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 border, __u32 size)
101 return (border & (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_of(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 for_each_set_bit(idx, lad->lad_bitmap, lad->lad_bitmap_count) {
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, __u64 end,
340 magic = le32_to_cpu(lmm->lmm_magic);
341 /* If magic crashed, keep it there. Sometime later, during OST-object
342 * orphan handling, if some OST-object(s) back-point to it, it can be
343 * verified and repaired. */
344 if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3) {
347 if ((magic & LOV_MAGIC_MASK) == LOV_MAGIC_MAGIC)
352 CDEBUG(D_LFSCK, "%s LOV EA magic 0x%X for the file "DFID"\n",
353 rc == -EINVAL ? "Unknown" : "Unsupported",
354 magic, PFID(lfsck_dto2fid(obj)));
359 pattern = le32_to_cpu(lmm->lmm_pattern);
360 *dom = !!(lov_pattern(pattern) == LOV_PATTERN_MDT);
362 /* XXX: DoM file verification will be supportted via LU-11081. */
363 if (lov_pattern(pattern) == LOV_PATTERN_MDT) {
366 CDEBUG(D_LFSCK, "The DoM entry for "DFID" is not "
367 "the first component in the mirror %x/%llu\n",
368 PFID(lfsck_dto2fid(obj)), comp_id, start);
373 } else if (!lov_pattern_supported_normal_comp(lov_pattern(pattern))) {
374 CDEBUG(D_LFSCK, "Unsupported LOV EA pattern %u for the file "
375 DFID" in the component %x\n",
376 pattern, PFID(lfsck_dto2fid(obj)), comp_id);
381 size = le32_to_cpu(lmm->lmm_stripe_size);
382 if (!ext && end != LUSTRE_EOF && start != end &&
383 !lfsck_comp_extent_aligned(end, size)){
384 CDEBUG(D_LFSCK, "not aligned border in PFL extent range "
385 "[%llu - %llu) stripesize %u for the file "DFID
386 " at idx %d\n", start, end, size,
387 PFID(lfsck_dto2fid(obj)), comp_id);
395 static int lfsck_layout_verify_header_foreign(struct dt_object *obj,
396 struct lov_foreign_md *lfm,
399 /* magic has been verified already */
400 __u32 value_len = le32_to_cpu(lfm->lfm_length);
401 /* type and flags are not checked for instance */
403 CDEBUG(D_INFO, "foreign LOV EA, magic %x, len %u, type %x, flags %x, for file "DFID"\n",
404 le32_to_cpu(lfm->lfm_magic), value_len,
405 le32_to_cpu(lfm->lfm_type), le32_to_cpu(lfm->lfm_flags),
406 PFID(lfsck_dto2fid(obj)));
408 if (len != value_len + offsetof(typeof(*lfm), lfm_value))
409 CDEBUG(D_LFSCK, "foreign LOV EA internal size %u does not match EA full size %zu for file "DFID"\n",
410 value_len, len, PFID(lfsck_dto2fid(obj)));
412 /* nothing to repair */
416 static int lfsck_layout_verify_header(struct dt_object *obj,
417 struct lov_mds_md_v1 *lmm, size_t len)
422 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1 ||
423 le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_SEL) {
424 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
427 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
429 if (unlikely(count == 0)) {
430 CDEBUG(D_LFSCK, "the PFL file "DFID" contains invalid "
431 "components count 0\n",
432 PFID(lfsck_dto2fid(obj)));
437 for (i = 0; i < count && !rc; i++) {
438 struct lov_comp_md_entry_v1 *lcme =
439 &lcm->lcm_entries[i];
440 __u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
441 __u64 end = le64_to_cpu(lcme->lcme_extent.e_end);
442 __u32 comp_id = le32_to_cpu(lcme->lcme_id);
443 bool ext, inited, zero;
446 if (unlikely(comp_id == LCME_ID_INVAL ||
447 comp_id > LCME_ID_MAX)) {
448 CDEBUG(D_LFSCK, "found invalid PFL ID %u "
449 "for the file "DFID" at idx %d\n",
450 comp_id, PFID(lfsck_dto2fid(obj)), i);
455 flags = le32_to_cpu(lcme->lcme_flags);
456 ext = flags & LCME_FL_EXTENSION;
457 inited = flags & LCME_FL_INIT;
458 zero = !!(start == end);
460 if ((i == 0) && zero) {
461 CDEBUG(D_LFSCK, "invalid PFL comp %d: [%llu "
462 "- %llu) for "DFID"\n", i, start, end,
463 PFID(lfsck_dto2fid(obj)));
467 if ((zero && (inited || (i + 1 == count))) ||
469 CDEBUG(D_LFSCK, "invalid PFL comp %d/%d: "
470 "[%llu, %llu) for "DFID", %sinited\n",
471 i, count, start, end,
472 PFID(lfsck_dto2fid(obj)),
473 inited ? "" : "NOT ");
477 if (!ext && p_zero) {
478 CDEBUG(D_LFSCK, "invalid PFL comp %d: [%llu, "
479 "%llu) for "DFID": NOT extension "
480 "after 0-length component\n", i,
481 start, end, PFID(lfsck_dto2fid(obj)));
485 if (ext && (inited || p_dom || zero)) {
486 CDEBUG(D_LFSCK, "invalid PFL comp %d: [%llu, "
487 "%llu) for "DFID": %s\n", i,
488 start, end, PFID(lfsck_dto2fid(obj)),
489 inited ? "inited extension" :
490 p_dom ? "extension follows DOM" :
491 zero ? "zero length extension" : "");
495 rc = lfsck_layout_verify_header_v1v3(obj,
496 (struct lov_mds_md_v1 *)((char *)lmm +
497 le32_to_cpu(lcme->lcme_offset)), start,
498 end, comp_id, ext, &p_dom);
502 } else if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_FOREIGN) {
503 rc = lfsck_layout_verify_header_foreign(obj,
504 (struct lov_foreign_md *)lmm,
507 rc = lfsck_layout_verify_header_v1v3(obj, lmm, 0, LUSTRE_EOF,
514 static int lfsck_layout_get_lovea(const struct lu_env *env,
515 struct dt_object *obj, struct lu_buf *buf)
521 rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LOV);
523 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV);
525 return !rc ? -ENODATA : rc;
527 lu_buf_realloc(buf, rc);
528 if (buf->lb_buf == NULL)
535 return !rc ? -ENODATA : rc;
537 if (unlikely(buf->lb_buf == NULL)) {
538 lu_buf_alloc(buf, rc);
539 if (buf->lb_buf == NULL)
545 rc1 = lfsck_layout_verify_header(obj, buf->lb_buf, rc);
547 return rc1 ? rc1 : rc;
550 #define LFSCK_RBTREE_BITMAP_SIZE PAGE_SIZE
551 #define LFSCK_RBTREE_BITMAP_WIDTH (LFSCK_RBTREE_BITMAP_SIZE << 3)
552 #define LFSCK_RBTREE_BITMAP_MASK (LFSCK_RBTREE_BITMAP_WIDTH - 1)
554 struct lfsck_rbtree_node {
555 struct rb_node lrn_node;
558 atomic_t lrn_known_count;
559 atomic_t lrn_accessed_count;
560 void *lrn_known_bitmap;
561 void *lrn_accessed_bitmap;
564 static inline int lfsck_rbtree_cmp(struct lfsck_rbtree_node *lrn,
565 __u64 seq, __u32 oid)
567 if (seq < lrn->lrn_seq)
570 if (seq > lrn->lrn_seq)
573 if (oid < lrn->lrn_first_oid)
576 if (oid - lrn->lrn_first_oid >= LFSCK_RBTREE_BITMAP_WIDTH)
582 /* The caller should hold llsd->llsd_rb_lock. */
583 static struct lfsck_rbtree_node *
584 lfsck_rbtree_search(struct lfsck_layout_slave_data *llsd,
585 const struct lu_fid *fid, bool *exact)
587 struct rb_node *node = llsd->llsd_rb_root.rb_node;
588 struct rb_node *prev = NULL;
589 struct lfsck_rbtree_node *lrn = NULL;
595 while (node != NULL) {
597 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
598 rc = lfsck_rbtree_cmp(lrn, fid_seq(fid), fid_oid(fid));
600 node = node->rb_left;
602 node = node->rb_right;
610 /* If there is no exactly matched one, then to the next valid one. */
613 /* The rbtree is empty. */
620 node = rb_next(prev);
622 /* The end of the rbtree. */
626 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
631 static struct lfsck_rbtree_node *lfsck_rbtree_new(const struct lu_env *env,
632 const struct lu_fid *fid)
634 struct lfsck_rbtree_node *lrn;
638 return ERR_PTR(-ENOMEM);
640 OBD_ALLOC(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
641 if (lrn->lrn_known_bitmap == NULL) {
644 return ERR_PTR(-ENOMEM);
647 OBD_ALLOC(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
648 if (lrn->lrn_accessed_bitmap == NULL) {
649 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
652 return ERR_PTR(-ENOMEM);
655 RB_CLEAR_NODE(&lrn->lrn_node);
656 lrn->lrn_seq = fid_seq(fid);
657 lrn->lrn_first_oid = fid_oid(fid) & ~LFSCK_RBTREE_BITMAP_MASK;
658 atomic_set(&lrn->lrn_known_count, 0);
659 atomic_set(&lrn->lrn_accessed_count, 0);
664 static void lfsck_rbtree_free(struct lfsck_rbtree_node *lrn)
666 OBD_FREE(lrn->lrn_accessed_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
667 OBD_FREE(lrn->lrn_known_bitmap, LFSCK_RBTREE_BITMAP_SIZE);
671 /* The caller should hold lock. */
672 static struct lfsck_rbtree_node *
673 lfsck_rbtree_insert(struct lfsck_layout_slave_data *llsd,
674 struct lfsck_rbtree_node *lrn)
676 struct rb_node **pos = &llsd->llsd_rb_root.rb_node;
677 struct rb_node *parent = NULL;
678 struct lfsck_rbtree_node *tmp;
681 while (*pos != NULL) {
683 tmp = rb_entry(parent, struct lfsck_rbtree_node, lrn_node);
684 rc = lfsck_rbtree_cmp(tmp, lrn->lrn_seq, lrn->lrn_first_oid);
686 pos = &(*pos)->rb_left;
688 pos = &(*pos)->rb_right;
693 rb_link_node(&lrn->lrn_node, parent, pos);
694 rb_insert_color(&lrn->lrn_node, &llsd->llsd_rb_root);
699 static const struct dt_index_operations lfsck_orphan_index_ops;
701 static int lfsck_rbtree_setup(const struct lu_env *env,
702 struct lfsck_component *com)
704 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
705 struct lfsck_instance *lfsck = com->lc_lfsck;
706 struct dt_device *dev = lfsck->li_bottom;
707 struct lfsck_layout_slave_data *llsd = com->lc_data;
708 struct dt_object *obj;
710 fid->f_seq = FID_SEQ_LAYOUT_RBTREE;
711 fid->f_oid = lfsck_dev_idx(lfsck);
713 obj = dt_locate(env, dev, fid);
715 RETURN(PTR_ERR(obj));
717 /* Generate an in-RAM object to stand for the layout rbtree.
718 * Scanning the layout rbtree will be via the iteration over
719 * the object. In the future, the rbtree may be written onto
720 * disk with the object.
722 * Mark the object to be as exist. */
723 obj->do_lu.lo_header->loh_attr |= LOHA_EXISTS;
724 obj->do_index_ops = &lfsck_orphan_index_ops;
725 llsd->llsd_rb_obj = obj;
726 llsd->llsd_rbtree_valid = 1;
727 dev->dd_record_fid_accessed = 1;
729 CDEBUG(D_LFSCK, "%s: layout LFSCK init OST-objects accessing bitmap\n",
730 lfsck_lfsck2name(lfsck));
735 static void lfsck_rbtree_cleanup(const struct lu_env *env,
736 struct lfsck_component *com)
738 struct lfsck_instance *lfsck = com->lc_lfsck;
739 struct lfsck_layout_slave_data *llsd = com->lc_data;
740 struct rb_node *node = rb_first(&llsd->llsd_rb_root);
741 struct rb_node *next;
742 struct lfsck_rbtree_node *lrn;
744 lfsck->li_bottom->dd_record_fid_accessed = 0;
745 /* Invalid the rbtree, then no others will use it. */
746 down_write(&llsd->llsd_rb_rwsem);
747 llsd->llsd_rbtree_valid = 0;
748 up_write(&llsd->llsd_rb_rwsem);
750 while (node != NULL) {
751 next = rb_next(node);
752 lrn = rb_entry(node, struct lfsck_rbtree_node, lrn_node);
753 rb_erase(node, &llsd->llsd_rb_root);
754 lfsck_rbtree_free(lrn);
758 if (llsd->llsd_rb_obj != NULL) {
759 lfsck_object_put(env, llsd->llsd_rb_obj);
760 llsd->llsd_rb_obj = NULL;
763 CDEBUG(D_LFSCK, "%s: layout LFSCK fini OST-objects accessing bitmap\n",
764 lfsck_lfsck2name(lfsck));
767 static void lfsck_rbtree_update_bitmap(const struct lu_env *env,
768 struct lfsck_component *com,
769 const struct lu_fid *fid,
772 struct lfsck_layout_slave_data *llsd = com->lc_data;
773 struct lfsck_rbtree_node *lrn;
779 if (unlikely(!fid_is_sane(fid) || fid_is_last_id(fid)))
782 if (!fid_is_idif(fid) && !fid_is_norm(fid))
785 down_read(&llsd->llsd_rb_rwsem);
786 if (!llsd->llsd_rbtree_valid)
787 GOTO(unlock, rc = 0);
789 lrn = lfsck_rbtree_search(llsd, fid, NULL);
791 struct lfsck_rbtree_node *tmp;
795 up_read(&llsd->llsd_rb_rwsem);
796 tmp = lfsck_rbtree_new(env, fid);
798 GOTO(out, rc = PTR_ERR(tmp));
801 down_write(&llsd->llsd_rb_rwsem);
802 if (!llsd->llsd_rbtree_valid) {
803 lfsck_rbtree_free(tmp);
804 GOTO(unlock, rc = 0);
807 lrn = lfsck_rbtree_insert(llsd, tmp);
809 lfsck_rbtree_free(tmp);
812 idx = fid_oid(fid) & LFSCK_RBTREE_BITMAP_MASK;
813 /* Any accessed object must be a known object. */
814 if (!test_and_set_bit(idx, lrn->lrn_known_bitmap))
815 atomic_inc(&lrn->lrn_known_count);
816 if (accessed && !test_and_set_bit(idx, lrn->lrn_accessed_bitmap))
817 atomic_inc(&lrn->lrn_accessed_count);
819 GOTO(unlock, rc = 0);
823 up_write(&llsd->llsd_rb_rwsem);
825 up_read(&llsd->llsd_rb_rwsem);
827 if (rc != 0 && accessed) {
828 struct lfsck_layout *lo = com->lc_file_ram;
830 CDEBUG(D_LFSCK, "%s: fail to update OST-objects accessing "
831 "bitmap, and will cause incorrect LFSCK OST-object "
832 "handling, so disable it to cancel orphan handling "
833 "for related device. rc = %d\n",
834 lfsck_lfsck2name(com->lc_lfsck), rc);
836 lo->ll_flags |= LF_INCOMPLETE;
837 lfsck_rbtree_cleanup(env, com);
841 static inline void lldk_le_to_cpu(struct lfsck_layout_dangling_key *des,
842 const struct lfsck_layout_dangling_key *src)
844 fid_le_to_cpu(&des->lldk_fid, &src->lldk_fid);
845 des->lldk_comp_id = le32_to_cpu(src->lldk_comp_id);
846 des->lldk_ea_off = le32_to_cpu(src->lldk_ea_off);
849 static inline void lldk_cpu_to_le(struct lfsck_layout_dangling_key *des,
850 const struct lfsck_layout_dangling_key *src)
852 fid_cpu_to_le(&des->lldk_fid, &src->lldk_fid);
853 des->lldk_comp_id = cpu_to_le32(src->lldk_comp_id);
854 des->lldk_ea_off = cpu_to_le32(src->lldk_ea_off);
857 static inline void lldk_be_to_cpu(struct lfsck_layout_dangling_key *des,
858 const struct lfsck_layout_dangling_key *src)
860 fid_be_to_cpu(&des->lldk_fid, &src->lldk_fid);
861 des->lldk_comp_id = be32_to_cpu(src->lldk_comp_id);
862 des->lldk_ea_off = be32_to_cpu(src->lldk_ea_off);
865 static inline void lldk_cpu_to_be(struct lfsck_layout_dangling_key *des,
866 const struct lfsck_layout_dangling_key *src)
868 fid_cpu_to_be(&des->lldk_fid, &src->lldk_fid);
869 des->lldk_comp_id = cpu_to_be32(src->lldk_comp_id);
870 des->lldk_ea_off = cpu_to_be32(src->lldk_ea_off);
873 static void lfsck_layout_le_to_cpu(struct lfsck_layout *des,
874 const struct lfsck_layout *src)
878 des->ll_magic = le32_to_cpu(src->ll_magic);
879 des->ll_status = le32_to_cpu(src->ll_status);
880 des->ll_flags = le32_to_cpu(src->ll_flags);
881 des->ll_success_count = le32_to_cpu(src->ll_success_count);
882 des->ll_run_time_phase1 = le64_to_cpu(src->ll_run_time_phase1);
883 des->ll_run_time_phase2 = le64_to_cpu(src->ll_run_time_phase2);
884 des->ll_time_last_complete = le64_to_cpu(src->ll_time_last_complete);
885 des->ll_time_latest_start = le64_to_cpu(src->ll_time_latest_start);
886 des->ll_time_last_checkpoint =
887 le64_to_cpu(src->ll_time_last_checkpoint);
888 des->ll_pos_latest_start = le64_to_cpu(src->ll_pos_latest_start);
889 des->ll_pos_last_checkpoint = le64_to_cpu(src->ll_pos_last_checkpoint);
890 des->ll_pos_first_inconsistent =
891 le64_to_cpu(src->ll_pos_first_inconsistent);
892 des->ll_objs_checked_phase1 = le64_to_cpu(src->ll_objs_checked_phase1);
893 des->ll_objs_failed_phase1 = le64_to_cpu(src->ll_objs_failed_phase1);
894 des->ll_objs_checked_phase2 = le64_to_cpu(src->ll_objs_checked_phase2);
895 des->ll_objs_failed_phase2 = le64_to_cpu(src->ll_objs_failed_phase2);
896 for (i = 0; i < LLIT_MAX; i++)
897 des->ll_objs_repaired[i] =
898 le64_to_cpu(src->ll_objs_repaired[i]);
899 des->ll_objs_skipped = le64_to_cpu(src->ll_objs_skipped);
900 des->ll_bitmap_size = le32_to_cpu(src->ll_bitmap_size);
901 lldk_le_to_cpu(&des->ll_lldk_latest_scanned_phase2,
902 &src->ll_lldk_latest_scanned_phase2);
905 static void lfsck_layout_cpu_to_le(struct lfsck_layout *des,
906 const struct lfsck_layout *src)
910 des->ll_magic = cpu_to_le32(src->ll_magic);
911 des->ll_status = cpu_to_le32(src->ll_status);
912 des->ll_flags = cpu_to_le32(src->ll_flags);
913 des->ll_success_count = cpu_to_le32(src->ll_success_count);
914 des->ll_run_time_phase1 = cpu_to_le64(src->ll_run_time_phase1);
915 des->ll_run_time_phase2 = cpu_to_le64(src->ll_run_time_phase2);
916 des->ll_time_last_complete = cpu_to_le64(src->ll_time_last_complete);
917 des->ll_time_latest_start = cpu_to_le64(src->ll_time_latest_start);
918 des->ll_time_last_checkpoint =
919 cpu_to_le64(src->ll_time_last_checkpoint);
920 des->ll_pos_latest_start = cpu_to_le64(src->ll_pos_latest_start);
921 des->ll_pos_last_checkpoint = cpu_to_le64(src->ll_pos_last_checkpoint);
922 des->ll_pos_first_inconsistent =
923 cpu_to_le64(src->ll_pos_first_inconsistent);
924 des->ll_objs_checked_phase1 = cpu_to_le64(src->ll_objs_checked_phase1);
925 des->ll_objs_failed_phase1 = cpu_to_le64(src->ll_objs_failed_phase1);
926 des->ll_objs_checked_phase2 = cpu_to_le64(src->ll_objs_checked_phase2);
927 des->ll_objs_failed_phase2 = cpu_to_le64(src->ll_objs_failed_phase2);
928 for (i = 0; i < LLIT_MAX; i++)
929 des->ll_objs_repaired[i] =
930 cpu_to_le64(src->ll_objs_repaired[i]);
931 des->ll_objs_skipped = cpu_to_le64(src->ll_objs_skipped);
932 des->ll_bitmap_size = cpu_to_le32(src->ll_bitmap_size);
933 lldk_cpu_to_le(&des->ll_lldk_latest_scanned_phase2,
934 &src->ll_lldk_latest_scanned_phase2);
938 * Load the OST bitmap from the lfsck_layout trace file.
940 * \param[in] env pointer to the thread context
941 * \param[in] com pointer to the lfsck component
943 * \retval 0 for success
944 * \retval negative error number on failure or data corruption
946 static int lfsck_layout_load_bitmap(const struct lu_env *env,
947 struct lfsck_component *com)
949 struct dt_object *obj = com->lc_obj;
950 struct lfsck_assistant_data *lad = com->lc_data;
951 struct lfsck_layout *lo = com->lc_file_ram;
952 unsigned long *bitmap = lad->lad_bitmap;
953 loff_t pos = com->lc_file_size;
959 if (com->lc_lfsck->li_ost_descs.ltd_tgts_mask_len > lo->ll_bitmap_size)
960 nbits = com->lc_lfsck->li_ost_descs.ltd_tgts_mask_len;
962 nbits = lo->ll_bitmap_size;
964 if (unlikely(nbits < BITS_PER_LONG))
965 nbits = BITS_PER_LONG;
967 if (nbits > lad->lad_bitmap_count) {
968 u32 new_bits = lad->lad_bitmap_count;
969 unsigned long *new_bitmap;
971 while (new_bits < nbits)
974 new_bitmap = bitmap_zalloc(new_bits, GFP_KERNEL);
975 if (new_bitmap == NULL)
978 lad->lad_bitmap = new_bitmap;
979 lad->lad_bitmap_count = new_bits;
984 if (lo->ll_bitmap_size == 0) {
985 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
986 bitmap_zero(bitmap, lad->lad_bitmap_count);
990 size = (lo->ll_bitmap_size + 7) >> 3;
991 rc = dt_read(env, obj, lfsck_buf_get(env, bitmap, size), &pos);
993 RETURN(rc >= 0 ? -EINVAL : rc);
995 if (bitmap_empty(bitmap, lad->lad_bitmap_count))
996 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
998 set_bit(LAD_INCOMPLETE, &lad->lad_flags);
1004 * Load the layout LFSCK trace file from disk.
1006 * The layout LFSCK trace file records the layout LFSCK status information
1007 * and other statistics, such as how many objects have been scanned, and how
1008 * many objects have been repaired, and etc. It also contains the bitmap for
1009 * failed OSTs during the layout LFSCK. All these information will be loaded
1010 * from disk to RAM when the layout LFSCK component setup.
1012 * \param[in] env pointer to the thread context
1013 * \param[in] com pointer to the lfsck component
1015 * \retval positive number for file data corruption, the caller
1016 * should reset the layout LFSCK trace file
1017 * \retval 0 for success
1018 * \retval negative error number on failure
1020 static int lfsck_layout_load(const struct lu_env *env,
1021 struct lfsck_component *com)
1023 struct lfsck_layout *lo = com->lc_file_ram;
1024 ssize_t size = com->lc_file_size;
1028 rc = dt_read(env, com->lc_obj,
1029 lfsck_buf_get(env, com->lc_file_disk, size), &pos);
1032 } else if (rc < 0) {
1033 CDEBUG(D_LFSCK, "%s: failed to load lfsck_layout: rc = %d\n",
1034 lfsck_lfsck2name(com->lc_lfsck), rc);
1036 } else if (rc != size) {
1037 CDEBUG(D_LFSCK, "%s: lfsck_layout size %u != %u; reset it\n",
1038 lfsck_lfsck2name(com->lc_lfsck), rc, (unsigned int)size);
1042 lfsck_layout_le_to_cpu(lo, com->lc_file_disk);
1043 if (lo->ll_magic != LFSCK_LAYOUT_MAGIC) {
1044 CDEBUG(D_LFSCK, "%s: invalid lfsck_layout magic %#x != %#x, "
1045 "to be reset\n", lfsck_lfsck2name(com->lc_lfsck),
1046 lo->ll_magic, LFSCK_LAYOUT_MAGIC);
1054 * Store the layout LFSCK trace file on disk.
1056 * The layout LFSCK trace file records the layout LFSCK status information
1057 * and other statistics, such as how many objects have been scanned, and how
1058 * many objects have been repaired, and etc. It also contains the bitmap for
1059 * failed OSTs during the layout LFSCK. All these information will be synced
1060 * from RAM to disk periodically.
1062 * \param[in] env pointer to the thread context
1063 * \param[in] com pointer to the lfsck component
1065 * \retval 0 for success
1066 * \retval negative error number on failure
1068 static int lfsck_layout_store(const struct lu_env *env,
1069 struct lfsck_component *com)
1071 struct dt_object *obj = com->lc_obj;
1072 struct lfsck_instance *lfsck = com->lc_lfsck;
1073 struct lfsck_layout *lo_ram = com->lc_file_ram;
1074 struct lfsck_layout *lo = com->lc_file_disk;
1076 struct dt_device *dev = lfsck_obj2dev(obj);
1077 unsigned long *bitmap = NULL;
1079 ssize_t size = com->lc_file_size;
1084 if (lfsck->li_master) {
1085 struct lfsck_assistant_data *lad = com->lc_data;
1087 bitmap = lad->lad_bitmap;
1088 nbits = lad->lad_bitmap_count;
1091 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
1094 lo_ram->ll_bitmap_size = nbits;
1095 lfsck_layout_cpu_to_le(lo, lo_ram);
1096 th = dt_trans_create(env, dev);
1098 GOTO(log, rc = PTR_ERR(th));
1100 rc = dt_declare_record_write(env, obj, lfsck_buf_get(env, lo, size),
1105 if (bitmap != NULL) {
1106 rc = dt_declare_record_write(env, obj,
1107 lfsck_buf_get(env, bitmap, nbits >> 3),
1113 rc = dt_trans_start_local(env, dev, th);
1118 rc = dt_record_write(env, obj, lfsck_buf_get(env, lo, size), &pos, th);
1122 if (bitmap != NULL) {
1124 rc = dt_record_write(env, obj,
1125 lfsck_buf_get(env, bitmap, nbits >> 3),
1132 dt_trans_stop(env, dev, th);
1136 CDEBUG(D_LFSCK, "%s: fail to store lfsck_layout: rc = %d\n",
1137 lfsck_lfsck2name(lfsck), rc);
1142 static int lfsck_layout_init(const struct lu_env *env,
1143 struct lfsck_component *com)
1145 struct lfsck_layout *lo = com->lc_file_ram;
1148 memset(lo, 0, com->lc_file_size);
1149 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
1150 lo->ll_status = LS_INIT;
1151 down_write(&com->lc_sem);
1152 rc = lfsck_layout_store(env, com);
1153 if (rc == 0 && com->lc_lfsck->li_master)
1154 rc = lfsck_load_sub_trace_files(env, com,
1155 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
1156 up_write(&com->lc_sem);
1161 static int fid_is_for_ostobj(const struct lu_env *env,
1162 struct lfsck_instance *lfsck,
1163 struct dt_object *obj, const struct lu_fid *fid)
1165 struct seq_server_site *ss = lfsck_dev_site(lfsck);
1166 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
1167 struct lustre_ost_attrs *loa;
1170 fld_range_set_any(range);
1171 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1173 if (fld_range_is_ost(range))
1179 loa = &lfsck_env_info(env)->lti_loa;
1180 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, loa, sizeof(*loa)),
1182 if (rc >= (int)sizeof(struct lustre_mdt_attrs)) {
1183 lustre_lma_swab(&loa->loa_lma);
1185 return loa->loa_lma.lma_compat & LMAC_FID_ON_OST ? 1 : 0;
1188 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_FID);
1193 static struct lfsck_layout_seq *
1194 lfsck_layout_seq_lookup(struct lfsck_layout_slave_data *llsd, __u64 seq)
1196 struct lfsck_layout_seq *lls;
1198 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1199 if (lls->lls_seq == seq)
1202 if (lls->lls_seq > seq)
1210 lfsck_layout_seq_insert(struct lfsck_layout_slave_data *llsd,
1211 struct lfsck_layout_seq *lls)
1213 struct lfsck_layout_seq *tmp;
1214 struct list_head *pos = &llsd->llsd_seq_list;
1216 list_for_each_entry(tmp, &llsd->llsd_seq_list, lls_list) {
1217 if (lls->lls_seq < tmp->lls_seq) {
1218 pos = &tmp->lls_list;
1222 list_add_tail(&lls->lls_list, pos);
1226 lfsck_layout_lastid_create(const struct lu_env *env,
1227 struct lfsck_instance *lfsck,
1228 struct dt_object *obj)
1230 struct lfsck_thread_info *info = lfsck_env_info(env);
1231 struct lu_attr *la = &info->lti_la;
1232 struct dt_object_format *dof = &info->lti_dof;
1233 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1234 struct dt_device *dt = lfsck_obj2dev(obj);
1241 if (bk->lb_param & LPF_DRYRUN)
1244 memset(la, 0, sizeof(*la));
1245 la->la_mode = S_IFREG | S_IRUGO | S_IWUSR;
1246 la->la_valid = LA_MODE | LA_UID | LA_GID;
1247 memset(dof, 0, sizeof(*dof));
1248 dof->dof_type = dt_mode_to_dft(S_IFREG);
1250 th = lfsck_trans_create(env, dt, lfsck);
1252 GOTO(log, rc = PTR_ERR(th));
1254 rc = dt_declare_create(env, obj, la, NULL, dof, th);
1258 rc = dt_declare_record_write(env, obj,
1259 lfsck_buf_get(env, &lastid,
1265 rc = dt_trans_start_local(env, dt, th);
1269 dt_write_lock(env, obj, 0);
1270 if (likely(dt_object_exists(obj) == 0)) {
1271 rc = dt_create(env, obj, la, NULL, dof, th);
1273 rc = dt_record_write(env, obj,
1274 lfsck_buf_get(env, &lastid, sizeof(lastid)),
1277 dt_write_unlock(env, obj);
1282 dt_trans_stop(env, dt, th);
1285 CDEBUG(D_LFSCK, "%s: layout LFSCK will create LAST_ID for <seq> "
1287 lfsck_lfsck2name(lfsck), fid_seq(lfsck_dto2fid(obj)), rc);
1293 lfsck_layout_lastid_reload(const struct lu_env *env,
1294 struct lfsck_component *com,
1295 struct lfsck_layout_seq *lls)
1301 dt_read_lock(env, lls->lls_lastid_obj, 0);
1302 rc = dt_record_read(env, lls->lls_lastid_obj,
1303 lfsck_buf_get(env, &lastid, sizeof(lastid)), &pos);
1304 dt_read_unlock(env, lls->lls_lastid_obj);
1305 if (unlikely(rc != 0))
1308 lastid = le64_to_cpu(lastid);
1309 if (lastid < lls->lls_lastid_known) {
1310 struct lfsck_instance *lfsck = com->lc_lfsck;
1311 struct lfsck_layout *lo = com->lc_file_ram;
1313 lls->lls_lastid = lls->lls_lastid_known;
1315 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1316 LASSERT(lfsck->li_out_notify != NULL);
1318 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1319 LE_LASTID_REBUILDING);
1320 lo->ll_flags |= LF_CRASHED_LASTID;
1322 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
1323 "LAST_ID file (1) for the sequence %#llx"
1324 ", old value %llu, known value %llu\n",
1325 lfsck_lfsck2name(lfsck), lls->lls_seq,
1326 lastid, lls->lls_lastid);
1328 } else if (lastid >= lls->lls_lastid) {
1329 lls->lls_lastid = lastid;
1337 lfsck_layout_lastid_store(const struct lu_env *env,
1338 struct lfsck_component *com)
1340 struct lfsck_instance *lfsck = com->lc_lfsck;
1341 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1342 struct dt_device *dt = lfsck->li_bottom;
1343 struct lfsck_layout_slave_data *llsd = com->lc_data;
1344 struct lfsck_layout_seq *lls;
1350 list_for_each_entry(lls, &llsd->llsd_seq_list, lls_list) {
1353 if (!lls->lls_dirty)
1356 CDEBUG(D_LFSCK, "%s: layout LFSCK will sync the LAST_ID for "
1357 "<seq> %#llx as <oid> %llu\n",
1358 lfsck_lfsck2name(lfsck), lls->lls_seq, lls->lls_lastid);
1360 if (bk->lb_param & LPF_DRYRUN) {
1365 th = lfsck_trans_create(env, dt, lfsck);
1368 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1369 "the LAST_ID for <seq> %#llx(1): rc = %d\n",
1370 lfsck_lfsck2name(com->lc_lfsck),
1375 lastid = cpu_to_le64(lls->lls_lastid);
1376 rc = dt_declare_record_write(env, lls->lls_lastid_obj,
1377 lfsck_buf_get(env, &lastid,
1383 rc = dt_trans_start_local(env, dt, th);
1387 dt_write_lock(env, lls->lls_lastid_obj, 0);
1388 rc = dt_record_write(env, lls->lls_lastid_obj,
1389 lfsck_buf_get(env, &lastid,
1390 sizeof(lastid)), &pos, th);
1391 dt_write_unlock(env, lls->lls_lastid_obj);
1396 dt_trans_stop(env, dt, th);
1399 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to store "
1400 "the LAST_ID for <seq> %#llx(2): rc = %d\n",
1401 lfsck_lfsck2name(com->lc_lfsck),
1410 lfsck_layout_lastid_load(const struct lu_env *env,
1411 struct lfsck_component *com,
1412 struct lfsck_layout_seq *lls)
1414 struct lfsck_instance *lfsck = com->lc_lfsck;
1415 struct lfsck_layout *lo = com->lc_file_ram;
1416 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
1417 struct dt_object *obj;
1422 lu_last_id_fid(fid, lls->lls_seq, lfsck_dev_idx(lfsck));
1423 obj = dt_locate(env, lfsck->li_bottom, fid);
1425 RETURN(PTR_ERR(obj));
1427 /* LAST_ID crashed, to be rebuilt */
1428 if (dt_object_exists(obj) == 0) {
1429 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
1430 LASSERT(lfsck->li_out_notify != NULL);
1432 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1433 LE_LASTID_REBUILDING);
1434 lo->ll_flags |= LF_CRASHED_LASTID;
1436 CDEBUG(D_LFSCK, "%s: layout LFSCK cannot find the "
1437 "LAST_ID file for sequence %#llx\n",
1438 lfsck_lfsck2name(lfsck), lls->lls_seq);
1440 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY4) &&
1442 struct ptlrpc_thread *thread =
1445 up_write(&com->lc_sem);
1446 wait_event_idle_timeout(
1447 thread->t_ctl_waitq,
1448 !thread_is_running(thread),
1449 cfs_time_seconds(cfs_fail_val));
1450 down_write(&com->lc_sem);
1454 rc = lfsck_layout_lastid_create(env, lfsck, obj);
1456 dt_read_lock(env, obj, 0);
1457 rc = dt_read(env, obj,
1458 lfsck_buf_get(env, &lls->lls_lastid, sizeof(__u64)),
1460 dt_read_unlock(env, obj);
1461 if (rc != 0 && rc != sizeof(__u64))
1462 GOTO(out, rc = (rc > 0 ? -EFAULT : rc));
1464 if (rc == 0 && !(lo->ll_flags & LF_CRASHED_LASTID)) {
1465 LASSERT(lfsck->li_out_notify != NULL);
1467 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
1468 LE_LASTID_REBUILDING);
1469 lo->ll_flags |= LF_CRASHED_LASTID;
1471 CDEBUG(D_LFSCK, "%s: layout LFSCK finds invalid "
1472 "LAST_ID file for the sequence %#llx"
1474 lfsck_lfsck2name(lfsck), lls->lls_seq, rc);
1477 lls->lls_lastid = le64_to_cpu(lls->lls_lastid);
1485 lfsck_object_put(env, obj);
1487 lls->lls_lastid_obj = obj;
1492 static void lfsck_layout_record_failure(const struct lu_env *env,
1493 struct lfsck_instance *lfsck,
1494 struct lfsck_layout *lo)
1498 lo->ll_objs_failed_phase1++;
1499 cookie = lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
1501 if (lo->ll_pos_first_inconsistent == 0 ||
1502 lo->ll_pos_first_inconsistent < cookie) {
1503 lo->ll_pos_first_inconsistent = cookie;
1505 CDEBUG(D_LFSCK, "%s: layout LFSCK hit first non-repaired "
1506 "inconsistency at the pos [%llu]\n",
1507 lfsck_lfsck2name(lfsck),
1508 lo->ll_pos_first_inconsistent);
1512 static int lfsck_layout_double_scan_result(const struct lu_env *env,
1513 struct lfsck_component *com,
1516 struct lfsck_instance *lfsck = com->lc_lfsck;
1517 struct lfsck_layout *lo = com->lc_file_ram;
1519 CDEBUG(D_LFSCK, "%s: layout LFSCK double scan: rc = %d\n",
1520 lfsck_lfsck2name(lfsck), rc);
1522 down_write(&com->lc_sem);
1523 lo->ll_run_time_phase2 += ktime_get_seconds() -
1524 com->lc_time_last_checkpoint;
1525 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
1526 lo->ll_objs_checked_phase2 += com->lc_new_checked;
1529 if (lo->ll_flags & LF_INCOMPLETE) {
1530 lo->ll_status = LS_PARTIAL;
1532 if (lfsck->li_master) {
1533 struct lfsck_assistant_data *lad = com->lc_data;
1535 if (test_bit(LAD_INCOMPLETE, &lad->lad_flags))
1536 lo->ll_status = LS_PARTIAL;
1538 lo->ll_status = LS_COMPLETED;
1540 lo->ll_status = LS_COMPLETED;
1543 lo->ll_flags &= ~LF_SCANNED_ONCE;
1544 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
1545 lo->ll_flags &= ~LF_INCONSISTENT;
1546 lo->ll_time_last_complete = lo->ll_time_last_checkpoint;
1547 lo->ll_success_count++;
1548 } else if (rc == 0) {
1549 if (lfsck->li_status != 0)
1550 lo->ll_status = lfsck->li_status;
1552 lo->ll_status = LS_STOPPED;
1554 lo->ll_status = LS_FAILED;
1557 rc = lfsck_layout_store(env, com);
1558 up_write(&com->lc_sem);
1560 CDEBUG(D_LFSCK, "%s: layout LFSCK double scan result %u: rc = %d\n",
1561 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
1566 static int lfsck_layout_trans_stop(const struct lu_env *env,
1567 struct dt_device *dev,
1568 struct thandle *handle, int result)
1572 /* XXX: If there is something worng or it needs to repair nothing,
1573 * then notify the lower to stop the modification. Currently,
1574 * we use th_result for such purpose, that may be replaced by
1575 * some rollback mechanism in the future. */
1576 handle->th_result = result;
1577 rc = dt_trans_stop(env, dev, handle);
1579 return result > 0 ? 0 : result;
1581 return rc == 0 ? 1 : rc;
1584 static int lfsck_layout_ins_dangling_rec(const struct lu_env *env,
1585 struct lfsck_component *com,
1586 const struct lu_fid *pfid,
1587 const struct lu_fid *cfid,
1588 __u32 comp_id, __u32 ea_off,
1591 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1592 struct lu_fid *rec = &lfsck_env_info(env)->lti_fid3;
1593 struct dt_device *dev;
1594 struct dt_object *obj;
1595 struct thandle *th = NULL;
1600 idx = lfsck_sub_trace_file_fid2idx(pfid);
1601 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1602 dev = lfsck_obj2dev(obj);
1604 fid_cpu_to_be(&key->lldk_fid, pfid);
1605 key->lldk_comp_id = cpu_to_be32(comp_id);
1606 key->lldk_ea_off = cpu_to_be32(ea_off);
1608 fid_cpu_to_be(rec, cfid);
1609 rec->f_ver = cpu_to_be32(ost_idx);
1611 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1613 th = lfsck_trans_create(env, dev, com->lc_lfsck);
1615 GOTO(unlock, rc = PTR_ERR(th));
1617 rc = dt_declare_insert(env, obj,
1618 (const struct dt_rec *)rec,
1619 (const struct dt_key *)key, th);
1623 rc = dt_trans_start_local(env, dev, th);
1627 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1628 (const struct dt_key *)key, th);
1633 if (th && !IS_ERR(th))
1634 dt_trans_stop(env, dev, th);
1636 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1638 CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", comp_id = %u, "
1639 "ea_off = %u, ost_idx = %u, into the trace file for further "
1640 "dangling check: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1641 PFID(pfid), PFID(cfid), comp_id, ea_off, ost_idx, rc);
1646 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1647 struct lfsck_component *com,
1648 const struct lu_fid *fid,
1649 __u32 comp_id, __u32 ea_off)
1651 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1652 struct dt_device *dev;
1653 struct dt_object *obj;
1654 struct thandle *th = NULL;
1659 idx = lfsck_sub_trace_file_fid2idx(fid);
1660 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1661 dev = lfsck_obj2dev(obj);
1663 fid_cpu_to_be(&key->lldk_fid, fid);
1664 key->lldk_comp_id = cpu_to_be32(comp_id);
1665 key->lldk_ea_off = cpu_to_be32(ea_off);
1667 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1669 th = lfsck_trans_create(env, dev, com->lc_lfsck);
1671 GOTO(unlock, rc = PTR_ERR(th));
1673 rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1677 rc = dt_trans_start_local(env, dev, th);
1681 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1686 if (th && !IS_ERR(th))
1687 dt_trans_stop(env, dev, th);
1689 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1691 CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1692 ", comp_id = %u, ea_off = %u from the trace file: rc = %d\n",
1693 lfsck_lfsck2name(com->lc_lfsck), PFID(fid), comp_id, ea_off, rc);
1699 * Get the system default stripe size.
1701 * \param[in] env pointer to the thread context
1702 * \param[in] lfsck pointer to the lfsck instance
1703 * \param[out] size pointer to the default stripe size
1705 * \retval 0 for success
1706 * \retval negative error number on failure
1708 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1709 struct lfsck_instance *lfsck,
1712 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1713 struct dt_object *root;
1716 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1718 return PTR_ERR(root);
1720 /* Get the default stripe size via xattr_get on the backend root. */
1721 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1724 /* The lum->lmm_stripe_size is LE mode. The *size also
1725 * should be LE mode. So it is unnecessary to convert. */
1726 *size = lum->lmm_stripe_size;
1728 } else if (unlikely(rc == 0)) {
1732 lfsck_object_put(env, root);
1738 * \retval +1: repaired
1739 * \retval 0: did nothing
1740 * \retval -ve: on error
1742 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1743 struct lfsck_instance *lfsck,
1744 struct thandle *handle,
1745 struct dt_object *parent,
1746 const struct lu_fid *cfid,
1748 struct lov_mds_md_v1 *lmm,
1749 struct lov_ost_data_v1 *slot,
1750 int fl, __u32 ost_idx, int size)
1752 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1753 struct lu_buf ea_buf;
1760 magic = le32_to_cpu(lmm->lmm_magic);
1761 pattern = le32_to_cpu(lmm->lmm_pattern);
1762 count = le16_to_cpu(lmm->lmm_stripe_count);
1764 fid_to_ostid(cfid, oi);
1765 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1766 slot->l_ost_gen = cpu_to_le32(0);
1767 slot->l_ost_idx = cpu_to_le32(ost_idx);
1769 if (pattern & LOV_PATTERN_F_HOLE) {
1770 struct lov_ost_data_v1 *objs;
1773 if (magic == LOV_MAGIC_V1)
1774 objs = &lmm->lmm_objects[0];
1776 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1777 for (i = 0; i < count; i++, objs++) {
1778 if (lovea_slot_is_dummy(objs))
1782 /* If the @slot is the last dummy slot to be refilled,
1783 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1786 cpu_to_le32(pattern & ~LOV_PATTERN_F_HOLE);
1788 CDEBUG(D_LFSCK, "%s: remove layout HOLE for "DFID
1789 ": parent "DFID"\n", lfsck_lfsck2name(lfsck),
1790 PFID(cfid), PFID(lfsck_dto2fid(parent)));
1794 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
1795 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1802 static struct lov_ost_data_v1 *
1803 __lfsck_layout_new_v1_lovea(struct lov_mds_md_v1 *lmm,
1804 const struct lu_fid *pfid,
1805 __u32 stripe_size, __u32 ea_off,
1806 __u32 pattern, __u16 count)
1808 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1809 lmm->lmm_pattern = cpu_to_le32(pattern);
1810 fid_to_lmm_oi(pfid, &lmm->lmm_oi);
1811 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1812 lmm->lmm_stripe_size = cpu_to_le32(stripe_size);
1813 lmm->lmm_stripe_count = cpu_to_le16(count);
1814 lmm->lmm_layout_gen = cpu_to_le16(1);
1815 memset(&lmm->lmm_objects[0], 0,
1816 sizeof(struct lov_ost_data_v1) * count);
1818 return &lmm->lmm_objects[ea_off];
1821 static int lfsck_layout_new_v1_lovea(const struct lu_env *env,
1822 struct lfsck_instance *lfsck,
1823 struct ost_layout *ol,
1824 struct dt_object *parent,
1825 struct lu_buf *buf, __u32 ea_off,
1826 struct lov_mds_md_v1 **lmm,
1827 struct lov_ost_data_v1 **objs)
1830 __u32 stripe_size = ol->ol_stripe_size;
1831 __u32 pattern = LOV_PATTERN_RAID0;
1834 if (ol->ol_stripe_count != 0)
1835 count = ol->ol_stripe_count;
1839 size = lov_mds_md_size(count, LOV_MAGIC_V1);
1840 LASSERTF(buf->lb_len >= size,
1841 "buffer len %d is less than real size %d\n",
1842 (int)buf->lb_len, size);
1844 if (stripe_size == 0) {
1847 rc = lfsck_layout_get_def_stripesize(env, lfsck, &stripe_size);
1853 if (ol->ol_stripe_count > 1 ||
1854 (ol->ol_stripe_count == 0 && ea_off != 0)) {
1855 pattern |= LOV_PATTERN_F_HOLE;
1856 memset(&(*lmm)->lmm_objects[0], 0,
1857 count * sizeof(struct lov_ost_data_v1));
1860 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1861 stripe_size, ea_off, pattern, count);
1866 static int lfsck_layout_new_comp_lovea(const struct lu_env *env,
1867 struct lu_orphan_rec_v3 *rec,
1868 struct dt_object *parent,
1869 struct lu_buf *buf, __u32 ea_off,
1870 struct lov_mds_md_v1 **lmm,
1871 struct lov_ost_data_v1 **objs)
1873 struct ost_layout *ol = &rec->lor_layout;
1874 struct lov_comp_md_v1 *lcm;
1875 struct lov_comp_md_entry_v1 *lcme;
1876 __u32 pattern = LOV_PATTERN_RAID0;
1877 __u32 offset = sizeof(*lcm) + sizeof(*lcme);
1878 int lcme_size = lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1879 int size = offset + lcme_size;
1881 LASSERTF(buf->lb_len >= size,
1882 "buffer len %d is less than real size %d\n",
1883 (int)buf->lb_len, size);
1886 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1887 lcm->lcm_size = cpu_to_le32(size);
1888 if (rec->lor_range) {
1889 lcm->lcm_layout_gen = cpu_to_le32(rec->lor_layout_version +
1891 lcm->lcm_flags = cpu_to_le16(LCM_FL_WRITE_PENDING);
1892 } else if (rec->lor_layout_version) {
1893 lcm->lcm_layout_gen = cpu_to_le32(rec->lor_layout_version +
1895 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
1897 lcm->lcm_layout_gen = cpu_to_le32(1);
1898 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
1900 lcm->lcm_entry_count = cpu_to_le16(1);
1901 /* Currently, we do not know how many mirrors will be, set it as zero
1902 * at the beginning. It will be updated when more mirrors are found. */
1903 lcm->lcm_mirror_count = 0;
1905 lcme = &lcm->lcm_entries[0];
1906 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1907 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1908 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1909 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1910 lcme->lcme_offset = cpu_to_le32(offset);
1911 lcme->lcme_size = cpu_to_le32(lcme_size);
1912 lcme->lcme_layout_gen = lcm->lcm_layout_gen;
1913 if (ol->ol_stripe_count > 1)
1914 pattern |= LOV_PATTERN_F_HOLE;
1916 *lmm = buf->lb_buf + offset;
1917 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1918 ol->ol_stripe_size, ea_off,
1919 pattern, ol->ol_stripe_count);
1924 static void lfsck_layout_update_lcm(struct lov_comp_md_v1 *lcm,
1925 struct lov_comp_md_entry_v1 *lcme,
1926 __u32 version, __u32 range)
1928 struct lov_comp_md_entry_v1 *tmp;
1929 __u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
1930 __u64 end = le64_to_cpu(lcme->lcme_extent.e_end);
1931 __u32 gen = version + range;
1934 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1935 __u16 flags = le16_to_cpu(lcm->lcm_flags);
1939 lcme->lcme_layout_gen = cpu_to_le32(gen);
1940 if (le32_to_cpu(lcm->lcm_layout_gen) < gen)
1941 lcm->lcm_layout_gen = cpu_to_le32(gen);
1944 lcm->lcm_flags = cpu_to_le16(LCM_FL_WRITE_PENDING);
1945 else if (flags == LCM_FL_NONE && le16_to_cpu(lcm->lcm_mirror_count) > 0)
1946 lcm->lcm_flags = cpu_to_le16(LCM_FL_RDONLY);
1948 for (i = 0; i < count; i++) {
1949 tmp = &lcm->lcm_entries[i];
1950 if (le64_to_cpu(tmp->lcme_extent.e_end) <= start)
1953 if (le64_to_cpu(tmp->lcme_extent.e_start) >= end)
1956 if (le32_to_cpu(tmp->lcme_flags) & LCME_FL_STALE)
1959 tmp_gen = le32_to_cpu(tmp->lcme_layout_gen);
1960 /* "lcme_layout_gen == 0" but without LCME_FL_STALE flag,
1961 * then it should be the latest version of all mirrors. */
1962 if (tmp_gen == 0 || tmp_gen > gen) {
1963 lcme->lcme_flags = cpu_to_le32(
1964 le32_to_cpu(lcme->lcme_flags) | LCME_FL_STALE);
1969 tmp->lcme_flags = cpu_to_le32(
1970 le32_to_cpu(tmp->lcme_flags) | LCME_FL_STALE);
1974 static int lfsck_layout_add_comp(const struct lu_env *env,
1975 struct lfsck_instance *lfsck,
1976 struct thandle *handle,
1977 struct lu_orphan_rec_v3 *rec,
1978 struct dt_object *parent,
1979 const struct lu_fid *cfid,
1980 struct lu_buf *buf, __u32 ost_idx,
1981 __u32 ea_off, int pos, bool new_mirror)
1983 struct ost_layout *ol = &rec->lor_layout;
1984 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1985 struct lov_comp_md_entry_v1 *lcme;
1986 struct lov_mds_md_v1 *lmm;
1987 struct lov_ost_data_v1 *objs;
1988 int added = sizeof(*lcme) +
1989 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1990 int size = le32_to_cpu(lcm->lcm_size) + added;
1994 __u32 pattern = LOV_PATTERN_RAID0;
1995 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1998 lu_buf_check_and_grow(buf, size);
1999 /* set the lcm again because lu_buf_check_and_grow() may
2000 * have reallocated the buf. */
2002 lcm->lcm_size = cpu_to_le32(size);
2003 lcm->lcm_entry_count = cpu_to_le16(count + 1);
2005 le16_add_cpu(&lcm->lcm_mirror_count, 1);
2007 /* 1. Move the component bodies from [pos, count-1] to [pos+1, count]
2008 * with distance of 'added'. */
2011 for (i = pos; i < count; i++) {
2012 lcme = &lcm->lcm_entries[i];
2013 size += le32_to_cpu(lcme->lcme_size);
2016 offset = le32_to_cpu(lcm->lcm_entries[pos].lcme_offset);
2017 memmove(buf->lb_buf + offset + added,
2018 buf->lb_buf + offset, size);
2022 /* 2. Move the component header [0, pos-1] to [0, pos-1] with distance
2023 * of 'sizeof(struct lov_comp_md_entry_v1)' */
2025 for (i = 0; i < pos; i++) {
2026 lcme = &lcm->lcm_entries[i];
2027 size += le32_to_cpu(lcme->lcme_size);
2030 offset = le32_to_cpu(lcm->lcm_entries[0].lcme_offset);
2031 memmove(buf->lb_buf + offset + sizeof(*lcme),
2032 buf->lb_buf + offset, size);
2035 /* 3. Recalculate the enter offset for the component [pos, count-1] */
2036 for (i = count - 1; i >= pos; i--) {
2037 lcm->lcm_entries[i + 1] = lcm->lcm_entries[i];
2038 lcm->lcm_entries[i + 1].lcme_offset =
2039 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i + 1].
2040 lcme_offset) + added);
2043 /* 4. Recalculate the enter offset for the component [0, pos) */
2044 for (i = 0; i < pos; i++) {
2045 lcm->lcm_entries[i].lcme_offset =
2046 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i].
2047 lcme_offset) + sizeof(*lcme));
2050 offset = sizeof(*lcm) + sizeof(*lcme) * (count + 1) + size;
2051 /* 4. Insert the new component header (entry) at the slot 'pos'. */
2052 lcme = &lcm->lcm_entries[pos];
2053 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
2054 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
2055 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
2056 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
2057 lcme->lcme_offset = cpu_to_le32(offset);
2058 lcme->lcme_size = cpu_to_le32(lov_mds_md_size(ol->ol_stripe_count,
2061 if (ol->ol_stripe_count > 1)
2062 pattern |= LOV_PATTERN_F_HOLE;
2064 lmm = buf->lb_buf + offset;
2065 /* 5. Insert teh new component body at the 'offset'. */
2066 objs = __lfsck_layout_new_v1_lovea(lmm, lfsck_dto2fid(parent),
2067 ol->ol_stripe_size, ea_off,
2068 pattern, ol->ol_stripe_count);
2070 /* 6. Update mirror related flags and version. */
2071 lfsck_layout_update_lcm(lcm, lcme, rec->lor_layout_version,
2074 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
2075 lmm, objs, LU_XATTR_REPLACE, ost_idx,
2076 le32_to_cpu(lcm->lcm_size));
2078 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant add new COMP for "
2079 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2080 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2081 "comp_end %llu, layout version %u, range %u, "
2082 "%s LOV EA hole: rc = %d\n",
2083 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2084 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2085 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2086 rec->lor_layout_version, rec->lor_range,
2087 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2088 "with" : "without", rc);
2093 static int lfsck_layout_extend_v1v3_lovea(const struct lu_env *env,
2094 struct lfsck_instance *lfsck,
2095 struct thandle *handle,
2096 struct ost_layout *ol,
2097 struct dt_object *parent,
2098 const struct lu_fid *cfid,
2099 struct lu_buf *buf, __u32 ost_idx,
2102 struct lov_mds_md_v1 *lmm = buf->lb_buf;
2103 struct lov_ost_data_v1 *objs;
2104 __u16 count = le16_to_cpu(lmm->lmm_stripe_count);
2105 __u32 magic = le32_to_cpu(lmm->lmm_magic);
2111 /* The original LOVEA maybe re-generated via old filter_fid, at
2112 * that time, we do not know the stripe count and stripe size. */
2113 if (ol->ol_stripe_count > count)
2114 count = ol->ol_stripe_count;
2115 if (ol->ol_stripe_size != 0 &&
2116 ol->ol_stripe_size != le32_to_cpu(lmm->lmm_stripe_size))
2117 lmm->lmm_stripe_size = cpu_to_le32(ol->ol_stripe_size);
2119 if (magic == LOV_MAGIC_V1)
2120 objs = &lmm->lmm_objects[count];
2122 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[count];
2124 gap = ea_off - count;
2128 size = lov_mds_md_size(count, magic);
2129 LASSERTF(buf->lb_len >= size,
2130 "buffer len %d is less than real size %d\n",
2131 (int)buf->lb_len, size);
2134 memset(objs, 0, gap * sizeof(*objs));
2135 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
2138 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2139 lmm->lmm_stripe_count = cpu_to_le16(count);
2142 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
2143 lmm, objs, LU_XATTR_REPLACE, ost_idx, size);
2145 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
2146 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2147 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2148 "comp_end %llu, %s LOV EA hole: rc = %d\n",
2149 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2150 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2151 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2152 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2153 "with" : "without", rc);
2159 * \retval +1: repaired
2160 * \retval 0: did nothing
2161 * \retval -ve: on error
2163 static int lfsck_layout_update_lovea(const struct lu_env *env,
2164 struct lfsck_instance *lfsck,
2165 struct thandle *handle,
2166 struct lu_orphan_rec_v3 *rec,
2167 struct dt_object *parent,
2168 const struct lu_fid *cfid,
2169 struct lu_buf *buf, int fl,
2170 __u32 ost_idx, __u32 ea_off)
2172 struct ost_layout *ol = &rec->lor_layout;
2173 struct lov_mds_md_v1 *lmm = NULL;
2174 struct lov_ost_data_v1 *objs = NULL;
2178 if (ol->ol_comp_id != 0)
2179 rc = lfsck_layout_new_comp_lovea(env, rec, parent, buf, ea_off,
2182 rc = lfsck_layout_new_v1_lovea(env, lfsck, &rec->lor_layout,
2183 parent, buf, ea_off, &lmm,
2186 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid,
2187 buf, lmm, objs, fl, ost_idx, rc);
2189 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant created layout EA for "
2190 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2191 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2192 "comp_end %llu, layout version %u, range %u, fl %d, "
2193 "%s LOV EA hole: rc = %d\n",
2194 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2195 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2196 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2197 rec->lor_layout_version, rec->lor_range, fl,
2198 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2199 "with" : "without", rc);
2204 static int __lfsck_layout_update_pfid(const struct lu_env *env,
2205 struct lfsck_component *com,
2206 struct dt_object *child,
2207 const struct lu_fid *pfid,
2208 const struct ost_layout *ol, __u32 offset,
2209 __u32 version, __u32 range)
2211 struct dt_device *dev = lfsck_obj2dev(child);
2212 struct filter_fid *ff = &lfsck_env_info(env)->lti_ff;
2213 struct thandle *handle;
2214 struct lu_buf buf = { NULL };
2217 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2218 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2219 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2220 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2221 * parent MDT-object's layout EA. */
2222 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
2223 ost_layout_cpu_to_le(&ff->ff_layout, ol);
2224 ff->ff_layout_version = cpu_to_le32(version);
2225 ff->ff_range = cpu_to_le32(range);
2226 lfsck_buf_init(&buf, ff, sizeof(*ff));
2228 handle = lfsck_trans_create(env, dev, com->lc_lfsck);
2230 RETURN(PTR_ERR(handle));
2232 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2236 rc = dt_trans_start_local(env, dev, handle);
2240 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2245 dt_trans_stop(env, dev, handle);
2251 * \retval +1: repaired
2252 * \retval 0: did nothing
2253 * \retval -ve: on error
2255 static int lfsck_layout_update_pfid(const struct lu_env *env,
2256 struct lfsck_component *com,
2257 struct dt_object *parent,
2258 struct lu_fid *cfid,
2259 struct dt_device *cdev,
2260 struct lu_orphan_rec_v3 *rec, __u32 ea_off)
2262 struct dt_object *child;
2266 child = lfsck_object_find_by_dev(env, cdev, cfid);
2268 RETURN(PTR_ERR(child));
2270 rc = __lfsck_layout_update_pfid(env, com, child,
2271 lu_object_fid(&parent->do_lu),
2272 &rec->lor_layout, ea_off,
2273 rec->lor_layout_version,
2275 lfsck_object_put(env, child);
2277 RETURN(rc == 0 ? 1 : rc);
2280 static int lfsck_lovea_size(struct ost_layout *ol, __u32 ea_off)
2282 if (ol->ol_comp_id != 0)
2283 return sizeof(struct lov_comp_md_v1) +
2284 sizeof(struct lov_comp_md_entry_v1) +
2285 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2287 if (ol->ol_stripe_count != 0)
2288 return lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2290 return lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2294 * This function will create the MDT-object with the given (partial) LOV EA.
2296 * Under some data corruption cases, the MDT-object of the file may be lost,
2297 * but its OST-objects, or some of them are there. The layout LFSCK needs to
2298 * re-create the MDT-object with the orphan OST-object(s) information.
2300 * On the other hand, the LFSCK may has created some OST-object for repairing
2301 * dangling LOV EA reference, but as the LFSCK processing, it may find that
2302 * the old OST-object is there and should replace the former new created OST
2303 * object. Unfortunately, some others have modified such newly created object.
2304 * To keep the data (both new and old), the LFSCK will create MDT-object with
2305 * new FID to reference the original OST-object.
2307 * \param[in] env pointer to the thread context
2308 * \param[in] com pointer to the lfsck component
2309 * \param[in] ltd pointer to target device descriptor
2310 * \param[in] rec pointer to the record for the orphan OST-object
2311 * \param[in] cfid pointer to FID for the orphan OST-object
2312 * \param[in] infix additional information, such as the FID for original
2313 * MDT-object and the stripe offset in the LOV EA
2314 * \param[in] type the type for describing why the orphan MDT-object is
2315 * created. The rules are as following:
2317 * type "C": Multiple OST-objects claim the same MDT-object and the
2318 * same slot in the layout EA. Then the LFSCK will create
2319 * new MDT-object(s) to hold the conflict OST-object(s).
2321 * type "N": The orphan OST-object does not know which one was the
2322 * real parent MDT-object, so the LFSCK uses new FID for
2323 * its parent MDT-object.
2325 * type "R": The orphan OST-object knows its parent MDT-object FID,
2326 * but does not know the position (the file name) in the
2329 * type "D": The MDT-object is a directory, it may knows its parent
2330 * but because there is no valid linkEA, the LFSCK cannot
2331 * know where to put it back to the namespace.
2332 * type "O": The MDT-object has no linkEA, and there is no name
2333 * entry that references the MDT-object.
2335 * type "P": The orphan object to be created was a parent directory
2336 * of some MDT-object which linkEA shows that the @orphan
2337 * object is missing.
2339 * The orphan name will be like:
2340 * ${FID}-${infix}-${type}-${conflict_version}
2342 * \param[in] ea_off the stripe offset in the LOV EA
2344 * \retval positive on repaired something
2345 * \retval 0 if needs to repair nothing
2346 * \retval negative error number on failure
2348 static int lfsck_layout_recreate_parent(const struct lu_env *env,
2349 struct lfsck_component *com,
2350 struct lfsck_tgt_desc *ltd,
2351 struct lu_orphan_rec_v3 *rec,
2352 struct lu_fid *cfid,
2357 struct lfsck_thread_info *info = lfsck_env_info(env);
2358 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
2359 char *name = info->lti_key;
2360 struct lu_attr *la = &info->lti_la2;
2361 struct dt_object_format *dof = &info->lti_dof;
2362 struct lfsck_instance *lfsck = com->lc_lfsck;
2363 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
2364 struct lu_fid *tfid = &info->lti_fid3;
2365 struct dt_device *dev = lfsck->li_bottom;
2366 struct dt_object *lpf = lfsck->li_lpf_obj;
2367 struct dt_object *pobj = NULL;
2368 struct dt_object *cobj = NULL;
2369 struct thandle *th = NULL;
2370 struct lu_buf *ea_buf = &info->lti_big_buf;
2371 struct lu_buf lov_buf;
2372 struct lfsck_lock_handle *llh = &info->lti_llh;
2373 struct linkea_data ldata = { NULL };
2374 struct lu_buf linkea_buf;
2375 const struct lu_name *pname;
2381 if (lfsck_is_dryrun(lfsck))
2384 if (unlikely(lpf == NULL))
2385 GOTO(log, rc = -ENXIO);
2387 /* We use two separated transactions to repair the inconsistency.
2389 * 1) create the MDT-object locally.
2390 * 2) update the OST-object's PFID EA if necessary.
2392 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
2393 * updated when the layout LFSCK run next time.
2395 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
2396 * when the layout LFSCK run next time. */
2398 if (fid_is_zero(pfid)) {
2399 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
2403 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2405 GOTO(log, rc = PTR_ERR(cobj));
2408 pobj = lfsck_object_find_by_dev(env, dev, pfid);
2410 GOTO(log, rc = PTR_ERR(pobj));
2412 LASSERT(infix != NULL);
2413 LASSERT(type != NULL);
2415 memset(la, 0, sizeof(*la));
2416 la->la_uid = rec->lor_rec.lor_uid;
2417 la->la_gid = rec->lor_rec.lor_gid;
2418 la->la_mode = S_IFREG | S_IRUSR;
2419 la->la_valid = LA_MODE | LA_UID | LA_GID;
2421 memset(dof, 0, sizeof(*dof));
2422 dof->dof_type = dt_mode_to_dft(S_IFREG);
2423 /* Because the dof->dof_reg.striped = 0, the LOD will not create
2424 * the stripe(s). The LFSCK will specify the LOV EA via
2425 * lfsck_layout_update_lovea(). */
2427 size = lfsck_lovea_size(&rec->lor_layout, ea_off);
2428 if (ea_buf->lb_len < size) {
2429 lu_buf_realloc(ea_buf, size);
2430 if (ea_buf->lb_buf == NULL)
2431 GOTO(log, rc = -ENOMEM);
2436 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
2438 rc = dt_lookup_dir(env, lfsck->li_lpf_obj, name, tfid);
2439 if (rc != 0 && rc != -ENOENT)
2443 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
2444 MDS_INODELOCK_UPDATE, LCK_PW);
2448 /* Re-check whether the name conflict with othrs after taken
2450 rc = dt_lookup_dir(env, lfsck->li_lpf_obj, name, tfid);
2451 if (unlikely(rc == 0)) {
2459 pname = lfsck_name_get_const(env, name, strlen(name));
2460 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
2461 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
2465 /* The 1st transaction. */
2466 th = lfsck_trans_create(env, dev, lfsck);
2468 GOTO(unlock, rc = PTR_ERR(th));
2470 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
2474 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
2475 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
2476 LU_XATTR_CREATE, th);
2480 dtrec->rec_fid = pfid;
2481 dtrec->rec_type = S_IFREG;
2482 rc = dt_declare_insert(env, lpf,
2483 (const struct dt_rec *)dtrec,
2484 (const struct dt_key *)name, th);
2488 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2489 ldata.ld_leh->leh_len);
2490 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
2491 XATTR_NAME_LINK, 0, th);
2495 rc = dt_trans_start_local(env, dev, th);
2499 dt_write_lock(env, pobj, 0);
2500 rc = dt_create(env, pobj, la, NULL, dof, th);
2502 rc = lfsck_layout_update_lovea(env, lfsck, th, rec, pobj, cfid,
2503 &lov_buf, LU_XATTR_CREATE, ltd->ltd_index, ea_off);
2504 dt_write_unlock(env, pobj);
2508 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
2509 (const struct dt_key *)name, th);
2513 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
2514 if (rc == 0 && cobj != NULL) {
2515 dt_trans_stop(env, dev, th);
2518 /* The 2nd transaction. */
2519 rc = __lfsck_layout_update_pfid(env, com, cobj, pfid,
2520 &rec->lor_layout, ea_off,
2521 rec->lor_layout_version,
2529 dt_trans_stop(env, dev, th);
2535 if (cobj != NULL && !IS_ERR(cobj))
2536 lfsck_object_put(env, cobj);
2537 if (pobj != NULL && !IS_ERR(pobj))
2538 lfsck_object_put(env, pobj);
2541 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2542 "recreate the lost MDT-object: parent "DFID
2543 ", child "DFID", OST-index %u, stripe-index %u, "
2544 "infix %s, type %s: rc = %d\n",
2545 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2546 ltd->ltd_index, ea_off, infix, type, rc);
2548 return rc >= 0 ? 1 : rc;
2551 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2552 struct lfsck_component *com,
2553 const struct lu_fid *fid,
2556 struct lfsck_thread_info *info = lfsck_env_info(env);
2557 struct lfsck_request *lr = &info->lti_lr;
2558 struct lfsck_instance *lfsck = com->lc_lfsck;
2559 struct lfsck_tgt_desc *ltd;
2560 struct ptlrpc_request *req;
2561 struct lfsck_request *tmp;
2562 struct obd_export *exp;
2566 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2567 if (unlikely(ltd == NULL))
2571 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2572 GOTO(put, rc = -EOPNOTSUPP);
2574 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2576 GOTO(put, rc = -ENOMEM);
2578 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2580 ptlrpc_request_free(req);
2585 memset(lr, 0, sizeof(*lr));
2586 lr->lr_event = LE_CONDITIONAL_DESTROY;
2587 lr->lr_active = LFSCK_TYPE_LAYOUT;
2590 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2592 ptlrpc_request_set_replen(req);
2594 rc = ptlrpc_queue_wait(req);
2595 ptlrpc_req_finished(req);
2605 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2606 struct lfsck_component *com,
2607 struct lfsck_request *lr)
2609 struct lfsck_thread_info *info = lfsck_env_info(env);
2610 struct lu_attr *la = &info->lti_la;
2611 union ldlm_policy_data *policy = &info->lti_policy;
2612 struct ldlm_res_id *resid = &info->lti_resid;
2613 struct lfsck_instance *lfsck = com->lc_lfsck;
2614 struct dt_device *dev = lfsck->li_bottom;
2615 struct lu_fid *fid = &lr->lr_fid;
2616 struct dt_object *obj;
2617 struct thandle *th = NULL;
2618 struct lustre_handle lh = { 0 };
2623 obj = lfsck_object_find_by_dev(env, dev, fid);
2625 RETURN(PTR_ERR(obj));
2627 dt_read_lock(env, obj, 0);
2628 if (dt_object_exists(obj) == 0 ||
2629 lfsck_is_dead_obj(obj)) {
2630 dt_read_unlock(env, obj);
2632 GOTO(put, rc = -ENOENT);
2635 /* Get obj's attr without lock firstly. */
2636 rc = dt_attr_get(env, obj, la);
2637 dt_read_unlock(env, obj);
2641 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2642 GOTO(put, rc = -ETXTBSY);
2644 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2645 LASSERT(lfsck->li_namespace != NULL);
2647 memset(policy, 0, sizeof(*policy));
2648 policy->l_extent.end = OBD_OBJECT_EOF;
2649 ost_fid_build_resid(fid, resid);
2650 rc = ldlm_cli_enqueue_local(env, lfsck->li_namespace, resid,
2651 LDLM_EXTENT, policy, LCK_EX, &flags,
2652 ldlm_blocking_ast, ldlm_completion_ast,
2653 NULL, NULL, 0, LVB_T_NONE, NULL, &lh);
2655 GOTO(put, rc = -EIO);
2657 dt_write_lock(env, obj, 0);
2658 /* Get obj's attr within lock again. */
2659 rc = dt_attr_get(env, obj, la);
2663 if (la->la_ctime != 0)
2664 GOTO(unlock, rc = -ETXTBSY);
2666 th = lfsck_trans_create(env, dev, lfsck);
2668 GOTO(unlock, rc = PTR_ERR(th));
2670 rc = dt_declare_ref_del(env, obj, th);
2674 rc = dt_declare_destroy(env, obj, th);
2678 rc = dt_trans_start_local(env, dev, th);
2682 rc = dt_ref_del(env, obj, th);
2686 rc = dt_destroy(env, obj, th);
2688 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2689 "OST-object "DFID" that was created for reparing "
2690 "dangling referenced case. But the original missing "
2691 "OST-object is found now.\n",
2692 lfsck_lfsck2name(lfsck), PFID(fid));
2697 dt_trans_stop(env, dev, th);
2700 dt_write_unlock(env, obj);
2701 ldlm_lock_decref(&lh, LCK_EX);
2704 lfsck_object_put(env, obj);
2710 * Some OST-object has occupied the specified layout EA slot.
2711 * Such OST-object may be generated by the LFSCK when repair
2712 * dangling referenced MDT-object, which can be indicated by
2713 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2714 * is true and such OST-object has not been modified yet, we
2715 * will replace it with the orphan OST-object; otherwise the
2716 * LFSCK will create new MDT-object to reference the orphan.
2718 * \retval +1: repaired
2719 * \retval 0: did nothing
2720 * \retval -ve: on error
2722 static int lfsck_layout_conflict_create(const struct lu_env *env,
2723 struct lfsck_component *com,
2724 struct lfsck_tgt_desc *ltd,
2725 struct lu_orphan_rec_v3 *rec,
2726 struct dt_object *parent,
2727 struct lu_fid *cfid,
2728 struct lu_buf *ea_buf,
2729 struct lov_mds_md_v1 *lmm,
2730 struct lov_ost_data_v1 *slot,
2731 __u32 ea_off, int lovea_size)
2733 struct lfsck_thread_info *info = lfsck_env_info(env);
2734 struct lu_fid *cfid2 = &info->lti_fid2;
2735 struct ost_id *oi = &info->lti_oi;
2736 struct dt_device *dev = lfsck_obj2dev(parent);
2737 struct thandle *th = NULL;
2738 struct lustre_handle lh = { 0 };
2739 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2743 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2744 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2748 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2749 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2753 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2754 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2759 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2761 /* If the conflict OST-obejct is not created for fixing dangling
2762 * referenced MDT-object in former LFSCK check/repair, or it has
2763 * been modified by others, then we cannot destroy it. Re-create
2764 * a new MDT-object for the orphan OST-object. */
2765 if (rc == -ETXTBSY) {
2766 /* No need the layout lock on the original parent. */
2767 lfsck_ibits_unlock(&lh, LCK_EX);
2769 fid_zero(&rec->lor_rec.lor_fid);
2770 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2771 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2773 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2774 info->lti_tmpbuf, "C", ea_off);
2779 if (rc != 0 && rc != -ENOENT)
2782 if (lfsck_is_dryrun(com->lc_lfsck))
2783 GOTO(unlock, rc = 0);
2785 th = lfsck_trans_create(env, dev, com->lc_lfsck);
2787 GOTO(unlock, rc = PTR_ERR(th));
2789 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2790 LU_XATTR_REPLACE, th);
2794 rc = dt_trans_start_local(env, dev, th);
2798 dt_write_lock(env, parent, 0);
2799 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2800 rc = lfsck_layout_refill_lovea(env, com->lc_lfsck, th, parent, cfid,
2801 ea_buf, lmm, slot, LU_XATTR_REPLACE,
2802 ltd->ltd_index, lovea_size);
2803 dt_write_unlock(env, parent);
2808 dt_trans_stop(env, dev, th);
2811 lfsck_ibits_unlock(&lh, LCK_EX);
2814 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2815 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2816 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2817 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2818 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2821 return rc >= 0 ? 1 : rc;
2825 * \retval +1: repaired
2826 * \retval 0: did nothing
2827 * \retval -ve: on error
2829 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2830 struct lfsck_component *com,
2831 struct lfsck_tgt_desc *ltd,
2832 struct lu_orphan_rec_v3 *rec,
2833 struct dt_object *parent,
2834 struct lu_fid *cfid,
2835 __u32 ost_idx, __u32 ea_off)
2837 struct lfsck_thread_info *info = lfsck_env_info(env);
2838 struct lu_buf *buf = &info->lti_big_buf;
2839 struct lu_fid *fid = &info->lti_fid2;
2840 struct ost_id *oi = &info->lti_oi;
2841 struct lfsck_instance *lfsck = com->lc_lfsck;
2842 struct dt_device *dt = lfsck_obj2dev(parent);
2843 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2844 struct ost_layout *ol = &rec->lor_layout;
2845 struct lov_comp_md_v1 *lcm = NULL;
2846 struct lov_comp_md_entry_v1 *lcme = NULL;
2847 struct thandle *handle = NULL;
2849 struct lov_mds_md_v1 *lmm;
2850 struct lov_ost_data_v1 *objs;
2851 struct lustre_handle lh = { 0 };
2860 bool locked = false;
2861 bool new_mirror = true;
2864 if (lfsck_is_dryrun(lfsck))
2867 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2868 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2871 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2872 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2873 "stripe-index %u, comp_id %u, comp_start %llu, "
2874 "comp_end %llu, layout version %u, range %u: rc = %d\n",
2875 lfsck_lfsck2name(lfsck), PFID(cfid),
2876 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off,
2877 ol->ol_comp_id, ol->ol_comp_start,
2878 ol->ol_comp_end, rec->lor_layout_version,
2879 rec->lor_range, rc);
2886 dt_write_unlock(env, parent);
2890 if (handle != NULL) {
2891 dt_trans_stop(env, dt, handle);
2896 GOTO(unlock_layout, rc);
2899 if (buf->lb_len < lovea_size) {
2900 lu_buf_realloc(buf, lovea_size);
2901 if (buf->lb_buf == NULL)
2902 GOTO(unlock_layout, rc = -ENOMEM);
2905 if (!(bk->lb_param & LPF_DRYRUN)) {
2906 handle = lfsck_trans_create(env, dt, lfsck);
2908 GOTO(unlock_layout, rc = PTR_ERR(handle));
2910 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2915 rc = dt_trans_start_local(env, dt, handle);
2920 dt_write_lock(env, parent, 0);
2922 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2923 if (rc == -ERANGE) {
2924 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2927 } else if (rc == -ENODATA || rc == 0) {
2928 lovea_size = lfsck_lovea_size(ol, ea_off);
2929 /* If the declared is not big enough, re-try. */
2930 if (buf->lb_len < lovea_size) {
2934 fl = LU_XATTR_CREATE;
2935 } else if (rc < 0) {
2936 GOTO(unlock_parent, rc);
2937 } else if (unlikely(buf->lb_len == 0)) {
2940 fl = LU_XATTR_REPLACE;
2944 if (fl == LU_XATTR_CREATE) {
2945 if (bk->lb_param & LPF_DRYRUN)
2946 GOTO(unlock_parent, rc = 1);
2948 LASSERT(buf->lb_len >= lovea_size);
2950 rc = lfsck_layout_update_lovea(env, lfsck, handle, rec, parent,
2951 cfid, buf, fl, ost_idx, ea_off);
2953 GOTO(unlock_parent, rc);
2957 rc1 = lfsck_layout_verify_header(parent, lmm, lovea_size);
2959 /* If the LOV EA crashed, the rebuild it. */
2960 if (rc1 == -EINVAL) {
2961 if (bk->lb_param & LPF_DRYRUN)
2962 GOTO(unlock_parent, rc = 1);
2964 LASSERT(buf->lb_len >= lovea_size);
2966 rc = lfsck_layout_update_lovea(env, lfsck, handle, rec, parent,
2967 cfid, buf, fl, ost_idx, ea_off);
2969 GOTO(unlock_parent, rc);
2972 /* For other unknown magic/pattern, keep the current LOV EA. */
2973 if (rc1 == -EOPNOTSUPP)
2974 GOTO(unlock_parent, rc1 = 0);
2977 GOTO(unlock_parent, rc = rc1);
2979 magic = le32_to_cpu(lmm->lmm_magic);
2980 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
2983 __u16 mirror_id0 = mirror_id_of(ol->ol_comp_id);
2986 if (bk->lb_param & LPF_DRYRUN)
2987 GOTO(unlock_parent, rc = 1);
2990 count = le16_to_cpu(lcm->lcm_entry_count);
2991 for (i = 0; i < count; pos = ++i) {
2992 lcme = &lcm->lcm_entries[i];
2993 start = le64_to_cpu(lcme->lcme_extent.e_start);
2994 end = le64_to_cpu(lcme->lcme_extent.e_end);
2995 mirror_id1 = mirror_id_of(le32_to_cpu(lcme->lcme_id));
2997 if (mirror_id0 > mirror_id1)
3000 if (mirror_id0 < mirror_id1)
3004 if (end <= ol->ol_comp_start)
3007 if (start >= ol->ol_comp_end)
3010 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
3011 magic = le32_to_cpu(lmm->lmm_magic);
3012 flags = le32_to_cpu(lcme->lcme_flags);
3016 rc = lfsck_layout_add_comp(env, lfsck, handle, rec, parent,
3017 cfid, buf, ost_idx, ea_off, pos, new_mirror);
3019 GOTO(unlock_parent, rc);
3023 count = le16_to_cpu(lmm->lmm_stripe_count);
3025 GOTO(unlock_parent, rc = -EINVAL);
3028 /* Exceed the current end of MDT-object layout EA. Then extend it. */
3029 if (count <= ea_off) {
3030 if (bk->lb_param & LPF_DRYRUN)
3031 GOTO(unlock_parent, rc = 1);
3033 lovea_size = lov_mds_md_size(ea_off + 1, magic);
3034 /* If the declared is not big enough, re-try. */
3035 if (buf->lb_len < lovea_size) {
3043 lcme->lcme_flags = cpu_to_le32(flags | LCME_FL_INIT);
3044 lfsck_layout_update_lcm(lcm, lcme,
3045 rec->lor_layout_version,
3049 rc = lfsck_layout_extend_v1v3_lovea(env, lfsck, handle, ol,
3050 parent, cfid, buf, ost_idx, ea_off);
3052 GOTO(unlock_parent, rc);
3055 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
3057 if (magic == LOV_MAGIC_V1) {
3058 objs = &lmm->lmm_objects[0];
3060 LASSERT(magic == LOV_MAGIC_V3);
3061 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3064 for (i = 0; i < count; i++, objs++) {
3065 /* The MDT-object was created via lfsck_layout_recover_create()
3066 * by others before, and we fill the dummy layout EA. */
3067 if ((lcme && !(flags & LCME_FL_INIT)) ||
3068 lovea_slot_is_dummy(objs)) {
3072 if (bk->lb_param & LPF_DRYRUN)
3073 GOTO(unlock_parent, rc = 1);
3075 lmm->lmm_layout_gen =
3076 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3080 if (le32_to_cpu(lmm->lmm_stripe_size) !=
3081 ol->ol_stripe_size ||
3082 le16_to_cpu(lmm->lmm_stripe_count) !=
3083 ol->ol_stripe_count ||
3084 le64_to_cpu(lcme->lcme_extent.e_start) !=
3085 ol->ol_comp_start ||
3086 le64_to_cpu(lcme->lcme_extent.e_end) !=
3088 CDEBUG(D_LFSCK, "%s: found invalid "
3089 "component for "DFID ": parent "DFID
3090 ", stripe-index %u, stripe_size %u, "
3091 "stripe_count %u, comp_id %u, "
3092 "comp_start %llu, comp_end %llu, "
3093 "cur_stripe_size %u, "
3094 "cur_stripe_count %u, "
3095 "cur_comp_start %llu, "
3096 "cur_comp_end %llu\n",
3097 lfsck_lfsck2name(lfsck), PFID(cfid),
3098 PFID(lfsck_dto2fid(parent)), ea_off,
3100 ol->ol_stripe_count, ol->ol_comp_id,
3101 ol->ol_comp_start, ol->ol_comp_end,
3102 le32_to_cpu(lmm->lmm_stripe_size),
3103 le16_to_cpu(lmm->lmm_stripe_count),
3104 le64_to_cpu(lcme->lcme_extent.e_start),
3105 le64_to_cpu(lcme->lcme_extent.e_end));
3107 GOTO(unlock_parent, rc = -EINVAL);
3110 lovea_size = le32_to_cpu(lcm->lcm_size);
3111 lcme->lcme_flags = cpu_to_le32(flags |
3113 lfsck_layout_update_lcm(lcm, lcme,
3114 rec->lor_layout_version,
3118 LASSERTF(buf->lb_len >= lovea_size,
3119 "buffer len %d is less than real size %d\n",
3120 (int)buf->lb_len, (int)lovea_size);
3122 rc = lfsck_layout_refill_lovea(env, lfsck, handle,
3123 parent, cfid, buf, lmm, objs,
3124 fl, ost_idx, lovea_size);
3126 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
3127 "dummy layout slot for "DFID": parent "DFID
3128 ", OST-index %u, stripe-index %u: rc = %d\n",
3129 lfsck_lfsck2name(lfsck), PFID(cfid),
3130 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
3132 GOTO(unlock_parent, rc);
3135 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3136 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
3138 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3139 "invalid layout EA at the slot %d, index %u\n",
3140 lfsck_lfsck2name(lfsck),
3141 PFID(lfsck_dto2fid(parent)), i,
3142 le32_to_cpu(objs->l_ost_idx));
3144 GOTO(unlock_parent, rc);
3147 /* It should be rare case, the slot is there, but the LFSCK
3148 * does not handle it during the first-phase cycle scanning. */
3149 if (unlikely(lu_fid_eq(fid, cfid))) {
3151 GOTO(unlock_parent, rc = 0);
3153 /* Rare case that the OST-object index
3154 * does not match the parent MDT-object
3155 * layout EA. We trust the later one. */
3156 if (bk->lb_param & LPF_DRYRUN)
3157 GOTO(unlock_parent, rc = 1);
3159 dt_write_unlock(env, parent);
3161 dt_trans_stop(env, dt, handle);
3162 lfsck_ibits_unlock(&lh, LCK_EX);
3163 rc = lfsck_layout_update_pfid(env, com, parent,
3167 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
3168 "updated OST-object's pfid for "DFID
3169 ": parent "DFID", OST-index %u, "
3170 "stripe-index %u: rc = %d\n",
3171 lfsck_lfsck2name(lfsck), PFID(cfid),
3172 PFID(lfsck_dto2fid(parent)),
3173 ltd->ltd_index, i, rc);
3180 /* The MDT-object exists, but related layout EA slot is occupied
3182 if (bk->lb_param & LPF_DRYRUN)
3183 GOTO(unlock_parent, rc = 1);
3185 dt_write_unlock(env, parent);
3187 dt_trans_stop(env, dt, handle);
3188 lfsck_ibits_unlock(&lh, LCK_EX);
3189 if (magic == LOV_MAGIC_V1)
3190 objs = &lmm->lmm_objects[ea_off];
3192 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
3193 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
3194 buf, lmm, objs, ea_off, lovea_size);
3200 dt_write_unlock(env, parent);
3204 dt_trans_stop(env, dt, handle);
3207 lfsck_ibits_unlock(&lh, LCK_EX);
3212 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
3213 struct lfsck_component *com,
3214 struct lfsck_tgt_desc *ltd,
3215 struct lu_orphan_rec_v3 *rec,
3216 struct lu_fid *cfid)
3218 struct lfsck_layout *lo = com->lc_file_ram;
3219 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
3220 struct dt_object *parent = NULL;
3221 __u32 ea_off = pfid->f_stripe_idx;
3225 if (!fid_is_sane(cfid))
3226 GOTO(out, rc = -EINVAL);
3229 if (fid_is_zero(pfid)) {
3230 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3235 if (!fid_is_sane(pfid))
3236 GOTO(out, rc = -EINVAL);
3238 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
3240 GOTO(out, rc = PTR_ERR(parent));
3242 if (unlikely(dt_object_remote(parent) != 0))
3243 GOTO(put, rc = -EXDEV);
3245 if (dt_object_exists(parent) == 0) {
3246 lfsck_object_put(env, parent);
3247 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3252 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
3253 GOTO(put, rc = -EISDIR);
3255 /* The orphan OST-object claims to be the parent's stripe, then
3256 * related dangling record in the trace file is meaningless. */
3257 rc = lfsck_layout_del_dangling_rec(env, com, pfid,
3258 rec->lor_layout.ol_comp_id, ea_off);
3259 if (rc && rc != -ENOENT)
3262 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
3263 ltd->ltd_index, ea_off);
3269 lfsck_object_put(env, parent);
3271 /* The layout EA is changed, need to be reloaded next time. */
3272 dt_object_put_nocache(env, parent);
3275 down_write(&com->lc_sem);
3276 com->lc_new_scanned++;
3277 com->lc_new_checked++;
3279 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
3281 } else if (rc < 0) {
3282 lo->ll_objs_failed_phase2++;
3284 up_write(&com->lc_sem);
3289 static int lfsck_layout_scan_orphan(const struct lu_env *env,
3290 struct lfsck_component *com,
3291 struct lfsck_tgt_desc *ltd)
3293 struct lfsck_assistant_data *lad = com->lc_data;
3294 struct lfsck_instance *lfsck = com->lc_lfsck;
3295 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3296 struct lfsck_thread_info *info = lfsck_env_info(env);
3297 struct lu_fid *fid = &info->lti_fid;
3298 struct dt_object *obj;
3299 const struct dt_it_ops *iops;
3304 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
3305 "scanning for OST%04x\n",
3306 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3308 if (test_bit(ltd->ltd_index, lad->lad_bitmap)) {
3309 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
3310 "scanning for OST%04x\n",
3311 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3316 fid->f_seq = fid_idif_seq(0, ltd->ltd_index);
3317 fid->f_oid = fid->f_ver = 0;
3319 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
3320 if (unlikely(IS_ERR(obj)))
3321 GOTO(log, rc = PTR_ERR(obj));
3323 rc = obj->do_ops->do_index_try(env, obj,
3324 &dt_lfsck_layout_orphan_features);
3328 iops = &obj->do_index_ops->dio_it;
3329 di = iops->init(env, obj, 0);
3331 GOTO(put, rc = PTR_ERR(di));
3333 rc = iops->load(env, di, 0);
3335 /* -ESRCH means that the orphan OST-objects rbtree has been
3336 * cleanup because of the OSS server restart or other errors. */
3337 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
3342 rc = iops->next(env, di);
3354 struct lu_orphan_rec_v3 *rec = &info->lti_rec;
3356 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3357 unlikely(!thread_is_running(&lfsck->li_thread)))
3360 key = iops->key(env, di);
3361 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
3362 /* Remote target OST may be runnning old LFSCK */
3363 memset(rec, 0, sizeof(*rec));
3364 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
3366 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
3367 &com->lc_fid_latest_scanned_phase2);
3368 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3371 lfsck_control_speed_by_self(com);
3373 rc = iops->next(env, di);
3374 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
3381 iops->fini(env, di);
3383 lfsck_object_put(env, obj);
3386 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
3387 "scanning for OST%04x: rc = %d\n",
3388 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
3390 return rc > 0 ? 0 : rc;
3393 static int lfsck_lov2layout(struct lov_mds_md_v1 *lmm, struct filter_fid *ff,
3396 struct ost_layout *ol = &ff->ff_layout;
3397 __u32 magic = le32_to_cpu(lmm->lmm_magic);
3401 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3) {
3402 ol->ol_stripe_size = lmm->lmm_stripe_size;
3403 ol->ol_stripe_count = lmm->lmm_stripe_count;
3404 ol->ol_comp_start = 0;
3405 ol->ol_comp_end = 0;
3407 ff->ff_layout_version = 0;
3409 } else if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
3410 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
3411 struct lov_comp_md_entry_v1 *lcme = NULL;
3412 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3415 for (i = 0; i < count; i++) {
3416 lcme = &lcm->lcm_entries[i];
3417 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3418 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3425 /* The comp has been removed, do nothing. */
3429 lmm = (void *)lmm + le32_to_cpu(lcme->lcme_offset);
3430 ol->ol_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
3431 ol->ol_stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
3432 ol->ol_comp_start = le64_to_cpu(lcme->lcme_extent.e_start);
3433 ol->ol_comp_end = le64_to_cpu(lcme->lcme_extent.e_end);
3434 ol->ol_comp_id = le32_to_cpu(lcme->lcme_id);
3435 ff->ff_layout_version = le32_to_cpu(lcme->lcme_layout_gen);
3438 GOTO(out, rc = -EINVAL);
3448 * Repair the MDT-object with dangling LOV EA reference.
3450 * we need to repair the inconsistency according to the users' requirement:
3452 * 1) Keep the inconsistency there and report the inconsistency case,
3453 * then give the chance to the application to find related issues,
3454 * and the users can make the decision about how to handle it with
3455 * more human knownledge. (by default)
3457 * 2) Re-create the missing OST-object with the FID/owner information.
3459 * \param[in] env pointer to the thread context
3460 * \param[in] com the layout LFSCK component
3461 * \param[in] parent the MDT-object with dangling LOV EA reference
3462 * \param[in] child the OST-object to be created
3463 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3464 * \param[in] ea_off the offset of the OST-object in the LOV EA
3465 * \param[in] ost_idx the index of OST on which the OST-object resides
3467 * \retval +1 for repair successfully
3468 * \retval 0 for did nothing
3469 * \retval negative error number on failure
3471 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
3472 struct lfsck_component *com,
3473 struct dt_object *parent,
3474 struct dt_object *child,
3475 __u32 comp_id, __u32 ea_off,
3476 __u32 ost_idx, bool log)
3478 struct lfsck_thread_info *info = lfsck_env_info(env);
3479 struct filter_fid *ff = &info->lti_ff;
3480 struct dt_object_format *dof = &info->lti_dof;
3481 struct lu_attr *la = &info->lti_la;
3482 struct lfsck_instance *lfsck = com->lc_lfsck;
3483 struct dt_device *dev = lfsck_obj2dev(child);
3484 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3485 const struct lu_fid *cfid = lfsck_dto2fid(child);
3486 struct lu_buf *tbuf = &info->lti_big_buf;
3487 struct thandle *handle;
3489 struct lustre_handle lh = { 0 };
3493 if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
3496 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3497 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3502 rc = dt_attr_get(env, parent, la);
3506 la->la_mode = S_IFREG | 0666;
3507 la->la_atime = la->la_mtime = la->la_ctime = 0;
3508 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
3509 LA_ATIME | LA_MTIME | LA_CTIME;
3510 memset(dof, 0, sizeof(*dof));
3511 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
3512 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
3513 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3514 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3515 * parent MDT-object's layout EA. */
3516 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
3518 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3519 if (unlikely(rc == -ENODATA))
3524 rc = lfsck_lov2layout(tbuf->lb_buf, ff, comp_id);
3528 buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
3529 handle = lfsck_trans_create(env, dev, lfsck);
3531 GOTO(unlock1, rc = PTR_ERR(handle));
3533 rc = dt_declare_create(env, child, la, NULL, dof, handle);
3537 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
3538 LU_XATTR_CREATE, handle);
3542 rc = dt_trans_start_local(env, dev, handle);
3546 dt_read_lock(env, parent, 0);
3547 if (unlikely(lfsck_is_dead_obj(parent)))
3548 GOTO(unlock2, rc = 0);
3550 if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
3551 struct ost_id *oi = &info->lti_oi;
3552 struct lu_fid *tfid = &info->lti_fid2;
3553 struct lu_buf *lovea = &info->lti_big_buf;
3554 struct lov_mds_md_v1 *lmm;
3555 struct lov_ost_data_v1 *objs;
3560 rc = lfsck_layout_get_lovea(env, parent, lovea);
3561 if (unlikely(rc == -ENODATA))
3566 lmm = lovea->lb_buf;
3567 magic = le32_to_cpu(lmm->lmm_magic);
3568 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
3569 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3570 struct lov_comp_md_entry_v1 *lcme;
3571 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3574 for (i = 0; i < count; i++) {
3575 lcme = &lcm->lcm_entries[i];
3576 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3577 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3580 lmm = lovea->lb_buf +
3581 le32_to_cpu(lcme->lcme_offset);
3582 magic = le32_to_cpu(lmm->lmm_magic);
3587 /* Someone removed the component, do nothing. */
3588 GOTO(unlock2, rc = 0);
3592 count = le16_to_cpu(lmm->lmm_stripe_count);
3593 /* Someone changed the LOV EA, do nothing. */
3594 if (count <= ea_off)
3595 GOTO(unlock2, rc = 0);
3597 if (magic == LOV_MAGIC_V1) {
3598 objs = &lmm->lmm_objects[ea_off];
3600 LASSERT(magic == LOV_MAGIC_V3);
3602 objs = &((struct lov_mds_md_v3 *)lmm)->\
3603 lmm_objects[ea_off];
3606 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3607 idx2 = le32_to_cpu(objs->l_ost_idx);
3608 rc = ostid_to_fid(tfid, oi, idx2);
3609 /* Someone changed the LOV EA, do nothing. */
3610 if (rc != 0 || !lu_fid_eq(tfid, cfid))
3614 rc = dt_create(env, child, la, NULL, dof, handle);
3618 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
3624 dt_read_unlock(env, parent);
3627 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3630 lfsck_ibits_unlock(&lh, LCK_EX);
3634 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3635 "dangling reference for: parent "DFID", child "
3636 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: "
3638 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3639 comp_id, ea_off, ost_idx,
3640 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3641 "Create the lost OST-object as required" :
3642 "Keep the MDT-object there by default", rc);
3648 * Repair the MDT-object with dangling LOV EA reference.
3650 * Prepare parameters and call __lfsck_layout_repair_dangling()
3651 * to repair the dangling LOV EA reference.
3653 * \param[in] env pointer to the thread context
3654 * \param[in] com the layout LFSCK component
3655 * \param[in] pfid the MDT-object's FID
3656 * \param[in] cfid the FID for the OST-object to be created
3657 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3658 * \param[in] ea_off the offset of the OST-object in the LOV EA
3659 * \param[in] ost_idx the index of OST on which the OST-object resides
3661 * \retval +1 for repair successfully
3662 * \retval 0 for did nothing
3663 * \retval negative error number on failure
3665 static int lfsck_layout_repair_dangling(const struct lu_env *env,
3666 struct lfsck_component *com,
3667 const struct lu_fid *pfid,
3668 const struct lu_fid *cfid,
3669 __u32 comp_id, __u32 ea_off,
3672 struct lfsck_instance *lfsck = com->lc_lfsck;
3673 struct dt_object *parent = NULL;
3674 struct dt_object *child = NULL;
3675 struct lfsck_tgt_desc *ltd;
3679 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3681 GOTO(log, rc = PTR_ERR(parent));
3683 /* The MDT-object has been removed. */
3684 if (dt_object_exists(parent) == 0)
3687 ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
3688 if (unlikely(ltd == NULL))
3689 GOTO(log, rc = -ENODEV);
3691 child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
3693 GOTO(log, rc = PTR_ERR(child));
3695 /* The OST-object has been created. */
3696 if (unlikely(dt_object_exists(child) != 0))
3699 rc = __lfsck_layout_repair_dangling(env, com, parent, child,
3700 comp_id, ea_off, ost_idx, false);
3705 if (child != NULL && !IS_ERR(child))
3706 lfsck_object_put(env, child);
3708 if (parent != NULL && !IS_ERR(parent))
3709 lfsck_object_put(env, parent);
3712 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3713 "dangling reference for: parent "DFID", child "
3714 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: rc = %d\n",
3715 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3716 comp_id, ea_off, ost_idx,
3717 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3718 "Create the lost OST-object as required" :
3719 "Keep the MDT-object there by default", rc);
3724 /* If the OST-object does not recognize the MDT-object as its parent, and
3725 * there is no other MDT-object claims as its parent, then just trust the
3726 * given MDT-object as its parent. So update the OST-object filter_fid. */
3727 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
3728 struct lfsck_component *com,
3729 struct dt_object *parent,
3730 struct lfsck_layout_req *llr,
3733 struct lfsck_thread_info *info = lfsck_env_info(env);
3734 struct filter_fid *ff = &info->lti_ff;
3735 struct dt_object *child = llr->llr_child;
3736 struct dt_device *dev = lfsck_obj2dev(child);
3737 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
3738 struct lu_buf *tbuf = &info->lti_big_buf;
3739 struct thandle *handle;
3741 struct lustre_handle lh = { 0 };
3745 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3746 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3751 ff->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3752 ff->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3753 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3754 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3755 * parent MDT-object's layout EA. */
3756 ff->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3758 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3759 if (unlikely(rc == -ENODATA))
3764 rc = lfsck_lov2layout(tbuf->lb_buf, ff, llr->llr_comp_id);
3768 buf = lfsck_buf_get(env, ff, sizeof(*ff));
3770 handle = lfsck_trans_create(env, dev, com->lc_lfsck);
3772 GOTO(unlock1, rc = PTR_ERR(handle));
3774 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3778 rc = dt_attr_get(env, parent, la);
3782 la->la_valid = LA_UID | LA_GID;
3783 rc = dt_declare_attr_set(env, child, la, handle);
3787 rc = dt_trans_start_local(env, dev, handle);
3791 dt_write_lock(env, parent, 0);
3792 if (unlikely(lfsck_is_dead_obj(parent)))
3793 GOTO(unlock2, rc = 1);
3795 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3799 /* Get the latest parent's owner. */
3800 rc = dt_attr_get(env, parent, la);
3804 la->la_valid = LA_UID | LA_GID;
3805 rc = dt_attr_set(env, child, la, handle);
3810 dt_write_unlock(env, parent);
3813 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3816 lfsck_ibits_unlock(&lh, LCK_EX);
3820 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3821 "unmatched MDT-OST pair for: parent "DFID
3822 ", child "DFID", comp_id %u, OST-index %u, "
3823 "stripe-index %u, owner %u/%u: rc = %d\n",
3824 lfsck_lfsck2name(com->lc_lfsck),
3825 PFID(lfsck_dto2fid(parent)),
3826 PFID(lfsck_dto2fid(child)),
3827 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3828 la->la_uid, la->la_gid, rc);
3833 /* If there are more than one MDT-objects claim as the OST-object's parent,
3834 * and the OST-object only recognizes one of them, then we need to generate
3835 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3836 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3837 struct lfsck_component *com,
3838 struct dt_object *parent,
3839 struct lfsck_layout_req *llr,
3842 struct lfsck_thread_info *info = lfsck_env_info(env);
3843 struct dt_allocation_hint *hint = &info->lti_hint;
3844 struct dt_object_format *dof = &info->lti_dof;
3845 struct ost_id *oi = &info->lti_oi;
3846 struct lu_buf *buf = &info->lti_big_buf;
3847 struct lfsck_instance *lfsck = com->lc_lfsck;
3848 struct dt_device *dev;
3849 struct lu_device *d =
3850 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3851 struct lu_object *o;
3852 struct lu_object *n;
3853 struct dt_object *child = NULL;
3854 struct thandle *handle = NULL;
3855 struct lov_mds_md_v1 *lmm;
3856 struct lov_ost_data_v1 *objs;
3857 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3859 struct lustre_handle lh = { 0 };
3865 /* We use two separated transactions to repair the inconsistency.
3867 * 1) create the child (OST-object).
3868 * 2) update the parent LOV EA according to the child's FID.
3870 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3871 * handled as orphan when the layout LFSCK run next time.
3873 * If 1) failed, but 2) succeed, then such OST-object will be re-created
3874 * as dangling referened case when the layout LFSCK run next time. */
3876 /* The 1st transaction. */
3877 o = lu_object_anon(env, d, NULL);
3879 GOTO(log, rc = PTR_ERR(o));
3881 n = lu_object_locate(o->lo_header, d->ld_type);
3882 if (unlikely(n == NULL)) {
3883 lu_object_put_nocache(env, o);
3885 GOTO(log, rc = -EINVAL);
3888 child = container_of(n, struct dt_object, do_lu);
3889 memset(hint, 0, sizeof(*hint));
3890 rc = dt_attr_get(env, parent, la);
3894 la->la_valid = LA_UID | LA_GID;
3895 memset(dof, 0, sizeof(*dof));
3897 dev = lfsck_obj2dev(child);
3898 handle = lfsck_trans_create(env, dev, lfsck);
3900 GOTO(log, rc = PTR_ERR(handle));
3902 rc = dt_declare_create(env, child, la, hint, dof, handle);
3906 rc = dt_trans_start_local(env, dev, handle);
3910 rc = dt_create(env, child, la, hint, dof, handle);
3911 dt_trans_stop(env, dev, handle);
3916 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3917 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3922 /* The 2nd transaction. */
3924 /* XXX: Generally, we should use bottom device (OSD) to update parent
3925 * LOV EA. But because the LOD-object still references the wrong
3926 * OSP-object that should be detached after the parent's LOV EA
3927 * refreshed. Unfortunately, there is no suitable API for that.
3928 * So we have to make the LOD to re-load the OSP-object(s) via
3929 * replacing the LOV EA against the LOD-object.
3931 * Once the DNE2 patches have been landed, we can replace the
3932 * LOD device with the OSD device. LU-6230. */
3934 dev = lfsck->li_next;
3935 parent = lfsck_object_locate(dev, parent);
3937 GOTO(log, rc = PTR_ERR(parent));
3939 handle = lfsck_trans_create(env, dev, lfsck);
3941 GOTO(log, rc = PTR_ERR(handle));
3943 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3944 LU_XATTR_REPLACE, handle);
3948 rc = dt_trans_start_local(env, dev, handle);
3952 dt_write_lock(env, parent, 0);
3953 if (unlikely(lfsck_is_dead_obj(parent)))
3954 GOTO(unlock, rc = 0);
3956 rc = lfsck_layout_get_lovea(env, parent, buf);
3957 if (unlikely(rc == -ENODATA))
3963 magic = le32_to_cpu(lmm->lmm_magic);
3964 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
3965 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3966 struct lov_comp_md_entry_v1 *lcme;
3967 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3970 LASSERT(llr->llr_comp_id != 0);
3972 for (i = 0; i < count; i++) {
3973 lcme = &lcm->lcm_entries[i];
3974 if (le32_to_cpu(lcme->lcme_id) == llr->llr_comp_id) {
3975 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3978 le32_add_cpu(&lcm->lcm_layout_gen, 1);
3980 le32_to_cpu(lcme->lcme_offset);
3981 magic = le32_to_cpu(lmm->lmm_magic);
3986 GOTO(unlock, rc = 0);
3990 if (magic == LOV_MAGIC_V1) {
3991 objs = &lmm->lmm_objects[llr->llr_lov_idx];
3993 LASSERT(magic == LOV_MAGIC_V3);
3995 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
3998 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3999 index = le32_to_cpu(objs->l_ost_idx);
4000 rc = ostid_to_fid(&tfid, oi, index);
4001 /* Someone changed layout during the LFSCK, no need to repair then. */
4002 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
4003 GOTO(unlock, rc = 0);
4005 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
4006 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
4007 ostid_cpu_to_le(oi, &objs->l_ost_oi);
4008 objs->l_ost_gen = cpu_to_le32(0);
4009 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
4010 rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
4011 LU_XATTR_REPLACE, handle);
4013 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
4016 dt_write_unlock(env, parent);
4020 dt_trans_stop(env, dev, handle);
4023 lfsck_ibits_unlock(&lh, LCK_EX);
4025 lfsck_object_put(env, child);
4028 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
4029 "multiple references for: parent "DFID", comp_id %u, "
4030 "OST-index %u, stripe-index %u, owner %u/%u: rc = %d\n",
4031 lfsck_lfsck2name(lfsck), PFID(pfid),
4032 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
4033 la->la_uid, la->la_gid, rc);
4038 /* If the MDT-object and the OST-object have different owner information,
4039 * then trust the MDT-object, because the normal chown/chgrp handle order
4040 * is from MDT to OST, and it is possible that some chown/chgrp operation
4041 * is partly done. */
4042 static int lfsck_layout_repair_owner(const struct lu_env *env,
4043 struct lfsck_component *com,
4044 struct dt_object *parent,
4045 struct lfsck_layout_req *llr,
4046 struct lu_attr *pla,
4047 const struct lu_attr *cla)
4049 struct lfsck_thread_info *info = lfsck_env_info(env);
4050 struct lu_attr *tla = &info->lti_la2;
4051 struct dt_object *child = llr->llr_child;
4052 struct dt_device *dev = lfsck_obj2dev(child);
4053 struct thandle *handle;
4055 dt_obj_version_t version;
4058 tla->la_uid = pla->la_uid;
4059 tla->la_gid = pla->la_gid;
4060 tla->la_valid = LA_UID | LA_GID;
4061 handle = lfsck_trans_create(env, dev, com->lc_lfsck);
4063 GOTO(log, rc = PTR_ERR(handle));
4065 rc = dt_declare_attr_set(env, child, tla, handle);
4069 rc = dt_trans_start_local(env, dev, handle);
4073 /* Use the dt_object lock to serialize with destroy and attr_set. */
4074 dt_read_lock(env, parent, 0);
4075 if (unlikely(lfsck_is_dead_obj(parent)))
4076 GOTO(unlock, rc = 1);
4078 version = dt_version_get(env, child);
4079 if (version == -EOPNOTSUPP)
4082 /* Get the latest parent's owner. */
4083 rc = dt_attr_get(env, parent, pla);
4087 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
4088 if (unlikely((!version && tla->la_ctime == 0) ||
4089 tla->la_uid != pla->la_uid || tla->la_gid != pla->la_gid))
4092 rc = dt_attr_set(env, child, tla, handle);
4097 dt_read_unlock(env, parent);
4100 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
4104 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
4105 "inconsistent file owner for: parent "DFID", child "DFID
4106 ", OST-index %u, stripe-index %u, old owner %u/%u, "
4107 "new owner %u/%u: rc = %d\n",
4108 lfsck_lfsck2name(com->lc_lfsck),
4109 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
4110 llr->llr_ost_idx, llr->llr_lov_idx,
4111 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
4116 #define CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid, msg) \
4117 CDEBUG(D_LFSCK, "%s:("DFID"|"DFID")/"DFID":XATTR %s: %s\n", \
4118 lfsck_lfsck2name(lfsck), PFID(&lso->lso_fid), PFID(pfid), \
4119 PFID(cfid), XATTR_NAME_FID, msg);
4121 /* Check whether the OST-object correctly back points to the
4122 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
4123 static int lfsck_layout_check_parent(const struct lu_env *env,
4124 struct lfsck_component *com,
4125 struct lfsck_assistant_object *lso,
4126 struct filter_fid *ff,
4127 const struct lu_fid *cfid,
4128 const struct lu_attr *cla,
4129 struct lfsck_layout_req *llr)
4131 struct lfsck_thread_info *info = lfsck_env_info(env);
4132 struct lu_buf *buf = &info->lti_big_buf;
4133 struct lu_fid *pfid = &info->lti_fid;
4134 struct dt_object *tobj;
4135 struct lov_mds_md_v1 *lmm;
4136 struct lov_ost_data_v1 *objs;
4137 struct lustre_handle lh = { 0 };
4138 struct lfsck_instance *lfsck = com->lc_lfsck;
4146 *pfid = ff->ff_parent;
4147 idx = pfid->f_stripe_idx;
4150 if (unlikely(!fid_is_sane(pfid))) {
4151 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4152 "the parent FID is invalid");
4154 RETURN(LLIT_UNMATCHED_PAIR);
4157 if (lu_fid_eq(pfid, &lso->lso_fid)) {
4158 if (likely(llr->llr_lov_idx == idx))
4161 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4162 "the stripe index is unmatched");
4164 RETURN(LLIT_UNMATCHED_PAIR);
4167 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4169 RETURN(PTR_ERR(tobj));
4171 if (dt_object_exists(tobj) == 0) {
4172 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4173 "the parent is nonexistent");
4175 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4178 if (lfsck_is_dead_obj(tobj)) {
4179 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4180 "the parent is dead object");
4182 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4185 if (!S_ISREG(lfsck_object_type(tobj))) {
4186 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4187 "the parent is not a regular file");
4189 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4192 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
4193 * remote one on another MDT. Then check whether the given OST-object
4194 * is in such layout. If yes, it is multiple referenced, otherwise it
4195 * is unmatched referenced case. */
4196 rc = lfsck_layout_get_lovea(env, tobj, buf);
4197 if (rc == 0 || rc == -ENODATA || rc == -ENOENT) {
4198 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4199 "the parent has no stripe data");
4201 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4204 if (unlikely(rc == -EOPNOTSUPP))
4205 GOTO(out, rc = LLIT_NONE);
4211 magic = le32_to_cpu(lmm->lmm_magic);
4212 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
4213 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4214 struct lov_comp_md_entry_v1 *lcme;
4216 if (ff->ff_layout.ol_comp_id == 0) {
4217 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4218 "the parent has incorrect comp_id");
4220 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4223 count = le16_to_cpu(lcm->lcm_entry_count);
4224 for (i = 0; i < count; i++) {
4225 lcme = &lcm->lcm_entries[i];
4226 if (le32_to_cpu(lcme->lcme_id) ==
4227 ff->ff_layout.ol_comp_id) {
4229 le32_to_cpu(lcme->lcme_offset);
4230 magic = le32_to_cpu(lmm->lmm_magic);
4231 if (!(le32_to_cpu(lcme->lcme_flags) &
4233 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid,
4235 "the parent has uninitialized component");
4237 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4244 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4245 "the parent has no matched comp_id");
4247 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4251 if (magic == LOV_MAGIC_V1) {
4252 objs = &lmm->lmm_objects[0];
4254 LASSERT(magic == LOV_MAGIC_V3);
4255 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4258 count = le16_to_cpu(lmm->lmm_stripe_count);
4259 for (i = 0; i < count; i++, objs++) {
4260 struct lu_fid *tfid = &info->lti_fid2;
4261 struct ost_id *oi = &info->lti_oi;
4264 if (lovea_slot_is_dummy(objs))
4267 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4268 idx2 = le32_to_cpu(objs->l_ost_idx);
4269 rc = ostid_to_fid(tfid, oi, idx2);
4271 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
4272 "invalid layout EA at the slot %d, index %u\n",
4273 lfsck_lfsck2name(com->lc_lfsck),
4274 PFID(pfid), i, idx2);
4276 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4279 if (lu_fid_eq(cfid, tfid)) {
4280 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
4281 MDS_INODELOCK_UPDATE |
4282 MDS_INODELOCK_LAYOUT |
4283 MDS_INODELOCK_XATTR,
4288 dt_read_lock(env, tobj, 0);
4290 /* For local MDT-object, re-check existence
4291 * after taken the lock. */
4292 if (!dt_object_remote(tobj)) {
4293 if (dt_object_exists(tobj) == 0 ||
4294 lfsck_is_dead_obj(tobj)) {
4295 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid,
4297 "the parent doesn't exist anymore after lock");
4299 rc = LLIT_UNMATCHED_PAIR;
4301 rc = LLIT_MULTIPLE_REFERENCED;
4307 /* For migration case, the new MDT-object and old
4308 * MDT-object may reference the same OST-object at
4309 * some migration internal time.
4311 * For remote MDT-object, the local MDT may not know
4312 * whether it has been removed or not. Try checking
4313 * for a non-existent xattr to check if this object
4314 * has been been removed or not. */
4315 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
4317 if (unlikely(rc == -ENOENT || rc >= 0)) {
4318 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4319 "the parent is remote object and nonexistent after lock");
4321 rc = LLIT_UNMATCHED_PAIR;
4322 } else if (rc == -ENODATA) {
4323 rc = LLIT_MULTIPLE_REFERENCED;
4330 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4331 "the parent has no matched stripe");
4333 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4336 if (lustre_handle_is_used(&lh)) {
4337 dt_read_unlock(env, tobj);
4338 lfsck_ibits_unlock(&lh, LCK_EX);
4342 lfsck_object_put(env, tobj);
4347 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
4348 struct lfsck_component *com,
4349 struct lfsck_assistant_req *lar)
4351 struct lfsck_layout_req *llr =
4352 container_of(lar, struct lfsck_layout_req, llr_lar);
4353 struct lfsck_assistant_object *lso = lar->lar_parent;
4354 struct lfsck_layout *lo = com->lc_file_ram;
4355 struct lfsck_thread_info *info = lfsck_env_info(env);
4356 struct filter_fid *ff = &info->lti_ff;
4357 struct lu_buf buf = { .lb_buf = ff,
4358 .lb_len = sizeof(*ff) };
4359 struct dt_object *parent = NULL;
4360 struct dt_object *child = llr->llr_child;
4361 struct lu_attr *pla = &lso->lso_attr;
4362 struct lu_attr *cla = &info->lti_la;
4363 struct lfsck_instance *lfsck = com->lc_lfsck;
4364 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4365 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
4372 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ENGINE_DELAY, cfs_fail_val);
4374 rc = dt_attr_get(env, child, cla);
4375 if (rc == -ENOENT) {
4376 parent = lfsck_assistant_object_load(env, lfsck, lso);
4377 if (IS_ERR(parent)) {
4378 rc = PTR_ERR(parent);
4380 RETURN(rc == -ENOENT ? 0 : rc);
4383 type = LLIT_DANGLING;
4390 lfsck_buf_init(&buf, ff, sizeof(*ff));
4391 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
4392 if (unlikely(rc > 0 && rc < sizeof(struct lu_fid))) {
4393 CDEBUG(D_LFSCK, "%s:"DFID"/"DFID": "
4394 "the child object's %s is corrupted\n",
4395 lfsck_lfsck2name(lfsck), PFID(&lso->lso_fid),
4396 PFID(lu_object_fid(&child->do_lu)),
4399 type = LLIT_UNMATCHED_PAIR;
4403 if (rc < 0 && rc != -ENODATA)
4406 if (rc == 0 || rc == -ENODATA)
4407 GOTO(check_owner, rc = 0);
4409 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
4410 rc = lfsck_layout_check_parent(env, com, lso, ff,
4411 lu_object_fid(&child->do_lu), cla, llr);
4421 /* Someone may has changed the owner after the parent attr pre-loaded.
4422 * It can be handled later inside the lfsck_layout_repair_owner(). */
4423 if (unlikely(cla->la_uid != pla->la_uid ||
4424 cla->la_gid != pla->la_gid)) {
4425 type = LLIT_INCONSISTENT_OWNER;
4430 if (type == LLIT_NONE)
4433 if (bk->lb_param & LPF_DRYRUN)
4436 if (parent == NULL) {
4437 parent = lfsck_assistant_object_load(env, lfsck, lso);
4438 if (IS_ERR(parent)) {
4439 rc = PTR_ERR(parent);
4450 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
4451 rc = lfsck_layout_ins_dangling_rec(env, com,
4452 lfsck_dto2fid(parent), lfsck_dto2fid(child),
4453 llr->llr_comp_id, llr->llr_lov_idx,
4456 rc = __lfsck_layout_repair_dangling(env, com, parent,
4463 case LLIT_UNMATCHED_PAIR:
4464 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
4467 case LLIT_MULTIPLE_REFERENCED:
4468 rc = lfsck_layout_repair_multiple_references(env, com, parent,
4471 case LLIT_INCONSISTENT_OWNER:
4472 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
4482 down_write(&com->lc_sem);
4484 struct lfsck_assistant_data *lad = com->lc_data;
4486 if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags))) {
4488 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
4489 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
4490 rc == -EHOSTUNREACH) {
4491 /* If cannot touch the target server,
4492 * mark the LFSCK as INCOMPLETE. */
4493 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
4494 "talk with OST %x: rc = %d\n",
4495 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
4496 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
4497 lo->ll_objs_skipped++;
4500 lfsck_layout_record_failure(env, lfsck, lo);
4502 } else if (rc > 0 && (type != LLIT_DANGLING ||
4503 !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
4504 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
4505 "unknown type = %d\n", type);
4507 lo->ll_objs_repaired[type - 1]++;
4508 if (bk->lb_param & LPF_DRYRUN &&
4509 unlikely(lo->ll_pos_first_inconsistent == 0))
4510 lo->ll_pos_first_inconsistent =
4511 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
4514 up_write(&com->lc_sem);
4516 if (parent != NULL && !IS_ERR(parent))
4517 lfsck_object_put(env, parent);
4523 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
4524 struct lfsck_component *com,
4525 struct dt_object *obj, bool first)
4527 struct lfsck_instance *lfsck = com->lc_lfsck;
4528 struct ptlrpc_thread *thread = &lfsck->li_thread;
4529 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4530 struct lfsck_layout *lo = com->lc_file_ram;
4531 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
4534 struct lfsck_layout_dangling_key *parent =
4535 &lfsck_env_info(env)->lti_lldk;
4536 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
4541 di = iops->init(env, obj, 0);
4543 RETURN(PTR_ERR(di));
4546 lldk_cpu_to_be(parent, &lo->ll_lldk_latest_scanned_phase2);
4548 memset(parent, 0, sizeof(*parent));
4549 rc = iops->get(env, di, (const struct dt_key *)parent);
4554 /* The start one either has been processed or does not exist,
4556 rc = iops->next(env, di);
4562 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
4563 unlikely(!thread_is_running(thread)))
4566 key = iops->key(env, di);
4575 lldk_be_to_cpu(parent,
4576 (const struct lfsck_layout_dangling_key *)key);
4577 if (!fid_is_sane(&parent->lldk_fid)) {
4582 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
4584 fid_be_to_cpu(cfid, cfid);
4585 ost_idx = cfid->f_ver;
4587 if (!fid_is_sane(cfid)) {
4592 rc = lfsck_layout_repair_dangling(env, com,
4593 &parent->lldk_fid, cfid,
4594 parent->lldk_comp_id,
4595 parent->lldk_ea_off, ost_idx);
4599 down_write(&com->lc_sem);
4600 com->lc_new_checked++;
4601 com->lc_new_scanned++;
4603 lo->ll_lldk_latest_scanned_phase2 = *parent;
4606 lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
4608 lo->ll_objs_failed_phase2++;
4609 up_write(&com->lc_sem);
4611 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4614 if (unlikely(com->lc_time_next_checkpoint <=
4615 ktime_get_seconds()) &&
4616 com->lc_new_checked != 0) {
4617 down_write(&com->lc_sem);
4618 lo->ll_run_time_phase2 += ktime_get_seconds() -
4619 com->lc_time_last_checkpoint;
4620 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
4621 lo->ll_objs_checked_phase2 += com->lc_new_checked;
4622 com->lc_new_checked = 0;
4623 lfsck_layout_store(env, com);
4624 up_write(&com->lc_sem);
4626 com->lc_time_last_checkpoint = ktime_get_seconds();
4627 com->lc_time_next_checkpoint =
4628 com->lc_time_last_checkpoint +
4629 LFSCK_CHECKPOINT_INTERVAL;
4632 lfsck_control_speed_by_self(com);
4633 if (unlikely(!thread_is_running(thread)))
4636 rc = iops->next(env, di);
4645 iops->fini(env, di);
4650 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
4651 struct lfsck_component *com)
4653 struct lfsck_assistant_data *lad = com->lc_data;
4654 struct lfsck_instance *lfsck = com->lc_lfsck;
4655 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4656 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4657 struct lfsck_tgt_desc *ltd;
4661 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
4662 lfsck_lfsck2name(lfsck));
4664 spin_lock(<ds->ltd_lock);
4665 while (!list_empty(&lad->lad_ost_phase2_list)) {
4666 ltd = list_entry(lad->lad_ost_phase2_list.next,
4667 struct lfsck_tgt_desc,
4668 ltd_layout_phase_list);
4669 list_del_init(<d->ltd_layout_phase_list);
4670 if (bk->lb_param & LPF_OST_ORPHAN) {
4671 spin_unlock(<ds->ltd_lock);
4672 rc = lfsck_layout_scan_orphan(env, com, ltd);
4673 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
4676 if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags) ||
4677 !thread_is_running(&lfsck->li_thread)))
4679 spin_lock(<ds->ltd_lock);
4683 if (list_empty(&lad->lad_ost_phase1_list))
4687 spin_unlock(<ds->ltd_lock);
4689 if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
4690 struct lfsck_layout *lo = com->lc_file_ram;
4693 com->lc_new_checked = 0;
4694 com->lc_new_scanned = 0;
4695 com->lc_time_last_checkpoint = ktime_get_seconds();
4696 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4697 LFSCK_CHECKPOINT_INTERVAL;
4699 i = lfsck_sub_trace_file_fid2idx(
4700 &lo->ll_lldk_latest_scanned_phase2.lldk_fid);
4701 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4702 com->lc_sub_trace_objs[i].lsto_obj, true);
4703 while (rc > 0 && ++i < LFSCK_STF_COUNT)
4704 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4705 com->lc_sub_trace_objs[i].lsto_obj, false);
4707 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
4708 "at the No. %d trace file: rc = %d\n",
4709 lfsck_lfsck2name(lfsck), i, rc);
4712 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
4713 lfsck_lfsck2name(lfsck), rc);
4719 lfsck_layout_slave_async_interpret(const struct lu_env *env,
4720 struct ptlrpc_request *req,
4723 struct lfsck_layout_slave_async_args *llsaa = args;
4724 struct obd_export *exp = llsaa->llsaa_exp;
4725 struct lfsck_component *com = llsaa->llsaa_com;
4726 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
4727 struct lfsck_layout_slave_data *llsd = com->lc_data;
4728 struct lfsck_reply *lr = NULL;
4732 /* It is probably caused by network trouble, or target crash,
4733 * it will try several times (depends on the obd_timeout, and
4734 * will not less than 3 times). But to make the LFSCK can go
4735 * ahead, we should not try for ever. After some try but still
4736 * hit failure, it will assume that the target exit the LFSCK
4737 * prcoessing and stop try. */
4738 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
4739 int max_try = max_t(int, obd_timeout / 30, 3);
4741 if (++(llst->llst_failures) > max_try)
4747 llst->llst_failures = 0;
4748 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
4749 if (lr->lr_status != LS_SCANNING_PHASE1 &&
4750 lr->lr_status != LS_SCANNING_PHASE2)
4755 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
4756 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
4757 llst->llst_index, lr != NULL ? lr->lr_status : rc,
4758 llst->llst_failures);
4760 lfsck_layout_llst_del(llsd, llst);
4763 lfsck_layout_llst_put(llst);
4764 lfsck_component_put(env, com);
4765 class_export_put(exp);
4770 static int lfsck_layout_async_query(const struct lu_env *env,
4771 struct lfsck_component *com,
4772 struct obd_export *exp,
4773 struct lfsck_layout_slave_target *llst,
4774 struct lfsck_request *lr,
4775 struct ptlrpc_request_set *set)
4777 struct lfsck_layout_slave_async_args *llsaa;
4778 struct ptlrpc_request *req;
4779 struct lfsck_request *tmp;
4783 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
4787 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
4789 ptlrpc_request_free(req);
4793 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4795 ptlrpc_request_set_replen(req);
4797 llsaa = ptlrpc_req_async_args(llsaa, req);
4798 llsaa->llsaa_exp = exp;
4799 llsaa->llsaa_com = lfsck_component_get(com);
4800 llsaa->llsaa_llst = llst;
4801 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
4802 req->rq_allow_intr = 1;
4803 req->rq_no_delay = 1;
4804 ptlrpc_set_add_req(set, req);
4809 static int lfsck_layout_async_notify(const struct lu_env *env,
4810 struct obd_export *exp,
4811 struct lfsck_request *lr,
4812 struct ptlrpc_request_set *set)
4814 struct ptlrpc_request *req;
4815 struct lfsck_request *tmp;
4819 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4823 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4825 ptlrpc_request_free(req);
4829 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4831 ptlrpc_request_set_replen(req);
4832 req->rq_allow_intr = 1;
4833 req->rq_no_delay = 1;
4834 ptlrpc_set_add_req(set, req);
4840 lfsck_layout_slave_query_master(const struct lu_env *env,
4841 struct lfsck_component *com)
4843 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4844 struct lfsck_instance *lfsck = com->lc_lfsck;
4845 struct lfsck_layout_slave_data *llsd = com->lc_data;
4846 struct lfsck_layout_slave_target *llst;
4847 struct obd_export *exp;
4848 struct ptlrpc_request_set *set;
4853 set = ptlrpc_prep_set();
4855 GOTO(log, rc = -ENOMEM);
4857 memset(lr, 0, sizeof(*lr));
4858 lr->lr_event = LE_QUERY;
4859 lr->lr_active = LFSCK_TYPE_LAYOUT;
4861 llsd->llsd_touch_gen++;
4862 spin_lock(&llsd->llsd_lock);
4863 while (!list_empty(&llsd->llsd_master_list)) {
4864 llst = list_entry(llsd->llsd_master_list.next,
4865 struct lfsck_layout_slave_target,
4867 if (llst->llst_gen == llsd->llsd_touch_gen)
4870 llst->llst_gen = llsd->llsd_touch_gen;
4871 list_move_tail(&llst->llst_list,
4872 &llsd->llsd_master_list);
4873 atomic_inc(&llst->llst_ref);
4874 spin_unlock(&llsd->llsd_lock);
4876 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4879 lfsck_layout_llst_del(llsd, llst);
4880 lfsck_layout_llst_put(llst);
4881 spin_lock(&llsd->llsd_lock);
4885 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
4887 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4888 "query %s for layout: rc = %d\n",
4889 lfsck_lfsck2name(lfsck),
4890 exp->exp_obd->obd_name, rc);
4893 lfsck_layout_llst_put(llst);
4894 class_export_put(exp);
4896 spin_lock(&llsd->llsd_lock);
4898 spin_unlock(&llsd->llsd_lock);
4900 rc = ptlrpc_set_wait(env, set);
4901 ptlrpc_set_destroy(set);
4903 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
4906 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
4907 lfsck_lfsck2name(com->lc_lfsck), rc);
4913 lfsck_layout_slave_notify_master(const struct lu_env *env,
4914 struct lfsck_component *com,
4915 enum lfsck_events event, int result)
4917 struct lfsck_layout *lo = com->lc_file_ram;
4918 struct lfsck_instance *lfsck = com->lc_lfsck;
4919 struct lfsck_layout_slave_data *llsd = com->lc_data;
4920 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4921 struct lfsck_layout_slave_target *llst;
4922 struct obd_export *exp;
4923 struct ptlrpc_request_set *set;
4927 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
4928 lfsck_lfsck2name(com->lc_lfsck));
4930 set = ptlrpc_prep_set();
4934 memset(lr, 0, sizeof(*lr));
4935 lr->lr_event = event;
4936 lr->lr_flags = LEF_FROM_OST;
4937 lr->lr_status = result;
4938 lr->lr_index = lfsck_dev_idx(lfsck);
4939 lr->lr_active = LFSCK_TYPE_LAYOUT;
4940 lr->lr_flags2 = lo->ll_flags;
4941 llsd->llsd_touch_gen++;
4942 spin_lock(&llsd->llsd_lock);
4943 while (!list_empty(&llsd->llsd_master_list)) {
4944 llst = list_entry(llsd->llsd_master_list.next,
4945 struct lfsck_layout_slave_target,
4947 if (llst->llst_gen == llsd->llsd_touch_gen)
4950 llst->llst_gen = llsd->llsd_touch_gen;
4951 list_move_tail(&llst->llst_list,
4952 &llsd->llsd_master_list);
4953 atomic_inc(&llst->llst_ref);
4954 spin_unlock(&llsd->llsd_lock);
4956 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4959 lfsck_layout_llst_del(llsd, llst);
4960 lfsck_layout_llst_put(llst);
4961 spin_lock(&llsd->llsd_lock);
4965 rc = lfsck_layout_async_notify(env, exp, lr, set);
4967 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
4968 "notify %s for layout: rc = %d\n",
4969 lfsck_lfsck2name(lfsck),
4970 exp->exp_obd->obd_name, rc);
4972 lfsck_layout_llst_put(llst);
4973 class_export_put(exp);
4974 spin_lock(&llsd->llsd_lock);
4976 spin_unlock(&llsd->llsd_lock);
4978 ptlrpc_set_wait(env, set);
4979 ptlrpc_set_destroy(set);
4985 * \ret -ENODATA: unrecognized stripe
4986 * \ret = 0 : recognized stripe
4987 * \ret < 0 : other failures
4989 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
4990 struct lfsck_component *com,
4991 struct lu_fid *cfid,
4992 struct lu_fid *pfid, __u32 comp_id)
4994 struct lfsck_thread_info *info = lfsck_env_info(env);
4995 struct lu_buf *buf = &info->lti_big_buf;
4996 struct ost_id *oi = &info->lti_oi;
4997 struct dt_object *obj;
4998 struct lov_mds_md_v1 *lmm;
4999 struct lov_ost_data_v1 *objs;
5000 __u32 idx = pfid->f_stripe_idx;
5008 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
5010 RETURN(PTR_ERR(obj));
5012 dt_read_lock(env, obj, 0);
5013 if (unlikely(dt_object_exists(obj) == 0 ||
5014 lfsck_is_dead_obj(obj)))
5015 GOTO(unlock, rc = -ENOENT);
5017 if (!S_ISREG(lfsck_object_type(obj)))
5018 GOTO(unlock, rc = -ENODATA);
5020 rc = lfsck_layout_get_lovea(env, obj, buf);
5025 magic = le32_to_cpu(lmm->lmm_magic);
5026 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
5027 struct lov_comp_md_v1 *lcm = buf->lb_buf;
5028 struct lov_comp_md_entry_v1 *lcme;
5031 GOTO(unlock, rc = -ENODATA);
5033 count = le16_to_cpu(lcm->lcm_entry_count);
5034 for (i = 0; i < count; i++) {
5035 lcme = &lcm->lcm_entries[i];
5036 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
5038 le32_to_cpu(lcme->lcme_offset);
5039 magic = le32_to_cpu(lmm->lmm_magic);
5040 if (!(le32_to_cpu(lcme->lcme_flags) &
5042 GOTO(unlock, rc = -ENODATA);
5048 GOTO(unlock, rc = -ENODATA);
5052 if (magic == LOV_MAGIC_V1) {
5053 objs = &lmm->lmm_objects[0];
5055 LASSERT(magic == LOV_MAGIC_V3);
5056 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5059 fid_to_ostid(cfid, oi);
5060 count = le16_to_cpu(lmm->lmm_stripe_count);
5061 for (i = 0; i < count; i++, objs++) {
5064 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
5065 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
5066 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
5069 GOTO(unlock, rc = -ENODATA);
5072 dt_read_unlock(env, obj);
5073 lfsck_object_put(env, obj);
5079 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
5080 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
5081 * layout EA from MDT to OST. On one hand, the OST no need to understand
5082 * the layout EA structure; on the other hand, it may cause trouble when
5083 * transfer large layout EA from MDT to OST via normal OUT RPC.
5085 * \ret > 0: unrecognized stripe
5086 * \ret = 0: recognized stripe
5087 * \ret < 0: other failures
5089 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
5090 struct lfsck_component *com,
5091 struct lu_fid *cfid,
5092 struct lu_fid *pfid, __u32 comp_id)
5094 struct lfsck_instance *lfsck = com->lc_lfsck;
5095 struct obd_device *obd = lfsck->li_obd;
5096 struct seq_server_site *ss = lfsck_dev_site(lfsck);
5097 struct obd_export *exp = NULL;
5098 struct ptlrpc_request *req = NULL;
5099 struct lfsck_request *lr;
5100 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
5104 if (unlikely(fid_is_idif(pfid)))
5107 fld_range_set_any(range);
5108 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
5110 RETURN(rc == -ENOENT ? 1 : rc);
5112 if (unlikely(!fld_range_is_mdt(range)))
5115 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
5116 if (unlikely(exp == NULL))
5119 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
5120 GOTO(out, rc = -EOPNOTSUPP);
5122 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
5124 GOTO(out, rc = -ENOMEM);
5126 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
5128 ptlrpc_request_free(req);
5133 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
5134 memset(lr, 0, sizeof(*lr));
5135 lr->lr_event = LE_PAIRS_VERIFY;
5136 lr->lr_active = LFSCK_TYPE_LAYOUT;
5137 lr->lr_fid = *cfid; /* OST-object itself FID. */
5138 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
5139 lr->lr_comp_id = comp_id;
5141 ptlrpc_request_set_replen(req);
5142 rc = ptlrpc_queue_wait(req);
5143 ptlrpc_req_finished(req);
5145 if (rc == -ENOENT || rc == -ENODATA)
5152 class_export_put(exp);
5157 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
5158 struct lfsck_component *com,
5159 struct lfsck_req_local *lrl)
5161 struct dt_object *obj;
5165 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lrl->lrl_fid);
5167 GOTO(log, rc = PTR_ERR(obj));
5169 dt_write_lock(env, obj, 0);
5170 if (unlikely(dt_object_exists(obj) == 0 ||
5171 lfsck_is_dead_obj(obj)))
5172 GOTO(unlock, rc = 0);
5174 rc = __lfsck_layout_update_pfid(env, com, obj,
5175 &lrl->lrl_ff_client.ff_parent,
5176 &lrl->lrl_ff_client.ff_layout,
5177 lrl->lrl_ff_client.ff_layout_version,
5178 lrl->lrl_ff_client.ff_range,
5179 lrl->lrl_ff_client.ff_parent.f_ver);
5184 dt_write_unlock(env, obj);
5185 lfsck_object_put(env, obj);
5188 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
5189 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
5190 PFID(&lrl->lrl_fid), PFID(&lrl->lrl_ff_client.ff_parent), rc);
5197 static void lfsck_layout_slave_quit(const struct lu_env *env,
5198 struct lfsck_component *com);
5200 static int lfsck_layout_reset(const struct lu_env *env,
5201 struct lfsck_component *com, bool init)
5203 struct lfsck_layout *lo = com->lc_file_ram;
5206 down_write(&com->lc_sem);
5208 memset(lo, 0, com->lc_file_size);
5210 __u32 count = lo->ll_success_count;
5211 time64_t last_time = lo->ll_time_last_complete;
5213 memset(lo, 0, com->lc_file_size);
5214 lo->ll_success_count = count;
5215 lo->ll_time_last_complete = last_time;
5218 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
5219 lo->ll_status = LS_INIT;
5221 if (com->lc_lfsck->li_master) {
5222 struct lfsck_assistant_data *lad = com->lc_data;
5224 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
5225 bitmap_zero(lad->lad_bitmap, lad->lad_bitmap_count);
5228 rc = lfsck_layout_store(env, com);
5229 if (rc == 0 && com->lc_lfsck->li_master)
5230 rc = lfsck_load_sub_trace_files(env, com,
5231 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
5232 up_write(&com->lc_sem);
5234 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
5235 lfsck_lfsck2name(com->lc_lfsck), rc);
5240 static void lfsck_layout_fail(const struct lu_env *env,
5241 struct lfsck_component *com, bool new_checked)
5243 struct lfsck_layout *lo = com->lc_file_ram;
5245 down_write(&com->lc_sem);
5247 com->lc_new_checked++;
5248 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
5249 up_write(&com->lc_sem);
5252 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
5253 struct lfsck_component *com, bool init)
5255 struct lfsck_instance *lfsck = com->lc_lfsck;
5256 struct lfsck_layout *lo = com->lc_file_ram;
5260 rc = lfsck_checkpoint_generic(env, com);
5262 return rc > 0 ? 0 : rc;
5265 down_write(&com->lc_sem);
5267 lo->ll_pos_latest_start =
5268 lfsck->li_pos_checkpoint.lp_oit_cookie;
5270 lo->ll_pos_last_checkpoint =
5271 lfsck->li_pos_checkpoint.lp_oit_cookie;
5272 lo->ll_run_time_phase1 += ktime_get_seconds() -
5273 lfsck->li_time_last_checkpoint;
5274 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5275 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5276 com->lc_new_checked = 0;
5279 rc = lfsck_layout_store(env, com);
5280 up_write(&com->lc_sem);
5282 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
5283 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5284 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5289 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
5290 struct lfsck_component *com, bool init)
5292 struct lfsck_instance *lfsck = com->lc_lfsck;
5293 struct lfsck_layout *lo = com->lc_file_ram;
5296 if (com->lc_new_checked == 0 && !init)
5299 down_write(&com->lc_sem);
5301 lo->ll_pos_latest_start =
5302 lfsck->li_pos_checkpoint.lp_oit_cookie;
5304 lo->ll_pos_last_checkpoint =
5305 lfsck->li_pos_checkpoint.lp_oit_cookie;
5306 lo->ll_run_time_phase1 += ktime_get_seconds() -
5307 lfsck->li_time_last_checkpoint;
5308 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5309 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5310 com->lc_new_checked = 0;
5313 rc = lfsck_layout_store(env, com);
5314 up_write(&com->lc_sem);
5316 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
5317 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5318 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5323 static int lfsck_layout_prep(const struct lu_env *env,
5324 struct lfsck_component *com,
5325 struct lfsck_start *start)
5327 struct lfsck_instance *lfsck = com->lc_lfsck;
5328 struct lfsck_layout *lo = com->lc_file_ram;
5329 struct lfsck_position *pos = &com->lc_pos_start;
5331 fid_zero(&pos->lp_dir_parent);
5332 pos->lp_dir_cookie = 0;
5333 if (lo->ll_status == LS_COMPLETED ||
5334 lo->ll_status == LS_PARTIAL ||
5335 /* To handle orphan, must scan from the beginning. */
5336 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
5339 rc = lfsck_layout_reset(env, com, false);
5341 rc = lfsck_set_param(env, lfsck, start, true);
5344 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
5345 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
5351 down_write(&com->lc_sem);
5352 lo->ll_time_latest_start = ktime_get_real_seconds();
5353 spin_lock(&lfsck->li_lock);
5354 if (lo->ll_flags & LF_SCANNED_ONCE) {
5355 if (!lfsck->li_drop_dryrun ||
5356 lo->ll_pos_first_inconsistent == 0) {
5357 lo->ll_status = LS_SCANNING_PHASE2;
5358 list_move_tail(&com->lc_link,
5359 &lfsck->li_list_double_scan);
5360 pos->lp_oit_cookie = 0;
5364 lo->ll_status = LS_SCANNING_PHASE1;
5365 lo->ll_run_time_phase1 = 0;
5366 lo->ll_run_time_phase2 = 0;
5367 lo->ll_objs_checked_phase1 = 0;
5368 lo->ll_objs_checked_phase2 = 0;
5369 lo->ll_objs_failed_phase1 = 0;
5370 lo->ll_objs_failed_phase2 = 0;
5371 for (i = 0; i < LLIT_MAX; i++)
5372 lo->ll_objs_repaired[i] = 0;
5374 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5375 fid_zero(&com->lc_fid_latest_scanned_phase2);
5378 lo->ll_status = LS_SCANNING_PHASE1;
5379 if (!lfsck->li_drop_dryrun ||
5380 lo->ll_pos_first_inconsistent == 0)
5381 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
5383 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5385 spin_unlock(&lfsck->li_lock);
5386 up_write(&com->lc_sem);
5391 static int lfsck_layout_slave_prep(const struct lu_env *env,
5392 struct lfsck_component *com,
5393 struct lfsck_start_param *lsp)
5395 struct lfsck_layout_slave_data *llsd = com->lc_data;
5396 struct lfsck_instance *lfsck = com->lc_lfsck;
5397 struct lfsck_layout *lo = com->lc_file_ram;
5398 struct lfsck_start *start = lsp->lsp_start;
5401 rc = lfsck_layout_prep(env, com, start);
5405 if (lo->ll_flags & LF_CRASHED_LASTID &&
5406 list_empty(&llsd->llsd_master_list)) {
5407 LASSERT(lfsck->li_out_notify != NULL);
5409 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5410 LE_LASTID_REBUILDING);
5413 if (!lsp->lsp_index_valid)
5416 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5417 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
5418 LASSERT(!llsd->llsd_rbtree_valid);
5420 down_write(&llsd->llsd_rb_rwsem);
5421 rc = lfsck_rbtree_setup(env, com);
5422 up_write(&llsd->llsd_rb_rwsem);
5425 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
5426 "%llu]\n", lfsck_lfsck2name(lfsck),
5427 com->lc_pos_start.lp_oit_cookie);
5432 static int lfsck_layout_master_prep(const struct lu_env *env,
5433 struct lfsck_component *com,
5434 struct lfsck_start_param *lsp)
5439 rc = lfsck_layout_load_bitmap(env, com);
5441 rc = lfsck_layout_reset(env, com, false);
5443 rc = lfsck_set_param(env, com->lc_lfsck,
5444 lsp->lsp_start, true);
5450 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
5454 rc = lfsck_start_assistant(env, com, lsp);
5459 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
5460 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
5461 com->lc_pos_start.lp_oit_cookie);
5466 /* Pre-fetch the attribute for each stripe in the given layout EA. */
5467 static int lfsck_layout_scan_stripes(const struct lu_env *env,
5468 struct lfsck_component *com,
5469 struct dt_object *parent,
5470 struct lov_mds_md_v1 *lmm, __u32 comp_id)
5472 struct lfsck_thread_info *info = lfsck_env_info(env);
5473 struct lfsck_instance *lfsck = com->lc_lfsck;
5474 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5475 struct lfsck_layout *lo = com->lc_file_ram;
5476 struct lfsck_assistant_data *lad = com->lc_data;
5477 struct lfsck_assistant_object *lso = NULL;
5478 struct lov_ost_data_v1 *objs;
5479 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
5480 struct ptlrpc_thread *mthread = &lfsck->li_thread;
5481 struct ptlrpc_thread *athread = &lad->lad_thread;
5489 lfsck_buf_init(&buf, &info->lti_ff, sizeof(struct filter_fid));
5490 magic = le32_to_cpu(lmm->lmm_magic);
5491 if (magic == LOV_MAGIC_V1) {
5492 objs = &lmm->lmm_objects[0];
5494 LASSERT(magic == LOV_MAGIC_V3);
5495 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5498 count = le16_to_cpu(lmm->lmm_stripe_count);
5499 for (i = 0; i < count; i++, objs++) {
5500 struct lu_fid *fid = &info->lti_fid;
5501 struct ost_id *oi = &info->lti_oi;
5502 struct lfsck_layout_req *llr;
5503 struct lfsck_tgt_desc *tgt = NULL;
5504 struct dt_object *cobj = NULL;
5506 bool wakeup = false;
5508 if (unlikely(lovea_slot_is_dummy(objs)))
5511 wait_event_idle(mthread->t_ctl_waitq,
5512 lad->lad_prefetched < bk->lb_async_windows ||
5513 !thread_is_running(mthread) ||
5514 thread_is_stopped(athread));
5516 if (unlikely(!thread_is_running(mthread)) ||
5517 thread_is_stopped(athread))
5520 if (unlikely(lfsck_is_dead_obj(parent)))
5523 ostid_le_to_cpu(&objs->l_ost_oi, oi);
5524 index = le32_to_cpu(objs->l_ost_idx);
5525 rc = ostid_to_fid(fid, oi, index);
5527 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
5528 ": "DOSTID", idx %u, comp_id %u\n",
5529 lfsck_lfsck2name(lfsck),
5530 PFID(lfsck_dto2fid(parent)), POSTID(oi),
5535 tgt = lfsck_tgt_get(ltds, index);
5536 if (unlikely(tgt == NULL)) {
5537 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
5538 "did not join the layout LFSCK, comp_id %u\n",
5539 lfsck_lfsck2name(lfsck), index, comp_id);
5540 lfsck_lad_set_bitmap(env, com, index);
5544 /* There is potential deadlock race condition between object
5545 * destroy and layout LFSCK. Consider the following scenario:
5547 * 1) The LFSCK thread obtained the parent object firstly, at
5548 * that time, the parent object has not been destroyed yet.
5550 * 2) One RPC service thread destroyed the parent and all its
5551 * children objects. Because the LFSCK is referencing the
5552 * parent object, then the parent object will be marked as
5553 * dying in RAM. On the other hand, the parent object is
5554 * referencing all its children objects, then all children
5555 * objects will be marked as dying in RAM also.
5557 * 3) The LFSCK thread tries to find some child object with
5558 * the parent object referenced. Then it will find that the
5559 * child object is dying. According to the object visibility
5560 * rules: the object with dying flag cannot be returned to
5561 * others. So the LFSCK thread has to wait until the dying
5562 * object has been purged from RAM, then it can allocate a
5563 * new object (with the same FID) in RAM. Unfortunately, the
5564 * LFSCK thread itself is referencing the parent object, and
5565 * cause the parent object cannot be purged, then cause the
5566 * child object cannot be purged also. So the LFSCK thread
5567 * will fall into deadlock.
5569 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
5571 if (lfsck_is_dead_obj(parent)) {
5581 rc = dt_declare_attr_get(env, cobj);
5585 rc = dt_declare_xattr_get(env, cobj, &buf, XATTR_NAME_FID);
5590 struct lu_attr *attr = &info->lti_la;
5592 rc = dt_attr_get(env, parent, attr);
5596 lso = lfsck_assistant_object_init(env,
5597 lfsck_dto2fid(parent), attr,
5598 lfsck->li_pos_current.lp_oit_cookie, false);
5607 llr = lfsck_layout_assistant_req_init(lso, cobj, comp_id,
5615 spin_lock(&lad->lad_lock);
5616 if (lad->lad_assistant_status < 0) {
5617 spin_unlock(&lad->lad_lock);
5618 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
5620 RETURN(lad->lad_assistant_status);
5623 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
5624 if (lad->lad_prefetched == 0)
5627 lad->lad_prefetched++;
5628 spin_unlock(&lad->lad_lock);
5630 wake_up(&athread->t_ctl_waitq);
5633 down_write(&com->lc_sem);
5634 com->lc_new_checked++;
5636 lfsck_layout_record_failure(env, lfsck, lo);
5637 up_write(&com->lc_sem);
5639 if (cobj != NULL && !IS_ERR(cobj))
5640 lfsck_object_put(env, cobj);
5642 if (likely(tgt != NULL))
5645 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5653 lfsck_assistant_object_put(env, lso);
5658 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
5659 * the OST-object's attribute and generate an structure lfsck_layout_req on the
5660 * list ::lad_req_list.
5662 * For each request on above list, the lfsck_layout_assistant thread compares
5663 * the OST side attribute with local attribute, if inconsistent, then repair it.
5665 * All above processing is async mode with pipeline. */
5666 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
5667 struct lfsck_component *com,
5668 struct dt_object *obj)
5670 struct lfsck_thread_info *info = lfsck_env_info(env);
5671 struct ost_id *oi = &info->lti_oi;
5672 struct lfsck_layout *lo = com->lc_file_ram;
5673 struct lfsck_assistant_data *lad = com->lc_data;
5674 struct lfsck_instance *lfsck = com->lc_lfsck;
5675 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5676 struct thandle *handle = NULL;
5677 struct lu_buf *buf = &info->lti_big_buf;
5678 struct lov_mds_md_v1 *lmm = NULL;
5679 struct dt_device *dev = lfsck_obj2dev(obj);
5680 struct lustre_handle lh = { 0 };
5681 struct lu_buf ea_buf = { NULL };
5682 struct lov_comp_md_v1 *lcm = NULL;
5683 struct lov_comp_md_entry_v1 *lcme = NULL;
5688 bool locked = false;
5689 bool stripe = false;
5690 bool bad_oi = false;
5693 if (!S_ISREG(lfsck_object_type(obj)))
5696 if (lad->lad_assistant_status < 0)
5697 GOTO(out, rc = -ESRCH);
5699 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
5700 lmm_oi_cpu_to_le(oi, oi);
5701 dt_read_lock(env, obj, 0);
5706 if (dt_object_exists(obj) == 0 ||
5707 lfsck_is_dead_obj(obj))
5710 rc = lfsck_layout_get_lovea(env, obj, buf);
5711 if (rc == -EINVAL || rc == -ENODATA || rc == -EOPNOTSUPP)
5712 /* Skip bad lov EA during the 1st cycle scanning, and
5713 * try to recover it via orphan in the 2nd scanning. */
5720 magic = le32_to_cpu(lmm->lmm_magic);
5721 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
5722 struct lov_mds_md_v1 *v1;
5726 count = le16_to_cpu(lcm->lcm_entry_count);
5727 for (i = 0; i < count; i++) {
5728 lcme = &lcm->lcm_entries[i];
5729 v1 = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5730 if (memcmp(oi, &v1->lmm_oi, sizeof(*oi)) != 0)
5734 GOTO(out, stripe = true);
5735 } else if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0) {
5736 GOTO(out, stripe = true);
5740 /* Inconsistent lmm_oi, should be repaired. */
5743 if (bk->lb_param & LPF_DRYRUN) {
5744 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5746 GOTO(out, stripe = true);
5749 if (!lustre_handle_is_used(&lh)) {
5750 dt_read_unlock(env, obj);
5752 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5753 MDS_INODELOCK_LAYOUT |
5754 MDS_INODELOCK_XATTR, LCK_EX);
5758 handle = lfsck_trans_create(env, dev, lfsck);
5760 GOTO(out, rc = PTR_ERR(handle));
5762 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
5763 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5764 LU_XATTR_REPLACE, handle);
5768 rc = dt_trans_start_local(env, dev, handle);
5772 dt_write_lock(env, obj, 0);
5778 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
5779 struct lov_mds_md_v1 *v1;
5782 for (i = 0; i < count; i++) {
5783 lcme = &lcm->lcm_entries[i];
5784 v1 = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5791 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5792 LU_XATTR_REPLACE, handle);
5796 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5798 GOTO(out, stripe = true);
5802 if (lustre_handle_is_used(&lh))
5803 dt_write_unlock(env, obj);
5805 dt_read_unlock(env, obj);
5808 if (handle != NULL && !IS_ERR(handle))
5809 dt_trans_stop(env, dev, handle);
5811 lfsck_ibits_unlock(&lh, LCK_EX);
5814 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
5815 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
5816 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
5817 PFID(lfsck_dto2fid(obj)), rc);
5820 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
5823 for (i = 0; i < count; i++) {
5824 lcme = &lcm->lcm_entries[i];
5825 if (!(le32_to_cpu(lcme->lcme_flags) &
5829 rc = lfsck_layout_scan_stripes(env, com, obj,
5830 (struct lov_mds_md_v1 *)(buf->lb_buf +
5831 le32_to_cpu(lcme->lcme_offset)),
5832 le32_to_cpu(lcme->lcme_id));
5835 rc = lfsck_layout_scan_stripes(env, com, obj, lmm, 0);
5838 down_write(&com->lc_sem);
5839 com->lc_new_checked++;
5841 lfsck_layout_record_failure(env, lfsck, lo);
5842 up_write(&com->lc_sem);
5848 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
5849 struct lfsck_component *com,
5850 struct dt_object *obj)
5852 struct lfsck_instance *lfsck = com->lc_lfsck;
5853 struct lfsck_layout *lo = com->lc_file_ram;
5854 const struct lu_fid *fid = lfsck_dto2fid(obj);
5855 struct lfsck_layout_slave_data *llsd = com->lc_data;
5856 struct lfsck_layout_seq *lls;
5862 LASSERT(llsd != NULL);
5864 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
5865 cfs_fail_val == lfsck_dev_idx(lfsck)) {
5866 struct ptlrpc_thread *thread = &lfsck->li_thread;
5868 wait_event_idle_timeout(thread->t_ctl_waitq,
5869 !thread_is_running(thread),
5870 cfs_time_seconds(1));
5873 lfsck_rbtree_update_bitmap(env, com, fid, false);
5875 down_write(&com->lc_sem);
5876 if (fid_is_idif(fid))
5878 else if (!fid_is_norm(fid) ||
5879 !fid_is_for_ostobj(env, lfsck, obj, fid))
5880 GOTO(unlock, rc = 0);
5883 com->lc_new_checked++;
5885 lls = lfsck_layout_seq_lookup(llsd, seq);
5888 if (unlikely(lls == NULL))
5889 GOTO(unlock, rc = -ENOMEM);
5891 INIT_LIST_HEAD(&lls->lls_list);
5893 rc = lfsck_layout_lastid_load(env, com, lls);
5895 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5896 "load LAST_ID for %#llx: rc = %d\n",
5897 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
5898 lo->ll_objs_failed_phase1++;
5903 lfsck_layout_seq_insert(llsd, lls);
5906 if (unlikely(fid_is_last_id(fid)))
5907 GOTO(unlock, rc = 0);
5909 if (fid_is_idif(fid))
5910 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
5914 if (oid > lls->lls_lastid_known)
5915 lls->lls_lastid_known = oid;
5917 if (oid > lls->lls_lastid) {
5918 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
5919 /* OFD may create new objects during LFSCK scanning. */
5920 rc = lfsck_layout_lastid_reload(env, com, lls);
5921 if (unlikely(rc != 0)) {
5922 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
5923 "reload LAST_ID for %#llx: rc = %d\n",
5924 lfsck_lfsck2name(com->lc_lfsck),
5930 if (oid <= lls->lls_lastid ||
5931 lo->ll_flags & LF_CRASHED_LASTID)
5932 GOTO(unlock, rc = 0);
5934 LASSERT(lfsck->li_out_notify != NULL);
5936 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5937 LE_LASTID_REBUILDING);
5938 lo->ll_flags |= LF_CRASHED_LASTID;
5940 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
5941 "LAST_ID file (2) for the sequence %#llx"
5942 ", old value %llu, known value %llu\n",
5943 lfsck_lfsck2name(lfsck), lls->lls_seq,
5944 lls->lls_lastid, oid);
5947 lls->lls_lastid = oid;
5951 GOTO(unlock, rc = 0);
5954 up_write(&com->lc_sem);
5959 static int lfsck_layout_exec_dir(const struct lu_env *env,
5960 struct lfsck_component *com,
5961 struct lfsck_assistant_object *lso,
5962 struct lu_dirent *ent, __u16 type)
5967 static int lfsck_layout_master_post(const struct lu_env *env,
5968 struct lfsck_component *com,
5969 int result, bool init)
5971 struct lfsck_instance *lfsck = com->lc_lfsck;
5972 struct lfsck_layout *lo = com->lc_file_ram;
5976 lfsck_post_generic(env, com, &result);
5978 down_write(&com->lc_sem);
5979 spin_lock(&lfsck->li_lock);
5981 lo->ll_pos_last_checkpoint =
5982 lfsck->li_pos_checkpoint.lp_oit_cookie;
5985 if (lo->ll_flags & LF_INCOMPLETE)
5986 lo->ll_status = LS_PARTIAL;
5988 lo->ll_status = LS_SCANNING_PHASE2;
5989 lo->ll_flags |= LF_SCANNED_ONCE;
5990 lo->ll_flags &= ~LF_UPGRADE;
5991 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
5992 } else if (result == 0) {
5993 if (lfsck->li_status != 0)
5994 lo->ll_status = lfsck->li_status;
5996 lo->ll_status = LS_STOPPED;
5997 if (lo->ll_status != LS_PAUSED)
5998 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6000 lo->ll_status = LS_FAILED;
6001 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6003 spin_unlock(&lfsck->li_lock);
6006 lo->ll_run_time_phase1 += ktime_get_seconds() -
6007 lfsck->li_time_last_checkpoint;
6008 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
6009 lo->ll_objs_checked_phase1 += com->lc_new_checked;
6010 com->lc_new_checked = 0;
6013 rc = lfsck_layout_store(env, com);
6014 up_write(&com->lc_sem);
6016 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
6017 lfsck_lfsck2name(lfsck), rc);
6022 static int lfsck_layout_slave_post(const struct lu_env *env,
6023 struct lfsck_component *com,
6024 int result, bool init)
6026 struct lfsck_instance *lfsck = com->lc_lfsck;
6027 struct lfsck_layout *lo = com->lc_file_ram;
6031 down_write(&com->lc_sem);
6032 rc = lfsck_layout_lastid_store(env, com);
6036 LASSERT(lfsck->li_out_notify != NULL);
6038 spin_lock(&lfsck->li_lock);
6040 lo->ll_pos_last_checkpoint =
6041 lfsck->li_pos_checkpoint.lp_oit_cookie;
6044 lo->ll_status = LS_SCANNING_PHASE2;
6045 lo->ll_flags |= LF_SCANNED_ONCE;
6046 if (lo->ll_flags & LF_CRASHED_LASTID) {
6048 lo->ll_flags &= ~LF_CRASHED_LASTID;
6050 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
6051 "crashed LAST_ID files successfully\n",
6052 lfsck_lfsck2name(lfsck));
6054 lo->ll_flags &= ~LF_UPGRADE;
6055 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
6056 } else if (result == 0) {
6057 if (lfsck->li_status != 0)
6058 lo->ll_status = lfsck->li_status;
6060 lo->ll_status = LS_STOPPED;
6061 if (lo->ll_status != LS_PAUSED)
6062 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6064 lo->ll_status = LS_FAILED;
6065 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6067 spin_unlock(&lfsck->li_lock);
6070 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6074 lo->ll_run_time_phase1 += ktime_get_seconds() -
6075 lfsck->li_time_last_checkpoint;
6076 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
6077 lo->ll_objs_checked_phase1 += com->lc_new_checked;
6078 com->lc_new_checked = 0;
6081 rc = lfsck_layout_store(env, com);
6082 up_write(&com->lc_sem);
6084 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
6086 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
6087 lfsck_lfsck2name(lfsck), rc);
6092 static void lfsck_layout_dump(const struct lu_env *env,
6093 struct lfsck_component *com, struct seq_file *m)
6095 struct lfsck_instance *lfsck = com->lc_lfsck;
6096 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6097 struct lfsck_layout *lo = com->lc_file_ram;
6100 down_read(&com->lc_sem);
6101 if (bk->lb_param & LPF_DRYRUN)
6102 prefix = "inconsistent";
6104 prefix = "repaired";
6106 seq_printf(m, "name: lfsck_layout\n"
6112 lfsck_status2name(lo->ll_status));
6114 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
6116 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
6118 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
6120 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
6122 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
6124 seq_printf(m, "latest_start_position: %llu\n"
6125 "last_checkpoint_position: %llu\n"
6126 "first_failure_position: %llu\n",
6127 lo->ll_pos_latest_start,
6128 lo->ll_pos_last_checkpoint,
6129 lo->ll_pos_first_inconsistent);
6131 seq_printf(m, "success_count: %u\n"
6132 "%s_dangling: %llu\n"
6133 "%s_unmatched_pair: %llu\n"
6134 "%s_multiple_referenced: %llu\n"
6136 "%s_inconsistent_owner: %llu\n"
6139 "failed_phase1: %llu\n"
6140 "failed_phase2: %llu\n",
6141 lo->ll_success_count,
6142 prefix, lo->ll_objs_repaired[LLIT_DANGLING - 1],
6143 prefix, lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
6144 prefix, lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
6145 prefix, lo->ll_objs_repaired[LLIT_ORPHAN - 1],
6146 prefix, lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
6147 prefix, lo->ll_objs_repaired[LLIT_OTHERS - 1],
6148 lo->ll_objs_skipped,
6149 lo->ll_objs_failed_phase1,
6150 lo->ll_objs_failed_phase2);
6152 if (lo->ll_status == LS_SCANNING_PHASE1) {
6153 time64_t duration = ktime_get_seconds() -
6154 lfsck->li_time_last_checkpoint;
6155 u64 checked = lo->ll_objs_checked_phase1 +
6156 com->lc_new_checked;
6157 u64 speed = checked;
6158 u64 new_checked = com->lc_new_checked;
6159 time64_t rtime = lo->ll_run_time_phase1 + duration;
6163 new_checked = div64_s64(new_checked, duration);
6165 speed = div64_s64(speed, rtime);
6166 seq_printf(m, "checked_phase1: %llu\n"
6167 "checked_phase2: %llu\n"
6168 "run_time_phase1: %lld seconds\n"
6169 "run_time_phase2: %lld seconds\n"
6170 "average_speed_phase1: %llu items/sec\n"
6171 "average_speed_phase2: N/A\n"
6172 "real_time_speed_phase1: %llu items/sec\n"
6173 "real_time_speed_phase2: N/A\n",
6175 lo->ll_objs_checked_phase2,
6177 lo->ll_run_time_phase2,
6181 if (likely(lfsck->li_di_oit)) {
6182 const struct dt_it_ops *iops =
6183 &lfsck->li_obj_oit->do_index_ops->dio_it;
6185 /* The low layer otable-based iteration position may NOT
6186 * exactly match the layout-based directory traversal
6187 * cookie. Generally, it is not a serious issue. But the
6188 * caller should NOT make assumption on that. */
6189 pos = iops->store(env, lfsck->li_di_oit);
6190 if (!lfsck->li_current_oit_processed)
6193 pos = lo->ll_pos_last_checkpoint;
6196 seq_printf(m, "current_position: %llu\n", pos);
6197 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
6198 time64_t duration = ktime_get_seconds() -
6199 com->lc_time_last_checkpoint;
6200 u64 checked = lo->ll_objs_checked_phase2 +
6201 com->lc_new_checked;
6202 u64 speed1 = lo->ll_objs_checked_phase1;
6203 u64 speed2 = checked;
6204 u64 new_checked = com->lc_new_checked;
6205 time64_t rtime = lo->ll_run_time_phase2 + duration;
6208 new_checked = div64_s64(new_checked, duration);
6209 if (lo->ll_run_time_phase1 != 0)
6210 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
6212 speed2 = div64_s64(speed2, rtime);
6213 seq_printf(m, "checked_phase1: %llu\n"
6214 "checked_phase2: %llu\n"
6215 "run_time_phase1: %lld seconds\n"
6216 "run_time_phase2: %lld seconds\n"
6217 "average_speed_phase1: %llu items/sec\n"
6218 "average_speed_phase2: %llu items/sec\n"
6219 "real_time_speed_phase1: N/A\n"
6220 "real_time_speed_phase2: %llu items/sec\n"
6221 "current_position: "DFID"\n",
6222 lo->ll_objs_checked_phase1,
6224 lo->ll_run_time_phase1,
6229 PFID(&com->lc_fid_latest_scanned_phase2));
6231 __u64 speed1 = lo->ll_objs_checked_phase1;
6232 __u64 speed2 = lo->ll_objs_checked_phase2;
6234 if (lo->ll_run_time_phase1 != 0)
6235 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
6236 if (lo->ll_run_time_phase2 != 0)
6237 speed2 = div64_s64(speed2, lo->ll_run_time_phase2);
6238 seq_printf(m, "checked_phase1: %llu\n"
6239 "checked_phase2: %llu\n"
6240 "run_time_phase1: %lld seconds\n"
6241 "run_time_phase2: %lld seconds\n"
6242 "average_speed_phase1: %llu items/sec\n"
6243 "average_speed_phase2: %llu objs/sec\n"
6244 "real_time_speed_phase1: N/A\n"
6245 "real_time_speed_phase2: N/A\n"
6246 "current_position: N/A\n",
6247 lo->ll_objs_checked_phase1,
6248 lo->ll_objs_checked_phase2,
6249 lo->ll_run_time_phase1,
6250 lo->ll_run_time_phase2,
6255 up_read(&com->lc_sem);
6258 static int lfsck_layout_master_double_scan(const struct lu_env *env,
6259 struct lfsck_component *com)
6261 struct lfsck_layout *lo = com->lc_file_ram;
6262 struct lfsck_assistant_data *lad = com->lc_data;
6263 struct lfsck_instance *lfsck = com->lc_lfsck;
6264 struct lfsck_tgt_descs *ltds;
6265 struct lfsck_tgt_desc *ltd;
6266 struct lfsck_tgt_desc *next;
6269 rc = lfsck_double_scan_generic(env, com, lo->ll_status);
6271 if (thread_is_stopped(&lad->lad_thread)) {
6272 LASSERT(list_empty(&lad->lad_req_list));
6273 LASSERT(list_empty(&lad->lad_ost_phase1_list));
6274 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
6276 ltds = &lfsck->li_ost_descs;
6277 spin_lock(<ds->ltd_lock);
6278 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6279 ltd_layout_phase_list) {
6280 list_del_init(<d->ltd_layout_phase_list);
6282 spin_unlock(<ds->ltd_lock);
6284 ltds = &lfsck->li_mdt_descs;
6285 spin_lock(<ds->ltd_lock);
6286 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6287 ltd_layout_phase_list) {
6288 list_del_init(<d->ltd_layout_phase_list);
6290 spin_unlock(<ds->ltd_lock);
6296 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
6297 struct lfsck_component *com)
6299 struct lfsck_instance *lfsck = com->lc_lfsck;
6300 struct lfsck_layout_slave_data *llsd = com->lc_data;
6301 struct lfsck_layout *lo = com->lc_file_ram;
6302 struct ptlrpc_thread *thread = &lfsck->li_thread;
6306 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
6307 lfsck_lfsck2name(lfsck));
6309 atomic_inc(&lfsck->li_double_scan_count);
6311 if (lo->ll_flags & LF_INCOMPLETE)
6314 com->lc_new_checked = 0;
6315 com->lc_new_scanned = 0;
6316 com->lc_time_last_checkpoint = ktime_get_seconds();
6317 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6318 LFSCK_CHECKPOINT_INTERVAL;
6321 rc = lfsck_layout_slave_query_master(env, com);
6322 if (list_empty(&llsd->llsd_master_list)) {
6323 if (unlikely(!thread_is_running(thread)))
6334 rc = wait_event_idle_timeout(
6335 thread->t_ctl_waitq,
6336 !thread_is_running(thread) ||
6337 lo->ll_flags & LF_INCOMPLETE ||
6338 list_empty(&llsd->llsd_master_list),
6339 cfs_time_seconds(30));
6340 if (unlikely(!thread_is_running(thread)))
6343 if (lo->ll_flags & LF_INCOMPLETE)
6353 rc = lfsck_layout_double_scan_result(env, com, rc);
6354 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
6355 (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
6356 lfsck_layout_slave_quit(env, com);
6357 if (atomic_dec_and_test(&lfsck->li_double_scan_count))
6358 wake_up(&lfsck->li_thread.t_ctl_waitq);
6360 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
6361 "status %d: rc = %d\n",
6362 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
6367 static void lfsck_layout_master_data_release(const struct lu_env *env,
6368 struct lfsck_component *com)
6370 struct lfsck_assistant_data *lad = com->lc_data;
6371 struct lfsck_instance *lfsck = com->lc_lfsck;
6372 struct lfsck_tgt_descs *ltds;
6373 struct lfsck_tgt_desc *ltd;
6374 struct lfsck_tgt_desc *next;
6376 LASSERT(lad != NULL);
6377 LASSERT(thread_is_init(&lad->lad_thread) ||
6378 thread_is_stopped(&lad->lad_thread));
6379 LASSERT(list_empty(&lad->lad_req_list));
6381 com->lc_data = NULL;
6383 ltds = &lfsck->li_ost_descs;
6384 spin_lock(<ds->ltd_lock);
6385 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6386 ltd_layout_phase_list) {
6387 list_del_init(<d->ltd_layout_phase_list);
6389 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6390 ltd_layout_phase_list) {
6391 list_del_init(<d->ltd_layout_phase_list);
6393 list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
6395 list_del_init(<d->ltd_layout_list);
6397 spin_unlock(<ds->ltd_lock);
6399 ltds = &lfsck->li_mdt_descs;
6400 spin_lock(<ds->ltd_lock);
6401 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6402 ltd_layout_phase_list) {
6403 list_del_init(<d->ltd_layout_phase_list);
6405 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6406 ltd_layout_phase_list) {
6407 list_del_init(<d->ltd_layout_phase_list);
6409 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
6411 list_del_init(<d->ltd_layout_list);
6413 spin_unlock(<ds->ltd_lock);
6415 bitmap_free(lad->lad_bitmap);
6420 static void lfsck_layout_slave_data_release(const struct lu_env *env,
6421 struct lfsck_component *com)
6423 struct lfsck_layout_slave_data *llsd = com->lc_data;
6425 lfsck_layout_slave_quit(env, com);
6426 com->lc_data = NULL;
6430 static void lfsck_layout_master_quit(const struct lu_env *env,
6431 struct lfsck_component *com)
6433 struct lfsck_assistant_data *lad = com->lc_data;
6434 struct lfsck_instance *lfsck = com->lc_lfsck;
6435 struct lfsck_tgt_descs *ltds;
6436 struct lfsck_tgt_desc *ltd;
6437 struct lfsck_tgt_desc *next;
6439 LASSERT(lad != NULL);
6441 lfsck_quit_generic(env, com);
6443 LASSERT(thread_is_init(&lad->lad_thread) ||
6444 thread_is_stopped(&lad->lad_thread));
6445 LASSERT(list_empty(&lad->lad_req_list));
6447 ltds = &lfsck->li_ost_descs;
6448 spin_lock(<ds->ltd_lock);
6449 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6450 ltd_layout_phase_list) {
6451 list_del_init(<d->ltd_layout_phase_list);
6453 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6454 ltd_layout_phase_list) {
6455 list_del_init(<d->ltd_layout_phase_list);
6457 spin_unlock(<ds->ltd_lock);
6459 ltds = &lfsck->li_mdt_descs;
6460 spin_lock(<ds->ltd_lock);
6461 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6462 ltd_layout_phase_list) {
6463 list_del_init(<d->ltd_layout_phase_list);
6465 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6466 ltd_layout_phase_list) {
6467 list_del_init(<d->ltd_layout_phase_list);
6469 spin_unlock(<ds->ltd_lock);
6472 static void lfsck_layout_slave_quit(const struct lu_env *env,
6473 struct lfsck_component *com)
6475 struct lfsck_layout_slave_data *llsd = com->lc_data;
6476 struct lfsck_layout_seq *lls;
6477 struct lfsck_layout_seq *next;
6478 struct lfsck_layout_slave_target *llst;
6480 LASSERT(llsd != NULL);
6482 down_write(&com->lc_sem);
6483 list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
6485 list_del_init(&lls->lls_list);
6486 lfsck_object_put(env, lls->lls_lastid_obj);
6489 up_write(&com->lc_sem);
6491 spin_lock(&llsd->llsd_lock);
6492 while (!list_empty(&llsd->llsd_master_list)) {
6493 llst = list_entry(llsd->llsd_master_list.next,
6494 struct lfsck_layout_slave_target, llst_list);
6495 list_del_init(&llst->llst_list);
6496 spin_unlock(&llsd->llsd_lock);
6497 lfsck_layout_llst_put(llst);
6498 spin_lock(&llsd->llsd_lock);
6500 spin_unlock(&llsd->llsd_lock);
6502 lfsck_rbtree_cleanup(env, com);
6505 static int lfsck_layout_master_in_notify(const struct lu_env *env,
6506 struct lfsck_component *com,
6507 struct lfsck_request *lr)
6509 struct lfsck_instance *lfsck = com->lc_lfsck;
6510 struct lfsck_layout *lo = com->lc_file_ram;
6511 struct lfsck_assistant_data *lad = com->lc_data;
6512 struct lfsck_tgt_descs *ltds;
6513 struct lfsck_tgt_desc *ltd;
6517 if (lr->lr_event == LE_PAIRS_VERIFY) {
6520 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
6527 CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
6528 "from %s %x, status %d, flags %x, flags2 %x\n",
6529 lfsck_lfsck2name(lfsck), lr->lr_event,
6530 (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
6531 lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
6533 if (lr->lr_event != LE_PHASE1_DONE &&
6534 lr->lr_event != LE_PHASE2_DONE &&
6535 lr->lr_event != LE_PEER_EXIT)
6538 if (lr->lr_flags & LEF_FROM_OST)
6539 ltds = &lfsck->li_ost_descs;
6541 ltds = &lfsck->li_mdt_descs;
6542 spin_lock(<ds->ltd_lock);
6543 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
6545 spin_unlock(<ds->ltd_lock);
6550 list_del_init(<d->ltd_layout_phase_list);
6551 switch (lr->lr_event) {
6552 case LE_PHASE1_DONE:
6553 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
6554 if (lr->lr_flags2 & LF_INCOMPLETE) {
6555 if (lr->lr_flags & LEF_FROM_OST)
6556 lfsck_lad_set_bitmap(env, com,
6559 lo->ll_flags |= LF_INCOMPLETE;
6561 ltd->ltd_layout_done = 1;
6562 list_del_init(<d->ltd_layout_list);
6567 if (lr->lr_flags & LEF_FROM_OST) {
6568 if (list_empty(<d->ltd_layout_list))
6569 list_add_tail(<d->ltd_layout_list,
6570 &lad->lad_ost_list);
6571 list_add_tail(<d->ltd_layout_phase_list,
6572 &lad->lad_ost_phase2_list);
6574 if (list_empty(<d->ltd_layout_list))
6575 list_add_tail(<d->ltd_layout_list,
6576 &lad->lad_mdt_list);
6577 list_add_tail(<d->ltd_layout_phase_list,
6578 &lad->lad_mdt_phase2_list);
6581 case LE_PHASE2_DONE:
6582 ltd->ltd_layout_done = 1;
6583 if (!list_empty(<d->ltd_layout_list))
6584 list_del_init(<d->ltd_layout_list);
6586 if (lr->lr_flags2 & LF_INCOMPLETE) {
6587 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
6594 ltd->ltd_layout_done = 1;
6595 list_del_init(<d->ltd_layout_list);
6596 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
6597 !(lr->lr_flags & LEF_FROM_OST))
6598 lo->ll_flags |= LF_INCOMPLETE;
6603 spin_unlock(<ds->ltd_lock);
6605 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
6606 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6608 memset(stop, 0, sizeof(*stop));
6609 stop->ls_status = lr->lr_status;
6610 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6611 lfsck_stop(env, lfsck->li_bottom, stop);
6612 } else if (lfsck_phase2_next_ready(lad)) {
6613 wake_up(&lad->lad_thread.t_ctl_waitq);
6619 static int lfsck_layout_slave_in_notify_local(const struct lu_env *env,
6620 struct lfsck_component *com,
6621 struct lfsck_req_local *lrl,
6626 switch (lrl->lrl_event) {
6627 case LEL_FID_ACCESSED:
6628 lfsck_rbtree_update_bitmap(env, com, &lrl->lrl_fid, true);
6630 case LEL_PAIRS_VERIFY_LOCAL: {
6633 lrl->lrl_status = LPVS_INIT;
6634 /* Firstly, if the MDT-object which is claimed via OST-object
6635 * local stored PFID xattr recognizes the OST-object, then it
6636 * must be that the client given PFID is wrong. */
6637 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6638 &lrl->lrl_ff_local.ff_parent,
6639 lrl->lrl_ff_local.ff_layout.ol_comp_id);
6643 lrl->lrl_status = LPVS_INCONSISTENT;
6644 /* The OST-object local stored PFID xattr is stale. We need to
6645 * check whether the MDT-object that is claimed via the client
6646 * given PFID information recognizes the OST-object or not. If
6647 * matches, then need to update the OST-object's PFID xattr. */
6648 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6649 &lrl->lrl_ff_client.ff_parent,
6650 lrl->lrl_ff_client.ff_layout.ol_comp_id);
6652 * We are not sure whether the client given PFID information
6653 * is correct or not, do nothing to avoid improper fixing.
6656 * The client given PFID information is also invalid, we can
6657 * NOT fix the OST-object inconsistency.
6660 lrl->lrl_status = LPVS_INCONSISTENT_TOFIX;
6661 rc = lfsck_layout_slave_repair_pfid(env, com, lrl);
6670 RETURN(-EOPNOTSUPP);
6673 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
6674 struct lfsck_component *com,
6675 struct lfsck_request *lr)
6677 struct lfsck_instance *lfsck = com->lc_lfsck;
6678 struct lfsck_layout_slave_data *llsd = com->lc_data;
6679 struct lfsck_layout_slave_target *llst;
6683 switch (lr->lr_event) {
6684 case LE_CONDITIONAL_DESTROY:
6685 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
6687 case LE_PHASE1_DONE: {
6688 if (lr->lr_flags2 & LF_INCOMPLETE) {
6689 struct lfsck_layout *lo = com->lc_file_ram;
6691 lo->ll_flags |= LF_INCOMPLETE;
6692 llst = lfsck_layout_llst_find_and_del(llsd,
6696 lfsck_layout_llst_put(llst);
6697 wake_up(&lfsck->li_thread.t_ctl_waitq);
6703 case LE_PHASE2_DONE:
6705 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
6706 "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
6707 lr->lr_event, lr->lr_index, lr->lr_status);
6713 llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
6717 lfsck_layout_llst_put(llst);
6718 if (list_empty(&llsd->llsd_master_list))
6719 wake_up(&lfsck->li_thread.t_ctl_waitq);
6721 if (lr->lr_event == LE_PEER_EXIT &&
6722 (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
6723 (list_empty(&llsd->llsd_master_list) &&
6724 (lr->lr_status == LS_STOPPED ||
6725 lr->lr_status == LS_CO_STOPPED)))) {
6726 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6728 memset(stop, 0, sizeof(*stop));
6729 stop->ls_status = lr->lr_status;
6730 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6731 lfsck_stop(env, lfsck->li_bottom, stop);
6737 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
6741 for (i = 0; i < LLIT_MAX; i++)
6742 *count += lo->ll_objs_repaired[i];
6745 static int lfsck_layout_query_all(const struct lu_env *env,
6746 struct lfsck_component *com,
6747 __u32 *mdts_count, __u32 *osts_count,
6750 struct lfsck_layout *lo = com->lc_file_ram;
6751 struct lfsck_tgt_descs *ltds;
6752 struct lfsck_tgt_desc *ltd;
6757 rc = lfsck_query_all(env, com);
6761 ltds = &com->lc_lfsck->li_mdt_descs;
6762 down_read(<ds->ltd_rw_sem);
6763 for_each_set_bit(idx, ltds->ltd_tgts_bitmap, ltds->ltd_tgts_mask_len) {
6764 ltd = lfsck_ltd2tgt(ltds, idx);
6765 LASSERT(ltd != NULL);
6767 mdts_count[ltd->ltd_layout_status]++;
6768 *repaired += ltd->ltd_layout_repaired;
6770 up_read(<ds->ltd_rw_sem);
6772 ltds = &com->lc_lfsck->li_ost_descs;
6773 down_read(<ds->ltd_rw_sem);
6774 for_each_set_bit(idx, ltds->ltd_tgts_bitmap, ltds->ltd_tgts_mask_len) {
6775 ltd = lfsck_ltd2tgt(ltds, idx);
6776 LASSERT(ltd != NULL);
6778 osts_count[ltd->ltd_layout_status]++;
6779 *repaired += ltd->ltd_layout_repaired;
6781 up_read(<ds->ltd_rw_sem);
6783 down_read(&com->lc_sem);
6784 mdts_count[lo->ll_status]++;
6785 lfsck_layout_repaired(lo, repaired);
6786 up_read(&com->lc_sem);
6791 static int lfsck_layout_query(const struct lu_env *env,
6792 struct lfsck_component *com,
6793 struct lfsck_request *req,
6794 struct lfsck_reply *rep,
6795 struct lfsck_query *que, int idx)
6797 struct lfsck_layout *lo = com->lc_file_ram;
6801 LASSERT(com->lc_lfsck->li_master);
6803 rc = lfsck_layout_query_all(env, com,
6804 que->lu_mdts_count[idx],
6805 que->lu_osts_count[idx],
6806 &que->lu_repaired[idx]);
6808 down_read(&com->lc_sem);
6809 rep->lr_status = lo->ll_status;
6810 if (req->lr_flags & LEF_QUERY_ALL)
6811 lfsck_layout_repaired(lo, &rep->lr_repaired);
6812 up_read(&com->lc_sem);
6818 /* with lfsck::li_lock held */
6819 static int lfsck_layout_slave_join(const struct lu_env *env,
6820 struct lfsck_component *com,
6821 struct lfsck_start_param *lsp)
6823 struct lfsck_instance *lfsck = com->lc_lfsck;
6824 struct lfsck_layout_slave_data *llsd = com->lc_data;
6825 struct lfsck_layout_slave_target *llst;
6826 struct lfsck_start *start = lsp->lsp_start;
6830 if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
6833 if (!lsp->lsp_index_valid)
6836 /* If someone is running the LFSCK without orphan handling,
6837 * it will not maintain the object accessing rbtree. So we
6838 * cannot join it for orphan handling. */
6839 if (!llsd->llsd_rbtree_valid)
6842 spin_unlock(&lfsck->li_lock);
6843 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
6844 spin_lock(&lfsck->li_lock);
6845 if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
6846 spin_unlock(&lfsck->li_lock);
6847 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
6850 lfsck_layout_llst_put(llst);
6851 spin_lock(&lfsck->li_lock);
6858 static const struct lfsck_operations lfsck_layout_master_ops = {
6859 .lfsck_reset = lfsck_layout_reset,
6860 .lfsck_fail = lfsck_layout_fail,
6861 .lfsck_checkpoint = lfsck_layout_master_checkpoint,
6862 .lfsck_prep = lfsck_layout_master_prep,
6863 .lfsck_exec_oit = lfsck_layout_master_exec_oit,
6864 .lfsck_exec_dir = lfsck_layout_exec_dir,
6865 .lfsck_post = lfsck_layout_master_post,
6866 .lfsck_dump = lfsck_layout_dump,
6867 .lfsck_double_scan = lfsck_layout_master_double_scan,
6868 .lfsck_data_release = lfsck_layout_master_data_release,
6869 .lfsck_quit = lfsck_layout_master_quit,
6870 .lfsck_in_notify = lfsck_layout_master_in_notify,
6871 .lfsck_query = lfsck_layout_query,
6874 static const struct lfsck_operations lfsck_layout_slave_ops = {
6875 .lfsck_reset = lfsck_layout_reset,
6876 .lfsck_fail = lfsck_layout_fail,
6877 .lfsck_checkpoint = lfsck_layout_slave_checkpoint,
6878 .lfsck_prep = lfsck_layout_slave_prep,
6879 .lfsck_exec_oit = lfsck_layout_slave_exec_oit,
6880 .lfsck_exec_dir = lfsck_layout_exec_dir,
6881 .lfsck_post = lfsck_layout_slave_post,
6882 .lfsck_dump = lfsck_layout_dump,
6883 .lfsck_double_scan = lfsck_layout_slave_double_scan,
6884 .lfsck_data_release = lfsck_layout_slave_data_release,
6885 .lfsck_quit = lfsck_layout_slave_quit,
6886 .lfsck_in_notify_local = lfsck_layout_slave_in_notify_local,
6887 .lfsck_in_notify = lfsck_layout_slave_in_notify,
6888 .lfsck_query = lfsck_layout_query,
6889 .lfsck_join = lfsck_layout_slave_join,
6892 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
6893 struct lfsck_component *com,
6894 struct lfsck_position *pos)
6896 struct lfsck_assistant_data *lad = com->lc_data;
6897 struct lfsck_layout_req *llr;
6899 if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
6903 if (list_empty(&lad->lad_req_list))
6906 llr = list_entry(lad->lad_req_list.next,
6907 struct lfsck_layout_req,
6909 pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
6912 const struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
6913 .la_handler_p1 = lfsck_layout_assistant_handler_p1,
6914 .la_handler_p2 = lfsck_layout_assistant_handler_p2,
6915 .la_fill_pos = lfsck_layout_assistant_fill_pos,
6916 .la_double_scan_result = lfsck_layout_double_scan_result,
6917 .la_req_fini = lfsck_layout_assistant_req_fini,
6918 .la_sync_failures = lfsck_layout_assistant_sync_failures,
6921 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
6923 struct lfsck_component *com;
6924 struct lfsck_layout *lo;
6925 struct dt_object *root = NULL;
6926 struct dt_object *obj;
6935 INIT_LIST_HEAD(&com->lc_link);
6936 INIT_LIST_HEAD(&com->lc_link_dir);
6937 init_rwsem(&com->lc_sem);
6938 atomic_set(&com->lc_ref, 1);
6939 com->lc_lfsck = lfsck;
6940 com->lc_type = LFSCK_TYPE_LAYOUT;
6941 if (lfsck->li_master) {
6942 com->lc_ops = &lfsck_layout_master_ops;
6943 com->lc_data = lfsck_assistant_data_init(
6944 &lfsck_layout_assistant_ops,
6946 if (com->lc_data == NULL)
6947 GOTO(out, rc = -ENOMEM);
6949 for (i = 0; i < LFSCK_STF_COUNT; i++)
6950 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
6952 struct lfsck_layout_slave_data *llsd;
6954 com->lc_ops = &lfsck_layout_slave_ops;
6955 OBD_ALLOC_PTR(llsd);
6957 GOTO(out, rc = -ENOMEM);
6959 INIT_LIST_HEAD(&llsd->llsd_seq_list);
6960 INIT_LIST_HEAD(&llsd->llsd_master_list);
6961 spin_lock_init(&llsd->llsd_lock);
6962 llsd->llsd_rb_root = RB_ROOT;
6963 init_rwsem(&llsd->llsd_rb_rwsem);
6964 com->lc_data = llsd;
6966 com->lc_file_size = sizeof(*lo);
6967 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6968 if (com->lc_file_ram == NULL)
6969 GOTO(out, rc = -ENOMEM);
6971 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6972 if (com->lc_file_disk == NULL)
6973 GOTO(out, rc = -ENOMEM);
6975 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6977 GOTO(out, rc = PTR_ERR(root));
6979 if (unlikely(!dt_try_as_dir(env, root, true)))
6980 GOTO(out, rc = -ENOTDIR);
6982 obj = local_file_find_or_create(env, lfsck->li_los, root,
6984 S_IFREG | S_IRUGO | S_IWUSR);
6986 GOTO(out, rc = PTR_ERR(obj));
6989 rc = lfsck_layout_load(env, com);
6991 rc = lfsck_layout_reset(env, com, true);
6992 } else if (rc == -ENOENT) {
6993 rc = lfsck_layout_init(env, com);
6994 } else if (lfsck->li_master) {
6995 rc = lfsck_load_sub_trace_files(env, com,
6996 &dt_lfsck_layout_dangling_features,
6997 LFSCK_LAYOUT, false);
6999 rc = lfsck_layout_reset(env, com, true);
7005 lo = com->lc_file_ram;
7006 switch (lo->ll_status) {
7012 spin_lock(&lfsck->li_lock);
7013 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
7014 spin_unlock(&lfsck->li_lock);
7017 CERROR("%s: unknown lfsck_layout status %d\n",
7018 lfsck_lfsck2name(lfsck), lo->ll_status);
7020 case LS_SCANNING_PHASE1:
7021 case LS_SCANNING_PHASE2:
7022 /* No need to store the status to disk right now.
7023 * If the system crashed before the status stored,
7024 * it will be loaded back when next time. */
7025 lo->ll_status = LS_CRASHED;
7026 if (!lfsck->li_master)
7027 lo->ll_flags |= LF_INCOMPLETE;
7034 spin_lock(&lfsck->li_lock);
7035 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
7036 spin_unlock(&lfsck->li_lock);
7040 if (lo->ll_flags & LF_CRASHED_LASTID) {
7041 LASSERT(lfsck->li_out_notify != NULL);
7043 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
7044 LE_LASTID_REBUILDING);
7050 if (root != NULL && !IS_ERR(root))
7051 lfsck_object_put(env, root);
7054 lfsck_component_cleanup(env, com);
7055 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
7056 lfsck_lfsck2name(lfsck), rc);
7062 struct lfsck_orphan_it {
7063 struct lfsck_component *loi_com;
7064 struct lfsck_rbtree_node *loi_lrn;
7065 struct lfsck_layout_slave_target *loi_llst;
7066 struct lu_fid loi_key;
7067 struct lu_orphan_rec_v3 loi_rec;
7069 unsigned int loi_over:1;
7072 static int lfsck_fid_match_idx(const struct lu_env *env,
7073 struct lfsck_instance *lfsck,
7074 const struct lu_fid *fid, int idx)
7076 struct seq_server_site *ss;
7077 struct lu_server_fld *sf;
7078 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
7081 /* All abnormal cases will be returned to MDT0. */
7082 if (!fid_is_norm(fid)) {
7089 ss = lfsck_dev_site(lfsck);
7090 if (unlikely(ss == NULL))
7093 sf = ss->ss_server_fld;
7094 LASSERT(sf != NULL);
7096 fld_range_set_any(range);
7097 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
7101 if (!fld_range_is_mdt(range))
7104 if (range->lsr_index == idx)
7110 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
7111 struct lfsck_instance *lfsck,
7112 struct dt_object *obj)
7114 struct dt_device *dev = lfsck_obj2dev(obj);
7115 struct thandle *handle;
7119 handle = lfsck_trans_create(env, dev, lfsck);
7123 rc = dt_declare_ref_del(env, obj, handle);
7127 rc = dt_declare_destroy(env, obj, handle);
7131 rc = dt_trans_start_local(env, dev, handle);
7135 dt_write_lock(env, obj, 0);
7136 rc = dt_ref_del(env, obj, handle);
7138 rc = dt_destroy(env, obj, handle);
7139 dt_write_unlock(env, obj);
7144 dt_trans_stop(env, dev, handle);
7146 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
7147 PFID(lfsck_dto2fid(obj)), rc);
7152 static int lfsck_orphan_index_lookup(const struct lu_env *env,
7153 struct dt_object *dt,
7155 const struct dt_key *key)
7160 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
7161 struct dt_object *dt,
7162 const struct dt_rec *rec,
7163 const struct dt_key *key,
7164 struct thandle *handle)
7169 static int lfsck_orphan_index_insert(const struct lu_env *env,
7170 struct dt_object *dt,
7171 const struct dt_rec *rec,
7172 const struct dt_key *key,
7173 struct thandle *handle)
7178 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
7179 struct dt_object *dt,
7180 const struct dt_key *key,
7181 struct thandle *handle)
7186 static int lfsck_orphan_index_delete(const struct lu_env *env,
7187 struct dt_object *dt,
7188 const struct dt_key *key,
7189 struct thandle *handle)
7194 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
7195 struct dt_object *dt,
7198 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
7199 struct lfsck_instance *lfsck;
7200 struct lfsck_component *com = NULL;
7201 struct lfsck_layout_slave_data *llsd;
7202 struct lfsck_orphan_it *it = NULL;
7203 struct lfsck_layout *lo;
7207 lfsck = lfsck_instance_find(dev, true, false);
7208 if (unlikely(lfsck == NULL))
7209 RETURN(ERR_PTR(-ENXIO));
7211 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
7212 if (unlikely(com == NULL))
7213 GOTO(out, rc = -ENOENT);
7215 lo = com->lc_file_ram;
7216 if (lo->ll_flags & LF_INCOMPLETE)
7217 GOTO(out, rc = -ESRCH);
7219 llsd = com->lc_data;
7220 if (!llsd->llsd_rbtree_valid)
7221 GOTO(out, rc = -ESRCH);
7225 GOTO(out, rc = -ENOMEM);
7227 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
7228 if (it->loi_llst == NULL)
7229 GOTO(out, rc = -ENXIO);
7231 if (dev->dd_record_fid_accessed) {
7232 /* The first iteration against the rbtree, scan the whole rbtree
7233 * to remove the nodes which do NOT need to be handled. */
7234 down_write(&llsd->llsd_rb_rwsem);
7235 if (dev->dd_record_fid_accessed) {
7236 struct rb_node *node;
7237 struct rb_node *next;
7238 struct lfsck_rbtree_node *lrn;
7240 /* No need to record the fid accessing anymore. */
7241 dev->dd_record_fid_accessed = 0;
7243 node = rb_first(&llsd->llsd_rb_root);
7244 while (node != NULL) {
7245 next = rb_next(node);
7246 lrn = rb_entry(node, struct lfsck_rbtree_node,
7248 if (atomic_read(&lrn->lrn_known_count) <=
7249 atomic_read(&lrn->lrn_accessed_count)) {
7250 rb_erase(node, &llsd->llsd_rb_root);
7251 lfsck_rbtree_free(lrn);
7256 up_write(&llsd->llsd_rb_rwsem);
7259 /* read lock the rbtree when init, and unlock when fini */
7260 down_read(&llsd->llsd_rb_rwsem);
7268 lfsck_component_put(env, com);
7270 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
7271 lfsck_lfsck2name(lfsck), rc);
7273 lfsck_instance_put(env, lfsck);
7278 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
7281 return (struct dt_it *)it;
7284 static void lfsck_orphan_it_fini(const struct lu_env *env,
7287 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7288 struct lfsck_component *com = it->loi_com;
7289 struct lfsck_layout_slave_data *llsd;
7290 struct lfsck_layout_slave_target *llst;
7293 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
7294 lfsck_lfsck2name(com->lc_lfsck));
7296 llsd = com->lc_data;
7297 up_read(&llsd->llsd_rb_rwsem);
7298 llst = it->loi_llst;
7299 LASSERT(llst != NULL);
7301 /* Save the key and hash for iterate next. */
7302 llst->llst_fid = it->loi_key;
7303 llst->llst_hash = it->loi_hash;
7304 lfsck_layout_llst_put(llst);
7305 lfsck_component_put(env, com);
7311 * \retval +1: the iteration finished
7312 * \retval 0: on success, not finished
7313 * \retval -ve: on error
7315 static int lfsck_orphan_it_next(const struct lu_env *env,
7318 struct lfsck_thread_info *info = lfsck_env_info(env);
7319 struct filter_fid *ff = &info->lti_ff;
7320 struct lu_attr *la = &info->lti_la;
7321 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7322 struct lu_fid *key = &it->loi_key;
7323 struct lu_orphan_rec_v3 *rec = &it->loi_rec;
7324 struct ost_layout *ol = &rec->lor_layout;
7325 struct lfsck_component *com = it->loi_com;
7326 struct lfsck_instance *lfsck = com->lc_lfsck;
7327 struct lfsck_layout_slave_data *llsd = com->lc_data;
7328 struct dt_object *obj;
7329 struct lfsck_rbtree_node *lrn;
7333 __u32 idx = it->loi_llst->llst_index;
7343 lrn = lfsck_rbtree_search(llsd, key, &exact);
7351 key->f_seq = lrn->lrn_seq;
7352 key->f_oid = lrn->lrn_first_oid;
7357 if (unlikely(key->f_oid == 0)) {
7364 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
7370 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
7371 atomic_read(&lrn->lrn_accessed_count))) {
7372 struct rb_node *next = rb_next(&lrn->lrn_node);
7374 while (next != NULL) {
7375 lrn = rb_entry(next, struct lfsck_rbtree_node,
7377 if (atomic_read(&lrn->lrn_known_count) >
7378 atomic_read(&lrn->lrn_accessed_count))
7380 next = rb_next(next);
7389 key->f_seq = lrn->lrn_seq;
7390 key->f_oid = lrn->lrn_first_oid;
7394 pos = key->f_oid - lrn->lrn_first_oid;
7397 pos = find_next_bit(lrn->lrn_known_bitmap,
7398 LFSCK_RBTREE_BITMAP_WIDTH, pos);
7399 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
7400 key->f_oid = lrn->lrn_first_oid + pos;
7401 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
7409 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
7414 key->f_oid = lrn->lrn_first_oid + pos;
7415 obj = lfsck_object_find_bottom(env, lfsck, key);
7418 if (rc == -ENOENT) {
7425 dt_read_lock(env, obj, 0);
7426 if (dt_object_exists(obj) == 0 ||
7427 lfsck_is_dead_obj(obj)) {
7428 dt_read_unlock(env, obj);
7429 lfsck_object_put(env, obj);
7434 rc = dt_attr_get(env, obj, la);
7438 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, ff, sizeof(*ff)),
7440 if (rc == -ENODATA) {
7441 /* For the pre-created OST-object, update the bitmap to avoid
7442 * others LFSCK (second phase) iteration to touch it again. */
7443 if (la->la_ctime == 0) {
7444 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
7445 atomic_inc(&lrn->lrn_accessed_count);
7447 /* For the race between repairing dangling referenced
7448 * MDT-object and unlink the file, it may left orphan
7449 * OST-object there. Destroy it now! */
7450 if (unlikely(!(la->la_mode & S_ISUID))) {
7451 dt_read_unlock(env, obj);
7452 lfsck_layout_destroy_orphan(env, lfsck, obj);
7453 lfsck_object_put(env, obj);
7457 } else if (idx == 0) {
7458 /* If the orphan OST-object has no parent information,
7459 * regard it as referenced by the MDT-object on MDT0. */
7460 fid_zero(&rec->lor_rec.lor_fid);
7461 rec->lor_rec.lor_uid = la->la_uid;
7462 rec->lor_rec.lor_gid = la->la_gid;
7463 memset(ol, 0, sizeof(*ol));
7464 rec->lor_layout_version = 0;
7470 dt_read_unlock(env, obj);
7471 lfsck_object_put(env, obj);
7476 if (rc < sizeof(struct lu_fid))
7477 GOTO(out, rc = (rc < 0 ? rc : -EINVAL));
7479 fid_le_to_cpu(&rec->lor_rec.lor_fid, &ff->ff_parent);
7480 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
7481 * MDT-object's FID::f_ver, instead it is the OST-object index in its
7482 * parent MDT-object's layout EA. */
7483 save = rec->lor_rec.lor_fid.f_stripe_idx;
7484 rec->lor_rec.lor_fid.f_ver = 0;
7485 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_rec.lor_fid, idx);
7486 /* If the orphan OST-object does not claim the MDT, then next.
7488 * If we do not know whether it matches or not, then return it
7489 * to the MDT for further check. */
7491 dt_read_unlock(env, obj);
7492 lfsck_object_put(env, obj);
7497 rec->lor_rec.lor_fid.f_stripe_idx = save;
7498 rec->lor_rec.lor_uid = la->la_uid;
7499 rec->lor_rec.lor_gid = la->la_gid;
7500 ost_layout_le_to_cpu(ol, &ff->ff_layout);
7501 rec->lor_layout_version =
7502 le32_to_cpu(ff->ff_layout_version & ~LU_LAYOUT_RESYNC);
7503 rec->lor_range = le32_to_cpu(ff->ff_range);
7505 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u, "
7506 "stripe size %u, stripe count %u, COMP id %u, COMP start %llu, "
7507 "COMP end %llu, layout version %u, range %u\n",
7508 lfsck_lfsck2name(com->lc_lfsck), PFID(key),
7509 PFID(&rec->lor_rec.lor_fid), rec->lor_rec.lor_uid,
7510 rec->lor_rec.lor_gid, ol->ol_stripe_size, ol->ol_stripe_count,
7511 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
7512 rec->lor_layout_version, rec->lor_range);
7517 dt_read_unlock(env, obj);
7518 lfsck_object_put(env, obj);
7526 * \retval +1: locate to the exactly position
7527 * \retval 0: cannot locate to the exactly position,
7528 * call next() to move to a valid position.
7529 * \retval -ve: on error
7531 static int lfsck_orphan_it_get(const struct lu_env *env,
7533 const struct dt_key *key)
7535 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7538 it->loi_key = *(struct lu_fid *)key;
7539 rc = lfsck_orphan_it_next(env, di);
7549 static void lfsck_orphan_it_put(const struct lu_env *env,
7554 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
7555 const struct dt_it *di)
7557 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7559 return (struct dt_key *)&it->loi_key;
7562 static int lfsck_orphan_it_key_size(const struct lu_env *env,
7563 const struct dt_it *di)
7565 return sizeof(struct lu_fid);
7568 static int lfsck_orphan_it_rec(const struct lu_env *env,
7569 const struct dt_it *di,
7573 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7575 *(struct lu_orphan_rec_v3 *)rec = it->loi_rec;
7580 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
7581 const struct dt_it *di)
7583 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7585 return it->loi_hash;
7589 * \retval +1: locate to the exactly position
7590 * \retval 0: cannot locate to the exactly position,
7591 * call next() to move to a valid position.
7592 * \retval -ve: on error
7594 static int lfsck_orphan_it_load(const struct lu_env *env,
7595 const struct dt_it *di,
7598 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7599 struct lfsck_layout_slave_target *llst = it->loi_llst;
7602 LASSERT(llst != NULL);
7604 if (hash != llst->llst_hash) {
7605 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
7606 "iteration does not match the one when fini "
7607 "%llu, to be reset.\n",
7608 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
7610 fid_zero(&llst->llst_fid);
7611 llst->llst_hash = 0;
7614 it->loi_key = llst->llst_fid;
7615 it->loi_hash = llst->llst_hash;
7616 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
7626 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
7627 const struct dt_it *di,
7633 static const struct dt_index_operations lfsck_orphan_index_ops = {
7634 .dio_lookup = lfsck_orphan_index_lookup,
7635 .dio_declare_insert = lfsck_orphan_index_declare_insert,
7636 .dio_insert = lfsck_orphan_index_insert,
7637 .dio_declare_delete = lfsck_orphan_index_declare_delete,
7638 .dio_delete = lfsck_orphan_index_delete,
7640 .init = lfsck_orphan_it_init,
7641 .fini = lfsck_orphan_it_fini,
7642 .get = lfsck_orphan_it_get,
7643 .put = lfsck_orphan_it_put,
7644 .next = lfsck_orphan_it_next,
7645 .key = lfsck_orphan_it_key,
7646 .key_size = lfsck_orphan_it_key_size,
7647 .rec = lfsck_orphan_it_rec,
7648 .store = lfsck_orphan_it_store,
7649 .load = lfsck_orphan_it_load,
7650 .key_rec = lfsck_orphan_it_key_rec,