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 (CFS_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 if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1603 idx = lfsck_sub_trace_file_fid2idx(pfid);
1604 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1605 dev = lfsck_obj2dev(obj);
1607 fid_cpu_to_be(&key->lldk_fid, pfid);
1608 key->lldk_comp_id = cpu_to_be32(comp_id);
1609 key->lldk_ea_off = cpu_to_be32(ea_off);
1611 fid_cpu_to_be(rec, cfid);
1612 rec->f_ver = cpu_to_be32(ost_idx);
1614 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1616 th = lfsck_trans_create(env, dev, com->lc_lfsck);
1618 GOTO(unlock, rc = PTR_ERR(th));
1620 rc = dt_declare_insert(env, obj,
1621 (const struct dt_rec *)rec,
1622 (const struct dt_key *)key, th);
1626 rc = dt_trans_start_local(env, dev, th);
1630 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
1631 (const struct dt_key *)key, th);
1636 if (th && !IS_ERR(th))
1637 dt_trans_stop(env, dev, th);
1639 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1642 CDEBUG(D_LFSCK, "%s: insert the paris "DFID" => "DFID", comp_id = %u, "
1643 "ea_off = %u, ost_idx = %u, into the trace file for further "
1644 "dangling check: rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
1645 PFID(pfid), PFID(cfid), comp_id, ea_off, ost_idx, rc);
1650 static int lfsck_layout_del_dangling_rec(const struct lu_env *env,
1651 struct lfsck_component *com,
1652 const struct lu_fid *fid,
1653 __u32 comp_id, __u32 ea_off)
1655 struct lfsck_layout_dangling_key *key = &lfsck_env_info(env)->lti_lldk;
1656 struct dt_device *dev;
1657 struct dt_object *obj;
1658 struct thandle *th = NULL;
1663 if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1666 idx = lfsck_sub_trace_file_fid2idx(fid);
1667 obj = com->lc_sub_trace_objs[idx].lsto_obj;
1668 dev = lfsck_obj2dev(obj);
1670 fid_cpu_to_be(&key->lldk_fid, fid);
1671 key->lldk_comp_id = cpu_to_be32(comp_id);
1672 key->lldk_ea_off = cpu_to_be32(ea_off);
1674 mutex_lock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1676 th = lfsck_trans_create(env, dev, com->lc_lfsck);
1678 GOTO(unlock, rc = PTR_ERR(th));
1680 rc = dt_declare_delete(env, obj, (const struct dt_key *)key, th);
1684 rc = dt_trans_start_local(env, dev, th);
1688 rc = dt_delete(env, obj, (const struct dt_key *)key, th);
1693 if (th && !IS_ERR(th))
1694 dt_trans_stop(env, dev, th);
1696 mutex_unlock(&com->lc_sub_trace_objs[idx].lsto_mutex);
1699 CDEBUG(D_LFSCK, "%s: delete the dangling record for "DFID
1700 ", comp_id = %u, ea_off = %u from the trace file: rc = %d\n",
1701 lfsck_lfsck2name(com->lc_lfsck), PFID(fid), comp_id, ea_off, rc);
1707 * Get the system default stripe size.
1709 * \param[in] env pointer to the thread context
1710 * \param[in] lfsck pointer to the lfsck instance
1711 * \param[out] size pointer to the default stripe size
1713 * \retval 0 for success
1714 * \retval negative error number on failure
1716 static int lfsck_layout_get_def_stripesize(const struct lu_env *env,
1717 struct lfsck_instance *lfsck,
1720 struct lov_user_md *lum = &lfsck_env_info(env)->lti_lum;
1721 struct dt_object *root;
1724 root = dt_locate(env, lfsck->li_next, &lfsck->li_local_root_fid);
1726 return PTR_ERR(root);
1728 /* Get the default stripe size via xattr_get on the backend root. */
1729 rc = dt_xattr_get(env, root, lfsck_buf_get(env, lum, sizeof(*lum)),
1732 /* The lum->lmm_stripe_size is LE mode. The *size also
1733 * should be LE mode. So it is unnecessary to convert. */
1734 *size = lum->lmm_stripe_size;
1736 } else if (unlikely(rc == 0)) {
1740 lfsck_object_put(env, root);
1746 * \retval +1: repaired
1747 * \retval 0: did nothing
1748 * \retval -ve: on error
1750 static int lfsck_layout_refill_lovea(const struct lu_env *env,
1751 struct lfsck_instance *lfsck,
1752 struct thandle *handle,
1753 struct dt_object *parent,
1754 const struct lu_fid *cfid,
1756 struct lov_mds_md_v1 *lmm,
1757 struct lov_ost_data_v1 *slot,
1758 int fl, __u32 ost_idx, int size)
1760 struct ost_id *oi = &lfsck_env_info(env)->lti_oi;
1761 struct lu_buf ea_buf;
1768 magic = le32_to_cpu(lmm->lmm_magic);
1769 pattern = le32_to_cpu(lmm->lmm_pattern);
1770 count = le16_to_cpu(lmm->lmm_stripe_count);
1772 fid_to_ostid(cfid, oi);
1773 ostid_cpu_to_le(oi, &slot->l_ost_oi);
1774 slot->l_ost_gen = cpu_to_le32(0);
1775 slot->l_ost_idx = cpu_to_le32(ost_idx);
1777 if (pattern & LOV_PATTERN_F_HOLE) {
1778 struct lov_ost_data_v1 *objs;
1781 if (magic == LOV_MAGIC_V1)
1782 objs = &lmm->lmm_objects[0];
1784 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
1785 for (i = 0; i < count; i++, objs++) {
1786 if (lovea_slot_is_dummy(objs))
1790 /* If the @slot is the last dummy slot to be refilled,
1791 * then drop LOV_PATTERN_F_HOLE from lmm::lmm_pattern. */
1794 cpu_to_le32(pattern & ~LOV_PATTERN_F_HOLE);
1796 CDEBUG(D_LFSCK, "%s: remove layout HOLE for "DFID
1797 ": parent "DFID"\n", lfsck_lfsck2name(lfsck),
1798 PFID(cfid), PFID(lfsck_dto2fid(parent)));
1802 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
1803 rc = dt_xattr_set(env, parent, &ea_buf, XATTR_NAME_LOV, fl, handle);
1810 static struct lov_ost_data_v1 *
1811 __lfsck_layout_new_v1_lovea(struct lov_mds_md_v1 *lmm,
1812 const struct lu_fid *pfid,
1813 __u32 stripe_size, __u32 ea_off,
1814 __u32 pattern, __u16 count)
1816 lmm->lmm_magic = cpu_to_le32(LOV_MAGIC_V1);
1817 lmm->lmm_pattern = cpu_to_le32(pattern);
1818 fid_to_lmm_oi(pfid, &lmm->lmm_oi);
1819 lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
1820 lmm->lmm_stripe_size = cpu_to_le32(stripe_size);
1821 lmm->lmm_stripe_count = cpu_to_le16(count);
1822 lmm->lmm_layout_gen = cpu_to_le16(1);
1823 memset(&lmm->lmm_objects[0], 0,
1824 sizeof(struct lov_ost_data_v1) * count);
1826 return &lmm->lmm_objects[ea_off];
1829 static int lfsck_layout_new_v1_lovea(const struct lu_env *env,
1830 struct lfsck_instance *lfsck,
1831 struct ost_layout *ol,
1832 struct dt_object *parent,
1833 struct lu_buf *buf, __u32 ea_off,
1834 struct lov_mds_md_v1 **lmm,
1835 struct lov_ost_data_v1 **objs)
1838 __u32 stripe_size = ol->ol_stripe_size;
1839 __u32 pattern = LOV_PATTERN_RAID0;
1842 if (ol->ol_stripe_count != 0)
1843 count = ol->ol_stripe_count;
1847 size = lov_mds_md_size(count, LOV_MAGIC_V1);
1848 LASSERTF(buf->lb_len >= size,
1849 "buffer len %d is less than real size %d\n",
1850 (int)buf->lb_len, size);
1852 if (stripe_size == 0) {
1855 rc = lfsck_layout_get_def_stripesize(env, lfsck, &stripe_size);
1861 if (ol->ol_stripe_count > 1 ||
1862 (ol->ol_stripe_count == 0 && ea_off != 0)) {
1863 pattern |= LOV_PATTERN_F_HOLE;
1864 memset(&(*lmm)->lmm_objects[0], 0,
1865 count * sizeof(struct lov_ost_data_v1));
1868 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1869 stripe_size, ea_off, pattern, count);
1874 static int lfsck_layout_new_comp_lovea(const struct lu_env *env,
1875 struct lu_orphan_rec_v3 *rec,
1876 struct dt_object *parent,
1877 struct lu_buf *buf, __u32 ea_off,
1878 struct lov_mds_md_v1 **lmm,
1879 struct lov_ost_data_v1 **objs)
1881 struct ost_layout *ol = &rec->lor_layout;
1882 struct lov_comp_md_v1 *lcm;
1883 struct lov_comp_md_entry_v1 *lcme;
1884 __u32 pattern = LOV_PATTERN_RAID0;
1885 __u32 offset = sizeof(*lcm) + sizeof(*lcme);
1886 int lcme_size = lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
1887 int size = offset + lcme_size;
1889 LASSERTF(buf->lb_len >= size,
1890 "buffer len %d is less than real size %d\n",
1891 (int)buf->lb_len, size);
1894 lcm->lcm_magic = cpu_to_le32(LOV_MAGIC_COMP_V1);
1895 lcm->lcm_size = cpu_to_le32(size);
1896 if (rec->lor_range) {
1897 lcm->lcm_layout_gen = cpu_to_le32(rec->lor_layout_version +
1899 lcm->lcm_flags = cpu_to_le16(LCM_FL_WRITE_PENDING);
1900 } else if (rec->lor_layout_version) {
1901 lcm->lcm_layout_gen = cpu_to_le32(rec->lor_layout_version +
1903 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
1906 * if OST doesn't provide layout version, then try
1907 * to inherit one from MDS's layout, but increment
1908 * it so the client notices and applies modified
1911 le32_add_cpu(&lcm->lcm_layout_gen, 1);
1912 lcm->lcm_flags = cpu_to_le16(LCM_FL_NONE);
1914 lcm->lcm_entry_count = cpu_to_le16(1);
1915 /* Currently, we do not know how many mirrors will be, set it as zero
1916 * at the beginning. It will be updated when more mirrors are found. */
1917 lcm->lcm_mirror_count = 0;
1919 lcme = &lcm->lcm_entries[0];
1920 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
1921 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
1922 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
1923 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
1924 lcme->lcme_offset = cpu_to_le32(offset);
1925 lcme->lcme_size = cpu_to_le32(lcme_size);
1926 lcme->lcme_layout_gen = lcm->lcm_layout_gen;
1927 if (ol->ol_stripe_count > 1)
1928 pattern |= LOV_PATTERN_F_HOLE;
1930 *lmm = buf->lb_buf + offset;
1931 *objs = __lfsck_layout_new_v1_lovea(*lmm, lfsck_dto2fid(parent),
1932 ol->ol_stripe_size, ea_off,
1933 pattern, ol->ol_stripe_count);
1938 static void lfsck_layout_update_lcm(struct lov_comp_md_v1 *lcm,
1939 struct lov_comp_md_entry_v1 *lcme,
1940 __u32 version, __u32 range)
1942 struct lov_comp_md_entry_v1 *tmp;
1943 __u64 start = le64_to_cpu(lcme->lcme_extent.e_start);
1944 __u64 end = le64_to_cpu(lcme->lcme_extent.e_end);
1945 __u32 gen = version + range;
1948 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
1949 __u16 flags = le16_to_cpu(lcm->lcm_flags);
1953 lcme->lcme_layout_gen = cpu_to_le32(gen);
1954 if (le32_to_cpu(lcm->lcm_layout_gen) < gen)
1955 lcm->lcm_layout_gen = cpu_to_le32(gen);
1958 lcm->lcm_flags = cpu_to_le16(LCM_FL_WRITE_PENDING);
1959 else if (flags == LCM_FL_NONE && le16_to_cpu(lcm->lcm_mirror_count) > 0)
1960 lcm->lcm_flags = cpu_to_le16(LCM_FL_RDONLY);
1962 for (i = 0; i < count; i++) {
1963 tmp = &lcm->lcm_entries[i];
1964 if (le64_to_cpu(tmp->lcme_extent.e_end) <= start)
1967 if (le64_to_cpu(tmp->lcme_extent.e_start) >= end)
1970 if (le32_to_cpu(tmp->lcme_flags) & LCME_FL_STALE)
1973 tmp_gen = le32_to_cpu(tmp->lcme_layout_gen);
1974 /* "lcme_layout_gen == 0" but without LCME_FL_STALE flag,
1975 * then it should be the latest version of all mirrors. */
1976 if (tmp_gen == 0 || tmp_gen > gen) {
1977 lcme->lcme_flags = cpu_to_le32(
1978 le32_to_cpu(lcme->lcme_flags) | LCME_FL_STALE);
1983 tmp->lcme_flags = cpu_to_le32(
1984 le32_to_cpu(tmp->lcme_flags) | LCME_FL_STALE);
1988 static int lfsck_layout_add_comp(const struct lu_env *env,
1989 struct lfsck_instance *lfsck,
1990 struct thandle *handle,
1991 struct lu_orphan_rec_v3 *rec,
1992 struct dt_object *parent,
1993 const struct lu_fid *cfid,
1994 struct lu_buf *buf, __u32 ost_idx,
1995 __u32 ea_off, int pos, bool new_mirror)
1997 struct ost_layout *ol = &rec->lor_layout;
1998 struct lov_comp_md_v1 *lcm = buf->lb_buf;
1999 struct lov_comp_md_entry_v1 *lcme;
2000 struct lov_mds_md_v1 *lmm;
2001 struct lov_ost_data_v1 *objs;
2002 int added = sizeof(*lcme) +
2003 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2004 int size = le32_to_cpu(lcm->lcm_size) + added;
2008 __u32 pattern = LOV_PATTERN_RAID0;
2009 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
2012 lu_buf_check_and_grow(buf, size);
2013 /* set the lcm again because lu_buf_check_and_grow() may
2014 * have reallocated the buf. */
2016 lcm->lcm_size = cpu_to_le32(size);
2017 lcm->lcm_entry_count = cpu_to_le16(count + 1);
2019 le16_add_cpu(&lcm->lcm_mirror_count, 1);
2021 /* 1. Move the component bodies from [pos, count-1] to [pos+1, count]
2022 * with distance of 'added'. */
2025 for (i = pos; i < count; i++) {
2026 lcme = &lcm->lcm_entries[i];
2027 size += le32_to_cpu(lcme->lcme_size);
2030 offset = le32_to_cpu(lcm->lcm_entries[pos].lcme_offset);
2031 memmove(buf->lb_buf + offset + added,
2032 buf->lb_buf + offset, size);
2036 /* 2. Move the component header [0, pos-1] to [0, pos-1] with distance
2037 * of 'sizeof(struct lov_comp_md_entry_v1)' */
2039 for (i = 0; i < pos; i++) {
2040 lcme = &lcm->lcm_entries[i];
2041 size += le32_to_cpu(lcme->lcme_size);
2044 offset = le32_to_cpu(lcm->lcm_entries[0].lcme_offset);
2045 memmove(buf->lb_buf + offset + sizeof(*lcme),
2046 buf->lb_buf + offset, size);
2049 /* 3. Recalculate the enter offset for the component [pos, count-1] */
2050 for (i = count - 1; i >= pos; i--) {
2051 lcm->lcm_entries[i + 1] = lcm->lcm_entries[i];
2052 lcm->lcm_entries[i + 1].lcme_offset =
2053 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i + 1].
2054 lcme_offset) + added);
2057 /* 4. Recalculate the enter offset for the component [0, pos) */
2058 for (i = 0; i < pos; i++) {
2059 lcm->lcm_entries[i].lcme_offset =
2060 cpu_to_le32(le32_to_cpu(lcm->lcm_entries[i].
2061 lcme_offset) + sizeof(*lcme));
2064 offset = sizeof(*lcm) + sizeof(*lcme) * (count + 1) + size;
2065 /* 4. Insert the new component header (entry) at the slot 'pos'. */
2066 lcme = &lcm->lcm_entries[pos];
2067 lcme->lcme_id = cpu_to_le32(ol->ol_comp_id);
2068 lcme->lcme_flags = cpu_to_le32(LCME_FL_INIT);
2069 lcme->lcme_extent.e_start = cpu_to_le64(ol->ol_comp_start);
2070 lcme->lcme_extent.e_end = cpu_to_le64(ol->ol_comp_end);
2071 lcme->lcme_offset = cpu_to_le32(offset);
2072 lcme->lcme_size = cpu_to_le32(lov_mds_md_size(ol->ol_stripe_count,
2075 if (ol->ol_stripe_count > 1)
2076 pattern |= LOV_PATTERN_F_HOLE;
2078 lmm = buf->lb_buf + offset;
2079 /* 5. Insert teh new component body at the 'offset'. */
2080 objs = __lfsck_layout_new_v1_lovea(lmm, lfsck_dto2fid(parent),
2081 ol->ol_stripe_size, ea_off,
2082 pattern, ol->ol_stripe_count);
2084 /* 6. Update mirror related flags and version. */
2085 lfsck_layout_update_lcm(lcm, lcme, rec->lor_layout_version,
2088 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
2089 lmm, objs, LU_XATTR_REPLACE, ost_idx,
2090 le32_to_cpu(lcm->lcm_size));
2092 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant add new COMP for "
2093 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2094 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2095 "comp_end %llu, layout version %u, range %u, "
2096 "%s LOV EA hole: rc = %d\n",
2097 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2098 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2099 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2100 rec->lor_layout_version, rec->lor_range,
2101 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2102 "with" : "without", rc);
2107 static int lfsck_layout_extend_v1v3_lovea(const struct lu_env *env,
2108 struct lfsck_instance *lfsck,
2109 struct thandle *handle,
2110 struct ost_layout *ol,
2111 struct dt_object *parent,
2112 const struct lu_fid *cfid,
2113 struct lu_buf *buf, __u32 ost_idx,
2116 struct lov_mds_md_v1 *lmm = buf->lb_buf;
2117 struct lov_ost_data_v1 *objs;
2118 __u16 count = le16_to_cpu(lmm->lmm_stripe_count);
2119 __u32 magic = le32_to_cpu(lmm->lmm_magic);
2125 /* The original LOVEA maybe re-generated via old filter_fid, at
2126 * that time, we do not know the stripe count and stripe size. */
2127 if (ol->ol_stripe_count > count)
2128 count = ol->ol_stripe_count;
2129 if (ol->ol_stripe_size != 0 &&
2130 ol->ol_stripe_size != le32_to_cpu(lmm->lmm_stripe_size))
2131 lmm->lmm_stripe_size = cpu_to_le32(ol->ol_stripe_size);
2133 if (magic == LOV_MAGIC_V1)
2134 objs = &lmm->lmm_objects[count];
2136 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[count];
2138 gap = ea_off - count;
2142 size = lov_mds_md_size(count, magic);
2143 LASSERTF(buf->lb_len >= size,
2144 "buffer len %d is less than real size %d\n",
2145 (int)buf->lb_len, size);
2148 memset(objs, 0, gap * sizeof(*objs));
2149 lmm->lmm_pattern |= cpu_to_le32(LOV_PATTERN_F_HOLE);
2152 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2153 lmm->lmm_stripe_count = cpu_to_le16(count);
2156 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid, buf,
2157 lmm, objs, LU_XATTR_REPLACE, ost_idx, size);
2159 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant extend layout EA for "
2160 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2161 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2162 "comp_end %llu, %s LOV EA hole: rc = %d\n",
2163 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2164 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2165 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2166 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2167 "with" : "without", rc);
2173 * \retval +1: repaired
2174 * \retval 0: did nothing
2175 * \retval -ve: on error
2177 static int lfsck_layout_update_lovea(const struct lu_env *env,
2178 struct lfsck_instance *lfsck,
2179 struct thandle *handle,
2180 struct lu_orphan_rec_v3 *rec,
2181 struct dt_object *parent,
2182 const struct lu_fid *cfid,
2183 struct lu_buf *buf, int fl,
2184 __u32 ost_idx, __u32 ea_off)
2186 struct ost_layout *ol = &rec->lor_layout;
2187 struct lov_mds_md_v1 *lmm = NULL;
2188 struct lov_ost_data_v1 *objs = NULL;
2192 if (ol->ol_comp_id != 0)
2193 rc = lfsck_layout_new_comp_lovea(env, rec, parent, buf, ea_off,
2196 rc = lfsck_layout_new_v1_lovea(env, lfsck, &rec->lor_layout,
2197 parent, buf, ea_off, &lmm,
2200 rc = lfsck_layout_refill_lovea(env, lfsck, handle, parent, cfid,
2201 buf, lmm, objs, fl, ost_idx, rc);
2203 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant created layout EA for "
2204 DFID": parent "DFID", OST-index %u, stripe-index %u, "
2205 "stripe_size %u, stripe_count %u, comp_id %u, comp_start %llu, "
2206 "comp_end %llu, layout version %u, range %u, fl %d, "
2207 "%s LOV EA hole: rc = %d\n",
2208 lfsck_lfsck2name(lfsck), PFID(cfid), PFID(lfsck_dto2fid(parent)),
2209 ost_idx, ea_off, ol->ol_stripe_size, ol->ol_stripe_count,
2210 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
2211 rec->lor_layout_version, rec->lor_range, fl,
2212 le32_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_F_HOLE ?
2213 "with" : "without", rc);
2218 static int __lfsck_layout_update_pfid(const struct lu_env *env,
2219 struct lfsck_component *com,
2220 struct dt_object *child,
2221 const struct lu_fid *pfid,
2222 const struct ost_layout *ol, __u32 offset,
2223 __u32 version, __u32 range)
2225 struct dt_device *dev = lfsck_obj2dev(child);
2226 struct filter_fid *ff = &lfsck_env_info(env)->lti_ff;
2227 struct thandle *handle;
2228 struct lu_buf buf = { NULL };
2231 if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2234 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
2235 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
2236 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
2237 * MDT-object's FID::f_ver, instead it is the OST-object index in its
2238 * parent MDT-object's layout EA. */
2239 ff->ff_parent.f_stripe_idx = cpu_to_le32(offset);
2240 ost_layout_cpu_to_le(&ff->ff_layout, ol);
2241 ff->ff_layout_version = cpu_to_le32(version);
2242 ff->ff_range = cpu_to_le32(range);
2243 lfsck_buf_init(&buf, ff, sizeof(*ff));
2245 if (!dt_object_exists(child) || lfsck_is_dead_obj(child))
2248 handle = lfsck_trans_create(env, dev, com->lc_lfsck);
2250 RETURN(PTR_ERR(handle));
2252 rc = dt_declare_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2256 rc = dt_trans_start_local(env, dev, handle);
2260 dt_write_lock(env, child, 0);
2261 if (dt_object_exists(child) && !lfsck_is_dead_obj(child))
2262 rc = dt_xattr_set(env, child, &buf, XATTR_NAME_FID, 0, handle);
2263 dt_write_unlock(env, child);
2268 dt_trans_stop(env, dev, handle);
2274 * \retval +1: repaired
2275 * \retval 0: did nothing
2276 * \retval -ve: on error
2278 static int lfsck_layout_update_pfid(const struct lu_env *env,
2279 struct lfsck_component *com,
2280 struct dt_object *parent,
2281 struct lu_fid *cfid,
2282 struct dt_device *cdev,
2283 struct lu_orphan_rec_v3 *rec, __u32 ea_off)
2285 struct dt_object *child;
2289 child = lfsck_object_find_by_dev(env, cdev, cfid);
2291 RETURN(PTR_ERR(child));
2293 rc = __lfsck_layout_update_pfid(env, com, child,
2294 lu_object_fid(&parent->do_lu),
2295 &rec->lor_layout, ea_off,
2296 rec->lor_layout_version,
2298 lfsck_object_put(env, child);
2300 RETURN(rc == 0 ? 1 : rc);
2303 static int lfsck_lovea_size(struct ost_layout *ol, __u32 ea_off)
2305 if (ol->ol_comp_id != 0)
2306 return sizeof(struct lov_comp_md_v1) +
2307 sizeof(struct lov_comp_md_entry_v1) +
2308 lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2310 if (ol->ol_stripe_count != 0)
2311 return lov_mds_md_size(ol->ol_stripe_count, LOV_MAGIC_V1);
2313 return lov_mds_md_size(ea_off + 1, LOV_MAGIC_V1);
2317 * This function will create the MDT-object with the given (partial) LOV EA.
2319 * Under some data corruption cases, the MDT-object of the file may be lost,
2320 * but its OST-objects, or some of them are there. The layout LFSCK needs to
2321 * re-create the MDT-object with the orphan OST-object(s) information.
2323 * On the other hand, the LFSCK may has created some OST-object for repairing
2324 * dangling LOV EA reference, but as the LFSCK processing, it may find that
2325 * the old OST-object is there and should replace the former new created OST
2326 * object. Unfortunately, some others have modified such newly created object.
2327 * To keep the data (both new and old), the LFSCK will create MDT-object with
2328 * new FID to reference the original OST-object.
2330 * \param[in] env pointer to the thread context
2331 * \param[in] com pointer to the lfsck component
2332 * \param[in] ltd pointer to target device descriptor
2333 * \param[in] rec pointer to the record for the orphan OST-object
2334 * \param[in] cfid pointer to FID for the orphan OST-object
2335 * \param[in] infix additional information, such as the FID for original
2336 * MDT-object and the stripe offset in the LOV EA
2337 * \param[in] type the type for describing why the orphan MDT-object is
2338 * created. The rules are as following:
2340 * type "C": Multiple OST-objects claim the same MDT-object and the
2341 * same slot in the layout EA. Then the LFSCK will create
2342 * new MDT-object(s) to hold the conflict OST-object(s).
2344 * type "N": The orphan OST-object does not know which one was the
2345 * real parent MDT-object, so the LFSCK uses new FID for
2346 * its parent MDT-object.
2348 * type "R": The orphan OST-object knows its parent MDT-object FID,
2349 * but does not know the position (the file name) in the
2352 * type "D": The MDT-object is a directory, it may knows its parent
2353 * but because there is no valid linkEA, the LFSCK cannot
2354 * know where to put it back to the namespace.
2355 * type "O": The MDT-object has no linkEA, and there is no name
2356 * entry that references the MDT-object.
2358 * type "P": The orphan object to be created was a parent directory
2359 * of some MDT-object which linkEA shows that the @orphan
2360 * object is missing.
2362 * The orphan name will be like:
2363 * ${FID}-${infix}-${type}-${conflict_version}
2365 * \param[in] ea_off the stripe offset in the LOV EA
2367 * \retval positive on repaired something
2368 * \retval 0 if needs to repair nothing
2369 * \retval negative error number on failure
2371 static int lfsck_layout_recreate_parent(const struct lu_env *env,
2372 struct lfsck_component *com,
2373 struct lfsck_tgt_desc *ltd,
2374 struct lu_orphan_rec_v3 *rec,
2375 struct lu_fid *cfid,
2380 struct lfsck_thread_info *info = lfsck_env_info(env);
2381 struct dt_insert_rec *dtrec = &info->lti_dt_rec;
2382 char *name = info->lti_key;
2383 struct lu_attr *la = &info->lti_la2;
2384 struct dt_object_format *dof = &info->lti_dof;
2385 struct lfsck_instance *lfsck = com->lc_lfsck;
2386 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
2387 struct lu_fid *tfid = &info->lti_fid3;
2388 struct dt_device *dev = lfsck->li_bottom;
2389 struct dt_object *lpf = lfsck->li_lpf_obj;
2390 struct dt_object *pobj = NULL;
2391 struct dt_object *cobj = NULL;
2392 struct thandle *th = NULL;
2393 struct lu_buf *ea_buf = &info->lti_big_buf;
2394 struct lu_buf lov_buf;
2395 struct lfsck_lock_handle *llh = &info->lti_llh;
2396 struct linkea_data ldata = { NULL };
2397 struct lu_buf linkea_buf;
2398 const struct lu_name *pname;
2404 if (lfsck_is_dryrun(lfsck))
2407 if (unlikely(lpf == NULL))
2408 GOTO(log, rc = -ENXIO);
2410 /* We use two separated transactions to repair the inconsistency.
2412 * 1) create the MDT-object locally.
2413 * 2) update the OST-object's PFID EA if necessary.
2415 * If 1) succeed, but 2) failed, then the OST-object's PFID EA will be
2416 * updated when the layout LFSCK run next time.
2418 * If 1) failed, but 2) succeed, then such MDT-object will be re-created
2419 * when the layout LFSCK run next time. */
2421 if (fid_is_zero(pfid)) {
2422 rc = lfsck_fid_alloc(env, lfsck, pfid, false);
2426 cobj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
2428 GOTO(log, rc = PTR_ERR(cobj));
2431 pobj = lfsck_object_find_by_dev(env, dev, pfid);
2433 GOTO(log, rc = PTR_ERR(pobj));
2435 LASSERT(infix != NULL);
2436 LASSERT(type != NULL);
2438 memset(la, 0, sizeof(*la));
2439 la->la_uid = rec->lor_rec.lor_uid;
2440 la->la_gid = rec->lor_rec.lor_gid;
2441 la->la_mode = S_IFREG | S_IRUSR;
2442 la->la_valid = LA_MODE | LA_UID | LA_GID;
2444 memset(dof, 0, sizeof(*dof));
2445 dof->dof_type = dt_mode_to_dft(S_IFREG);
2446 /* Because the dof->dof_reg.striped = 0, the LOD will not create
2447 * the stripe(s). The LFSCK will specify the LOV EA via
2448 * lfsck_layout_update_lovea(). */
2450 size = lfsck_lovea_size(&rec->lor_layout, ea_off);
2451 if (ea_buf->lb_len < size) {
2452 lu_buf_realloc(ea_buf, size);
2453 if (ea_buf->lb_buf == NULL)
2454 GOTO(log, rc = -ENOMEM);
2459 snprintf(name, NAME_MAX, DFID"%s-%s-%d", PFID(pfid), infix,
2461 rc = dt_lookup_dir(env, lfsck->li_lpf_obj, name, tfid);
2462 if (rc != 0 && rc != -ENOENT)
2466 rc = lfsck_lock(env, lfsck, lfsck->li_lpf_obj, name, llh,
2467 MDS_INODELOCK_UPDATE, LCK_PW);
2471 /* Re-check whether the name conflict with othrs after taken
2473 rc = dt_lookup_dir(env, lfsck->li_lpf_obj, name, tfid);
2474 if (unlikely(rc == 0)) {
2482 pname = lfsck_name_get_const(env, name, strlen(name));
2483 rc = linkea_links_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf,
2484 pname, lfsck_dto2fid(lfsck->li_lpf_obj));
2488 /* The 1st transaction. */
2489 th = lfsck_trans_create(env, dev, lfsck);
2491 GOTO(unlock, rc = PTR_ERR(th));
2493 rc = dt_declare_create(env, pobj, la, NULL, dof, th);
2497 lfsck_buf_init(&lov_buf, ea_buf->lb_buf, size);
2498 rc = dt_declare_xattr_set(env, pobj, &lov_buf, XATTR_NAME_LOV,
2499 LU_XATTR_CREATE, th);
2503 dtrec->rec_fid = pfid;
2504 dtrec->rec_type = S_IFREG;
2505 rc = dt_declare_insert(env, lpf,
2506 (const struct dt_rec *)dtrec,
2507 (const struct dt_key *)name, th);
2511 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2512 ldata.ld_leh->leh_len);
2513 rc = dt_declare_xattr_set(env, pobj, &linkea_buf,
2514 XATTR_NAME_LINK, 0, th);
2518 rc = dt_trans_start_local(env, dev, th);
2522 dt_write_lock(env, pobj, 0);
2523 rc = dt_create(env, pobj, la, NULL, dof, th);
2525 rc = lfsck_layout_update_lovea(env, lfsck, th, rec, pobj, cfid,
2526 &lov_buf, LU_XATTR_CREATE, ltd->ltd_index, ea_off);
2527 dt_write_unlock(env, pobj);
2531 rc = dt_insert(env, lpf, (const struct dt_rec *)dtrec,
2532 (const struct dt_key *)name, th);
2536 rc = dt_xattr_set(env, pobj, &linkea_buf, XATTR_NAME_LINK, 0, th);
2537 if (rc == 0 && cobj != NULL) {
2538 dt_trans_stop(env, dev, th);
2541 /* The 2nd transaction. */
2542 rc = __lfsck_layout_update_pfid(env, com, cobj, pfid,
2543 &rec->lor_layout, ea_off,
2544 rec->lor_layout_version,
2552 dt_trans_stop(env, dev, th);
2558 if (cobj != NULL && !IS_ERR(cobj))
2559 lfsck_object_put(env, cobj);
2560 if (pobj != NULL && !IS_ERR(pobj))
2561 lfsck_object_put(env, pobj);
2564 CDEBUG(D_LFSCK, "%s layout LFSCK assistant failed to "
2565 "recreate the lost MDT-object: parent "DFID
2566 ", child "DFID", OST-index %u, stripe-index %u, "
2567 "infix %s, type %s: rc = %d\n",
2568 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
2569 ltd->ltd_index, ea_off, infix, type, rc);
2571 return rc >= 0 ? 1 : rc;
2574 static int lfsck_layout_master_conditional_destroy(const struct lu_env *env,
2575 struct lfsck_component *com,
2576 const struct lu_fid *fid,
2579 struct lfsck_thread_info *info = lfsck_env_info(env);
2580 struct lfsck_request *lr = &info->lti_lr;
2581 struct lfsck_instance *lfsck = com->lc_lfsck;
2582 struct lfsck_tgt_desc *ltd;
2583 struct ptlrpc_request *req;
2584 struct lfsck_request *tmp;
2585 struct obd_export *exp;
2589 ltd = lfsck_tgt_get(&lfsck->li_ost_descs, index);
2590 if (unlikely(ltd == NULL))
2594 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
2595 GOTO(put, rc = -EOPNOTSUPP);
2597 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
2599 GOTO(put, rc = -ENOMEM);
2601 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
2603 ptlrpc_request_free(req);
2608 memset(lr, 0, sizeof(*lr));
2609 lr->lr_event = LE_CONDITIONAL_DESTROY;
2610 lr->lr_active = LFSCK_TYPE_LAYOUT;
2613 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
2615 ptlrpc_request_set_replen(req);
2617 rc = ptlrpc_queue_wait(req);
2618 ptlrpc_req_finished(req);
2628 static int lfsck_layout_slave_conditional_destroy(const struct lu_env *env,
2629 struct lfsck_component *com,
2630 struct lfsck_request *lr)
2632 struct lfsck_thread_info *info = lfsck_env_info(env);
2633 struct lu_attr *la = &info->lti_la;
2634 union ldlm_policy_data *policy = &info->lti_policy;
2635 struct ldlm_res_id *resid = &info->lti_resid;
2636 struct lfsck_instance *lfsck = com->lc_lfsck;
2637 struct dt_device *dev = lfsck->li_bottom;
2638 struct lu_fid *fid = &lr->lr_fid;
2639 struct dt_object *obj;
2640 struct thandle *th = NULL;
2641 struct lustre_handle lh = { 0 };
2646 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2649 obj = lfsck_object_find_by_dev(env, dev, fid);
2651 RETURN(PTR_ERR(obj));
2653 dt_read_lock(env, obj, 0);
2654 if (dt_object_exists(obj) == 0 ||
2655 lfsck_is_dead_obj(obj)) {
2656 dt_read_unlock(env, obj);
2658 GOTO(put, rc = -ENOENT);
2661 /* Get obj's attr without lock firstly. */
2662 rc = dt_attr_get(env, obj, la);
2663 dt_read_unlock(env, obj);
2667 if (likely(la->la_ctime != 0 || la->la_mode & S_ISUID))
2668 GOTO(put, rc = -ETXTBSY);
2670 /* Acquire extent lock on [0, EOF] to sync with all possible written. */
2671 LASSERT(lfsck->li_namespace != NULL);
2673 memset(policy, 0, sizeof(*policy));
2674 policy->l_extent.end = OBD_OBJECT_EOF;
2675 ost_fid_build_resid(fid, resid);
2676 rc = ldlm_cli_enqueue_local(env, lfsck->li_namespace, resid,
2677 LDLM_EXTENT, policy, LCK_EX, &flags,
2678 ldlm_blocking_ast, ldlm_completion_ast,
2679 NULL, NULL, 0, LVB_T_NONE, NULL, &lh);
2681 GOTO(put, rc = -EIO);
2683 dt_write_lock(env, obj, 0);
2684 /* Get obj's attr within lock again. */
2685 rc = dt_attr_get(env, obj, la);
2689 if (la->la_ctime != 0)
2690 GOTO(unlock, rc = -ETXTBSY);
2692 th = lfsck_trans_create(env, dev, lfsck);
2694 GOTO(unlock, rc = PTR_ERR(th));
2696 rc = dt_declare_ref_del(env, obj, th);
2700 rc = dt_declare_destroy(env, obj, th);
2704 rc = dt_trans_start_local(env, dev, th);
2708 rc = dt_ref_del(env, obj, th);
2712 rc = dt_destroy(env, obj, th);
2714 CDEBUG(D_LFSCK, "%s: layout LFSCK destroyed the empty "
2715 "OST-object "DFID" that was created for reparing "
2716 "dangling referenced case. But the original missing "
2717 "OST-object is found now.\n",
2718 lfsck_lfsck2name(lfsck), PFID(fid));
2723 dt_trans_stop(env, dev, th);
2726 dt_write_unlock(env, obj);
2727 ldlm_lock_decref(&lh, LCK_EX);
2730 lfsck_object_put(env, obj);
2736 * Some OST-object has occupied the specified layout EA slot.
2737 * Such OST-object may be generated by the LFSCK when repair
2738 * dangling referenced MDT-object, which can be indicated by
2739 * attr::la_ctime == 0 but without S_ISUID in la_mode. If it
2740 * is true and such OST-object has not been modified yet, we
2741 * will replace it with the orphan OST-object; otherwise the
2742 * LFSCK will create new MDT-object to reference the orphan.
2744 * \retval +1: repaired
2745 * \retval 0: did nothing
2746 * \retval -ve: on error
2748 static int lfsck_layout_conflict_create(const struct lu_env *env,
2749 struct lfsck_component *com,
2750 struct lfsck_tgt_desc *ltd,
2751 struct lu_orphan_rec_v3 *rec,
2752 struct dt_object *parent,
2753 struct lu_fid *cfid,
2754 struct lu_buf *ea_buf,
2755 struct lov_mds_md_v1 *lmm,
2756 struct lov_ost_data_v1 *slot,
2757 __u32 ea_off, int lovea_size)
2759 struct lfsck_thread_info *info = lfsck_env_info(env);
2760 struct lu_fid *cfid2 = &info->lti_fid2;
2761 struct ost_id *oi = &info->lti_oi;
2762 struct dt_device *dev = lfsck_obj2dev(parent);
2763 struct thandle *th = NULL;
2764 struct lustre_handle lh = { 0 };
2765 __u32 ost_idx2 = le32_to_cpu(slot->l_ost_idx);
2769 while (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val)) {
2770 if (unlikely(!thread_is_running(&com->lc_lfsck->li_thread)))
2774 ostid_le_to_cpu(&slot->l_ost_oi, oi);
2775 rc = ostid_to_fid(cfid2, oi, ost_idx2);
2779 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
2780 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2785 rc = lfsck_layout_master_conditional_destroy(env, com, cfid2, ost_idx2);
2787 /* If the conflict OST-obejct is not created for fixing dangling
2788 * referenced MDT-object in former LFSCK check/repair, or it has
2789 * been modified by others, then we cannot destroy it. Re-create
2790 * a new MDT-object for the orphan OST-object. */
2791 if (rc == -ETXTBSY) {
2792 /* No need the layout lock on the original parent. */
2793 lfsck_ibits_unlock(&lh, LCK_EX);
2795 fid_zero(&rec->lor_rec.lor_fid);
2796 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2797 "-"DFID"-%x", PFID(lu_object_fid(&parent->do_lu)),
2799 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
2800 info->lti_tmpbuf, "C", ea_off);
2805 if (rc != 0 && rc != -ENOENT)
2808 if (lfsck_is_dryrun(com->lc_lfsck))
2809 GOTO(unlock, rc = 0);
2811 th = lfsck_trans_create(env, dev, com->lc_lfsck);
2813 GOTO(unlock, rc = PTR_ERR(th));
2815 rc = dt_declare_xattr_set(env, parent, ea_buf, XATTR_NAME_LOV,
2816 LU_XATTR_REPLACE, th);
2820 rc = dt_trans_start_local(env, dev, th);
2824 dt_write_lock(env, parent, 0);
2825 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
2826 rc = lfsck_layout_refill_lovea(env, com->lc_lfsck, th, parent, cfid,
2827 ea_buf, lmm, slot, LU_XATTR_REPLACE,
2828 ltd->ltd_index, lovea_size);
2829 dt_write_unlock(env, parent);
2834 dt_trans_stop(env, dev, th);
2837 lfsck_ibits_unlock(&lh, LCK_EX);
2840 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant replaced the conflict "
2841 "OST-object "DFID" on the OST %x with the orphan "DFID" on "
2842 "the OST %x: parent "DFID", stripe-index %u: rc = %d\n",
2843 lfsck_lfsck2name(com->lc_lfsck), PFID(cfid2), ost_idx2,
2844 PFID(cfid), ltd->ltd_index, PFID(lfsck_dto2fid(parent)),
2847 return rc >= 0 ? 1 : rc;
2851 * \retval +1: repaired
2852 * \retval 0: did nothing
2853 * \retval -ve: on error
2855 static int lfsck_layout_recreate_lovea(const struct lu_env *env,
2856 struct lfsck_component *com,
2857 struct lfsck_tgt_desc *ltd,
2858 struct lu_orphan_rec_v3 *rec,
2859 struct dt_object *parent,
2860 struct lu_fid *cfid,
2861 __u32 ost_idx, __u32 ea_off)
2863 struct lfsck_thread_info *info = lfsck_env_info(env);
2864 struct lu_buf *buf = &info->lti_big_buf;
2865 struct lu_fid *fid = &info->lti_fid2;
2866 struct ost_id *oi = &info->lti_oi;
2867 struct lfsck_instance *lfsck = com->lc_lfsck;
2868 struct dt_device *dt = lfsck_obj2dev(parent);
2869 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2870 struct ost_layout *ol = &rec->lor_layout;
2871 struct lov_comp_md_v1 *lcm = NULL;
2872 struct lov_comp_md_entry_v1 *lcme = NULL;
2873 struct thandle *handle = NULL;
2875 struct lov_mds_md_v1 *lmm;
2876 struct lov_ost_data_v1 *objs;
2877 struct lustre_handle lh = { 0 };
2886 bool locked = false;
2887 bool new_mirror = true;
2890 if (lfsck_is_dryrun(lfsck))
2893 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2894 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
2897 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant failed to recreate "
2898 "LOV EA for "DFID": parent "DFID", OST-index %u, "
2899 "stripe-index %u, comp_id %u, comp_start %llu, "
2900 "comp_end %llu, layout version %u, range %u: rc = %d\n",
2901 lfsck_lfsck2name(lfsck), PFID(cfid),
2902 PFID(lfsck_dto2fid(parent)), ost_idx, ea_off,
2903 ol->ol_comp_id, ol->ol_comp_start,
2904 ol->ol_comp_end, rec->lor_layout_version,
2905 rec->lor_range, rc);
2912 dt_write_unlock(env, parent);
2916 if (handle != NULL) {
2917 dt_trans_stop(env, dt, handle);
2922 GOTO(unlock_layout, rc);
2925 if (buf->lb_len < lovea_size) {
2926 lu_buf_realloc(buf, lovea_size);
2927 if (buf->lb_buf == NULL)
2928 GOTO(unlock_layout, rc = -ENOMEM);
2931 if (!(bk->lb_param & LPF_DRYRUN)) {
2932 handle = lfsck_trans_create(env, dt, lfsck);
2934 GOTO(unlock_layout, rc = PTR_ERR(handle));
2936 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
2941 rc = dt_trans_start_local(env, dt, handle);
2946 dt_write_lock(env, parent, 0);
2948 rc = dt_xattr_get(env, parent, buf, XATTR_NAME_LOV);
2949 if (rc == -ERANGE) {
2950 rc = dt_xattr_get(env, parent, &LU_BUF_NULL, XATTR_NAME_LOV);
2953 } else if (rc == -ENODATA || rc == 0) {
2954 lovea_size = lfsck_lovea_size(ol, ea_off);
2955 /* If the declared is not big enough, re-try. */
2956 if (buf->lb_len < lovea_size) {
2960 fl = LU_XATTR_CREATE;
2961 } else if (rc < 0) {
2962 GOTO(unlock_parent, rc);
2963 } else if (unlikely(buf->lb_len == 0)) {
2966 fl = LU_XATTR_REPLACE;
2970 if (fl == LU_XATTR_CREATE) {
2971 if (bk->lb_param & LPF_DRYRUN)
2972 GOTO(unlock_parent, rc = 1);
2974 LASSERT(buf->lb_len >= lovea_size);
2976 rc = lfsck_layout_update_lovea(env, lfsck, handle, rec, parent,
2977 cfid, buf, fl, ost_idx, ea_off);
2979 GOTO(unlock_parent, rc);
2983 rc1 = lfsck_layout_verify_header(parent, lmm, lovea_size);
2985 /* If the LOV EA crashed, the rebuild it. */
2986 if (rc1 == -EINVAL) {
2987 if (bk->lb_param & LPF_DRYRUN)
2988 GOTO(unlock_parent, rc = 1);
2990 LASSERT(buf->lb_len >= lovea_size);
2992 rc = lfsck_layout_update_lovea(env, lfsck, handle, rec, parent,
2993 cfid, buf, fl, ost_idx, ea_off);
2995 GOTO(unlock_parent, rc);
2998 /* For other unknown magic/pattern, keep the current LOV EA. */
2999 if (rc1 == -EOPNOTSUPP)
3000 GOTO(unlock_parent, rc1 = 0);
3003 GOTO(unlock_parent, rc = rc1);
3005 magic = le32_to_cpu(lmm->lmm_magic);
3006 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
3009 __u16 mirror_id0 = mirror_id_of(ol->ol_comp_id);
3012 if (bk->lb_param & LPF_DRYRUN)
3013 GOTO(unlock_parent, rc = 1);
3016 count = le16_to_cpu(lcm->lcm_entry_count);
3017 for (i = 0; i < count; pos = ++i) {
3018 lcme = &lcm->lcm_entries[i];
3019 start = le64_to_cpu(lcme->lcme_extent.e_start);
3020 end = le64_to_cpu(lcme->lcme_extent.e_end);
3021 mirror_id1 = mirror_id_of(le32_to_cpu(lcme->lcme_id));
3023 if (mirror_id0 > mirror_id1)
3026 if (mirror_id0 < mirror_id1)
3030 if (end <= ol->ol_comp_start)
3033 if (start >= ol->ol_comp_end)
3036 lmm = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
3037 magic = le32_to_cpu(lmm->lmm_magic);
3038 flags = le32_to_cpu(lcme->lcme_flags);
3042 rc = lfsck_layout_add_comp(env, lfsck, handle, rec, parent,
3043 cfid, buf, ost_idx, ea_off, pos, new_mirror);
3045 GOTO(unlock_parent, rc);
3049 count = le16_to_cpu(lmm->lmm_stripe_count);
3051 GOTO(unlock_parent, rc = -EINVAL);
3054 /* Exceed the current end of MDT-object layout EA. Then extend it. */
3055 if (count <= ea_off) {
3056 if (bk->lb_param & LPF_DRYRUN)
3057 GOTO(unlock_parent, rc = 1);
3059 lovea_size = lov_mds_md_size(ea_off + 1, magic);
3060 /* If the declared is not big enough, re-try. */
3061 if (buf->lb_len < lovea_size) {
3069 lcme->lcme_flags = cpu_to_le32(flags | LCME_FL_INIT);
3070 lfsck_layout_update_lcm(lcm, lcme,
3071 rec->lor_layout_version,
3075 rc = lfsck_layout_extend_v1v3_lovea(env, lfsck, handle, ol,
3076 parent, cfid, buf, ost_idx, ea_off);
3078 GOTO(unlock_parent, rc);
3081 LASSERTF(rc > 0, "invalid rc = %d\n", rc);
3083 if (magic == LOV_MAGIC_V1) {
3084 objs = &lmm->lmm_objects[0];
3086 LASSERT(magic == LOV_MAGIC_V3);
3087 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
3090 for (i = 0; i < count; i++, objs++) {
3091 /* The MDT-object was created via lfsck_layout_recover_create()
3092 * by others before, and we fill the dummy layout EA. */
3093 if ((lcme && !(flags & LCME_FL_INIT)) ||
3094 lovea_slot_is_dummy(objs)) {
3098 if (bk->lb_param & LPF_DRYRUN)
3099 GOTO(unlock_parent, rc = 1);
3101 lmm->lmm_layout_gen =
3102 cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
3106 if (le32_to_cpu(lmm->lmm_stripe_size) !=
3107 ol->ol_stripe_size ||
3108 le16_to_cpu(lmm->lmm_stripe_count) !=
3109 ol->ol_stripe_count ||
3110 le64_to_cpu(lcme->lcme_extent.e_start) !=
3111 ol->ol_comp_start ||
3112 le64_to_cpu(lcme->lcme_extent.e_end) !=
3114 CDEBUG(D_LFSCK, "%s: found invalid "
3115 "component for "DFID ": parent "DFID
3116 ", stripe-index %u, stripe_size %u, "
3117 "stripe_count %u, comp_id %u, "
3118 "comp_start %llu, comp_end %llu, "
3119 "cur_stripe_size %u, "
3120 "cur_stripe_count %u, "
3121 "cur_comp_start %llu, "
3122 "cur_comp_end %llu\n",
3123 lfsck_lfsck2name(lfsck), PFID(cfid),
3124 PFID(lfsck_dto2fid(parent)), ea_off,
3126 ol->ol_stripe_count, ol->ol_comp_id,
3127 ol->ol_comp_start, ol->ol_comp_end,
3128 le32_to_cpu(lmm->lmm_stripe_size),
3129 le16_to_cpu(lmm->lmm_stripe_count),
3130 le64_to_cpu(lcme->lcme_extent.e_start),
3131 le64_to_cpu(lcme->lcme_extent.e_end));
3133 GOTO(unlock_parent, rc = -EINVAL);
3136 lovea_size = le32_to_cpu(lcm->lcm_size);
3137 lcme->lcme_flags = cpu_to_le32(flags |
3139 lfsck_layout_update_lcm(lcm, lcme,
3140 rec->lor_layout_version,
3144 LASSERTF(buf->lb_len >= lovea_size,
3145 "buffer len %d is less than real size %d\n",
3146 (int)buf->lb_len, (int)lovea_size);
3148 rc = lfsck_layout_refill_lovea(env, lfsck, handle,
3149 parent, cfid, buf, lmm, objs,
3150 fl, ost_idx, lovea_size);
3152 CDEBUG(D_LFSCK, "%s layout LFSCK assistant fill "
3153 "dummy layout slot for "DFID": parent "DFID
3154 ", OST-index %u, stripe-index %u: rc = %d\n",
3155 lfsck_lfsck2name(lfsck), PFID(cfid),
3156 PFID(lfsck_dto2fid(parent)), ost_idx, i, rc);
3158 GOTO(unlock_parent, rc);
3161 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3162 rc = ostid_to_fid(fid, oi, le32_to_cpu(objs->l_ost_idx));
3164 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
3165 "invalid layout EA at the slot %d, index %u\n",
3166 lfsck_lfsck2name(lfsck),
3167 PFID(lfsck_dto2fid(parent)), i,
3168 le32_to_cpu(objs->l_ost_idx));
3170 GOTO(unlock_parent, rc);
3173 /* It should be rare case, the slot is there, but the LFSCK
3174 * does not handle it during the first-phase cycle scanning. */
3175 if (unlikely(lu_fid_eq(fid, cfid))) {
3177 GOTO(unlock_parent, rc = 0);
3179 /* Rare case that the OST-object index
3180 * does not match the parent MDT-object
3181 * layout EA. We trust the later one. */
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 rc = lfsck_layout_update_pfid(env, com, parent,
3193 CDEBUG(D_LFSCK, "%s layout LFSCK assistant "
3194 "updated OST-object's pfid for "DFID
3195 ": parent "DFID", OST-index %u, "
3196 "stripe-index %u: rc = %d\n",
3197 lfsck_lfsck2name(lfsck), PFID(cfid),
3198 PFID(lfsck_dto2fid(parent)),
3199 ltd->ltd_index, i, rc);
3206 /* The MDT-object exists, but related layout EA slot is occupied
3208 if (bk->lb_param & LPF_DRYRUN)
3209 GOTO(unlock_parent, rc = 1);
3211 dt_write_unlock(env, parent);
3213 dt_trans_stop(env, dt, handle);
3214 lfsck_ibits_unlock(&lh, LCK_EX);
3215 if (magic == LOV_MAGIC_V1)
3216 objs = &lmm->lmm_objects[ea_off];
3218 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[ea_off];
3219 rc = lfsck_layout_conflict_create(env, com, ltd, rec, parent, cfid,
3220 buf, lmm, objs, ea_off, lovea_size);
3226 dt_write_unlock(env, parent);
3230 dt_trans_stop(env, dt, handle);
3233 lfsck_ibits_unlock(&lh, LCK_EX);
3238 static int lfsck_layout_scan_orphan_one(const struct lu_env *env,
3239 struct lfsck_component *com,
3240 struct lfsck_tgt_desc *ltd,
3241 struct lu_orphan_rec_v3 *rec,
3242 struct lu_fid *cfid)
3244 struct lfsck_layout *lo = com->lc_file_ram;
3245 struct lu_fid *pfid = &rec->lor_rec.lor_fid;
3246 struct dt_object *parent = NULL;
3247 __u32 ea_off = pfid->f_stripe_idx;
3251 if (!fid_is_sane(cfid))
3252 GOTO(out, rc = -EINVAL);
3255 if (fid_is_zero(pfid)) {
3256 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3261 if (!fid_is_sane(pfid))
3262 GOTO(out, rc = -EINVAL);
3264 parent = lfsck_object_find_by_dev(env, com->lc_lfsck->li_bottom, pfid);
3266 GOTO(out, rc = PTR_ERR(parent));
3268 if (unlikely(dt_object_remote(parent) != 0))
3269 GOTO(put, rc = -EXDEV);
3271 if (dt_object_exists(parent) == 0) {
3272 lfsck_object_put(env, parent);
3273 rc = lfsck_layout_recreate_parent(env, com, ltd, rec, cfid,
3278 if (!S_ISREG(lu_object_attr(&parent->do_lu)))
3279 GOTO(put, rc = -EISDIR);
3281 /* The orphan OST-object claims to be the parent's stripe, then
3282 * related dangling record in the trace file is meaningless. */
3283 rc = lfsck_layout_del_dangling_rec(env, com, pfid,
3284 rec->lor_layout.ol_comp_id, ea_off);
3285 if (rc && rc != -ENOENT)
3288 rc = lfsck_layout_recreate_lovea(env, com, ltd, rec, parent, cfid,
3289 ltd->ltd_index, ea_off);
3295 lfsck_object_put(env, parent);
3297 /* The layout EA is changed, need to be reloaded next time. */
3298 dt_object_put_nocache(env, parent);
3301 down_write(&com->lc_sem);
3302 com->lc_new_scanned++;
3303 com->lc_new_checked++;
3305 lo->ll_objs_repaired[LLIT_ORPHAN - 1]++;
3307 } else if (rc < 0) {
3308 lo->ll_objs_failed_phase2++;
3310 up_write(&com->lc_sem);
3315 static int lfsck_layout_scan_orphan(const struct lu_env *env,
3316 struct lfsck_component *com,
3317 struct lfsck_tgt_desc *ltd)
3319 struct lfsck_assistant_data *lad = com->lc_data;
3320 struct lfsck_instance *lfsck = com->lc_lfsck;
3321 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
3322 struct lfsck_thread_info *info = lfsck_env_info(env);
3323 struct lu_fid *fid = &info->lti_fid;
3324 struct dt_object *obj;
3325 const struct dt_it_ops *iops;
3330 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant starts the orphan "
3331 "scanning for OST%04x\n",
3332 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3334 if (test_bit(ltd->ltd_index, lad->lad_bitmap)) {
3335 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant skip the orphan "
3336 "scanning for OST%04x\n",
3337 lfsck_lfsck2name(lfsck), ltd->ltd_index);
3342 fid->f_seq = fid_idif_seq(0, ltd->ltd_index);
3343 fid->f_oid = fid->f_ver = 0;
3345 obj = lfsck_object_find_by_dev(env, ltd->ltd_tgt, fid);
3346 if (unlikely(IS_ERR(obj)))
3347 GOTO(log, rc = PTR_ERR(obj));
3349 rc = obj->do_ops->do_index_try(env, obj,
3350 &dt_lfsck_layout_orphan_features);
3354 iops = &obj->do_index_ops->dio_it;
3355 di = iops->init(env, obj, 0);
3357 GOTO(put, rc = PTR_ERR(di));
3359 rc = iops->load(env, di, 0);
3361 /* -ESRCH means that the orphan OST-objects rbtree has been
3362 * cleanup because of the OSS server restart or other errors. */
3363 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
3368 rc = iops->next(env, di);
3380 struct lu_orphan_rec_v3 *rec = &info->lti_rec;
3382 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
3383 unlikely(!thread_is_running(&lfsck->li_thread)))
3386 key = iops->key(env, di);
3387 com->lc_fid_latest_scanned_phase2 = *(struct lu_fid *)key;
3388 /* Remote target OST may be runnning old LFSCK */
3389 memset(rec, 0, sizeof(*rec));
3390 rc = iops->rec(env, di, (struct dt_rec *)rec, 0);
3392 rc = lfsck_layout_scan_orphan_one(env, com, ltd, rec,
3393 &com->lc_fid_latest_scanned_phase2);
3394 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
3397 lfsck_control_speed_by_self(com);
3399 rc = iops->next(env, di);
3400 } while (rc < 0 && !(bk->lb_param & LPF_FAILOUT));
3407 iops->fini(env, di);
3409 lfsck_object_put(env, obj);
3412 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant finished the orphan "
3413 "scanning for OST%04x: rc = %d\n",
3414 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
3416 return rc > 0 ? 0 : rc;
3419 static int lfsck_lov2layout(struct lov_mds_md_v1 *lmm, struct filter_fid *ff,
3422 struct ost_layout *ol = &ff->ff_layout;
3423 __u32 magic = le32_to_cpu(lmm->lmm_magic);
3427 if (magic == LOV_MAGIC_V1 || magic == LOV_MAGIC_V3) {
3428 ol->ol_stripe_size = lmm->lmm_stripe_size;
3429 ol->ol_stripe_count = lmm->lmm_stripe_count;
3430 ol->ol_comp_start = 0;
3431 ol->ol_comp_end = 0;
3433 ff->ff_layout_version = 0;
3435 } else if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
3436 struct lov_comp_md_v1 *lcm = (struct lov_comp_md_v1 *)lmm;
3437 struct lov_comp_md_entry_v1 *lcme = NULL;
3438 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3441 for (i = 0; i < count; i++) {
3442 lcme = &lcm->lcm_entries[i];
3443 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3444 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3451 /* The comp has been removed, do nothing. */
3455 lmm = (void *)lmm + le32_to_cpu(lcme->lcme_offset);
3456 ol->ol_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
3457 ol->ol_stripe_count = le32_to_cpu(lmm->lmm_stripe_count);
3458 ol->ol_comp_start = le64_to_cpu(lcme->lcme_extent.e_start);
3459 ol->ol_comp_end = le64_to_cpu(lcme->lcme_extent.e_end);
3460 ol->ol_comp_id = le32_to_cpu(lcme->lcme_id);
3461 ff->ff_layout_version = le32_to_cpu(lcme->lcme_layout_gen);
3464 GOTO(out, rc = -EINVAL);
3474 * Repair the MDT-object with dangling LOV EA reference.
3476 * we need to repair the inconsistency according to the users' requirement:
3478 * 1) Keep the inconsistency there and report the inconsistency case,
3479 * then give the chance to the application to find related issues,
3480 * and the users can make the decision about how to handle it with
3481 * more human knownledge. (by default)
3483 * 2) Re-create the missing OST-object with the FID/owner information.
3485 * \param[in] env pointer to the thread context
3486 * \param[in] com the layout LFSCK component
3487 * \param[in] parent the MDT-object with dangling LOV EA reference
3488 * \param[in] child the OST-object to be created
3489 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3490 * \param[in] ea_off the offset of the OST-object in the LOV EA
3491 * \param[in] ost_idx the index of OST on which the OST-object resides
3493 * \retval +1 for repair successfully
3494 * \retval 0 for did nothing
3495 * \retval negative error number on failure
3497 static int __lfsck_layout_repair_dangling(const struct lu_env *env,
3498 struct lfsck_component *com,
3499 struct dt_object *parent,
3500 struct dt_object *child,
3501 __u32 comp_id, __u32 ea_off,
3502 __u32 ost_idx, bool log)
3504 struct lfsck_thread_info *info = lfsck_env_info(env);
3505 struct filter_fid *ff = &info->lti_ff;
3506 struct dt_object_format *dof = &info->lti_dof;
3507 struct lu_attr *la = &info->lti_la;
3508 struct lfsck_instance *lfsck = com->lc_lfsck;
3509 struct dt_device *dev = lfsck_obj2dev(child);
3510 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3511 const struct lu_fid *cfid = lfsck_dto2fid(child);
3512 struct lu_buf *tbuf = &info->lti_big_buf;
3513 struct thandle *handle;
3515 struct lustre_handle lh = { 0 };
3519 if (!(lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ))
3522 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3525 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3526 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3531 rc = dt_attr_get(env, parent, la);
3535 la->la_mode = S_IFREG | 0666;
3536 la->la_atime = la->la_mtime = la->la_ctime = 0;
3537 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
3538 LA_ATIME | LA_MTIME | LA_CTIME;
3539 memset(dof, 0, sizeof(*dof));
3540 ff->ff_parent.f_seq = cpu_to_le64(pfid->f_seq);
3541 ff->ff_parent.f_oid = cpu_to_le32(pfid->f_oid);
3542 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3543 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3544 * parent MDT-object's layout EA. */
3545 ff->ff_parent.f_stripe_idx = cpu_to_le32(ea_off);
3547 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3548 if (unlikely(rc == -ENODATA))
3553 rc = lfsck_lov2layout(tbuf->lb_buf, ff, comp_id);
3557 buf = lfsck_buf_get(env, ff, sizeof(struct filter_fid));
3558 handle = lfsck_trans_create(env, dev, lfsck);
3560 GOTO(unlock1, rc = PTR_ERR(handle));
3562 rc = dt_declare_create(env, child, la, NULL, dof, handle);
3566 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID,
3567 LU_XATTR_CREATE, handle);
3571 rc = dt_trans_start_local(env, dev, handle);
3575 dt_read_lock(env, parent, 0);
3576 if (unlikely(lfsck_is_dead_obj(parent)))
3577 GOTO(unlock2, rc = 0);
3579 if (lfsck->li_bookmark_ram.lb_param & LPF_DELAY_CREATE_OSTOBJ) {
3580 struct ost_id *oi = &info->lti_oi;
3581 struct lu_fid *tfid = &info->lti_fid2;
3582 struct lu_buf *lovea = &info->lti_big_buf;
3583 struct lov_mds_md_v1 *lmm;
3584 struct lov_ost_data_v1 *objs;
3589 rc = lfsck_layout_get_lovea(env, parent, lovea);
3590 if (unlikely(rc == -ENODATA))
3595 lmm = lovea->lb_buf;
3596 magic = le32_to_cpu(lmm->lmm_magic);
3597 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
3598 struct lov_comp_md_v1 *lcm = buf->lb_buf;
3599 struct lov_comp_md_entry_v1 *lcme;
3600 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
3603 for (i = 0; i < count; i++) {
3604 lcme = &lcm->lcm_entries[i];
3605 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
3606 LASSERT(le32_to_cpu(lcme->lcme_flags) &
3609 lmm = lovea->lb_buf +
3610 le32_to_cpu(lcme->lcme_offset);
3611 magic = le32_to_cpu(lmm->lmm_magic);
3616 /* Someone removed the component, do nothing. */
3617 GOTO(unlock2, rc = 0);
3621 count = le16_to_cpu(lmm->lmm_stripe_count);
3622 /* Someone changed the LOV EA, do nothing. */
3623 if (count <= ea_off)
3624 GOTO(unlock2, rc = 0);
3626 if (magic == LOV_MAGIC_V1) {
3627 objs = &lmm->lmm_objects[ea_off];
3629 LASSERT(magic == LOV_MAGIC_V3);
3631 objs = &((struct lov_mds_md_v3 *)lmm)->\
3632 lmm_objects[ea_off];
3635 ostid_le_to_cpu(&objs->l_ost_oi, oi);
3636 idx2 = le32_to_cpu(objs->l_ost_idx);
3637 rc = ostid_to_fid(tfid, oi, idx2);
3638 /* Someone changed the LOV EA, do nothing. */
3639 if (rc != 0 || !lu_fid_eq(tfid, cfid))
3643 rc = dt_create(env, child, la, NULL, dof, handle);
3647 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, LU_XATTR_CREATE,
3653 dt_read_unlock(env, parent);
3656 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3659 lfsck_ibits_unlock(&lh, LCK_EX);
3663 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3664 "dangling reference for: parent "DFID", child "
3665 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: "
3667 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3668 comp_id, ea_off, ost_idx,
3669 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3670 "Create the lost OST-object as required" :
3671 "Keep the MDT-object there by default", rc);
3677 * Repair the MDT-object with dangling LOV EA reference.
3679 * Prepare parameters and call __lfsck_layout_repair_dangling()
3680 * to repair the dangling LOV EA reference.
3682 * \param[in] env pointer to the thread context
3683 * \param[in] com the layout LFSCK component
3684 * \param[in] pfid the MDT-object's FID
3685 * \param[in] cfid the FID for the OST-object to be created
3686 * \param[in] comp_id the component ID of the OST-object in the LOV EA
3687 * \param[in] ea_off the offset of the OST-object in the LOV EA
3688 * \param[in] ost_idx the index of OST on which the OST-object resides
3690 * \retval +1 for repair successfully
3691 * \retval 0 for did nothing
3692 * \retval negative error number on failure
3694 static int lfsck_layout_repair_dangling(const struct lu_env *env,
3695 struct lfsck_component *com,
3696 const struct lu_fid *pfid,
3697 const struct lu_fid *cfid,
3698 __u32 comp_id, __u32 ea_off,
3701 struct lfsck_instance *lfsck = com->lc_lfsck;
3702 struct dt_object *parent = NULL;
3703 struct dt_object *child = NULL;
3704 struct lfsck_tgt_desc *ltd;
3708 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3710 GOTO(log, rc = PTR_ERR(parent));
3712 /* The MDT-object has been removed. */
3713 if (dt_object_exists(parent) == 0)
3716 ltd = lfsck_ltd2tgt(&lfsck->li_ost_descs, ost_idx);
3717 if (unlikely(ltd == NULL))
3718 GOTO(log, rc = -ENODEV);
3720 child = lfsck_object_find_by_dev(env, ltd->ltd_tgt, cfid);
3722 GOTO(log, rc = PTR_ERR(child));
3724 /* The OST-object has been created. */
3725 if (unlikely(dt_object_exists(child) != 0))
3728 rc = __lfsck_layout_repair_dangling(env, com, parent, child,
3729 comp_id, ea_off, ost_idx, false);
3734 if (child != NULL && !IS_ERR(child))
3735 lfsck_object_put(env, child);
3737 if (parent != NULL && !IS_ERR(parent))
3738 lfsck_object_put(env, parent);
3741 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant found "
3742 "dangling reference for: parent "DFID", child "
3743 DFID", comp_id %u, ea_off %u, ost_idx %u, %s: rc = %d\n",
3744 lfsck_lfsck2name(lfsck), PFID(pfid), PFID(cfid),
3745 comp_id, ea_off, ost_idx,
3746 (lfsck->li_bookmark_ram.lb_param & LPF_CREATE_OSTOBJ) ?
3747 "Create the lost OST-object as required" :
3748 "Keep the MDT-object there by default", rc);
3753 /* If the OST-object does not recognize the MDT-object as its parent, and
3754 * there is no other MDT-object claims as its parent, then just trust the
3755 * given MDT-object as its parent. So update the OST-object filter_fid. */
3756 static int lfsck_layout_repair_unmatched_pair(const struct lu_env *env,
3757 struct lfsck_component *com,
3758 struct dt_object *parent,
3759 struct lfsck_layout_req *llr,
3762 struct lfsck_thread_info *info = lfsck_env_info(env);
3763 struct filter_fid *ff = &info->lti_ff;
3764 struct dt_object *child = llr->llr_child;
3765 struct dt_device *dev = lfsck_obj2dev(child);
3766 const struct lu_fid *tfid = lu_object_fid(&parent->do_lu);
3767 struct lu_buf *tbuf = &info->lti_big_buf;
3768 struct thandle *handle;
3770 struct lustre_handle lh = { 0 };
3774 if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3777 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
3778 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3783 ff->ff_parent.f_seq = cpu_to_le64(tfid->f_seq);
3784 ff->ff_parent.f_oid = cpu_to_le32(tfid->f_oid);
3785 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
3786 * MDT-object's FID::f_ver, instead it is the OST-object index in its
3787 * parent MDT-object's layout EA. */
3788 ff->ff_parent.f_stripe_idx = cpu_to_le32(llr->llr_lov_idx);
3790 rc = lfsck_layout_get_lovea(env, parent, tbuf);
3791 if (unlikely(rc == -ENODATA))
3796 rc = lfsck_lov2layout(tbuf->lb_buf, ff, llr->llr_comp_id);
3800 buf = lfsck_buf_get(env, ff, sizeof(*ff));
3802 handle = lfsck_trans_create(env, dev, com->lc_lfsck);
3804 GOTO(unlock1, rc = PTR_ERR(handle));
3806 rc = dt_declare_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3810 rc = dt_attr_get(env, parent, la);
3814 la->la_valid = LA_UID | LA_GID;
3815 rc = dt_declare_attr_set(env, child, la, handle);
3819 rc = dt_trans_start_local(env, dev, handle);
3823 dt_write_lock(env, parent, 0);
3824 if (unlikely(lfsck_is_dead_obj(parent)))
3825 GOTO(unlock2, rc = 1);
3827 rc = dt_xattr_set(env, child, buf, XATTR_NAME_FID, 0, handle);
3831 /* Get the latest parent's owner. */
3832 rc = dt_attr_get(env, parent, la);
3836 la->la_valid = LA_UID | LA_GID;
3837 rc = dt_attr_set(env, child, la, handle);
3842 dt_write_unlock(env, parent);
3845 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
3848 lfsck_ibits_unlock(&lh, LCK_EX);
3852 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
3853 "unmatched MDT-OST pair for: parent "DFID
3854 ", child "DFID", comp_id %u, OST-index %u, "
3855 "stripe-index %u, owner %u/%u: rc = %d\n",
3856 lfsck_lfsck2name(com->lc_lfsck),
3857 PFID(lfsck_dto2fid(parent)),
3858 PFID(lfsck_dto2fid(child)),
3859 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
3860 la->la_uid, la->la_gid, rc);
3865 /* If there are more than one MDT-objects claim as the OST-object's parent,
3866 * and the OST-object only recognizes one of them, then we need to generate
3867 * new OST-object(s) with new fid(s) for the non-recognized MDT-object(s). */
3868 static int lfsck_layout_repair_multiple_references(const struct lu_env *env,
3869 struct lfsck_component *com,
3870 struct dt_object *parent,
3871 struct lfsck_layout_req *llr,
3874 struct lfsck_thread_info *info = lfsck_env_info(env);
3875 struct dt_allocation_hint *hint = &info->lti_hint;
3876 struct dt_object_format *dof = &info->lti_dof;
3877 struct ost_id *oi = &info->lti_oi;
3878 struct lu_buf *buf = &info->lti_big_buf;
3879 struct lfsck_instance *lfsck = com->lc_lfsck;
3880 struct dt_device *dev;
3881 struct lu_device *d =
3882 &lfsck_obj2dev(llr->llr_child)->dd_lu_dev;
3883 struct lu_object *o;
3884 struct lu_object *n;
3885 struct dt_object *child = NULL;
3886 struct thandle *handle = NULL;
3887 struct lov_mds_md_v1 *lmm;
3888 struct lov_ost_data_v1 *objs;
3889 const struct lu_fid *pfid = lfsck_dto2fid(parent);
3891 struct lustre_handle lh = { 0 };
3897 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
3900 /* We use two separated transactions to repair the inconsistency.
3902 * 1) create the child (OST-object).
3903 * 2) update the parent LOV EA according to the child's FID.
3905 * If 1) succeed, but 2) failed or aborted, then such OST-object will be
3906 * handled as orphan when the layout LFSCK run next time.
3908 * If 1) failed, but 2) succeed, then such OST-object will be re-created
3909 * as dangling referened case when the layout LFSCK run next time. */
3911 /* The 1st transaction. */
3912 o = lu_object_anon(env, d, NULL);
3914 GOTO(log, rc = PTR_ERR(o));
3916 n = lu_object_locate(o->lo_header, d->ld_type);
3917 if (unlikely(n == NULL)) {
3918 lu_object_put_nocache(env, o);
3920 GOTO(log, rc = -EINVAL);
3923 child = container_of(n, struct dt_object, do_lu);
3924 memset(hint, 0, sizeof(*hint));
3925 rc = dt_attr_get(env, parent, la);
3929 la->la_valid = LA_UID | LA_GID;
3930 memset(dof, 0, sizeof(*dof));
3932 dev = lfsck_obj2dev(child);
3933 handle = lfsck_trans_create(env, dev, lfsck);
3935 GOTO(log, rc = PTR_ERR(handle));
3937 rc = dt_declare_create(env, child, la, hint, dof, handle);
3941 rc = dt_trans_start_local(env, dev, handle);
3945 rc = dt_create(env, child, la, hint, dof, handle);
3946 dt_trans_stop(env, dev, handle);
3951 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
3952 MDS_INODELOCK_LAYOUT | MDS_INODELOCK_XATTR,
3957 /* The 2nd transaction. */
3959 /* XXX: Generally, we should use bottom device (OSD) to update parent
3960 * LOV EA. But because the LOD-object still references the wrong
3961 * OSP-object that should be detached after the parent's LOV EA
3962 * refreshed. Unfortunately, there is no suitable API for that.
3963 * So we have to make the LOD to re-load the OSP-object(s) via
3964 * replacing the LOV EA against the LOD-object.
3966 * Once the DNE2 patches have been landed, we can replace the
3967 * LOD device with the OSD device. LU-6230. */
3969 dev = lfsck->li_next;
3970 parent = lfsck_object_locate(dev, parent);
3972 GOTO(log, rc = PTR_ERR(parent));
3974 handle = lfsck_trans_create(env, dev, lfsck);
3976 GOTO(log, rc = PTR_ERR(handle));
3978 rc = dt_declare_xattr_set(env, parent, buf, XATTR_NAME_LOV,
3979 LU_XATTR_REPLACE, handle);
3983 rc = dt_trans_start_local(env, dev, handle);
3987 dt_write_lock(env, parent, 0);
3988 if (unlikely(lfsck_is_dead_obj(parent)))
3989 GOTO(unlock, rc = 0);
3991 rc = lfsck_layout_get_lovea(env, parent, buf);
3992 if (unlikely(rc == -ENODATA))
3998 magic = le32_to_cpu(lmm->lmm_magic);
3999 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
4000 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4001 struct lov_comp_md_entry_v1 *lcme;
4002 __u16 count = le16_to_cpu(lcm->lcm_entry_count);
4005 LASSERT(llr->llr_comp_id != 0);
4007 for (i = 0; i < count; i++) {
4008 lcme = &lcm->lcm_entries[i];
4009 if (le32_to_cpu(lcme->lcme_id) == llr->llr_comp_id) {
4010 LASSERT(le32_to_cpu(lcme->lcme_flags) &
4013 le32_add_cpu(&lcm->lcm_layout_gen, 1);
4015 le32_to_cpu(lcme->lcme_offset);
4016 magic = le32_to_cpu(lmm->lmm_magic);
4021 GOTO(unlock, rc = 0);
4025 if (magic == LOV_MAGIC_V1) {
4026 objs = &lmm->lmm_objects[llr->llr_lov_idx];
4028 LASSERT(magic == LOV_MAGIC_V3);
4030 &((struct lov_mds_md_v3 *)lmm)->lmm_objects[llr->llr_lov_idx];
4033 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4034 index = le32_to_cpu(objs->l_ost_idx);
4035 rc = ostid_to_fid(&tfid, oi, index);
4036 /* Someone changed layout during the LFSCK, no need to repair then. */
4037 if (rc == 0 && !lu_fid_eq(&tfid, lu_object_fid(&llr->llr_child->do_lu)))
4038 GOTO(unlock, rc = 0);
4040 lmm->lmm_layout_gen = cpu_to_le16(le16_to_cpu(lmm->lmm_layout_gen) + 1);
4041 fid_to_ostid(lu_object_fid(&child->do_lu), oi);
4042 ostid_cpu_to_le(oi, &objs->l_ost_oi);
4043 objs->l_ost_gen = cpu_to_le32(0);
4044 objs->l_ost_idx = cpu_to_le32(llr->llr_ost_idx);
4045 rc = dt_xattr_set(env, parent, buf, XATTR_NAME_LOV,
4046 LU_XATTR_REPLACE, handle);
4048 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
4051 dt_write_unlock(env, parent);
4055 dt_trans_stop(env, dev, handle);
4058 lfsck_ibits_unlock(&lh, LCK_EX);
4060 lfsck_object_put(env, child);
4063 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
4064 "multiple references for: parent "DFID", comp_id %u, "
4065 "OST-index %u, stripe-index %u, owner %u/%u: rc = %d\n",
4066 lfsck_lfsck2name(lfsck), PFID(pfid),
4067 llr->llr_comp_id, llr->llr_ost_idx, llr->llr_lov_idx,
4068 la->la_uid, la->la_gid, rc);
4073 /* If the MDT-object and the OST-object have different owner information,
4074 * then trust the MDT-object, because the normal chown/chgrp handle order
4075 * is from MDT to OST, and it is possible that some chown/chgrp operation
4076 * is partly done. */
4077 static int lfsck_layout_repair_owner(const struct lu_env *env,
4078 struct lfsck_component *com,
4079 struct dt_object *parent,
4080 struct lfsck_layout_req *llr,
4081 struct lu_attr *pla,
4082 const struct lu_attr *cla)
4084 struct lfsck_thread_info *info = lfsck_env_info(env);
4085 struct lu_attr *tla = &info->lti_la2;
4086 struct dt_object *child = llr->llr_child;
4087 struct dt_device *dev = lfsck_obj2dev(child);
4088 struct thandle *handle;
4090 dt_obj_version_t version;
4093 if (com->lc_lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
4096 tla->la_uid = pla->la_uid;
4097 tla->la_gid = pla->la_gid;
4098 tla->la_valid = LA_UID | LA_GID;
4099 handle = lfsck_trans_create(env, dev, com->lc_lfsck);
4101 GOTO(log, rc = PTR_ERR(handle));
4103 rc = dt_declare_attr_set(env, child, tla, handle);
4107 rc = dt_trans_start_local(env, dev, handle);
4111 /* Use the dt_object lock to serialize with destroy and attr_set. */
4112 dt_read_lock(env, parent, 0);
4113 if (unlikely(lfsck_is_dead_obj(parent)))
4114 GOTO(unlock, rc = 1);
4116 version = dt_version_get(env, child);
4117 if (version == -EOPNOTSUPP)
4120 /* Get the latest parent's owner. */
4121 rc = dt_attr_get(env, parent, pla);
4125 /* Some others chown/chgrp during the LFSCK, needs to do nothing. */
4126 if (unlikely((!version && tla->la_ctime == 0) ||
4127 tla->la_uid != pla->la_uid || tla->la_gid != pla->la_gid))
4130 rc = dt_attr_set(env, child, tla, handle);
4135 dt_read_unlock(env, parent);
4138 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
4142 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant repaired "
4143 "inconsistent file owner for: parent "DFID", child "DFID
4144 ", OST-index %u, stripe-index %u, old owner %u/%u, "
4145 "new owner %u/%u: rc = %d\n",
4146 lfsck_lfsck2name(com->lc_lfsck),
4147 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
4148 llr->llr_ost_idx, llr->llr_lov_idx,
4149 cla->la_uid, cla->la_gid, tla->la_uid, tla->la_gid, rc);
4154 #define CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid, msg) \
4155 CDEBUG(D_LFSCK, "%s:("DFID"|"DFID")/"DFID":XATTR %s: %s\n", \
4156 lfsck_lfsck2name(lfsck), PFID(&lso->lso_fid), PFID(pfid), \
4157 PFID(cfid), XATTR_NAME_FID, msg);
4159 /* Check whether the OST-object correctly back points to the
4160 * MDT-object (@parent) via the XATTR_NAME_FID xattr (@pfid). */
4161 static int lfsck_layout_check_parent(const struct lu_env *env,
4162 struct lfsck_component *com,
4163 struct lfsck_assistant_object *lso,
4164 struct filter_fid *ff,
4165 const struct lu_fid *cfid,
4166 const struct lu_attr *cla,
4167 struct lfsck_layout_req *llr)
4169 struct lfsck_thread_info *info = lfsck_env_info(env);
4170 struct lu_buf *buf = &info->lti_big_buf;
4171 struct lu_fid *pfid = &info->lti_fid;
4172 struct dt_object *tobj;
4173 struct lov_mds_md_v1 *lmm;
4174 struct lov_ost_data_v1 *objs;
4175 struct lustre_handle lh = { 0 };
4176 struct lfsck_instance *lfsck = com->lc_lfsck;
4184 *pfid = ff->ff_parent;
4185 idx = pfid->f_stripe_idx;
4188 if (unlikely(!fid_is_sane(pfid))) {
4189 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4190 "the parent FID is invalid");
4192 RETURN(LLIT_UNMATCHED_PAIR);
4195 if (lu_fid_eq(pfid, &lso->lso_fid)) {
4196 if (likely(llr->llr_lov_idx == idx))
4199 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4200 "the stripe index is unmatched");
4202 RETURN(LLIT_UNMATCHED_PAIR);
4205 tobj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
4207 RETURN(PTR_ERR(tobj));
4209 if (dt_object_exists(tobj) == 0) {
4210 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4211 "the parent is nonexistent");
4213 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4216 if (lfsck_is_dead_obj(tobj)) {
4217 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4218 "the parent is dead object");
4220 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4223 if (!S_ISREG(lfsck_object_type(tobj))) {
4224 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4225 "the parent is not a regular file");
4227 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4230 /* Load the tobj's layout EA, in spite of it is a local MDT-object or
4231 * remote one on another MDT. Then check whether the given OST-object
4232 * is in such layout. If yes, it is multiple referenced, otherwise it
4233 * is unmatched referenced case. */
4234 rc = lfsck_layout_get_lovea(env, tobj, buf);
4235 if (rc == 0 || rc == -ENODATA || rc == -ENOENT) {
4236 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4237 "the parent has no stripe data");
4239 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4242 if (unlikely(rc == -EOPNOTSUPP))
4243 GOTO(out, rc = LLIT_NONE);
4249 magic = le32_to_cpu(lmm->lmm_magic);
4250 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
4251 struct lov_comp_md_v1 *lcm = buf->lb_buf;
4252 struct lov_comp_md_entry_v1 *lcme;
4254 if (ff->ff_layout.ol_comp_id == 0) {
4255 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4256 "the parent has incorrect comp_id");
4258 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4261 count = le16_to_cpu(lcm->lcm_entry_count);
4262 for (i = 0; i < count; i++) {
4263 lcme = &lcm->lcm_entries[i];
4264 if (le32_to_cpu(lcme->lcme_id) ==
4265 ff->ff_layout.ol_comp_id) {
4267 le32_to_cpu(lcme->lcme_offset);
4268 magic = le32_to_cpu(lmm->lmm_magic);
4269 if (!(le32_to_cpu(lcme->lcme_flags) &
4271 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid,
4273 "the parent has uninitialized component");
4275 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4282 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4283 "the parent has no matched comp_id");
4285 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4289 if (magic == LOV_MAGIC_V1) {
4290 objs = &lmm->lmm_objects[0];
4292 LASSERT(magic == LOV_MAGIC_V3);
4293 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
4296 count = le16_to_cpu(lmm->lmm_stripe_count);
4297 for (i = 0; i < count; i++, objs++) {
4298 struct lu_fid *tfid = &info->lti_fid2;
4299 struct ost_id *oi = &info->lti_oi;
4302 if (lovea_slot_is_dummy(objs))
4305 ostid_le_to_cpu(&objs->l_ost_oi, oi);
4306 idx2 = le32_to_cpu(objs->l_ost_idx);
4307 rc = ostid_to_fid(tfid, oi, idx2);
4309 CDEBUG(D_LFSCK, "%s: the parent "DFID" contains "
4310 "invalid layout EA at the slot %d, index %u\n",
4311 lfsck_lfsck2name(com->lc_lfsck),
4312 PFID(pfid), i, idx2);
4314 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4317 if (lu_fid_eq(cfid, tfid)) {
4318 rc = lfsck_ibits_lock(env, com->lc_lfsck, tobj, &lh,
4319 MDS_INODELOCK_UPDATE |
4320 MDS_INODELOCK_LAYOUT |
4321 MDS_INODELOCK_XATTR,
4326 dt_read_lock(env, tobj, 0);
4328 /* For local MDT-object, re-check existence
4329 * after taken the lock. */
4330 if (!dt_object_remote(tobj)) {
4331 if (dt_object_exists(tobj) == 0 ||
4332 lfsck_is_dead_obj(tobj)) {
4333 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid,
4335 "the parent doesn't exist anymore after lock");
4337 rc = LLIT_UNMATCHED_PAIR;
4339 rc = LLIT_MULTIPLE_REFERENCED;
4345 /* For migration case, the new MDT-object and old
4346 * MDT-object may reference the same OST-object at
4347 * some migration internal time.
4349 * For remote MDT-object, the local MDT may not know
4350 * whether it has been removed or not. Try checking
4351 * for a non-existent xattr to check if this object
4352 * has been been removed or not. */
4353 rc = dt_xattr_get(env, tobj, &LU_BUF_NULL,
4355 if (unlikely(rc == -ENOENT || rc >= 0)) {
4356 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4357 "the parent is remote object and nonexistent after lock");
4359 rc = LLIT_UNMATCHED_PAIR;
4360 } else if (rc == -ENODATA) {
4361 rc = LLIT_MULTIPLE_REFERENCED;
4368 CDEBUG_UNMATCHED_PAIR(lfsck, lso, pfid, cfid,
4369 "the parent has no matched stripe");
4371 GOTO(out, rc = LLIT_UNMATCHED_PAIR);
4374 if (lustre_handle_is_used(&lh)) {
4375 dt_read_unlock(env, tobj);
4376 lfsck_ibits_unlock(&lh, LCK_EX);
4380 lfsck_object_put(env, tobj);
4386 * If the MDT-object has the LUSTRE_ENCRYPT_FL flag, it needs to be set
4387 * on the OST-object as well.
4389 static int lfsck_layout_repair_encflag(const struct lu_env *env,
4390 struct lfsck_component *com,
4391 struct dt_object *parent,
4392 struct lfsck_layout_req *llr)
4394 struct lfsck_thread_info *info = lfsck_env_info(env);
4395 struct lu_attr *tla = &info->lti_la2;
4396 struct dt_object *child = llr->llr_child;
4397 struct dt_device *dev = lfsck_obj2dev(child);
4398 struct thandle *handle;
4403 tla->la_valid = LA_FLAGS;
4404 tla->la_flags = LUSTRE_ENCRYPT_FL;
4405 handle = lfsck_trans_create(env, dev, com->lc_lfsck);
4407 GOTO(log, rc = PTR_ERR(handle));
4409 rc = dt_declare_attr_set(env, child, tla, handle);
4413 rc = dt_trans_start_local(env, dev, handle);
4417 /* Use the dt_object lock to serialize with destroy and attr_set. */
4418 dt_read_lock(env, parent, 0);
4419 if (unlikely(lfsck_is_dead_obj(parent)))
4420 GOTO(unlock, rc = 1);
4422 rc = dt_attr_set(env, child, tla, handle);
4426 dt_read_unlock(env, parent);
4429 rc = lfsck_layout_trans_stop(env, dev, handle, rc);
4434 "%s: layout LFSCK assistant repair of inconsistent file enc flag for: parent "
4436 DFID", OST-index %u, stripe-index %u: rc = %d\n",
4437 lfsck_lfsck2name(com->lc_lfsck),
4438 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
4439 llr->llr_ost_idx, llr->llr_lov_idx, rc);
4444 static int lfsck_layout_assistant_handler_p1(const struct lu_env *env,
4445 struct lfsck_component *com,
4446 struct lfsck_assistant_req *lar)
4448 struct lfsck_layout_req *llr =
4449 container_of(lar, struct lfsck_layout_req, llr_lar);
4450 struct lfsck_assistant_object *lso = lar->lar_parent;
4451 struct lfsck_layout *lo = com->lc_file_ram;
4452 struct lfsck_thread_info *info = lfsck_env_info(env);
4453 struct filter_fid *ff = &info->lti_ff;
4454 struct lu_buf buf = { .lb_buf = ff,
4455 .lb_len = sizeof(*ff) };
4456 struct dt_object *parent = NULL;
4457 struct dt_object *child = llr->llr_child;
4458 struct lu_attr *pla = &lso->lso_attr;
4459 struct lu_attr *cla = &info->lti_la;
4460 struct lfsck_instance *lfsck = com->lc_lfsck;
4461 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4462 enum lfsck_layout_inconsistency_type type = LLIT_NONE;
4469 CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_ENGINE_DELAY, cfs_fail_val);
4471 rc = dt_attr_get(env, child, cla);
4472 if (rc == -ENOENT) {
4473 parent = lfsck_assistant_object_load(env, lfsck, lso);
4474 if (IS_ERR(parent)) {
4475 rc = PTR_ERR(parent);
4477 RETURN(rc == -ENOENT ? 0 : rc);
4480 type = LLIT_DANGLING;
4487 if (!(bk->lb_param & LPF_DRYRUN) &&
4488 pla->la_valid & LA_FLAGS && pla->la_flags & LUSTRE_ENCRYPT_FL) {
4489 /* MDT-inode is encrypted */
4490 struct lu_buf lb = { .lb_buf = NULL, .lb_len = 0 };
4492 /* if OST-inode is missing encryption.c xattr, fix it */
4493 if (dt_xattr_get(env, child, &lb,
4494 LL_XATTR_NAME_ENCRYPTION_CONTEXT) >= 0)
4498 parent = lfsck_assistant_object_load(env, lfsck, lso);
4499 if (!IS_ERR_OR_NULL(parent))
4500 rc = lfsck_layout_repair_encflag(env, com, parent, llr);
4501 down_write(&com->lc_sem);
4503 lfsck_layout_record_failure(env, lfsck, lo);
4505 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
4506 up_write(&com->lc_sem);
4510 lfsck_buf_init(&buf, ff, sizeof(*ff));
4511 rc = dt_xattr_get(env, child, &buf, XATTR_NAME_FID);
4512 if (unlikely(rc > 0 && rc < sizeof(struct lu_fid))) {
4513 CDEBUG(D_LFSCK, "%s:"DFID"/"DFID": "
4514 "the child object's %s is corrupted\n",
4515 lfsck_lfsck2name(lfsck), PFID(&lso->lso_fid),
4516 PFID(lu_object_fid(&child->do_lu)),
4519 type = LLIT_UNMATCHED_PAIR;
4523 if (rc < 0 && rc != -ENODATA)
4526 if (rc == 0 || rc == -ENODATA)
4527 GOTO(check_owner, rc = 0);
4529 filter_fid_le_to_cpu(ff, ff, sizeof(*ff));
4530 rc = lfsck_layout_check_parent(env, com, lso, ff,
4531 lu_object_fid(&child->do_lu), cla, llr);
4541 /* Someone may has changed the owner after the parent attr pre-loaded.
4542 * It can be handled later inside the lfsck_layout_repair_owner(). */
4543 if (unlikely(cla->la_uid != pla->la_uid ||
4544 cla->la_gid != pla->la_gid)) {
4545 type = LLIT_INCONSISTENT_OWNER;
4550 if (type == LLIT_NONE)
4553 if (bk->lb_param & LPF_DRYRUN)
4556 if (parent == NULL) {
4557 parent = lfsck_assistant_object_load(env, lfsck, lso);
4558 if (IS_ERR(parent)) {
4559 rc = PTR_ERR(parent);
4570 if (bk->lb_param & LPF_DELAY_CREATE_OSTOBJ)
4571 rc = lfsck_layout_ins_dangling_rec(env, com,
4572 lfsck_dto2fid(parent), lfsck_dto2fid(child),
4573 llr->llr_comp_id, llr->llr_lov_idx,
4576 rc = __lfsck_layout_repair_dangling(env, com, parent,
4583 case LLIT_UNMATCHED_PAIR:
4584 rc = lfsck_layout_repair_unmatched_pair(env, com, parent,
4587 case LLIT_MULTIPLE_REFERENCED:
4588 rc = lfsck_layout_repair_multiple_references(env, com, parent,
4591 case LLIT_INCONSISTENT_OWNER:
4592 rc = lfsck_layout_repair_owner(env, com, parent, llr, pla, cla);
4602 down_write(&com->lc_sem);
4604 struct lfsck_assistant_data *lad = com->lc_data;
4606 if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags))) {
4608 } else if (rc == -ENOTCONN || rc == -ESHUTDOWN ||
4609 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
4610 rc == -EHOSTUNREACH) {
4611 /* If cannot touch the target server,
4612 * mark the LFSCK as INCOMPLETE. */
4613 CDEBUG(D_LFSCK, "%s: layout LFSCK assistant fail to "
4614 "talk with OST %x: rc = %d\n",
4615 lfsck_lfsck2name(lfsck), llr->llr_ost_idx, rc);
4616 lfsck_lad_set_bitmap(env, com, llr->llr_ost_idx);
4617 lo->ll_objs_skipped++;
4620 lfsck_layout_record_failure(env, lfsck, lo);
4622 } else if (rc > 0 && (type != LLIT_DANGLING ||
4623 !(bk->lb_param & LPF_DELAY_CREATE_OSTOBJ))) {
4624 LASSERTF(type > LLIT_NONE && type <= LLIT_MAX,
4625 "unknown type = %d\n", type);
4627 lo->ll_objs_repaired[type - 1]++;
4628 if (bk->lb_param & LPF_DRYRUN &&
4629 unlikely(lo->ll_pos_first_inconsistent == 0))
4630 lo->ll_pos_first_inconsistent =
4631 lfsck->li_obj_oit->do_index_ops->dio_it.store(env,
4634 up_write(&com->lc_sem);
4636 if (parent != NULL && !IS_ERR(parent))
4637 lfsck_object_put(env, parent);
4643 lfsck_layout_double_scan_one_trace_file(const struct lu_env *env,
4644 struct lfsck_component *com,
4645 struct dt_object *obj, bool first)
4647 struct lfsck_instance *lfsck = com->lc_lfsck;
4648 struct ptlrpc_thread *thread = &lfsck->li_thread;
4649 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4650 struct lfsck_layout *lo = com->lc_file_ram;
4651 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
4654 struct lfsck_layout_dangling_key *parent =
4655 &lfsck_env_info(env)->lti_lldk;
4656 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
4661 di = iops->init(env, obj, 0);
4663 RETURN(PTR_ERR(di));
4666 lldk_cpu_to_be(parent, &lo->ll_lldk_latest_scanned_phase2);
4668 memset(parent, 0, sizeof(*parent));
4669 rc = iops->get(env, di, (const struct dt_key *)parent);
4674 /* The start one either has been processed or does not exist,
4676 rc = iops->next(env, di);
4682 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY3, cfs_fail_val) &&
4683 unlikely(!thread_is_running(thread)))
4686 key = iops->key(env, di);
4695 lldk_be_to_cpu(parent,
4696 (const struct lfsck_layout_dangling_key *)key);
4697 if (!fid_is_sane(&parent->lldk_fid)) {
4702 rc = iops->rec(env, di, (struct dt_rec *)cfid, 0);
4704 fid_be_to_cpu(cfid, cfid);
4705 ost_idx = cfid->f_ver;
4707 if (!fid_is_sane(cfid)) {
4712 rc = lfsck_layout_repair_dangling(env, com,
4713 &parent->lldk_fid, cfid,
4714 parent->lldk_comp_id,
4715 parent->lldk_ea_off, ost_idx);
4719 down_write(&com->lc_sem);
4720 com->lc_new_checked++;
4721 com->lc_new_scanned++;
4723 lo->ll_lldk_latest_scanned_phase2 = *parent;
4726 lo->ll_objs_repaired[LLIT_DANGLING - 1]++;
4728 lo->ll_objs_failed_phase2++;
4729 up_write(&com->lc_sem);
4731 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
4734 if (unlikely(com->lc_time_next_checkpoint <=
4735 ktime_get_seconds()) &&
4736 com->lc_new_checked != 0) {
4737 down_write(&com->lc_sem);
4738 lo->ll_run_time_phase2 += ktime_get_seconds() -
4739 com->lc_time_last_checkpoint;
4740 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
4741 lo->ll_objs_checked_phase2 += com->lc_new_checked;
4742 com->lc_new_checked = 0;
4743 lfsck_layout_store(env, com);
4744 up_write(&com->lc_sem);
4746 com->lc_time_last_checkpoint = ktime_get_seconds();
4747 com->lc_time_next_checkpoint =
4748 com->lc_time_last_checkpoint +
4749 LFSCK_CHECKPOINT_INTERVAL;
4752 lfsck_control_speed_by_self(com);
4753 if (unlikely(!thread_is_running(thread)))
4756 rc = iops->next(env, di);
4765 iops->fini(env, di);
4770 static int lfsck_layout_assistant_handler_p2(const struct lu_env *env,
4771 struct lfsck_component *com)
4773 struct lfsck_assistant_data *lad = com->lc_data;
4774 struct lfsck_instance *lfsck = com->lc_lfsck;
4775 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4776 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
4777 struct lfsck_tgt_desc *ltd;
4781 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan start\n",
4782 lfsck_lfsck2name(lfsck));
4784 spin_lock(<ds->ltd_lock);
4785 while (!list_empty(&lad->lad_ost_phase2_list)) {
4786 ltd = list_first_entry(&lad->lad_ost_phase2_list,
4787 struct lfsck_tgt_desc,
4788 ltd_layout_phase_list);
4789 list_del_init(<d->ltd_layout_phase_list);
4790 if (bk->lb_param & LPF_OST_ORPHAN) {
4791 spin_unlock(<ds->ltd_lock);
4792 rc = lfsck_layout_scan_orphan(env, com, ltd);
4793 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
4796 if (unlikely(test_bit(LAD_EXIT, &lad->lad_flags) ||
4797 !thread_is_running(&lfsck->li_thread)))
4799 spin_lock(<ds->ltd_lock);
4803 if (list_empty(&lad->lad_ost_phase1_list))
4807 spin_unlock(<ds->ltd_lock);
4809 if (rc == 1 && bk->lb_param & LPF_OST_ORPHAN) {
4810 struct lfsck_layout *lo = com->lc_file_ram;
4813 com->lc_new_checked = 0;
4814 com->lc_new_scanned = 0;
4815 com->lc_time_last_checkpoint = ktime_get_seconds();
4816 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
4817 LFSCK_CHECKPOINT_INTERVAL;
4819 i = lfsck_sub_trace_file_fid2idx(
4820 &lo->ll_lldk_latest_scanned_phase2.lldk_fid);
4821 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4822 com->lc_sub_trace_objs[i].lsto_obj, true);
4823 while (rc > 0 && ++i < LFSCK_STF_COUNT)
4824 rc = lfsck_layout_double_scan_one_trace_file(env, com,
4825 com->lc_sub_trace_objs[i].lsto_obj, false);
4827 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan dangling stop "
4828 "at the No. %d trace file: rc = %d\n",
4829 lfsck_lfsck2name(lfsck), i, rc);
4832 CDEBUG(D_LFSCK, "%s: layout LFSCK phase2 scan stop: rc = %d\n",
4833 lfsck_lfsck2name(lfsck), rc);
4839 lfsck_layout_slave_async_interpret(const struct lu_env *env,
4840 struct ptlrpc_request *req,
4843 struct lfsck_layout_slave_async_args *llsaa = args;
4844 struct obd_export *exp = llsaa->llsaa_exp;
4845 struct lfsck_component *com = llsaa->llsaa_com;
4846 struct lfsck_layout_slave_target *llst = llsaa->llsaa_llst;
4847 struct lfsck_layout_slave_data *llsd = com->lc_data;
4848 struct lfsck_reply *lr = NULL;
4852 /* It is probably caused by network trouble, or target crash,
4853 * it will try several times (depends on the obd_timeout, and
4854 * will not less than 3 times). But to make the LFSCK can go
4855 * ahead, we should not try for ever. After some try but still
4856 * hit failure, it will assume that the target exit the LFSCK
4857 * prcoessing and stop try. */
4858 if (rc == -ENOTCONN || rc == -ESHUTDOWN) {
4859 int max_try = max_t(int, obd_timeout / 30, 3);
4861 if (++(llst->llst_failures) > max_try)
4867 llst->llst_failures = 0;
4868 lr = req_capsule_server_get(&req->rq_pill, &RMF_LFSCK_REPLY);
4869 if (lr->lr_status != LS_SCANNING_PHASE1 &&
4870 lr->lr_status != LS_SCANNING_PHASE2)
4875 CDEBUG(D_LFSCK, "%s: layout LFSCK slave gets the MDT %x "
4876 "status %d, failures_try %d\n", lfsck_lfsck2name(com->lc_lfsck),
4877 llst->llst_index, lr != NULL ? lr->lr_status : rc,
4878 llst->llst_failures);
4880 lfsck_layout_llst_del(llsd, llst);
4883 lfsck_layout_llst_put(llst);
4884 lfsck_component_put(env, com);
4885 class_export_put(exp);
4890 static int lfsck_layout_async_query(const struct lu_env *env,
4891 struct lfsck_component *com,
4892 struct obd_export *exp,
4893 struct lfsck_layout_slave_target *llst,
4894 struct lfsck_request *lr,
4895 struct ptlrpc_request_set *set)
4897 struct lfsck_layout_slave_async_args *llsaa;
4898 struct ptlrpc_request *req;
4899 struct lfsck_request *tmp;
4903 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_QUERY);
4907 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_QUERY);
4909 ptlrpc_request_free(req);
4913 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4915 ptlrpc_request_set_replen(req);
4917 llsaa = ptlrpc_req_async_args(llsaa, req);
4918 llsaa->llsaa_exp = exp;
4919 llsaa->llsaa_com = lfsck_component_get(com);
4920 llsaa->llsaa_llst = llst;
4921 req->rq_interpret_reply = lfsck_layout_slave_async_interpret;
4922 req->rq_allow_intr = 1;
4923 req->rq_no_delay = 1;
4924 ptlrpc_set_add_req(set, req);
4929 static int lfsck_layout_async_notify(const struct lu_env *env,
4930 struct obd_export *exp,
4931 struct lfsck_request *lr,
4932 struct ptlrpc_request_set *set)
4934 struct ptlrpc_request *req;
4935 struct lfsck_request *tmp;
4939 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
4943 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
4945 ptlrpc_request_free(req);
4949 tmp = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
4951 ptlrpc_request_set_replen(req);
4952 req->rq_allow_intr = 1;
4953 req->rq_no_delay = 1;
4954 ptlrpc_set_add_req(set, req);
4960 lfsck_layout_slave_query_master(const struct lu_env *env,
4961 struct lfsck_component *com)
4963 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
4964 struct lfsck_instance *lfsck = com->lc_lfsck;
4965 struct lfsck_layout_slave_data *llsd = com->lc_data;
4966 struct lfsck_layout_slave_target *llst;
4967 struct obd_export *exp;
4968 struct ptlrpc_request_set *set;
4973 set = ptlrpc_prep_set();
4975 GOTO(log, rc = -ENOMEM);
4977 memset(lr, 0, sizeof(*lr));
4978 lr->lr_event = LE_QUERY;
4979 lr->lr_active = LFSCK_TYPE_LAYOUT;
4981 llsd->llsd_touch_gen++;
4982 spin_lock(&llsd->llsd_lock);
4983 while (!list_empty(&llsd->llsd_master_list)) {
4984 llst = list_first_entry(&llsd->llsd_master_list,
4985 struct lfsck_layout_slave_target,
4987 if (llst->llst_gen == llsd->llsd_touch_gen)
4990 llst->llst_gen = llsd->llsd_touch_gen;
4991 list_move_tail(&llst->llst_list,
4992 &llsd->llsd_master_list);
4993 atomic_inc(&llst->llst_ref);
4994 spin_unlock(&llsd->llsd_lock);
4996 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
4999 lfsck_layout_llst_del(llsd, llst);
5000 lfsck_layout_llst_put(llst);
5001 spin_lock(&llsd->llsd_lock);
5005 rc = lfsck_layout_async_query(env, com, exp, llst, lr, set);
5007 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
5008 "query %s for layout: rc = %d\n",
5009 lfsck_lfsck2name(lfsck),
5010 exp->exp_obd->obd_name, rc);
5013 lfsck_layout_llst_put(llst);
5014 class_export_put(exp);
5016 spin_lock(&llsd->llsd_lock);
5018 spin_unlock(&llsd->llsd_lock);
5020 rc = ptlrpc_set_wait(env, set);
5021 ptlrpc_set_destroy(set);
5023 GOTO(log, rc = (rc1 != 0 ? rc1 : rc));
5026 CDEBUG(D_LFSCK, "%s: layout LFSCK slave queries master: rc = %d\n",
5027 lfsck_lfsck2name(com->lc_lfsck), rc);
5033 lfsck_layout_slave_notify_master(const struct lu_env *env,
5034 struct lfsck_component *com,
5035 enum lfsck_events event, int result)
5037 struct lfsck_layout *lo = com->lc_file_ram;
5038 struct lfsck_instance *lfsck = com->lc_lfsck;
5039 struct lfsck_layout_slave_data *llsd = com->lc_data;
5040 struct lfsck_request *lr = &lfsck_env_info(env)->lti_lr;
5041 struct lfsck_layout_slave_target *llst;
5042 struct obd_export *exp;
5043 struct ptlrpc_request_set *set;
5047 CDEBUG(D_LFSCK, "%s: layout LFSCK slave notifies master\n",
5048 lfsck_lfsck2name(com->lc_lfsck));
5050 set = ptlrpc_prep_set();
5054 memset(lr, 0, sizeof(*lr));
5055 lr->lr_event = event;
5056 lr->lr_flags = LEF_FROM_OST;
5057 lr->lr_status = result;
5058 lr->lr_index = lfsck_dev_idx(lfsck);
5059 lr->lr_active = LFSCK_TYPE_LAYOUT;
5060 lr->lr_flags2 = lo->ll_flags;
5061 llsd->llsd_touch_gen++;
5062 spin_lock(&llsd->llsd_lock);
5063 while (!list_empty(&llsd->llsd_master_list)) {
5064 llst = list_first_entry(&llsd->llsd_master_list,
5065 struct lfsck_layout_slave_target,
5067 if (llst->llst_gen == llsd->llsd_touch_gen)
5070 llst->llst_gen = llsd->llsd_touch_gen;
5071 list_move_tail(&llst->llst_list,
5072 &llsd->llsd_master_list);
5073 atomic_inc(&llst->llst_ref);
5074 spin_unlock(&llsd->llsd_lock);
5076 exp = lustre_find_lwp_by_index(lfsck->li_obd->obd_name,
5079 lfsck_layout_llst_del(llsd, llst);
5080 lfsck_layout_llst_put(llst);
5081 spin_lock(&llsd->llsd_lock);
5085 rc = lfsck_layout_async_notify(env, exp, lr, set);
5087 CDEBUG(D_LFSCK, "%s: layout LFSCK slave fail to "
5088 "notify %s for layout: rc = %d\n",
5089 lfsck_lfsck2name(lfsck),
5090 exp->exp_obd->obd_name, rc);
5092 lfsck_layout_llst_put(llst);
5093 class_export_put(exp);
5094 spin_lock(&llsd->llsd_lock);
5096 spin_unlock(&llsd->llsd_lock);
5098 ptlrpc_set_wait(env, set);
5099 ptlrpc_set_destroy(set);
5105 * \ret -ENODATA: unrecognized stripe
5106 * \ret = 0 : recognized stripe
5107 * \ret < 0 : other failures
5109 static int lfsck_layout_master_check_pairs(const struct lu_env *env,
5110 struct lfsck_component *com,
5111 struct lu_fid *cfid,
5112 struct lu_fid *pfid, __u32 comp_id)
5114 struct lfsck_thread_info *info = lfsck_env_info(env);
5115 struct lu_buf *buf = &info->lti_big_buf;
5116 struct ost_id *oi = &info->lti_oi;
5117 struct dt_object *obj;
5118 struct lov_mds_md_v1 *lmm;
5119 struct lov_ost_data_v1 *objs;
5120 __u32 idx = pfid->f_stripe_idx;
5128 obj = lfsck_object_find_bottom(env, com->lc_lfsck, pfid);
5130 RETURN(PTR_ERR(obj));
5132 dt_read_lock(env, obj, 0);
5133 if (unlikely(dt_object_exists(obj) == 0 ||
5134 lfsck_is_dead_obj(obj)))
5135 GOTO(unlock, rc = -ENOENT);
5137 if (!S_ISREG(lfsck_object_type(obj)))
5138 GOTO(unlock, rc = -ENODATA);
5140 rc = lfsck_layout_get_lovea(env, obj, buf);
5145 magic = le32_to_cpu(lmm->lmm_magic);
5146 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
5147 struct lov_comp_md_v1 *lcm = buf->lb_buf;
5148 struct lov_comp_md_entry_v1 *lcme;
5151 GOTO(unlock, rc = -ENODATA);
5153 count = le16_to_cpu(lcm->lcm_entry_count);
5154 for (i = 0; i < count; i++) {
5155 lcme = &lcm->lcm_entries[i];
5156 if (le32_to_cpu(lcme->lcme_id) == comp_id) {
5158 le32_to_cpu(lcme->lcme_offset);
5159 magic = le32_to_cpu(lmm->lmm_magic);
5160 if (!(le32_to_cpu(lcme->lcme_flags) &
5162 GOTO(unlock, rc = -ENODATA);
5168 GOTO(unlock, rc = -ENODATA);
5172 if (magic == LOV_MAGIC_V1) {
5173 objs = &lmm->lmm_objects[0];
5175 LASSERT(magic == LOV_MAGIC_V3);
5176 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5179 fid_to_ostid(cfid, oi);
5180 count = le16_to_cpu(lmm->lmm_stripe_count);
5181 for (i = 0; i < count; i++, objs++) {
5184 ostid_le_to_cpu(&objs->l_ost_oi, &oi2);
5185 if (memcmp(oi, &oi2, sizeof(*oi)) == 0)
5186 GOTO(unlock, rc = (i != idx ? -ENODATA : 0));
5189 GOTO(unlock, rc = -ENODATA);
5192 dt_read_unlock(env, obj);
5193 lfsck_object_put(env, obj);
5199 * The LFSCK-on-OST will ask the LFSCK-on-MDT to check whether the given
5200 * MDT-object/OST-object pairs match or not to aviod transfer MDT-object
5201 * layout EA from MDT to OST. On one hand, the OST no need to understand
5202 * the layout EA structure; on the other hand, it may cause trouble when
5203 * transfer large layout EA from MDT to OST via normal OUT RPC.
5205 * \ret > 0: unrecognized stripe
5206 * \ret = 0: recognized stripe
5207 * \ret < 0: other failures
5209 static int lfsck_layout_slave_check_pairs(const struct lu_env *env,
5210 struct lfsck_component *com,
5211 struct lu_fid *cfid,
5212 struct lu_fid *pfid, __u32 comp_id)
5214 struct lfsck_instance *lfsck = com->lc_lfsck;
5215 struct obd_device *obd = lfsck->li_obd;
5216 struct seq_server_site *ss = lfsck_dev_site(lfsck);
5217 struct obd_export *exp = NULL;
5218 struct ptlrpc_request *req = NULL;
5219 struct lfsck_request *lr;
5220 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
5224 if (unlikely(fid_is_idif(pfid)))
5227 fld_range_set_any(range);
5228 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(pfid), range);
5230 RETURN(rc == -ENOENT ? 1 : rc);
5232 if (unlikely(!fld_range_is_mdt(range)))
5235 exp = lustre_find_lwp_by_index(obd->obd_name, range->lsr_index);
5236 if (unlikely(exp == NULL))
5239 if (!(exp_connect_flags(exp) & OBD_CONNECT_LFSCK))
5240 GOTO(out, rc = -EOPNOTSUPP);
5242 req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_LFSCK_NOTIFY);
5244 GOTO(out, rc = -ENOMEM);
5246 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
5248 ptlrpc_request_free(req);
5253 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
5254 memset(lr, 0, sizeof(*lr));
5255 lr->lr_event = LE_PAIRS_VERIFY;
5256 lr->lr_active = LFSCK_TYPE_LAYOUT;
5257 lr->lr_fid = *cfid; /* OST-object itself FID. */
5258 lr->lr_fid2 = *pfid; /* The claimed parent FID. */
5259 lr->lr_comp_id = comp_id;
5261 ptlrpc_request_set_replen(req);
5262 rc = ptlrpc_queue_wait(req);
5263 ptlrpc_req_finished(req);
5265 if (rc == -ENOENT || rc == -ENODATA)
5272 class_export_put(exp);
5277 static int lfsck_layout_slave_repair_pfid(const struct lu_env *env,
5278 struct lfsck_component *com,
5279 struct lfsck_req_local *lrl)
5281 struct dt_object *obj;
5285 obj = lfsck_object_find_bottom(env, com->lc_lfsck, &lrl->lrl_fid);
5287 GOTO(log, rc = PTR_ERR(obj));
5289 rc = __lfsck_layout_update_pfid(env, com, obj,
5290 &lrl->lrl_ff_client.ff_parent,
5291 &lrl->lrl_ff_client.ff_layout,
5292 lrl->lrl_ff_client.ff_layout_version,
5293 lrl->lrl_ff_client.ff_range,
5294 lrl->lrl_ff_client.ff_parent.f_ver);
5296 lfsck_object_put(env, obj);
5299 CDEBUG(D_LFSCK, "%s: layout LFSCK slave repaired pfid for "DFID
5300 ", parent "DFID": rc = %d\n", lfsck_lfsck2name(com->lc_lfsck),
5301 PFID(&lrl->lrl_fid), PFID(&lrl->lrl_ff_client.ff_parent), rc);
5308 static void lfsck_layout_slave_quit(const struct lu_env *env,
5309 struct lfsck_component *com);
5311 static int lfsck_layout_reset(const struct lu_env *env,
5312 struct lfsck_component *com, bool init)
5314 struct lfsck_layout *lo = com->lc_file_ram;
5317 down_write(&com->lc_sem);
5319 memset(lo, 0, com->lc_file_size);
5321 __u32 count = lo->ll_success_count;
5322 time64_t last_time = lo->ll_time_last_complete;
5324 memset(lo, 0, com->lc_file_size);
5325 lo->ll_success_count = count;
5326 lo->ll_time_last_complete = last_time;
5329 lo->ll_magic = LFSCK_LAYOUT_MAGIC;
5330 lo->ll_status = LS_INIT;
5332 if (com->lc_lfsck->li_master) {
5333 struct lfsck_assistant_data *lad = com->lc_data;
5335 clear_bit(LAD_INCOMPLETE, &lad->lad_flags);
5336 bitmap_zero(lad->lad_bitmap, lad->lad_bitmap_count);
5339 rc = lfsck_layout_store(env, com);
5340 if (rc == 0 && com->lc_lfsck->li_master)
5341 rc = lfsck_load_sub_trace_files(env, com,
5342 &dt_lfsck_layout_dangling_features, LFSCK_LAYOUT, true);
5343 up_write(&com->lc_sem);
5345 CDEBUG(D_LFSCK, "%s: layout LFSCK reset: rc = %d\n",
5346 lfsck_lfsck2name(com->lc_lfsck), rc);
5351 static void lfsck_layout_fail(const struct lu_env *env,
5352 struct lfsck_component *com, bool new_checked)
5354 struct lfsck_layout *lo = com->lc_file_ram;
5356 down_write(&com->lc_sem);
5358 com->lc_new_checked++;
5359 lfsck_layout_record_failure(env, com->lc_lfsck, lo);
5360 up_write(&com->lc_sem);
5363 static int lfsck_layout_master_checkpoint(const struct lu_env *env,
5364 struct lfsck_component *com, bool init)
5366 struct lfsck_instance *lfsck = com->lc_lfsck;
5367 struct lfsck_layout *lo = com->lc_file_ram;
5371 rc = lfsck_checkpoint_generic(env, com);
5373 return rc > 0 ? 0 : rc;
5376 down_write(&com->lc_sem);
5378 lo->ll_pos_latest_start =
5379 lfsck->li_pos_checkpoint.lp_oit_cookie;
5381 lo->ll_pos_last_checkpoint =
5382 lfsck->li_pos_checkpoint.lp_oit_cookie;
5383 lo->ll_run_time_phase1 += ktime_get_seconds() -
5384 lfsck->li_time_last_checkpoint;
5385 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5386 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5387 com->lc_new_checked = 0;
5390 rc = lfsck_layout_store(env, com);
5391 up_write(&com->lc_sem);
5393 CDEBUG(D_LFSCK, "%s: layout LFSCK master checkpoint at the pos ["
5394 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5395 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5400 static int lfsck_layout_slave_checkpoint(const struct lu_env *env,
5401 struct lfsck_component *com, bool init)
5403 struct lfsck_instance *lfsck = com->lc_lfsck;
5404 struct lfsck_layout *lo = com->lc_file_ram;
5407 if (com->lc_new_checked == 0 && !init)
5410 down_write(&com->lc_sem);
5412 lo->ll_pos_latest_start =
5413 lfsck->li_pos_checkpoint.lp_oit_cookie;
5415 lo->ll_pos_last_checkpoint =
5416 lfsck->li_pos_checkpoint.lp_oit_cookie;
5417 lo->ll_run_time_phase1 += ktime_get_seconds() -
5418 lfsck->li_time_last_checkpoint;
5419 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
5420 lo->ll_objs_checked_phase1 += com->lc_new_checked;
5421 com->lc_new_checked = 0;
5424 rc = lfsck_layout_store(env, com);
5425 up_write(&com->lc_sem);
5427 CDEBUG(D_LFSCK, "%s: layout LFSCK slave checkpoint at the pos ["
5428 "%llu], status = %d: rc = %d\n", lfsck_lfsck2name(lfsck),
5429 lfsck->li_pos_current.lp_oit_cookie, lo->ll_status, rc);
5434 static int lfsck_layout_prep(const struct lu_env *env,
5435 struct lfsck_component *com,
5436 struct lfsck_start *start)
5438 struct lfsck_instance *lfsck = com->lc_lfsck;
5439 struct lfsck_layout *lo = com->lc_file_ram;
5440 struct lfsck_position *pos = &com->lc_pos_start;
5442 fid_zero(&pos->lp_dir_parent);
5443 pos->lp_dir_cookie = 0;
5444 if (lo->ll_status == LS_COMPLETED ||
5445 lo->ll_status == LS_PARTIAL ||
5446 /* To handle orphan, must scan from the beginning. */
5447 (start != NULL && start->ls_flags & LPF_OST_ORPHAN)) {
5450 rc = lfsck_layout_reset(env, com, false);
5452 rc = lfsck_set_param(env, lfsck, start, true);
5455 CDEBUG(D_LFSCK, "%s: layout LFSCK prep failed: "
5456 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
5462 down_write(&com->lc_sem);
5463 lo->ll_time_latest_start = ktime_get_real_seconds();
5464 spin_lock(&lfsck->li_lock);
5465 if (lo->ll_flags & LF_SCANNED_ONCE) {
5466 if (!lfsck->li_drop_dryrun ||
5467 lo->ll_pos_first_inconsistent == 0) {
5468 lo->ll_status = LS_SCANNING_PHASE2;
5469 list_move_tail(&com->lc_link,
5470 &lfsck->li_list_double_scan);
5471 pos->lp_oit_cookie = 0;
5475 lo->ll_status = LS_SCANNING_PHASE1;
5476 lo->ll_run_time_phase1 = 0;
5477 lo->ll_run_time_phase2 = 0;
5478 lo->ll_objs_checked_phase1 = 0;
5479 lo->ll_objs_checked_phase2 = 0;
5480 lo->ll_objs_failed_phase1 = 0;
5481 lo->ll_objs_failed_phase2 = 0;
5482 for (i = 0; i < LLIT_MAX; i++)
5483 lo->ll_objs_repaired[i] = 0;
5485 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5486 fid_zero(&com->lc_fid_latest_scanned_phase2);
5489 lo->ll_status = LS_SCANNING_PHASE1;
5490 if (!lfsck->li_drop_dryrun ||
5491 lo->ll_pos_first_inconsistent == 0)
5492 pos->lp_oit_cookie = lo->ll_pos_last_checkpoint + 1;
5494 pos->lp_oit_cookie = lo->ll_pos_first_inconsistent;
5496 spin_unlock(&lfsck->li_lock);
5497 up_write(&com->lc_sem);
5502 static int lfsck_layout_slave_prep(const struct lu_env *env,
5503 struct lfsck_component *com,
5504 struct lfsck_start_param *lsp)
5506 struct lfsck_layout_slave_data *llsd = com->lc_data;
5507 struct lfsck_instance *lfsck = com->lc_lfsck;
5508 struct lfsck_layout *lo = com->lc_file_ram;
5509 struct lfsck_start *start = lsp->lsp_start;
5512 rc = lfsck_layout_prep(env, com, start);
5516 if (lo->ll_flags & LF_CRASHED_LASTID &&
5517 list_empty(&llsd->llsd_master_list)) {
5518 LASSERT(lfsck->li_out_notify != NULL);
5520 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
5521 LE_LASTID_REBUILDING);
5524 if (!lsp->lsp_index_valid)
5527 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
5528 if (rc == 0 && start != NULL && start->ls_flags & LPF_OST_ORPHAN) {
5529 LASSERT(!llsd->llsd_rbtree_valid);
5531 down_write(&llsd->llsd_rb_rwsem);
5532 rc = lfsck_rbtree_setup(env, com);
5533 up_write(&llsd->llsd_rb_rwsem);
5536 CDEBUG(D_LFSCK, "%s: layout LFSCK slave prep done, start pos ["
5537 "%llu]\n", lfsck_lfsck2name(lfsck),
5538 com->lc_pos_start.lp_oit_cookie);
5543 static int lfsck_layout_master_prep(const struct lu_env *env,
5544 struct lfsck_component *com,
5545 struct lfsck_start_param *lsp)
5550 rc = lfsck_layout_load_bitmap(env, com);
5552 rc = lfsck_layout_reset(env, com, false);
5554 rc = lfsck_set_param(env, com->lc_lfsck,
5555 lsp->lsp_start, true);
5561 rc = lfsck_layout_prep(env, com, lsp->lsp_start);
5565 rc = lfsck_start_assistant(env, com, lsp);
5570 CDEBUG(D_LFSCK, "%s: layout LFSCK master prep done, start pos ["
5571 "%llu]\n", lfsck_lfsck2name(com->lc_lfsck),
5572 com->lc_pos_start.lp_oit_cookie);
5577 /* Pre-fetch the attribute for each stripe in the given layout EA. */
5578 static int lfsck_layout_scan_stripes(const struct lu_env *env,
5579 struct lfsck_component *com,
5580 struct dt_object *parent,
5581 struct lov_mds_md_v1 *lmm, __u32 comp_id)
5583 struct lfsck_thread_info *info = lfsck_env_info(env);
5584 struct lfsck_instance *lfsck = com->lc_lfsck;
5585 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5586 struct lfsck_layout *lo = com->lc_file_ram;
5587 struct lfsck_assistant_data *lad = com->lc_data;
5588 struct lfsck_assistant_object *lso = NULL;
5589 struct lov_ost_data_v1 *objs;
5590 struct lfsck_tgt_descs *ltds = &lfsck->li_ost_descs;
5591 struct ptlrpc_thread *mthread = &lfsck->li_thread;
5592 struct ptlrpc_thread *athread = &lad->lad_thread;
5600 lfsck_buf_init(&buf, &info->lti_ff, sizeof(struct filter_fid));
5601 magic = le32_to_cpu(lmm->lmm_magic);
5602 if (magic == LOV_MAGIC_V1) {
5603 objs = &lmm->lmm_objects[0];
5605 LASSERT(magic == LOV_MAGIC_V3);
5606 objs = &((struct lov_mds_md_v3 *)lmm)->lmm_objects[0];
5609 count = le16_to_cpu(lmm->lmm_stripe_count);
5610 for (i = 0; i < count; i++, objs++) {
5611 struct lu_fid *fid = &info->lti_fid;
5612 struct ost_id *oi = &info->lti_oi;
5613 struct lfsck_layout_req *llr;
5614 struct lfsck_tgt_desc *tgt = NULL;
5615 struct dt_object *cobj = NULL;
5617 bool wakeup = false;
5619 if (unlikely(lovea_slot_is_dummy(objs)))
5622 wait_event_idle(mthread->t_ctl_waitq,
5623 lad->lad_prefetched < bk->lb_async_windows ||
5624 !thread_is_running(mthread) ||
5625 thread_is_stopped(athread));
5627 if (unlikely(!thread_is_running(mthread)) ||
5628 thread_is_stopped(athread))
5631 if (unlikely(lfsck_is_dead_obj(parent)))
5634 ostid_le_to_cpu(&objs->l_ost_oi, oi);
5635 index = le32_to_cpu(objs->l_ost_idx);
5636 rc = ostid_to_fid(fid, oi, index);
5638 CDEBUG(D_LFSCK, "%s: get invalid layout EA for "DFID
5639 ": "DOSTID", idx %u, comp_id %u\n",
5640 lfsck_lfsck2name(lfsck),
5641 PFID(lfsck_dto2fid(parent)), POSTID(oi),
5646 tgt = lfsck_tgt_get(ltds, index);
5647 if (unlikely(tgt == NULL)) {
5648 CDEBUG(D_LFSCK, "%s: cannot talk with OST %x which "
5649 "did not join the layout LFSCK, comp_id %u\n",
5650 lfsck_lfsck2name(lfsck), index, comp_id);
5651 lfsck_lad_set_bitmap(env, com, index);
5655 /* There is potential deadlock race condition between object
5656 * destroy and layout LFSCK. Consider the following scenario:
5658 * 1) The LFSCK thread obtained the parent object firstly, at
5659 * that time, the parent object has not been destroyed yet.
5661 * 2) One RPC service thread destroyed the parent and all its
5662 * children objects. Because the LFSCK is referencing the
5663 * parent object, then the parent object will be marked as
5664 * dying in RAM. On the other hand, the parent object is
5665 * referencing all its children objects, then all children
5666 * objects will be marked as dying in RAM also.
5668 * 3) The LFSCK thread tries to find some child object with
5669 * the parent object referenced. Then it will find that the
5670 * child object is dying. According to the object visibility
5671 * rules: the object with dying flag cannot be returned to
5672 * others. So the LFSCK thread has to wait until the dying
5673 * object has been purged from RAM, then it can allocate a
5674 * new object (with the same FID) in RAM. Unfortunately, the
5675 * LFSCK thread itself is referencing the parent object, and
5676 * cause the parent object cannot be purged, then cause the
5677 * child object cannot be purged also. So the LFSCK thread
5678 * will fall into deadlock.
5680 cobj = lfsck_object_find_by_dev(env, tgt->ltd_tgt, fid);
5682 if (lfsck_is_dead_obj(parent)) {
5692 rc = dt_declare_attr_get(env, cobj);
5696 rc = dt_declare_xattr_get(env, cobj, &buf, XATTR_NAME_FID);
5701 struct lu_attr *attr = &info->lti_la;
5703 rc = dt_attr_get(env, parent, attr);
5707 lso = lfsck_assistant_object_init(env,
5708 lfsck_dto2fid(parent), attr,
5709 lfsck->li_pos_current.lp_oit_cookie, false);
5718 llr = lfsck_layout_assistant_req_init(lso, cobj, comp_id,
5726 spin_lock(&lad->lad_lock);
5727 if (lad->lad_assistant_status < 0) {
5728 spin_unlock(&lad->lad_lock);
5729 lfsck_layout_assistant_req_fini(env, &llr->llr_lar);
5731 RETURN(lad->lad_assistant_status);
5734 list_add_tail(&llr->llr_lar.lar_list, &lad->lad_req_list);
5735 if (lad->lad_prefetched == 0)
5738 lad->lad_prefetched++;
5739 spin_unlock(&lad->lad_lock);
5741 wake_up(&athread->t_ctl_waitq);
5744 down_write(&com->lc_sem);
5745 com->lc_new_checked++;
5747 lfsck_layout_record_failure(env, lfsck, lo);
5748 up_write(&com->lc_sem);
5750 if (cobj != NULL && !IS_ERR(cobj))
5751 lfsck_object_put(env, cobj);
5753 if (likely(tgt != NULL))
5756 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5764 lfsck_assistant_object_put(env, lso);
5769 /* For the given object, read its layout EA locally. For each stripe, pre-fetch
5770 * the OST-object's attribute and generate an structure lfsck_layout_req on the
5771 * list ::lad_req_list.
5773 * For each request on above list, the lfsck_layout_assistant thread compares
5774 * the OST side attribute with local attribute, if inconsistent, then repair it.
5776 * All above processing is async mode with pipeline. */
5777 static int lfsck_layout_master_exec_oit(const struct lu_env *env,
5778 struct lfsck_component *com,
5779 struct dt_object *obj)
5781 struct lfsck_thread_info *info = lfsck_env_info(env);
5782 struct ost_id *oi = &info->lti_oi;
5783 struct lfsck_layout *lo = com->lc_file_ram;
5784 struct lfsck_assistant_data *lad = com->lc_data;
5785 struct lfsck_instance *lfsck = com->lc_lfsck;
5786 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5787 struct thandle *handle = NULL;
5788 struct lu_buf *buf = &info->lti_big_buf;
5789 struct lov_mds_md_v1 *lmm = NULL;
5790 struct dt_device *dev = lfsck_obj2dev(obj);
5791 struct lustre_handle lh = { 0 };
5792 struct lu_buf ea_buf = { NULL };
5793 struct lov_comp_md_v1 *lcm = NULL;
5794 struct lov_comp_md_entry_v1 *lcme = NULL;
5799 bool locked = false;
5800 bool stripe = false;
5801 bool bad_oi = false;
5804 if (!S_ISREG(lfsck_object_type(obj)))
5807 if (lad->lad_assistant_status < 0)
5808 GOTO(out, rc = -ESRCH);
5810 fid_to_lmm_oi(lfsck_dto2fid(obj), oi);
5811 lmm_oi_cpu_to_le(oi, oi);
5812 dt_read_lock(env, obj, 0);
5817 if (dt_object_exists(obj) == 0 ||
5818 lfsck_is_dead_obj(obj))
5821 rc = lfsck_layout_get_lovea(env, obj, buf);
5822 if (rc == -EINVAL || rc == -ENODATA || rc == -EOPNOTSUPP)
5823 /* Skip bad lov EA during the 1st cycle scanning, and
5824 * try to recover it via orphan in the 2nd scanning. */
5831 magic = le32_to_cpu(lmm->lmm_magic);
5832 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
5833 struct lov_mds_md_v1 *v1;
5837 count = le16_to_cpu(lcm->lcm_entry_count);
5838 for (i = 0; i < count; i++) {
5839 lcme = &lcm->lcm_entries[i];
5840 v1 = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5841 if (memcmp(oi, &v1->lmm_oi, sizeof(*oi)) != 0)
5845 GOTO(out, stripe = true);
5846 } else if (memcmp(oi, &lmm->lmm_oi, sizeof(*oi)) == 0) {
5847 GOTO(out, stripe = true);
5851 /* Inconsistent lmm_oi, should be repaired. */
5854 if (bk->lb_param & LPF_DRYRUN) {
5855 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5857 GOTO(out, stripe = true);
5860 if (!lustre_handle_is_used(&lh)) {
5861 dt_read_unlock(env, obj);
5863 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5864 MDS_INODELOCK_LAYOUT |
5865 MDS_INODELOCK_XATTR, LCK_EX);
5869 handle = lfsck_trans_create(env, dev, lfsck);
5871 GOTO(out, rc = PTR_ERR(handle));
5873 lfsck_buf_init(&ea_buf, buf->lb_buf, size);
5874 rc = dt_declare_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5875 LU_XATTR_REPLACE, handle);
5879 rc = dt_trans_start_local(env, dev, handle);
5883 dt_write_lock(env, obj, 0);
5889 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
5890 struct lov_mds_md_v1 *v1;
5893 for (i = 0; i < count; i++) {
5894 lcme = &lcm->lcm_entries[i];
5895 v1 = buf->lb_buf + le32_to_cpu(lcme->lcme_offset);
5902 rc = dt_xattr_set(env, obj, &ea_buf, XATTR_NAME_LOV,
5903 LU_XATTR_REPLACE, handle);
5907 lo->ll_objs_repaired[LLIT_OTHERS - 1]++;
5909 GOTO(out, stripe = true);
5913 if (lustre_handle_is_used(&lh))
5914 dt_write_unlock(env, obj);
5916 dt_read_unlock(env, obj);
5919 if (handle != NULL && !IS_ERR(handle))
5920 dt_trans_stop(env, dev, handle);
5922 lfsck_ibits_unlock(&lh, LCK_EX);
5925 CDEBUG(D_LFSCK, "%s: layout LFSCK master %s bad lmm_oi for "
5926 DFID": rc = %d\n", lfsck_lfsck2name(lfsck),
5927 bk->lb_param & LPF_DRYRUN ? "found" : "repaired",
5928 PFID(lfsck_dto2fid(obj)), rc);
5931 if (magic == LOV_MAGIC_COMP_V1 || magic == LOV_MAGIC_SEL) {
5934 for (i = 0; i < count; i++) {
5935 lcme = &lcm->lcm_entries[i];
5936 if (!(le32_to_cpu(lcme->lcme_flags) &
5940 rc = lfsck_layout_scan_stripes(env, com, obj,
5941 (struct lov_mds_md_v1 *)(buf->lb_buf +
5942 le32_to_cpu(lcme->lcme_offset)),
5943 le32_to_cpu(lcme->lcme_id));
5946 rc = lfsck_layout_scan_stripes(env, com, obj, lmm, 0);
5949 down_write(&com->lc_sem);
5950 com->lc_new_checked++;
5952 lfsck_layout_record_failure(env, lfsck, lo);
5953 up_write(&com->lc_sem);
5959 static int lfsck_layout_slave_exec_oit(const struct lu_env *env,
5960 struct lfsck_component *com,
5961 struct dt_object *obj)
5963 struct lfsck_instance *lfsck = com->lc_lfsck;
5964 struct lfsck_layout *lo = com->lc_file_ram;
5965 const struct lu_fid *fid = lfsck_dto2fid(obj);
5966 struct lfsck_layout_slave_data *llsd = com->lc_data;
5967 struct lfsck_layout_seq *lls;
5973 LASSERT(llsd != NULL);
5975 if (CFS_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY5) &&
5976 cfs_fail_val == lfsck_dev_idx(lfsck)) {
5977 struct ptlrpc_thread *thread = &lfsck->li_thread;
5979 wait_event_idle_timeout(thread->t_ctl_waitq,
5980 !thread_is_running(thread),
5981 cfs_time_seconds(1));
5984 lfsck_rbtree_update_bitmap(env, com, fid, false);
5986 down_write(&com->lc_sem);
5987 if (fid_is_idif(fid))
5989 else if (!fid_is_norm(fid) ||
5990 !fid_is_for_ostobj(env, lfsck, obj, fid))
5991 GOTO(unlock, rc = 0);
5994 com->lc_new_checked++;
5996 lls = lfsck_layout_seq_lookup(llsd, seq);
5999 if (unlikely(lls == NULL))
6000 GOTO(unlock, rc = -ENOMEM);
6002 INIT_LIST_HEAD(&lls->lls_list);
6004 rc = lfsck_layout_lastid_load(env, com, lls);
6006 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
6007 "load LAST_ID for %#llx: rc = %d\n",
6008 lfsck_lfsck2name(com->lc_lfsck), seq, rc);
6009 lo->ll_objs_failed_phase1++;
6014 lfsck_layout_seq_insert(llsd, lls);
6017 if (unlikely(fid_is_last_id(fid)))
6018 GOTO(unlock, rc = 0);
6020 if (fid_is_idif(fid))
6021 oid = fid_idif_id(fid_seq(fid), fid_oid(fid), fid_ver(fid));
6025 if (oid > lls->lls_lastid_known)
6026 lls->lls_lastid_known = oid;
6028 if (oid > lls->lls_lastid) {
6029 if (!(lo->ll_flags & LF_CRASHED_LASTID)) {
6030 /* OFD may create new objects during LFSCK scanning. */
6031 rc = lfsck_layout_lastid_reload(env, com, lls);
6032 if (unlikely(rc != 0)) {
6033 CDEBUG(D_LFSCK, "%s: layout LFSCK failed to "
6034 "reload LAST_ID for %#llx: rc = %d\n",
6035 lfsck_lfsck2name(com->lc_lfsck),
6041 if (oid <= lls->lls_lastid ||
6042 lo->ll_flags & LF_CRASHED_LASTID)
6043 GOTO(unlock, rc = 0);
6045 LASSERT(lfsck->li_out_notify != NULL);
6047 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6048 LE_LASTID_REBUILDING);
6049 lo->ll_flags |= LF_CRASHED_LASTID;
6051 CDEBUG(D_LFSCK, "%s: layout LFSCK finds crashed "
6052 "LAST_ID file (2) for the sequence %#llx"
6053 ", old value %llu, known value %llu\n",
6054 lfsck_lfsck2name(lfsck), lls->lls_seq,
6055 lls->lls_lastid, oid);
6058 lls->lls_lastid = oid;
6062 GOTO(unlock, rc = 0);
6065 up_write(&com->lc_sem);
6070 static int lfsck_layout_exec_dir(const struct lu_env *env,
6071 struct lfsck_component *com,
6072 struct lfsck_assistant_object *lso,
6073 struct lu_dirent *ent, __u16 type)
6078 static int lfsck_layout_master_post(const struct lu_env *env,
6079 struct lfsck_component *com,
6080 int result, bool init)
6082 struct lfsck_instance *lfsck = com->lc_lfsck;
6083 struct lfsck_layout *lo = com->lc_file_ram;
6087 lfsck_post_generic(env, com, &result);
6089 down_write(&com->lc_sem);
6090 spin_lock(&lfsck->li_lock);
6092 lo->ll_pos_last_checkpoint =
6093 lfsck->li_pos_checkpoint.lp_oit_cookie;
6096 if (lo->ll_flags & LF_INCOMPLETE)
6097 lo->ll_status = LS_PARTIAL;
6099 lo->ll_status = LS_SCANNING_PHASE2;
6100 lo->ll_flags |= LF_SCANNED_ONCE;
6101 lo->ll_flags &= ~LF_UPGRADE;
6102 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
6103 } else if (result == 0) {
6104 if (lfsck->li_status != 0)
6105 lo->ll_status = lfsck->li_status;
6107 lo->ll_status = LS_STOPPED;
6108 if (lo->ll_status != LS_PAUSED)
6109 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6111 lo->ll_status = LS_FAILED;
6112 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6114 spin_unlock(&lfsck->li_lock);
6117 lo->ll_run_time_phase1 += ktime_get_seconds() -
6118 lfsck->li_time_last_checkpoint;
6119 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
6120 lo->ll_objs_checked_phase1 += com->lc_new_checked;
6121 com->lc_new_checked = 0;
6124 rc = lfsck_layout_store(env, com);
6125 up_write(&com->lc_sem);
6127 CDEBUG(D_LFSCK, "%s: layout LFSCK master post done: rc = %d\n",
6128 lfsck_lfsck2name(lfsck), rc);
6133 static int lfsck_layout_slave_post(const struct lu_env *env,
6134 struct lfsck_component *com,
6135 int result, bool init)
6137 struct lfsck_instance *lfsck = com->lc_lfsck;
6138 struct lfsck_layout *lo = com->lc_file_ram;
6142 down_write(&com->lc_sem);
6143 rc = lfsck_layout_lastid_store(env, com);
6147 LASSERT(lfsck->li_out_notify != NULL);
6149 spin_lock(&lfsck->li_lock);
6151 lo->ll_pos_last_checkpoint =
6152 lfsck->li_pos_checkpoint.lp_oit_cookie;
6155 lo->ll_status = LS_SCANNING_PHASE2;
6156 lo->ll_flags |= LF_SCANNED_ONCE;
6157 if (lo->ll_flags & LF_CRASHED_LASTID) {
6159 lo->ll_flags &= ~LF_CRASHED_LASTID;
6161 CDEBUG(D_LFSCK, "%s: layout LFSCK has rebuilt "
6162 "crashed LAST_ID files successfully\n",
6163 lfsck_lfsck2name(lfsck));
6165 lo->ll_flags &= ~LF_UPGRADE;
6166 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
6167 } else if (result == 0) {
6168 if (lfsck->li_status != 0)
6169 lo->ll_status = lfsck->li_status;
6171 lo->ll_status = LS_STOPPED;
6172 if (lo->ll_status != LS_PAUSED)
6173 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6175 lo->ll_status = LS_FAILED;
6176 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
6178 spin_unlock(&lfsck->li_lock);
6181 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
6185 lo->ll_run_time_phase1 += ktime_get_seconds() -
6186 lfsck->li_time_last_checkpoint;
6187 lo->ll_time_last_checkpoint = ktime_get_real_seconds();
6188 lo->ll_objs_checked_phase1 += com->lc_new_checked;
6189 com->lc_new_checked = 0;
6192 rc = lfsck_layout_store(env, com);
6193 up_write(&com->lc_sem);
6195 lfsck_layout_slave_notify_master(env, com, LE_PHASE1_DONE, result);
6197 CDEBUG(D_LFSCK, "%s: layout LFSCK slave post done: rc = %d\n",
6198 lfsck_lfsck2name(lfsck), rc);
6203 static void lfsck_layout_dump(const struct lu_env *env,
6204 struct lfsck_component *com, struct seq_file *m)
6206 struct lfsck_instance *lfsck = com->lc_lfsck;
6207 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
6208 struct lfsck_layout *lo = com->lc_file_ram;
6211 down_read(&com->lc_sem);
6212 if (bk->lb_param & LPF_DRYRUN)
6213 prefix = "inconsistent";
6215 prefix = "repaired";
6217 seq_printf(m, "name: lfsck_layout\n"
6223 lfsck_status2name(lo->ll_status));
6225 lfsck_bits_dump(m, lo->ll_flags, lfsck_flags_names, "flags");
6227 lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
6229 lfsck_time_dump(m, lo->ll_time_last_complete, "last_completed");
6231 lfsck_time_dump(m, lo->ll_time_latest_start, "latest_start");
6233 lfsck_time_dump(m, lo->ll_time_last_checkpoint, "last_checkpoint");
6235 seq_printf(m, "latest_start_position: %llu\n"
6236 "last_checkpoint_position: %llu\n"
6237 "first_failure_position: %llu\n",
6238 lo->ll_pos_latest_start,
6239 lo->ll_pos_last_checkpoint,
6240 lo->ll_pos_first_inconsistent);
6242 seq_printf(m, "success_count: %u\n"
6243 "%s_dangling: %llu\n"
6244 "%s_unmatched_pair: %llu\n"
6245 "%s_multiple_referenced: %llu\n"
6247 "%s_inconsistent_owner: %llu\n"
6250 "failed_phase1: %llu\n"
6251 "failed_phase2: %llu\n",
6252 lo->ll_success_count,
6253 prefix, lo->ll_objs_repaired[LLIT_DANGLING - 1],
6254 prefix, lo->ll_objs_repaired[LLIT_UNMATCHED_PAIR - 1],
6255 prefix, lo->ll_objs_repaired[LLIT_MULTIPLE_REFERENCED - 1],
6256 prefix, lo->ll_objs_repaired[LLIT_ORPHAN - 1],
6257 prefix, lo->ll_objs_repaired[LLIT_INCONSISTENT_OWNER - 1],
6258 prefix, lo->ll_objs_repaired[LLIT_OTHERS - 1],
6259 lo->ll_objs_skipped,
6260 lo->ll_objs_failed_phase1,
6261 lo->ll_objs_failed_phase2);
6263 if (lo->ll_status == LS_SCANNING_PHASE1) {
6264 time64_t duration = ktime_get_seconds() -
6265 lfsck->li_time_last_checkpoint;
6266 u64 checked = lo->ll_objs_checked_phase1 +
6267 com->lc_new_checked;
6268 u64 speed = checked;
6269 u64 new_checked = com->lc_new_checked;
6270 time64_t rtime = lo->ll_run_time_phase1 + duration;
6274 new_checked = div64_s64(new_checked, duration);
6276 speed = div64_s64(speed, rtime);
6277 seq_printf(m, "checked_phase1: %llu\n"
6278 "checked_phase2: %llu\n"
6279 "run_time_phase1: %lld seconds\n"
6280 "run_time_phase2: %lld seconds\n"
6281 "average_speed_phase1: %llu items/sec\n"
6282 "average_speed_phase2: N/A\n"
6283 "real_time_speed_phase1: %llu items/sec\n"
6284 "real_time_speed_phase2: N/A\n",
6286 lo->ll_objs_checked_phase2,
6288 lo->ll_run_time_phase2,
6292 if (likely(lfsck->li_di_oit)) {
6293 const struct dt_it_ops *iops =
6294 &lfsck->li_obj_oit->do_index_ops->dio_it;
6296 /* The low layer otable-based iteration position may NOT
6297 * exactly match the layout-based directory traversal
6298 * cookie. Generally, it is not a serious issue. But the
6299 * caller should NOT make assumption on that. */
6300 pos = iops->store(env, lfsck->li_di_oit);
6301 if (!lfsck->li_current_oit_processed)
6304 pos = lo->ll_pos_last_checkpoint;
6307 seq_printf(m, "current_position: %llu\n", pos);
6308 } else if (lo->ll_status == LS_SCANNING_PHASE2) {
6309 time64_t duration = ktime_get_seconds() -
6310 com->lc_time_last_checkpoint;
6311 u64 checked = lo->ll_objs_checked_phase2 +
6312 com->lc_new_checked;
6313 u64 speed1 = lo->ll_objs_checked_phase1;
6314 u64 speed2 = checked;
6315 u64 new_checked = com->lc_new_checked;
6316 time64_t rtime = lo->ll_run_time_phase2 + duration;
6319 new_checked = div64_s64(new_checked, duration);
6320 if (lo->ll_run_time_phase1 != 0)
6321 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
6323 speed2 = div64_s64(speed2, rtime);
6324 seq_printf(m, "checked_phase1: %llu\n"
6325 "checked_phase2: %llu\n"
6326 "run_time_phase1: %lld seconds\n"
6327 "run_time_phase2: %lld seconds\n"
6328 "average_speed_phase1: %llu items/sec\n"
6329 "average_speed_phase2: %llu items/sec\n"
6330 "real_time_speed_phase1: N/A\n"
6331 "real_time_speed_phase2: %llu items/sec\n"
6332 "current_position: "DFID"\n",
6333 lo->ll_objs_checked_phase1,
6335 lo->ll_run_time_phase1,
6340 PFID(&com->lc_fid_latest_scanned_phase2));
6342 __u64 speed1 = lo->ll_objs_checked_phase1;
6343 __u64 speed2 = lo->ll_objs_checked_phase2;
6345 if (lo->ll_run_time_phase1 != 0)
6346 speed1 = div64_s64(speed1, lo->ll_run_time_phase1);
6347 if (lo->ll_run_time_phase2 != 0)
6348 speed2 = div64_s64(speed2, lo->ll_run_time_phase2);
6349 seq_printf(m, "checked_phase1: %llu\n"
6350 "checked_phase2: %llu\n"
6351 "run_time_phase1: %lld seconds\n"
6352 "run_time_phase2: %lld seconds\n"
6353 "average_speed_phase1: %llu items/sec\n"
6354 "average_speed_phase2: %llu objs/sec\n"
6355 "real_time_speed_phase1: N/A\n"
6356 "real_time_speed_phase2: N/A\n"
6357 "current_position: N/A\n",
6358 lo->ll_objs_checked_phase1,
6359 lo->ll_objs_checked_phase2,
6360 lo->ll_run_time_phase1,
6361 lo->ll_run_time_phase2,
6366 up_read(&com->lc_sem);
6369 static int lfsck_layout_master_double_scan(const struct lu_env *env,
6370 struct lfsck_component *com)
6372 struct lfsck_layout *lo = com->lc_file_ram;
6373 struct lfsck_assistant_data *lad = com->lc_data;
6374 struct lfsck_instance *lfsck = com->lc_lfsck;
6375 struct lfsck_tgt_descs *ltds;
6376 struct lfsck_tgt_desc *ltd;
6377 struct lfsck_tgt_desc *next;
6380 rc = lfsck_double_scan_generic(env, com, lo->ll_status);
6382 if (thread_is_stopped(&lad->lad_thread)) {
6383 LASSERT(list_empty(&lad->lad_req_list));
6384 LASSERT(list_empty(&lad->lad_ost_phase1_list));
6385 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
6387 ltds = &lfsck->li_ost_descs;
6388 spin_lock(<ds->ltd_lock);
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 spin_unlock(<ds->ltd_lock);
6395 ltds = &lfsck->li_mdt_descs;
6396 spin_lock(<ds->ltd_lock);
6397 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6398 ltd_layout_phase_list) {
6399 list_del_init(<d->ltd_layout_phase_list);
6401 spin_unlock(<ds->ltd_lock);
6407 static int lfsck_layout_slave_double_scan(const struct lu_env *env,
6408 struct lfsck_component *com)
6410 struct lfsck_instance *lfsck = com->lc_lfsck;
6411 struct lfsck_layout_slave_data *llsd = com->lc_data;
6412 struct lfsck_layout *lo = com->lc_file_ram;
6413 struct ptlrpc_thread *thread = &lfsck->li_thread;
6417 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan start\n",
6418 lfsck_lfsck2name(lfsck));
6420 atomic_inc(&lfsck->li_double_scan_count);
6422 if (lo->ll_flags & LF_INCOMPLETE)
6425 com->lc_new_checked = 0;
6426 com->lc_new_scanned = 0;
6427 com->lc_time_last_checkpoint = ktime_get_seconds();
6428 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
6429 LFSCK_CHECKPOINT_INTERVAL;
6432 rc = lfsck_layout_slave_query_master(env, com);
6433 if (list_empty(&llsd->llsd_master_list)) {
6434 if (unlikely(!thread_is_running(thread)))
6445 rc = wait_event_idle_timeout(
6446 thread->t_ctl_waitq,
6447 !thread_is_running(thread) ||
6448 lo->ll_flags & LF_INCOMPLETE ||
6449 list_empty(&llsd->llsd_master_list),
6450 cfs_time_seconds(30));
6451 if (unlikely(!thread_is_running(thread)))
6454 if (lo->ll_flags & LF_INCOMPLETE)
6464 rc = lfsck_layout_double_scan_result(env, com, rc);
6465 lfsck_layout_slave_notify_master(env, com, LE_PHASE2_DONE,
6466 (rc > 0 && lo->ll_flags & LF_INCOMPLETE) ? 0 : rc);
6467 lfsck_layout_slave_quit(env, com);
6468 if (atomic_dec_and_test(&lfsck->li_double_scan_count))
6469 wake_up(&lfsck->li_thread.t_ctl_waitq);
6471 CDEBUG(D_LFSCK, "%s: layout LFSCK slave phase2 scan finished, "
6472 "status %d: rc = %d\n",
6473 lfsck_lfsck2name(lfsck), lo->ll_status, rc);
6478 static void lfsck_layout_master_data_release(const struct lu_env *env,
6479 struct lfsck_component *com)
6481 struct lfsck_assistant_data *lad = com->lc_data;
6482 struct lfsck_instance *lfsck = com->lc_lfsck;
6483 struct lfsck_tgt_descs *ltds;
6484 struct lfsck_tgt_desc *ltd;
6485 struct lfsck_tgt_desc *next;
6487 LASSERT(lad != NULL);
6488 LASSERT(thread_is_init(&lad->lad_thread) ||
6489 thread_is_stopped(&lad->lad_thread));
6490 LASSERT(list_empty(&lad->lad_req_list));
6492 com->lc_data = NULL;
6494 ltds = &lfsck->li_ost_descs;
6495 spin_lock(<ds->ltd_lock);
6496 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6497 ltd_layout_phase_list) {
6498 list_del_init(<d->ltd_layout_phase_list);
6500 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6501 ltd_layout_phase_list) {
6502 list_del_init(<d->ltd_layout_phase_list);
6504 list_for_each_entry_safe(ltd, next, &lad->lad_ost_list,
6506 list_del_init(<d->ltd_layout_list);
6508 spin_unlock(<ds->ltd_lock);
6510 ltds = &lfsck->li_mdt_descs;
6511 spin_lock(<ds->ltd_lock);
6512 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6513 ltd_layout_phase_list) {
6514 list_del_init(<d->ltd_layout_phase_list);
6516 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6517 ltd_layout_phase_list) {
6518 list_del_init(<d->ltd_layout_phase_list);
6520 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
6522 list_del_init(<d->ltd_layout_list);
6524 spin_unlock(<ds->ltd_lock);
6526 bitmap_free(lad->lad_bitmap);
6531 static void lfsck_layout_slave_data_release(const struct lu_env *env,
6532 struct lfsck_component *com)
6534 struct lfsck_layout_slave_data *llsd = com->lc_data;
6536 lfsck_layout_slave_quit(env, com);
6537 com->lc_data = NULL;
6541 static void lfsck_layout_master_quit(const struct lu_env *env,
6542 struct lfsck_component *com)
6544 struct lfsck_assistant_data *lad = com->lc_data;
6545 struct lfsck_instance *lfsck = com->lc_lfsck;
6546 struct lfsck_tgt_descs *ltds;
6547 struct lfsck_tgt_desc *ltd;
6548 struct lfsck_tgt_desc *next;
6550 LASSERT(lad != NULL);
6552 lfsck_quit_generic(env, com);
6554 LASSERT(thread_is_init(&lad->lad_thread) ||
6555 thread_is_stopped(&lad->lad_thread));
6556 LASSERT(list_empty(&lad->lad_req_list));
6558 ltds = &lfsck->li_ost_descs;
6559 spin_lock(<ds->ltd_lock);
6560 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase1_list,
6561 ltd_layout_phase_list) {
6562 list_del_init(<d->ltd_layout_phase_list);
6564 list_for_each_entry_safe(ltd, next, &lad->lad_ost_phase2_list,
6565 ltd_layout_phase_list) {
6566 list_del_init(<d->ltd_layout_phase_list);
6568 spin_unlock(<ds->ltd_lock);
6570 ltds = &lfsck->li_mdt_descs;
6571 spin_lock(<ds->ltd_lock);
6572 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
6573 ltd_layout_phase_list) {
6574 list_del_init(<d->ltd_layout_phase_list);
6576 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
6577 ltd_layout_phase_list) {
6578 list_del_init(<d->ltd_layout_phase_list);
6580 spin_unlock(<ds->ltd_lock);
6583 static void lfsck_layout_slave_quit(const struct lu_env *env,
6584 struct lfsck_component *com)
6586 struct lfsck_layout_slave_data *llsd = com->lc_data;
6587 struct lfsck_layout_seq *lls;
6588 struct lfsck_layout_seq *next;
6589 struct lfsck_layout_slave_target *llst;
6591 LASSERT(llsd != NULL);
6593 down_write(&com->lc_sem);
6594 list_for_each_entry_safe(lls, next, &llsd->llsd_seq_list,
6596 list_del_init(&lls->lls_list);
6597 lfsck_object_put(env, lls->lls_lastid_obj);
6600 up_write(&com->lc_sem);
6602 spin_lock(&llsd->llsd_lock);
6603 while (!list_empty(&llsd->llsd_master_list)) {
6604 llst = list_first_entry(&llsd->llsd_master_list,
6605 struct lfsck_layout_slave_target,
6607 list_del_init(&llst->llst_list);
6608 spin_unlock(&llsd->llsd_lock);
6609 lfsck_layout_llst_put(llst);
6610 spin_lock(&llsd->llsd_lock);
6612 spin_unlock(&llsd->llsd_lock);
6614 lfsck_rbtree_cleanup(env, com);
6617 static int lfsck_layout_master_in_notify(const struct lu_env *env,
6618 struct lfsck_component *com,
6619 struct lfsck_request *lr)
6621 struct lfsck_instance *lfsck = com->lc_lfsck;
6622 struct lfsck_layout *lo = com->lc_file_ram;
6623 struct lfsck_assistant_data *lad = com->lc_data;
6624 struct lfsck_tgt_descs *ltds;
6625 struct lfsck_tgt_desc *ltd;
6629 if (lr->lr_event == LE_PAIRS_VERIFY) {
6632 rc = lfsck_layout_master_check_pairs(env, com, &lr->lr_fid,
6639 CDEBUG(D_LFSCK, "%s: layout LFSCK master handles notify %u "
6640 "from %s %x, status %d, flags %x, flags2 %x\n",
6641 lfsck_lfsck2name(lfsck), lr->lr_event,
6642 (lr->lr_flags & LEF_FROM_OST) ? "OST" : "MDT",
6643 lr->lr_index, lr->lr_status, lr->lr_flags, lr->lr_flags2);
6645 if (lr->lr_event != LE_PHASE1_DONE &&
6646 lr->lr_event != LE_PHASE2_DONE &&
6647 lr->lr_event != LE_PEER_EXIT)
6650 if (lr->lr_flags & LEF_FROM_OST)
6651 ltds = &lfsck->li_ost_descs;
6653 ltds = &lfsck->li_mdt_descs;
6654 spin_lock(<ds->ltd_lock);
6655 ltd = lfsck_ltd2tgt(ltds, lr->lr_index);
6657 spin_unlock(<ds->ltd_lock);
6662 list_del_init(<d->ltd_layout_phase_list);
6663 switch (lr->lr_event) {
6664 case LE_PHASE1_DONE:
6665 if (lr->lr_status <= 0 || lr->lr_flags2 & LF_INCOMPLETE) {
6666 if (lr->lr_flags2 & LF_INCOMPLETE) {
6667 if (lr->lr_flags & LEF_FROM_OST)
6668 lfsck_lad_set_bitmap(env, com,
6671 lo->ll_flags |= LF_INCOMPLETE;
6673 ltd->ltd_layout_done = 1;
6674 list_del_init(<d->ltd_layout_list);
6679 if (lr->lr_flags & LEF_FROM_OST) {
6680 if (list_empty(<d->ltd_layout_list))
6681 list_add_tail(<d->ltd_layout_list,
6682 &lad->lad_ost_list);
6683 list_add_tail(<d->ltd_layout_phase_list,
6684 &lad->lad_ost_phase2_list);
6686 if (list_empty(<d->ltd_layout_list))
6687 list_add_tail(<d->ltd_layout_list,
6688 &lad->lad_mdt_list);
6689 list_add_tail(<d->ltd_layout_phase_list,
6690 &lad->lad_mdt_phase2_list);
6693 case LE_PHASE2_DONE:
6694 ltd->ltd_layout_done = 1;
6695 if (!list_empty(<d->ltd_layout_list))
6696 list_del_init(<d->ltd_layout_list);
6698 if (lr->lr_flags2 & LF_INCOMPLETE) {
6699 lfsck_lad_set_bitmap(env, com, ltd->ltd_index);
6706 ltd->ltd_layout_done = 1;
6707 list_del_init(<d->ltd_layout_list);
6708 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) &&
6709 !(lr->lr_flags & LEF_FROM_OST))
6710 lo->ll_flags |= LF_INCOMPLETE;
6715 spin_unlock(<ds->ltd_lock);
6717 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
6718 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6720 memset(stop, 0, sizeof(*stop));
6721 stop->ls_status = lr->lr_status;
6722 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6723 lfsck_stop(env, lfsck->li_bottom, stop);
6724 } else if (lfsck_phase2_next_ready(lad)) {
6725 wake_up(&lad->lad_thread.t_ctl_waitq);
6731 static int lfsck_layout_slave_in_notify_local(const struct lu_env *env,
6732 struct lfsck_component *com,
6733 struct lfsck_req_local *lrl,
6738 switch (lrl->lrl_event) {
6739 case LEL_FID_ACCESSED:
6740 lfsck_rbtree_update_bitmap(env, com, &lrl->lrl_fid, true);
6742 case LEL_PAIRS_VERIFY_LOCAL: {
6745 lrl->lrl_status = LPVS_INIT;
6746 /* Firstly, if the MDT-object which is claimed via OST-object
6747 * local stored PFID xattr recognizes the OST-object, then it
6748 * must be that the client given PFID is wrong. */
6749 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6750 &lrl->lrl_ff_local.ff_parent,
6751 lrl->lrl_ff_local.ff_layout.ol_comp_id);
6755 lrl->lrl_status = LPVS_INCONSISTENT;
6756 /* The OST-object local stored PFID xattr is stale. We need to
6757 * check whether the MDT-object that is claimed via the client
6758 * given PFID information recognizes the OST-object or not. If
6759 * matches, then need to update the OST-object's PFID xattr. */
6760 rc = lfsck_layout_slave_check_pairs(env, com, &lrl->lrl_fid,
6761 &lrl->lrl_ff_client.ff_parent,
6762 lrl->lrl_ff_client.ff_layout.ol_comp_id);
6764 * We are not sure whether the client given PFID information
6765 * is correct or not, do nothing to avoid improper fixing.
6768 * The client given PFID information is also invalid, we can
6769 * NOT fix the OST-object inconsistency.
6772 lrl->lrl_status = LPVS_INCONSISTENT_TOFIX;
6773 rc = lfsck_layout_slave_repair_pfid(env, com, lrl);
6782 RETURN(-EOPNOTSUPP);
6785 static int lfsck_layout_slave_in_notify(const struct lu_env *env,
6786 struct lfsck_component *com,
6787 struct lfsck_request *lr)
6789 struct lfsck_instance *lfsck = com->lc_lfsck;
6790 struct lfsck_layout_slave_data *llsd = com->lc_data;
6791 struct lfsck_layout_slave_target *llst;
6795 switch (lr->lr_event) {
6796 case LE_CONDITIONAL_DESTROY:
6797 rc = lfsck_layout_slave_conditional_destroy(env, com, lr);
6799 case LE_PHASE1_DONE: {
6800 if (lr->lr_flags2 & LF_INCOMPLETE) {
6801 struct lfsck_layout *lo = com->lc_file_ram;
6803 lo->ll_flags |= LF_INCOMPLETE;
6804 llst = lfsck_layout_llst_find_and_del(llsd,
6808 lfsck_layout_llst_put(llst);
6809 wake_up(&lfsck->li_thread.t_ctl_waitq);
6815 case LE_PHASE2_DONE:
6817 CDEBUG(D_LFSCK, "%s: layout LFSCK slave handle notify %u "
6818 "from MDT %x, status %d\n", lfsck_lfsck2name(lfsck),
6819 lr->lr_event, lr->lr_index, lr->lr_status);
6825 llst = lfsck_layout_llst_find_and_del(llsd, lr->lr_index, true);
6829 lfsck_layout_llst_put(llst);
6830 if (list_empty(&llsd->llsd_master_list))
6831 wake_up(&lfsck->li_thread.t_ctl_waitq);
6833 if (lr->lr_event == LE_PEER_EXIT &&
6834 (lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT ||
6835 (list_empty(&llsd->llsd_master_list) &&
6836 (lr->lr_status == LS_STOPPED ||
6837 lr->lr_status == LS_CO_STOPPED)))) {
6838 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
6840 memset(stop, 0, sizeof(*stop));
6841 stop->ls_status = lr->lr_status;
6842 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
6843 lfsck_stop(env, lfsck->li_bottom, stop);
6849 static void lfsck_layout_repaired(struct lfsck_layout *lo, __u64 *count)
6853 for (i = 0; i < LLIT_MAX; i++)
6854 *count += lo->ll_objs_repaired[i];
6857 static int lfsck_layout_query_all(const struct lu_env *env,
6858 struct lfsck_component *com,
6859 __u32 *mdts_count, __u32 *osts_count,
6862 struct lfsck_layout *lo = com->lc_file_ram;
6863 struct lfsck_tgt_descs *ltds;
6864 struct lfsck_tgt_desc *ltd;
6869 rc = lfsck_query_all(env, com);
6873 ltds = &com->lc_lfsck->li_mdt_descs;
6874 down_read(<ds->ltd_rw_sem);
6875 for_each_set_bit(idx, ltds->ltd_tgts_bitmap, ltds->ltd_tgts_mask_len) {
6876 ltd = lfsck_ltd2tgt(ltds, idx);
6877 LASSERT(ltd != NULL);
6879 mdts_count[ltd->ltd_layout_status]++;
6880 *repaired += ltd->ltd_layout_repaired;
6882 up_read(<ds->ltd_rw_sem);
6884 ltds = &com->lc_lfsck->li_ost_descs;
6885 down_read(<ds->ltd_rw_sem);
6886 for_each_set_bit(idx, ltds->ltd_tgts_bitmap, ltds->ltd_tgts_mask_len) {
6887 ltd = lfsck_ltd2tgt(ltds, idx);
6888 LASSERT(ltd != NULL);
6890 osts_count[ltd->ltd_layout_status]++;
6891 *repaired += ltd->ltd_layout_repaired;
6893 up_read(<ds->ltd_rw_sem);
6895 down_read(&com->lc_sem);
6896 mdts_count[lo->ll_status]++;
6897 lfsck_layout_repaired(lo, repaired);
6898 up_read(&com->lc_sem);
6903 static int lfsck_layout_query(const struct lu_env *env,
6904 struct lfsck_component *com,
6905 struct lfsck_request *req,
6906 struct lfsck_reply *rep,
6907 struct lfsck_query *que, int idx)
6909 struct lfsck_layout *lo = com->lc_file_ram;
6913 LASSERT(com->lc_lfsck->li_master);
6915 rc = lfsck_layout_query_all(env, com,
6916 que->lu_mdts_count[idx],
6917 que->lu_osts_count[idx],
6918 &que->lu_repaired[idx]);
6920 down_read(&com->lc_sem);
6921 rep->lr_status = lo->ll_status;
6922 if (req->lr_flags & LEF_QUERY_ALL)
6923 lfsck_layout_repaired(lo, &rep->lr_repaired);
6924 up_read(&com->lc_sem);
6930 /* with lfsck::li_lock held */
6931 static int lfsck_layout_slave_join(const struct lu_env *env,
6932 struct lfsck_component *com,
6933 struct lfsck_start_param *lsp)
6935 struct lfsck_instance *lfsck = com->lc_lfsck;
6936 struct lfsck_layout_slave_data *llsd = com->lc_data;
6937 struct lfsck_layout_slave_target *llst;
6938 struct lfsck_start *start = lsp->lsp_start;
6942 if (start == NULL || !(start->ls_flags & LPF_OST_ORPHAN))
6945 if (!lsp->lsp_index_valid)
6948 /* If someone is running the LFSCK without orphan handling,
6949 * it will not maintain the object accessing rbtree. So we
6950 * cannot join it for orphan handling. */
6951 if (!llsd->llsd_rbtree_valid)
6954 spin_unlock(&lfsck->li_lock);
6955 rc = lfsck_layout_llst_add(llsd, lsp->lsp_index);
6956 spin_lock(&lfsck->li_lock);
6957 if (rc == 0 && !thread_is_running(&lfsck->li_thread)) {
6958 spin_unlock(&lfsck->li_lock);
6959 llst = lfsck_layout_llst_find_and_del(llsd, lsp->lsp_index,
6962 lfsck_layout_llst_put(llst);
6963 spin_lock(&lfsck->li_lock);
6970 static const struct lfsck_operations lfsck_layout_master_ops = {
6971 .lfsck_reset = lfsck_layout_reset,
6972 .lfsck_fail = lfsck_layout_fail,
6973 .lfsck_checkpoint = lfsck_layout_master_checkpoint,
6974 .lfsck_prep = lfsck_layout_master_prep,
6975 .lfsck_exec_oit = lfsck_layout_master_exec_oit,
6976 .lfsck_exec_dir = lfsck_layout_exec_dir,
6977 .lfsck_post = lfsck_layout_master_post,
6978 .lfsck_dump = lfsck_layout_dump,
6979 .lfsck_double_scan = lfsck_layout_master_double_scan,
6980 .lfsck_data_release = lfsck_layout_master_data_release,
6981 .lfsck_quit = lfsck_layout_master_quit,
6982 .lfsck_in_notify = lfsck_layout_master_in_notify,
6983 .lfsck_query = lfsck_layout_query,
6986 static const struct lfsck_operations lfsck_layout_slave_ops = {
6987 .lfsck_reset = lfsck_layout_reset,
6988 .lfsck_fail = lfsck_layout_fail,
6989 .lfsck_checkpoint = lfsck_layout_slave_checkpoint,
6990 .lfsck_prep = lfsck_layout_slave_prep,
6991 .lfsck_exec_oit = lfsck_layout_slave_exec_oit,
6992 .lfsck_exec_dir = lfsck_layout_exec_dir,
6993 .lfsck_post = lfsck_layout_slave_post,
6994 .lfsck_dump = lfsck_layout_dump,
6995 .lfsck_double_scan = lfsck_layout_slave_double_scan,
6996 .lfsck_data_release = lfsck_layout_slave_data_release,
6997 .lfsck_quit = lfsck_layout_slave_quit,
6998 .lfsck_in_notify_local = lfsck_layout_slave_in_notify_local,
6999 .lfsck_in_notify = lfsck_layout_slave_in_notify,
7000 .lfsck_query = lfsck_layout_query,
7001 .lfsck_join = lfsck_layout_slave_join,
7004 static void lfsck_layout_assistant_fill_pos(const struct lu_env *env,
7005 struct lfsck_component *com,
7006 struct lfsck_position *pos)
7008 struct lfsck_assistant_data *lad = com->lc_data;
7009 struct lfsck_layout_req *llr;
7011 if (((struct lfsck_layout *)(com->lc_file_ram))->ll_status !=
7015 if (list_empty(&lad->lad_req_list))
7018 llr = list_first_entry(&lad->lad_req_list,
7019 struct lfsck_layout_req,
7021 pos->lp_oit_cookie = llr->llr_lar.lar_parent->lso_oit_cookie - 1;
7024 const struct lfsck_assistant_operations lfsck_layout_assistant_ops = {
7025 .la_handler_p1 = lfsck_layout_assistant_handler_p1,
7026 .la_handler_p2 = lfsck_layout_assistant_handler_p2,
7027 .la_fill_pos = lfsck_layout_assistant_fill_pos,
7028 .la_double_scan_result = lfsck_layout_double_scan_result,
7029 .la_req_fini = lfsck_layout_assistant_req_fini,
7030 .la_sync_failures = lfsck_layout_assistant_sync_failures,
7033 int lfsck_layout_setup(const struct lu_env *env, struct lfsck_instance *lfsck)
7035 struct lfsck_component *com;
7036 struct lfsck_layout *lo;
7037 struct dt_object *root = NULL;
7038 struct dt_object *obj;
7047 INIT_LIST_HEAD(&com->lc_link);
7048 INIT_LIST_HEAD(&com->lc_link_dir);
7049 init_rwsem(&com->lc_sem);
7050 atomic_set(&com->lc_ref, 1);
7051 com->lc_lfsck = lfsck;
7052 com->lc_type = LFSCK_TYPE_LAYOUT;
7053 if (lfsck->li_master) {
7054 com->lc_ops = &lfsck_layout_master_ops;
7055 com->lc_data = lfsck_assistant_data_init(
7056 &lfsck_layout_assistant_ops,
7058 if (com->lc_data == NULL)
7059 GOTO(out, rc = -ENOMEM);
7061 for (i = 0; i < LFSCK_STF_COUNT; i++)
7062 mutex_init(&com->lc_sub_trace_objs[i].lsto_mutex);
7064 struct lfsck_layout_slave_data *llsd;
7066 com->lc_ops = &lfsck_layout_slave_ops;
7067 OBD_ALLOC_PTR(llsd);
7069 GOTO(out, rc = -ENOMEM);
7071 INIT_LIST_HEAD(&llsd->llsd_seq_list);
7072 INIT_LIST_HEAD(&llsd->llsd_master_list);
7073 spin_lock_init(&llsd->llsd_lock);
7074 llsd->llsd_rb_root = RB_ROOT;
7075 init_rwsem(&llsd->llsd_rb_rwsem);
7076 com->lc_data = llsd;
7078 com->lc_file_size = sizeof(*lo);
7079 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
7080 if (com->lc_file_ram == NULL)
7081 GOTO(out, rc = -ENOMEM);
7083 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
7084 if (com->lc_file_disk == NULL)
7085 GOTO(out, rc = -ENOMEM);
7087 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
7089 GOTO(out, rc = PTR_ERR(root));
7091 if (unlikely(!dt_try_as_dir(env, root, true)))
7092 GOTO(out, rc = -ENOTDIR);
7094 obj = local_file_find_or_create(env, lfsck->li_los, root,
7096 S_IFREG | S_IRUGO | S_IWUSR);
7098 GOTO(out, rc = PTR_ERR(obj));
7101 rc = lfsck_layout_load(env, com);
7103 rc = lfsck_layout_reset(env, com, true);
7104 } else if (rc == -ENOENT) {
7105 rc = lfsck_layout_init(env, com);
7106 } else if (lfsck->li_master) {
7107 rc = lfsck_load_sub_trace_files(env, com,
7108 &dt_lfsck_layout_dangling_features,
7109 LFSCK_LAYOUT, false);
7111 rc = lfsck_layout_reset(env, com, true);
7117 lo = com->lc_file_ram;
7118 switch (lo->ll_status) {
7124 spin_lock(&lfsck->li_lock);
7125 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
7126 spin_unlock(&lfsck->li_lock);
7129 CERROR("%s: unknown lfsck_layout status %d\n",
7130 lfsck_lfsck2name(lfsck), lo->ll_status);
7132 case LS_SCANNING_PHASE1:
7133 case LS_SCANNING_PHASE2:
7134 /* No need to store the status to disk right now.
7135 * If the system crashed before the status stored,
7136 * it will be loaded back when next time. */
7137 lo->ll_status = LS_CRASHED;
7138 if (!lfsck->li_master)
7139 lo->ll_flags |= LF_INCOMPLETE;
7146 spin_lock(&lfsck->li_lock);
7147 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
7148 spin_unlock(&lfsck->li_lock);
7152 if (lo->ll_flags & LF_CRASHED_LASTID) {
7153 LASSERT(lfsck->li_out_notify != NULL);
7155 lfsck->li_out_notify(env, lfsck->li_out_notify_data,
7156 LE_LASTID_REBUILDING);
7162 if (root != NULL && !IS_ERR(root))
7163 lfsck_object_put(env, root);
7166 lfsck_component_cleanup(env, com);
7167 CERROR("%s: fail to init layout LFSCK component: rc = %d\n",
7168 lfsck_lfsck2name(lfsck), rc);
7174 struct lfsck_orphan_it {
7175 struct lfsck_component *loi_com;
7176 struct lfsck_rbtree_node *loi_lrn;
7177 struct lfsck_layout_slave_target *loi_llst;
7178 struct lu_fid loi_key;
7179 struct lu_orphan_rec_v3 loi_rec;
7181 unsigned int loi_over:1;
7184 static int lfsck_fid_match_idx(const struct lu_env *env,
7185 struct lfsck_instance *lfsck,
7186 const struct lu_fid *fid, int idx)
7188 struct seq_server_site *ss;
7189 struct lu_server_fld *sf;
7190 struct lu_seq_range *range = &lfsck_env_info(env)->lti_range;
7193 /* All abnormal cases will be returned to MDT0. */
7194 if (!fid_is_norm(fid)) {
7201 ss = lfsck_dev_site(lfsck);
7202 if (unlikely(ss == NULL))
7205 sf = ss->ss_server_fld;
7206 LASSERT(sf != NULL);
7208 fld_range_set_any(range);
7209 rc = fld_server_lookup(env, sf, fid_seq(fid), range);
7213 if (!fld_range_is_mdt(range))
7216 if (range->lsr_index == idx)
7222 static void lfsck_layout_destroy_orphan(const struct lu_env *env,
7223 struct lfsck_instance *lfsck,
7224 struct dt_object *obj)
7226 struct dt_device *dev = lfsck_obj2dev(obj);
7227 struct thandle *handle;
7231 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
7234 handle = lfsck_trans_create(env, dev, lfsck);
7238 rc = dt_declare_ref_del(env, obj, handle);
7242 rc = dt_declare_destroy(env, obj, handle);
7246 rc = dt_trans_start_local(env, dev, handle);
7250 dt_write_lock(env, obj, 0);
7251 rc = dt_ref_del(env, obj, handle);
7253 rc = dt_destroy(env, obj, handle);
7254 dt_write_unlock(env, obj);
7259 dt_trans_stop(env, dev, handle);
7262 CDEBUG(D_LFSCK, "destroy orphan OST-object "DFID": rc = %d\n",
7263 PFID(lfsck_dto2fid(obj)), rc);
7268 static int lfsck_orphan_index_lookup(const struct lu_env *env,
7269 struct dt_object *dt,
7271 const struct dt_key *key)
7276 static int lfsck_orphan_index_declare_insert(const struct lu_env *env,
7277 struct dt_object *dt,
7278 const struct dt_rec *rec,
7279 const struct dt_key *key,
7280 struct thandle *handle)
7285 static int lfsck_orphan_index_insert(const struct lu_env *env,
7286 struct dt_object *dt,
7287 const struct dt_rec *rec,
7288 const struct dt_key *key,
7289 struct thandle *handle)
7294 static int lfsck_orphan_index_declare_delete(const struct lu_env *env,
7295 struct dt_object *dt,
7296 const struct dt_key *key,
7297 struct thandle *handle)
7302 static int lfsck_orphan_index_delete(const struct lu_env *env,
7303 struct dt_object *dt,
7304 const struct dt_key *key,
7305 struct thandle *handle)
7310 static struct dt_it *lfsck_orphan_it_init(const struct lu_env *env,
7311 struct dt_object *dt,
7314 struct dt_device *dev = lu2dt_dev(dt->do_lu.lo_dev);
7315 struct lfsck_instance *lfsck;
7316 struct lfsck_component *com = NULL;
7317 struct lfsck_layout_slave_data *llsd;
7318 struct lfsck_orphan_it *it = NULL;
7319 struct lfsck_layout *lo;
7323 lfsck = lfsck_instance_find(dev, true, false);
7324 if (unlikely(lfsck == NULL))
7325 RETURN(ERR_PTR(-ENXIO));
7327 com = lfsck_component_find(lfsck, LFSCK_TYPE_LAYOUT);
7328 if (unlikely(com == NULL))
7329 GOTO(out, rc = -ENOENT);
7331 lo = com->lc_file_ram;
7332 if (lo->ll_flags & LF_INCOMPLETE)
7333 GOTO(out, rc = -ESRCH);
7335 llsd = com->lc_data;
7336 if (!llsd->llsd_rbtree_valid)
7337 GOTO(out, rc = -ESRCH);
7341 GOTO(out, rc = -ENOMEM);
7343 it->loi_llst = lfsck_layout_llst_find_and_del(llsd, attr, false);
7344 if (it->loi_llst == NULL)
7345 GOTO(out, rc = -ENXIO);
7347 if (dev->dd_record_fid_accessed) {
7348 /* The first iteration against the rbtree, scan the whole rbtree
7349 * to remove the nodes which do NOT need to be handled. */
7350 down_write(&llsd->llsd_rb_rwsem);
7351 if (dev->dd_record_fid_accessed) {
7352 struct rb_node *node;
7353 struct rb_node *next;
7354 struct lfsck_rbtree_node *lrn;
7356 /* No need to record the fid accessing anymore. */
7357 dev->dd_record_fid_accessed = 0;
7359 node = rb_first(&llsd->llsd_rb_root);
7360 while (node != NULL) {
7361 next = rb_next(node);
7362 lrn = rb_entry(node, struct lfsck_rbtree_node,
7364 if (atomic_read(&lrn->lrn_known_count) <=
7365 atomic_read(&lrn->lrn_accessed_count)) {
7366 rb_erase(node, &llsd->llsd_rb_root);
7367 lfsck_rbtree_free(lrn);
7372 up_write(&llsd->llsd_rb_rwsem);
7375 /* read lock the rbtree when init, and unlock when fini */
7376 down_read(&llsd->llsd_rb_rwsem);
7384 lfsck_component_put(env, com);
7386 CDEBUG(D_LFSCK, "%s: init the orphan iteration: rc = %d\n",
7387 lfsck_lfsck2name(lfsck), rc);
7389 lfsck_instance_put(env, lfsck);
7394 it = (struct lfsck_orphan_it *)ERR_PTR(rc);
7397 return (struct dt_it *)it;
7400 static void lfsck_orphan_it_fini(const struct lu_env *env,
7403 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7404 struct lfsck_component *com = it->loi_com;
7405 struct lfsck_layout_slave_data *llsd;
7406 struct lfsck_layout_slave_target *llst;
7409 CDEBUG(D_LFSCK, "%s: fini the orphan iteration\n",
7410 lfsck_lfsck2name(com->lc_lfsck));
7412 llsd = com->lc_data;
7413 up_read(&llsd->llsd_rb_rwsem);
7414 llst = it->loi_llst;
7415 LASSERT(llst != NULL);
7417 /* Save the key and hash for iterate next. */
7418 llst->llst_fid = it->loi_key;
7419 llst->llst_hash = it->loi_hash;
7420 lfsck_layout_llst_put(llst);
7421 lfsck_component_put(env, com);
7427 * \retval +1: the iteration finished
7428 * \retval 0: on success, not finished
7429 * \retval -ve: on error
7431 static int lfsck_orphan_it_next(const struct lu_env *env,
7434 struct lfsck_thread_info *info = lfsck_env_info(env);
7435 struct filter_fid *ff = &info->lti_ff;
7436 struct lu_attr *la = &info->lti_la;
7437 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7438 struct lu_fid *key = &it->loi_key;
7439 struct lu_orphan_rec_v3 *rec = &it->loi_rec;
7440 struct ost_layout *ol = &rec->lor_layout;
7441 struct lfsck_component *com = it->loi_com;
7442 struct lfsck_instance *lfsck = com->lc_lfsck;
7443 struct lfsck_layout_slave_data *llsd = com->lc_data;
7444 struct dt_object *obj;
7445 struct lfsck_rbtree_node *lrn;
7449 __u32 idx = it->loi_llst->llst_index;
7459 lrn = lfsck_rbtree_search(llsd, key, &exact);
7467 key->f_seq = lrn->lrn_seq;
7468 key->f_oid = lrn->lrn_first_oid;
7473 if (unlikely(key->f_oid == 0)) {
7480 lrn->lrn_first_oid + LFSCK_RBTREE_BITMAP_WIDTH) {
7486 if (unlikely(atomic_read(&lrn->lrn_known_count) <=
7487 atomic_read(&lrn->lrn_accessed_count))) {
7488 struct rb_node *next = rb_next(&lrn->lrn_node);
7490 while (next != NULL) {
7491 lrn = rb_entry(next, struct lfsck_rbtree_node,
7493 if (atomic_read(&lrn->lrn_known_count) >
7494 atomic_read(&lrn->lrn_accessed_count))
7496 next = rb_next(next);
7505 key->f_seq = lrn->lrn_seq;
7506 key->f_oid = lrn->lrn_first_oid;
7510 pos = key->f_oid - lrn->lrn_first_oid;
7513 pos = find_next_bit(lrn->lrn_known_bitmap,
7514 LFSCK_RBTREE_BITMAP_WIDTH, pos);
7515 if (pos >= LFSCK_RBTREE_BITMAP_WIDTH) {
7516 key->f_oid = lrn->lrn_first_oid + pos;
7517 if (unlikely(key->f_oid < lrn->lrn_first_oid)) {
7525 if (test_bit(pos, lrn->lrn_accessed_bitmap)) {
7530 key->f_oid = lrn->lrn_first_oid + pos;
7531 obj = lfsck_object_find_bottom(env, lfsck, key);
7534 if (rc == -ENOENT) {
7541 dt_read_lock(env, obj, 0);
7542 if (dt_object_exists(obj) == 0 ||
7543 lfsck_is_dead_obj(obj)) {
7544 dt_read_unlock(env, obj);
7545 lfsck_object_put(env, obj);
7550 rc = dt_attr_get(env, obj, la);
7554 rc = dt_xattr_get(env, obj, lfsck_buf_get(env, ff, sizeof(*ff)),
7556 if (rc == -ENODATA) {
7557 /* For the pre-created OST-object, update the bitmap to avoid
7558 * others LFSCK (second phase) iteration to touch it again. */
7559 if (la->la_ctime == 0) {
7560 if (!test_and_set_bit(pos, lrn->lrn_accessed_bitmap))
7561 atomic_inc(&lrn->lrn_accessed_count);
7563 /* For the race between repairing dangling referenced
7564 * MDT-object and unlink the file, it may left orphan
7565 * OST-object there. Destroy it now! */
7566 if (unlikely(!(la->la_mode & S_ISUID))) {
7567 dt_read_unlock(env, obj);
7568 lfsck_layout_destroy_orphan(env, lfsck, obj);
7569 lfsck_object_put(env, obj);
7573 } else if (idx == 0) {
7574 /* If the orphan OST-object has no parent information,
7575 * regard it as referenced by the MDT-object on MDT0. */
7576 fid_zero(&rec->lor_rec.lor_fid);
7577 rec->lor_rec.lor_uid = la->la_uid;
7578 rec->lor_rec.lor_gid = la->la_gid;
7579 memset(ol, 0, sizeof(*ol));
7580 rec->lor_layout_version = 0;
7586 dt_read_unlock(env, obj);
7587 lfsck_object_put(env, obj);
7592 if (rc < sizeof(struct lu_fid))
7593 GOTO(out, rc = (rc < 0 ? rc : -EINVAL));
7595 fid_le_to_cpu(&rec->lor_rec.lor_fid, &ff->ff_parent);
7596 /* Currently, the filter_fid::ff_parent::f_ver is not the real parent
7597 * MDT-object's FID::f_ver, instead it is the OST-object index in its
7598 * parent MDT-object's layout EA. */
7599 save = rec->lor_rec.lor_fid.f_stripe_idx;
7600 rec->lor_rec.lor_fid.f_ver = 0;
7601 rc = lfsck_fid_match_idx(env, lfsck, &rec->lor_rec.lor_fid, idx);
7602 /* If the orphan OST-object does not claim the MDT, then next.
7604 * If we do not know whether it matches or not, then return it
7605 * to the MDT for further check. */
7607 dt_read_unlock(env, obj);
7608 lfsck_object_put(env, obj);
7613 rec->lor_rec.lor_fid.f_stripe_idx = save;
7614 rec->lor_rec.lor_uid = la->la_uid;
7615 rec->lor_rec.lor_gid = la->la_gid;
7616 ost_layout_le_to_cpu(ol, &ff->ff_layout);
7617 rec->lor_layout_version =
7618 le32_to_cpu(ff->ff_layout_version & ~LU_LAYOUT_RESYNC);
7619 rec->lor_range = le32_to_cpu(ff->ff_range);
7621 CDEBUG(D_LFSCK, "%s: return orphan "DFID", PFID "DFID", owner %u:%u, "
7622 "stripe size %u, stripe count %u, COMP id %u, COMP start %llu, "
7623 "COMP end %llu, layout version %u, range %u\n",
7624 lfsck_lfsck2name(com->lc_lfsck), PFID(key),
7625 PFID(&rec->lor_rec.lor_fid), rec->lor_rec.lor_uid,
7626 rec->lor_rec.lor_gid, ol->ol_stripe_size, ol->ol_stripe_count,
7627 ol->ol_comp_id, ol->ol_comp_start, ol->ol_comp_end,
7628 rec->lor_layout_version, rec->lor_range);
7633 dt_read_unlock(env, obj);
7634 lfsck_object_put(env, obj);
7642 * \retval +1: locate to the exactly position
7643 * \retval 0: cannot locate to the exactly position,
7644 * call next() to move to a valid position.
7645 * \retval -ve: on error
7647 static int lfsck_orphan_it_get(const struct lu_env *env,
7649 const struct dt_key *key)
7651 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7654 it->loi_key = *(struct lu_fid *)key;
7655 rc = lfsck_orphan_it_next(env, di);
7665 static void lfsck_orphan_it_put(const struct lu_env *env,
7670 static struct dt_key *lfsck_orphan_it_key(const struct lu_env *env,
7671 const struct dt_it *di)
7673 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7675 return (struct dt_key *)&it->loi_key;
7678 static int lfsck_orphan_it_key_size(const struct lu_env *env,
7679 const struct dt_it *di)
7681 return sizeof(struct lu_fid);
7684 static int lfsck_orphan_it_rec(const struct lu_env *env,
7685 const struct dt_it *di,
7689 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7691 *(struct lu_orphan_rec_v3 *)rec = it->loi_rec;
7696 static __u64 lfsck_orphan_it_store(const struct lu_env *env,
7697 const struct dt_it *di)
7699 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7701 return it->loi_hash;
7705 * \retval +1: locate to the exactly position
7706 * \retval 0: cannot locate to the exactly position,
7707 * call next() to move to a valid position.
7708 * \retval -ve: on error
7710 static int lfsck_orphan_it_load(const struct lu_env *env,
7711 const struct dt_it *di,
7714 struct lfsck_orphan_it *it = (struct lfsck_orphan_it *)di;
7715 struct lfsck_layout_slave_target *llst = it->loi_llst;
7718 LASSERT(llst != NULL);
7720 if (hash != llst->llst_hash) {
7721 CDEBUG(D_LFSCK, "%s: the given hash %llu for orphan "
7722 "iteration does not match the one when fini "
7723 "%llu, to be reset.\n",
7724 lfsck_lfsck2name(it->loi_com->lc_lfsck), hash,
7726 fid_zero(&llst->llst_fid);
7727 llst->llst_hash = 0;
7730 it->loi_key = llst->llst_fid;
7731 it->loi_hash = llst->llst_hash;
7732 rc = lfsck_orphan_it_next(env, (struct dt_it *)di);
7742 static int lfsck_orphan_it_key_rec(const struct lu_env *env,
7743 const struct dt_it *di,
7749 static const struct dt_index_operations lfsck_orphan_index_ops = {
7750 .dio_lookup = lfsck_orphan_index_lookup,
7751 .dio_declare_insert = lfsck_orphan_index_declare_insert,
7752 .dio_insert = lfsck_orphan_index_insert,
7753 .dio_declare_delete = lfsck_orphan_index_declare_delete,
7754 .dio_delete = lfsck_orphan_index_delete,
7756 .init = lfsck_orphan_it_init,
7757 .fini = lfsck_orphan_it_fini,
7758 .get = lfsck_orphan_it_get,
7759 .put = lfsck_orphan_it_put,
7760 .next = lfsck_orphan_it_next,
7761 .key = lfsck_orphan_it_key,
7762 .key_size = lfsck_orphan_it_key_size,
7763 .rec = lfsck_orphan_it_rec,
7764 .store = lfsck_orphan_it_store,
7765 .load = lfsck_orphan_it_load,
7766 .key_rec = lfsck_orphan_it_key_rec,