994 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6995 LE_LASTID_REBUILDING);
7001 if (root != NULL && !IS_ERR(root))
7002 lfsck_object_put(env, root);
7005 lfsck_component_cleanup(env, com);
7006 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
7007 lfsck_lfsck2name(lfsck), rc);
7013 struct lfsck_orphan_it {
7014 struct lfsck_component *loi_com;
7015 struct lfsck_rbtree_node *loi_lrn;
7016 struct lfsck_layout_slave_target *loi_llst;
7017 struct lu_fid loi_key;
7018 struct lu_orphan_rec_v3 loi_rec;
7020 unsigned int loi_over:1;
7023 static int lfsck_fid_match_idx(const struct lu_env *env,
7024 struct lfsck_instance *lfsck,
7025 const struct lu_fid *fid, int idx)
7027 struct seq_server_site *ss;
7028 struct lu_server_fld *sf;
7029 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
7032 /* All abnormal cases will be returned to MDT0. */
7033 if (!fid_is_norm(fid)) {
7040 ss = lfsck_dev_site(lfsck);
7041 if (unlikely(ss == NULL))
7044 sf = ss->ss_server_fld;
7045 LASSERT(sf != NULL);
7047 fld_range_set_any(range);
7048 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
7052 if (!fld_range_is_mdt(range))
7055 if (range->lsr_index == idx)
7061 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
7062 struct dt_object *obj)
7064 struct dt_device *dev = lfsck_obj2dev(obj);
7065 struct thandle *handle;
7069 handle = dt_trans_create(env, dev);
7073 rc = dt_declare_ref_del(env, obj, handle);
7077 rc = dt_declare_destroy(env, obj, handle);
7081 rc = dt_trans_start_local(env, dev, handle);
7085 dt_write_lock(env, obj, 0);
7086 rc = dt_ref_del(env, obj, handle);
7088 rc = dt_destroy(env, obj, handle);
7089 dt_write_unlock(env, obj);
7094 dt_trans_stop(env, dev, handle);
7096 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
7097 PFID(lfsck_dto2fid(obj)), rc);
7102 static int lfsck_orphan_index_lookup(const struct lu_env *env,
7103 struct dt_object *dt,
7105 const struct dt_key *key)
7110 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
7111 struct dt_object *dt,
7112 const struct dt_rec *rec,
7113 const struct dt_key *key,
7114 struct thandle *handle)
7119 static int lfsck_orphan_index_insert(const struct lu_env *env,
7120 struct dt_object *dt,
7121 const struct dt_rec *rec,
7122 const struct dt_key *key,
7123 struct thandle *handle)
7128 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
7129 struct dt_object *dt,
7130 const struct dt_key *key,
7131 struct thandle *handle)
7136 static int lfsck_orphan_index_delete(const struct lu_env *env,
7137 struct dt_object *dt,
7138 const struct dt_key *key,
7139 struct thandle *handle)
7144 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
7145 struct dt_object *dt,
7148 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
7149 struct lfsck_instance *lfsck;
7150 struct lfsck_component *com = NULL;
7151 struct lfsck_layout_slave_data *llsd;
7152 struct lfsck_orphan_it *it = NULL;
7153 struct lfsck_layout *lo;
7157 lfsck = lfsck_instance_find(dev, true, false);
7158 if (unlikely(lfsck == NULL))
7159 RETURN(ERR_PTR(-ENXIO));
7161 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
7162 if (unlikely(com == NULL))
7163 GOTO(out, rc = -ENOENT);
7165 lo = com->lc_file_ram;
7166 if (lo->ll_flags & LF_INCOMPLETE)
7167 GOTO(out, rc = -ESRCH);
7169 llsd = com->lc_data;
7170 if (!llsd->llsd_rbtree_valid)
7171 GOTO(out, rc = -ESRCH);
7175 GOTO(out, rc = -ENOMEM);
7177 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
7178 if (it->loi_llst == NULL)
7179 GOTO(out, rc = -ENXIO);
7181 if (dev->dd_record_fid_accessed) {
7182 /* The first iteration against the rbtree, scan the whole rbtree
7183 * to remove the nodes which do NOT need to be handled. */
7184 down_write(&llsd->llsd_rb_rwsem);
7185 if (dev->dd_record_fid_accessed) {
7186 struct rb_node *node;
7187 struct rb_node *next;
7188 struct lfsck_rbtree_node *lrn;
7190 /* No need to record the fid accessing anymore. */
7191 dev->dd_record_fid_accessed = 0;
7193 node = rb_first(&llsd->llsd_rb_root);
7194 while (node != NULL) {
7195 next = rb_next(node);
7196 lrn = rb_entry(node, struct lfsck_rbtree_node,
7198 if (atomic_read(&lrn->lrn_known_count) <=
7199 atomic_read(&lrn->lrn_accessed_count)) {
7200 rb_erase(node, &llsd->llsd_rb_root);
7201 lfsck_rbtree_free(lrn);
7206 up_write(&llsd->llsd_rb_rwsem);
7209 /* read lock the rbtree when init, and unlock when fini */
7210 down_read(&llsd->llsd_rb_rwsem);
7218 lfsck_component_put(env, com);
7220 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
7221 lfsck_lfsck2name(lfsck), rc);
7223 lfsck_instance_put(env, lfsck);
7228 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
7231 return (struct dt_it *)it;
7234 static void lfsck_orphan_it_fini(const struct lu_env *env,
7237 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7238 struct lfsck_component *com = it->loi_com;
7239 struct lfsck_layout_slave_data *llsd;
7240 struct lfsck_layout_slave_target *llst;
7243 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
7244 lfsck_lfsck2name(com->lc_lfsck));
7246 llsd = com->lc_data;
7247 up_read(&llsd->llsd_rb_rwsem);
7248 llst = it->loi_llst;
7249 LASSERT(llst != NULL);
7251 /* Save the key and hash for iterate next. */
7252 llst->llst_fid = it->loi_key;
7253 llst->llst_hash = it->loi_hash;
7254 lfsck_layout_llst_put(llst);
7255 lfsck_component_put(env, com);
7261 * \retval +1: the iteration finished
7262 * \retval 0: on success, not finished
7263 * \retval -ve: on error
7265 static int lfsck_orphan_it_next(const struct lu_env *env,
7268 struct lfsck_thread_info *info = lfsck_env_info(env);
7269 struct filter_fid *ff = &info->lti_ff;
7270 struct lu_attr *la = &info->lti_la;
7271 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7272 struct lu_fid *key = &it->loi_key;
7273 struct lu_orphan_rec_v3 *rec = &it->loi_rec;
7274 struct ost_layout *ol = &rec->lor_layout;
7275 struct lfsck_component *com = it->loi_com;
7276 struct lfsck_instance *lfsck = com->lc_lfsck;
7277 struct lfsck_layout_slave_data *llsd = com->lc_data;
7278 struct dt_object *obj;
7279 struct lfsck_rbtree_node *lrn;
7283 __u32 idx = it->loi_llst->llst_index;
7293 lrn = lfsck_rbtree_search(llsd, key, &exact);
7301 key->f_seq = lrn->lrn_seq;
7302 key->f_oid = lrn->lrn_first_oid;
7307 if (unlikely(key->f_oid == 0)) {
7314 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
7320 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
7321 atomic_read(&lrn->lrn_accessed_count))) {
7322 struct rb_node *next = rb_next(&lrn->lrn_node);
7324 while (next != NULL) {
7325 lrn = rb_entry(next, struct lfsck_rbtree_node,
7327 if (atomic_read(&lrn->lrn_known_count) >
7328 atomic_read(&lrn->lrn_accessed_count))
7330 next = rb_next(next);
7339 key->f_seq = lrn->lrn_seq;
7340 key->f_oid = lrn->lrn_first_oid;
7344 pos = key->f_oid - lrn->lrn_first_oid;
7347 pos = find_next_bit(lrn->lrn_known_bitmap,
7348 LFSCK_RBTREE_BITMAP_WIDTH, pos);
7349 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
7350 key->f_oid = lrn->lrn_first_oid + pos;
7351 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
7359 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
7364 key->f_oid = lrn->lrn_first_oid + pos;
7365 obj = lfsck_object_find_bottom(env, lfsck, key);
7368 if (rc == -ENOENT) {
7375 dt_read_lock(env, obj, 0);
7376 if (dt_object_exists(obj) == 0 ||
7377 lfsck_is_dead_obj(obj)) {
7378 dt_read_unlock(env, obj);
7379 lfsck_object_put(env, obj);
7384 rc = dt_attr_get(env, obj, la);
7388 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, ff, sizeof(*ff)),
7390 if (rc == -ENODATA) {
7391 /* For the pre-created OST-object, update the bitmap to avoid
7392 * others LFSCK (second phase) iteration to touch it again. */
7393 if (la->la_ctime == 0) {
7394 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
7395 atomic_inc(&lrn->lrn_accessed_count);
7397 /* For the race between repairing dangling referenced
7398 * MDT-object and unlink the file, it may left orphan
7399 * OST-object there. Destroy it now! */
7400 if (unlikely(!(la->la_mode & S_ISUID))) {
7401 dt_read_unlock(env, obj);
7402 lfsck_layout_destroy_orphan(env, obj);
7403 lfsck_object_put(env, obj);
7407 } else if (idx == 0) {
7408 /* If the orphan OST-object has no parent information,
7409 * regard it as referenced by the MDT-object on MDT0. */
7410 fid_zero(&rec->lor_rec.lor_fid);
7411 rec->lor_rec.lor_uid = la->la_uid;
7412 rec->lor_rec.lor_gid = la->la_gid;
7413 memset(ol, 0, sizeof(*ol));
7414 rec->lor_layout_version = 0;
7420 dt_read_unlock(env, obj);
7421 lfsck_object_put(env, obj);
7426 if (rc < sizeof(struct lu_fid))
7427 GOTO(out, rc = (rc < 0 ? rc : -EINVAL));
7429 fid_le_to_cpu(&rec->lor_rec.lor_fid, &ff->ff_parent);
7430 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
7431 * MDT-object's FID::f_ver, instead it is the OST-object index in its
7432 * parent MDT-object's layout EA. */
7433 save = rec->lor_rec.lor_fid.f_stripe_idx;
7434 rec->lor_rec.lor_fid.f_ver = 0;
7435 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_rec.lor_fid, idx);
7436 /* If the orphan OST-object does not claim the MDT, then next.
7438 * If we do not know whether it matches or not, then return it
7439 * to the MDT for further check. */
7441 dt_read_unlock(env, obj);
7442 lfsck_object_put(env, obj);
7447 rec->lor_rec.lor_fid.f_stripe_idx = save;
7448 rec->lor_rec.lor_uid = la->la_uid;
7449 rec->lor_rec.lor_gid = la->la_gid;
7450 ost_layout_le_to_cpu(ol, &ff->ff_layout);
7451 rec->lor_layout_version =
7452 le32_to_cpu(ff->ff_layout_version & ~LU_LAYOUT_RESYNC);
7453 rec->lor_range = le32_to_cpu(ff->ff_range);
7455 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u, "
7456 "stripe size %u, stripe count %u, COMP id %u, COMP start %llu, "
7457 "COMP end %llu, layout version %u, range %u\n",
7458 lfsck_lfsck2name(com->lc_lfsck), PFID(key),
7459 PFID(&rec->lor_rec.lor_fid), rec->lor_rec.lor_uid,
7460 rec->lor_rec.lor_gid, ol->ol_stripe_size, ol->ol_stripe_count,
7461 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
7462 rec->lor_layout_version, rec->lor_range);
7467 dt_read_unlock(env, obj);
7468 lfsck_object_put(env, obj);
7476 * \retval +1: locate to the exactly position
7477 * \retval 0: cannot locate to the exactly position,
7478 * call next() to move to a valid position.
7479 * \retval -ve: on error
7481 static int lfsck_orphan_it_get(const struct lu_env *env,
7483 const struct dt_key *key)
7485 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7488 it->loi_key = *(struct lu_fid *)key;
7489 rc = lfsck_orphan_it_next(env, di);
7499 static void lfsck_orphan_it_put(const struct lu_env *env,
7504 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
7505 const struct dt_it *di)
7507 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7509 return (struct dt_key *)&it->loi_key;
7512 static int lfsck_orphan_it_key_size(const struct lu_env *env,
7513 const struct dt_it *di)
7515 return sizeof(struct lu_fid);
7518 static int lfsck_orphan_it_rec(const struct lu_env *env,
7519 const struct dt_it *di,
7523 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7525 *(struct lu_orphan_rec_v3 *)rec = it->loi_rec;
7530 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
7531 const struct dt_it *di)
7533 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7535 return it->loi_hash;
7539 * \retval +1: locate to the exactly position
7540 * \retval 0: cannot locate to the exactly position,
7541 * call next() to move to a valid position.
7542 * \retval -ve: on error
7544 static int lfsck_orphan_it_load(const struct lu_env *env,
7545 const struct dt_it *di,
7548 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7549 struct lfsck_layout_slave_target *llst = it->loi_llst;
7552 LASSERT(llst != NULL);
7554 if (hash != llst->llst_hash) {
7555 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
7556 "iteration does not match the one when fini "
7557 "%llu, to be reset.\n",
7558 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
7560 fid_zero(&llst->llst_fid);
7561 llst->llst_hash = 0;
7564 it->loi_key = llst->llst_fid;
7565 it->loi_hash = llst->llst_hash;
7566 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
7576 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
7577 const struct dt_it *di,
7583 static const struct dt_index_operations lfsck_orphan_index_ops = {
7584 .dio_lookup = lfsck_orphan_index_lookup,
7585 .dio_declare_insert = lfsck_orphan_index_declare_insert,
7586 .dio_insert = lfsck_orphan_index_insert,
7587 .dio_declare_delete = lfsck_orphan_index_declare_delete,
7588 .dio_delete = lfsck_orphan_index_delete,
7590 .init = lfsck_orphan_it_init,
7591 .fini = lfsck_orphan_it_fini,
7592 .get = lfsck_orphan_it_get,
7593 .put = lfsck_orphan_it_put,
7594 .next = lfsck_orphan_it_next,
7595 .key = lfsck_orphan_it_key,
7596 .key_size = lfsck_orphan_it_key_size,
7597 .rec = lfsck_orphan_it_rec,
7598 .store = lfsck_orphan_it_store,
7599 .load = lfsck_orphan_it_load,
7600 .key_rec = lfsck_orphan_it_key_rec,