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, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2017, Intel Corporation.
26 * lustre/obdclass/scrub.c
28 * The OI scrub is used for checking and (re)building Object Index files
29 * that are usually backend special. Here are some general scrub related
30 * functions that can be shared by different backends for OI scrub.
32 * Author: Fan Yong <fan.yong@intel.com>
35 #define DEBUG_SUBSYSTEM S_LFSCK
37 #include <linux/kthread.h>
38 #include <lustre_scrub.h>
39 #include <lustre_lib.h>
40 #include <lustre_fid.h>
42 static inline struct dt_device *scrub_obj2dev(struct dt_object *obj)
44 return container_of_safe(obj->do_lu.lo_dev, struct dt_device,
48 static void scrub_file_to_cpu(struct scrub_file *des, struct scrub_file *src)
50 uuid_copy(&des->sf_uuid, &src->sf_uuid);
51 des->sf_flags = le64_to_cpu(src->sf_flags);
52 des->sf_magic = le32_to_cpu(src->sf_magic);
53 des->sf_status = le16_to_cpu(src->sf_status);
54 des->sf_param = le16_to_cpu(src->sf_param);
55 des->sf_time_last_complete =
56 le64_to_cpu(src->sf_time_last_complete);
57 des->sf_time_latest_start =
58 le64_to_cpu(src->sf_time_latest_start);
59 des->sf_time_last_checkpoint =
60 le64_to_cpu(src->sf_time_last_checkpoint);
61 des->sf_pos_latest_start =
62 le64_to_cpu(src->sf_pos_latest_start);
63 des->sf_pos_last_checkpoint =
64 le64_to_cpu(src->sf_pos_last_checkpoint);
65 des->sf_pos_first_inconsistent =
66 le64_to_cpu(src->sf_pos_first_inconsistent);
67 des->sf_items_checked =
68 le64_to_cpu(src->sf_items_checked);
69 des->sf_items_updated =
70 le64_to_cpu(src->sf_items_updated);
71 des->sf_items_failed =
72 le64_to_cpu(src->sf_items_failed);
73 des->sf_items_updated_prior =
74 le64_to_cpu(src->sf_items_updated_prior);
75 des->sf_run_time = le32_to_cpu(src->sf_run_time);
76 des->sf_success_count = le32_to_cpu(src->sf_success_count);
77 des->sf_oi_count = le16_to_cpu(src->sf_oi_count);
78 des->sf_internal_flags = le16_to_cpu(src->sf_internal_flags);
79 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
82 static void scrub_file_to_le(struct scrub_file *des, struct scrub_file *src)
84 uuid_copy(&des->sf_uuid, &src->sf_uuid);
85 des->sf_flags = cpu_to_le64(src->sf_flags);
86 des->sf_magic = cpu_to_le32(src->sf_magic);
87 des->sf_status = cpu_to_le16(src->sf_status);
88 des->sf_param = cpu_to_le16(src->sf_param);
89 des->sf_time_last_complete =
90 cpu_to_le64(src->sf_time_last_complete);
91 des->sf_time_latest_start =
92 cpu_to_le64(src->sf_time_latest_start);
93 des->sf_time_last_checkpoint =
94 cpu_to_le64(src->sf_time_last_checkpoint);
95 des->sf_pos_latest_start =
96 cpu_to_le64(src->sf_pos_latest_start);
97 des->sf_pos_last_checkpoint =
98 cpu_to_le64(src->sf_pos_last_checkpoint);
99 des->sf_pos_first_inconsistent =
100 cpu_to_le64(src->sf_pos_first_inconsistent);
101 des->sf_items_checked =
102 cpu_to_le64(src->sf_items_checked);
103 des->sf_items_updated =
104 cpu_to_le64(src->sf_items_updated);
105 des->sf_items_failed =
106 cpu_to_le64(src->sf_items_failed);
107 des->sf_items_updated_prior =
108 cpu_to_le64(src->sf_items_updated_prior);
109 des->sf_run_time = cpu_to_le32(src->sf_run_time);
110 des->sf_success_count = cpu_to_le32(src->sf_success_count);
111 des->sf_oi_count = cpu_to_le16(src->sf_oi_count);
112 des->sf_internal_flags = cpu_to_le16(src->sf_internal_flags);
113 memcpy(des->sf_oi_bitmap, src->sf_oi_bitmap, SCRUB_OI_BITMAP_SIZE);
116 void scrub_file_init(struct lustre_scrub *scrub, uuid_t uuid)
118 struct scrub_file *sf = &scrub->os_file;
120 memset(sf, 0, sizeof(*sf));
121 uuid_copy(&sf->sf_uuid, &uuid);
122 sf->sf_magic = SCRUB_MAGIC_V1;
123 sf->sf_status = SS_INIT;
125 EXPORT_SYMBOL(scrub_file_init);
127 void scrub_file_reset(struct lustre_scrub *scrub, uuid_t uuid, u64 flags)
129 struct scrub_file *sf = &scrub->os_file;
131 CDEBUG(D_LFSCK, "%s: reset OI scrub file, old flags = "
132 "%#llx, add flags = %#llx\n",
133 scrub->os_name, sf->sf_flags, flags);
135 uuid_copy(&sf->sf_uuid, &uuid);
136 sf->sf_status = SS_INIT;
137 sf->sf_flags |= flags;
138 sf->sf_flags &= ~SF_AUTO;
140 sf->sf_time_latest_start = 0;
141 sf->sf_time_last_checkpoint = 0;
142 sf->sf_pos_latest_start = 0;
143 sf->sf_pos_last_checkpoint = 0;
144 sf->sf_pos_first_inconsistent = 0;
145 sf->sf_items_checked = 0;
146 sf->sf_items_updated = 0;
147 sf->sf_items_failed = 0;
148 sf->sf_items_noscrub = 0;
149 sf->sf_items_igif = 0;
150 if (!scrub->os_in_join)
151 sf->sf_items_updated_prior = 0;
153 EXPORT_SYMBOL(scrub_file_reset);
155 int scrub_file_load(const struct lu_env *env, struct lustre_scrub *scrub)
157 struct scrub_file *sf = &scrub->os_file;
158 struct lu_buf buf = {
159 .lb_buf = &scrub->os_file_disk,
160 .lb_len = sizeof(scrub->os_file_disk)
165 rc = dt_read(env, scrub->os_obj, &buf, &pos);
168 CERROR("%s: fail to load scrub file: rc = %d\n",
178 if (rc < buf.lb_len) {
179 CDEBUG(D_LFSCK, "%s: fail to load scrub file, "
180 "expected = %d: rc = %d\n",
181 scrub->os_name, (int)buf.lb_len, rc);
185 scrub_file_to_cpu(sf, &scrub->os_file_disk);
186 if (sf->sf_magic != SCRUB_MAGIC_V1) {
187 CDEBUG(D_LFSCK, "%s: invalid scrub magic 0x%x != 0x%x\n",
188 scrub->os_name, sf->sf_magic, SCRUB_MAGIC_V1);
194 EXPORT_SYMBOL(scrub_file_load);
196 int scrub_file_store(const struct lu_env *env, struct lustre_scrub *scrub)
198 struct scrub_file *sf = &scrub->os_file_disk;
199 struct dt_object *obj = scrub->os_obj;
200 struct dt_device *dev = scrub_obj2dev(obj);
201 struct lu_buf buf = {
203 .lb_len = sizeof(*sf)
210 /* Skip store under rdonly mode. */
214 scrub_file_to_le(sf, &scrub->os_file);
215 th = dt_trans_create(env, dev);
217 GOTO(log, rc = PTR_ERR(th));
219 rc = dt_declare_record_write(env, obj, &buf, pos, th);
223 rc = dt_trans_start_local(env, dev, th);
227 rc = dt_record_write(env, obj, &buf, &pos, th);
232 dt_trans_stop(env, dev, th);
236 CERROR("%s: store scrub file: rc = %d\n",
239 CDEBUG(D_LFSCK, "%s: store scrub file: rc = %d\n",
242 scrub->os_time_last_checkpoint = ktime_get_seconds();
243 scrub->os_time_next_checkpoint = scrub->os_time_last_checkpoint +
244 SCRUB_CHECKPOINT_INTERVAL;
247 EXPORT_SYMBOL(scrub_file_store);
249 int scrub_checkpoint(const struct lu_env *env, struct lustre_scrub *scrub)
251 struct scrub_file *sf = &scrub->os_file;
252 time64_t now = ktime_get_seconds();
255 if (likely(now < scrub->os_time_next_checkpoint ||
256 scrub->os_new_checked == 0))
259 CDEBUG(D_LFSCK, "%s: OI scrub checkpoint at pos %llu\n",
260 scrub->os_name, scrub->os_pos_current);
262 down_write(&scrub->os_rwsem);
263 sf->sf_items_checked += scrub->os_new_checked;
264 scrub->os_new_checked = 0;
265 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
266 sf->sf_time_last_checkpoint = ktime_get_real_seconds();
267 sf->sf_run_time += now - scrub->os_time_last_checkpoint;
268 rc = scrub_file_store(env, scrub);
269 up_write(&scrub->os_rwsem);
273 EXPORT_SYMBOL(scrub_checkpoint);
275 int scrub_start(int (*threadfn)(void *data), struct lustre_scrub *scrub,
276 void *data, __u32 flags)
278 struct task_struct *task;
285 if (scrub->os_file.sf_status == SS_COMPLETED) {
286 if (!(flags & SS_SET_FAILOUT))
287 flags |= SS_CLEAR_FAILOUT;
289 if (!(flags & SS_SET_DRYRUN))
290 flags |= SS_CLEAR_DRYRUN;
295 task = kthread_create(threadfn, data, "OI_scrub");
298 CERROR("%s: cannot start iteration thread: rc = %d\n",
302 spin_lock(&scrub->os_lock);
303 if (scrub->os_task) {
305 spin_unlock(&scrub->os_lock);
309 scrub->os_start_flags = flags;
310 scrub->os_task = task;
311 wake_up_process(task);
312 spin_unlock(&scrub->os_lock);
313 wait_var_event(scrub, scrub->os_running || !scrub->os_task);
317 EXPORT_SYMBOL(scrub_start);
319 void scrub_stop(struct lustre_scrub *scrub)
321 struct task_struct *task;
323 spin_lock(&scrub->os_lock);
324 scrub->os_running = 0;
325 spin_unlock(&scrub->os_lock);
326 task = xchg(&scrub->os_task, NULL);
330 EXPORT_SYMBOL(scrub_stop);
332 const char *const scrub_status_names[] = {
343 const char *const scrub_flags_names[] = {
351 const char *const scrub_param_names[] = {
357 static void scrub_bits_dump(struct seq_file *m, int bits,
358 const char *const names[],
364 seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
366 for (i = 0, flag = 1; bits != 0; i++, flag = BIT(i)) {
369 seq_printf(m, "%s%c", names[i],
370 bits != 0 ? ',' : '\n');
375 static void scrub_time_dump(struct seq_file *m, time64_t time,
379 seq_printf(m, "%s: %llu seconds\n", prefix,
380 ktime_get_real_seconds() - time);
382 seq_printf(m, "%s: N/A\n", prefix);
385 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
388 seq_printf(m, "%s: %llu\n", prefix, pos);
390 seq_printf(m, "%s: N/A\n", prefix);
393 void scrub_dump(struct seq_file *m, struct lustre_scrub *scrub)
395 struct scrub_file *sf = &scrub->os_file;
399 down_read(&scrub->os_rwsem);
400 seq_printf(m, "name: OI_scrub\n"
404 sf->sf_magic, (int)sf->sf_oi_count,
405 scrub_status_names[sf->sf_status]);
407 scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
409 scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
411 scrub_time_dump(m, sf->sf_time_last_complete,
412 "time_since_last_completed");
414 scrub_time_dump(m, sf->sf_time_latest_start,
415 "time_since_latest_start");
417 scrub_time_dump(m, sf->sf_time_last_checkpoint,
418 "time_since_last_checkpoint");
420 scrub_pos_dump(m, sf->sf_pos_latest_start,
421 "latest_start_position");
423 scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
424 "last_checkpoint_position");
426 scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
427 "first_failure_position");
429 checked = sf->sf_items_checked + scrub->os_new_checked;
430 seq_printf(m, "checked: %llu\n"
436 "success_count: %u\n",
438 sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
439 sf->sf_items_updated, sf->sf_items_failed,
440 sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
441 sf->sf_items_updated_prior, sf->sf_items_noscrub,
442 sf->sf_items_igif, sf->sf_success_count);
445 if (scrub->os_running) {
446 s64 new_checked = scrub->os_new_checked;
450 /* Since the time resolution is in seconds for new system
451 * or small devices it ismore likely that duration will be
452 * zero which will lead to inaccurate results.
454 duration = ktime_get_seconds() -
455 scrub->os_time_last_checkpoint;
457 new_checked = div_s64(new_checked, duration);
459 rtime = sf->sf_run_time + duration;
461 speed = div_s64(speed, rtime);
463 seq_printf(m, "run_time: %lld seconds\n"
464 "average_speed: %lld objects/sec\n"
465 "real_time_speed: %lld objects/sec\n"
466 "current_position: %llu\n"
467 "scrub_in_prior: %s\n"
468 "scrub_full_speed: %s\n"
469 "partial_scan: %s\n",
470 rtime, speed, new_checked,
471 scrub->os_pos_current,
472 scrub->os_in_prior ? "yes" : "no",
473 scrub->os_full_speed ? "yes" : "no",
474 scrub->os_partial_scan ? "yes" : "no");
476 if (sf->sf_run_time != 0)
477 speed = div_s64(speed, sf->sf_run_time);
478 seq_printf(m, "run_time: %d seconds\n"
479 "average_speed: %lld objects/sec\n"
480 "real_time_speed: N/A\n"
481 "current_position: N/A\n",
482 sf->sf_run_time, speed);
485 up_read(&scrub->os_rwsem);
487 EXPORT_SYMBOL(scrub_dump);
489 int lustre_liru_new(struct list_head *head, const struct lu_fid *pfid,
490 const struct lu_fid *cfid, __u64 child,
491 const char *name, int namelen)
493 struct lustre_index_restore_unit *liru;
494 int len = sizeof(*liru) + namelen + 1;
496 OBD_ALLOC(liru, len);
500 INIT_LIST_HEAD(&liru->liru_link);
501 liru->liru_pfid = *pfid;
502 liru->liru_cfid = *cfid;
503 liru->liru_clid = child;
504 liru->liru_len = len;
505 memcpy(liru->liru_name, name, namelen);
506 liru->liru_name[namelen] = 0;
507 list_add_tail(&liru->liru_link, head);
511 EXPORT_SYMBOL(lustre_liru_new);
513 int lustre_index_register(struct dt_device *dev, const char *devname,
514 struct list_head *head, spinlock_t *lock, int *guard,
515 const struct lu_fid *fid,
516 __u32 keysize, __u32 recsize)
518 struct lustre_index_backup_unit *libu, *pos;
522 if (dev->dd_rdonly || *guard)
529 INIT_LIST_HEAD(&libu->libu_link);
530 libu->libu_keysize = keysize;
531 libu->libu_recsize = recsize;
532 libu->libu_fid = *fid;
535 if (unlikely(*guard)) {
542 list_for_each_entry_reverse(pos, head, libu_link) {
543 rc = lu_fid_cmp(&pos->libu_fid, fid);
545 list_add(&libu->libu_link, &pos->libu_link);
552 /* Registered already. But the former registered one
553 * has different keysize/recsize. It may because that
554 * the former values are from disk and corrupted, then
555 * replace it with new values. */
556 if (unlikely(keysize != pos->libu_keysize ||
557 recsize != pos->libu_recsize)) {
558 CWARN("%s: the index "DFID" has registered "
559 "with %u/%u, may be invalid, replace "
561 devname, PFID(fid), pos->libu_keysize,
562 pos->libu_recsize, keysize, recsize);
564 pos->libu_keysize = keysize;
565 pos->libu_recsize = recsize;
577 list_add(&libu->libu_link, head);
582 EXPORT_SYMBOL(lustre_index_register);
584 static void lustre_index_degister(struct list_head *head, spinlock_t *lock,
585 const struct lu_fid *fid)
587 struct lustre_index_backup_unit *libu;
591 list_for_each_entry_reverse(libu, head, libu_link) {
592 rc = lu_fid_cmp(&libu->libu_fid, fid);
593 /* NOT registered. */
598 list_del(&libu->libu_link);
609 lustre_index_backup_make_header(struct lustre_index_backup_header *header,
610 __u32 keysize, __u32 recsize,
611 const struct lu_fid *fid, __u32 count)
613 memset(header, 0, sizeof(*header));
614 header->libh_magic = cpu_to_le32(INDEX_BACKUP_MAGIC_V1);
615 header->libh_count = cpu_to_le32(count);
616 header->libh_keysize = cpu_to_le32(keysize);
617 header->libh_recsize = cpu_to_le32(recsize);
618 fid_cpu_to_le(&header->libh_owner, fid);
621 static int lustre_index_backup_body(const struct lu_env *env,
622 struct dt_object *obj, loff_t *pos,
623 void *buf, int bufsize)
625 struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
627 struct lu_buf lbuf = {
634 th = dt_trans_create(env, dev);
638 rc = dt_declare_record_write(env, obj, &lbuf, *pos, th);
642 rc = dt_trans_start_local(env, dev, th);
646 rc = dt_record_write(env, obj, &lbuf, pos, th);
651 dt_trans_stop(env, dev, th);
655 static int lustre_index_backup_header(const struct lu_env *env,
656 struct dt_object *obj,
657 const struct lu_fid *tgt_fid,
658 __u32 keysize, __u32 recsize,
659 void *buf, int bufsize, int count)
661 struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
662 struct lustre_index_backup_header *header = buf;
663 struct lu_attr *la = buf;
665 struct lu_buf lbuf = {
667 .lb_len = sizeof(*header)
669 loff_t size = sizeof(*header) + (keysize + recsize) * count;
675 LASSERT(sizeof(*la) <= bufsize);
676 LASSERT(sizeof(*header) <= bufsize);
678 rc = dt_attr_get(env, obj, la);
682 if (la->la_size > size)
685 lustre_index_backup_make_header(header, keysize, recsize,
687 th = dt_trans_create(env, dev);
691 rc = dt_declare_record_write(env, obj, &lbuf, pos, th);
696 rc = dt_declare_punch(env, obj, size, OBD_OBJECT_EOF, th);
701 rc = dt_trans_start_local(env, dev, th);
705 rc = dt_record_write(env, obj, &lbuf, &pos, th);
707 rc = dt_punch(env, obj, size, OBD_OBJECT_EOF, th);
712 dt_trans_stop(env, dev, th);
716 static int lustre_index_update_lma(const struct lu_env *env,
717 struct dt_object *obj,
718 void *buf, int bufsize)
720 struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
721 struct lustre_mdt_attrs *lma = buf;
722 struct lu_buf lbuf = {
724 .lb_len = sizeof(struct lustre_ost_attrs)
727 int fl = LU_XATTR_REPLACE;
731 LASSERT(bufsize >= lbuf.lb_len);
733 rc = dt_xattr_get(env, obj, &lbuf, XATTR_NAME_LMA);
734 if (unlikely(rc == -ENODATA)) {
735 fl = LU_XATTR_CREATE;
736 lustre_lma_init(lma, lu_object_fid(&obj->do_lu),
739 } else if (rc < sizeof(*lma)) {
740 RETURN(rc < 0 ? rc : -EFAULT);
742 lustre_lma_swab(lma);
743 if (lma->lma_compat & LMAC_IDX_BACKUP)
746 lma->lma_compat |= LMAC_IDX_BACKUP;
749 lustre_lma_swab(lma);
751 th = dt_trans_create(env, dev);
755 rc = dt_declare_xattr_set(env, obj, &lbuf, XATTR_NAME_LMA, fl, th);
759 rc = dt_trans_start_local(env, dev, th);
763 rc = dt_xattr_set(env, obj, &lbuf, XATTR_NAME_LMA, fl, th);
768 dt_trans_stop(env, dev, th);
772 static int lustre_index_backup_one(const struct lu_env *env,
773 struct local_oid_storage *los,
774 struct dt_object *parent,
775 struct lustre_index_backup_unit *libu,
776 char *buf, int bufsize)
778 struct dt_device *dev = scrub_obj2dev(parent);
779 struct dt_object *tgt_obj = NULL;
780 struct dt_object *bak_obj = NULL;
781 const struct dt_it_ops *iops;
783 loff_t pos = sizeof(struct lustre_index_backup_header);
789 tgt_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
790 &libu->libu_fid, NULL));
791 if (IS_ERR_OR_NULL(tgt_obj))
792 GOTO(out, rc = tgt_obj ? PTR_ERR(tgt_obj) : -ENOENT);
794 if (!dt_object_exists(tgt_obj))
797 if (!tgt_obj->do_index_ops) {
798 struct dt_index_features feat;
800 feat.dif_flags = DT_IND_UPDATE;
801 feat.dif_keysize_min = libu->libu_keysize;
802 feat.dif_keysize_max = libu->libu_keysize;
803 feat.dif_recsize_min = libu->libu_recsize;
804 feat.dif_recsize_max = libu->libu_recsize;
805 feat.dif_ptrsize = 4;
806 rc = tgt_obj->do_ops->do_index_try(env, tgt_obj, &feat);
811 lustre_fid2lbx(buf, &libu->libu_fid, bufsize);
812 bak_obj = local_file_find_or_create(env, los, parent, buf,
813 S_IFREG | S_IRUGO | S_IWUSR);
814 if (IS_ERR_OR_NULL(bak_obj))
815 GOTO(out, rc = bak_obj ? PTR_ERR(bak_obj) : -ENOENT);
817 iops = &tgt_obj->do_index_ops->dio_it;
818 di = iops->init(env, tgt_obj, 0);
820 GOTO(out, rc = PTR_ERR(di));
822 rc = iops->load(env, di, 0);
824 rc = iops->next(env, di);
832 key = iops->key(env, di);
833 memcpy(&buf[size], key, libu->libu_keysize);
834 size += libu->libu_keysize;
836 rc = iops->rec(env, di, rec, 0);
840 size += libu->libu_recsize;
842 if (size + libu->libu_keysize + libu->libu_recsize > bufsize) {
843 rc = lustre_index_backup_body(env, bak_obj, &pos,
851 rc = iops->next(env, di);
854 if (rc >= 0 && size > 0)
855 rc = lustre_index_backup_body(env, bak_obj, &pos, buf, size);
860 rc = lustre_index_backup_header(env, bak_obj, &libu->libu_fid,
861 libu->libu_keysize, libu->libu_recsize,
862 buf, bufsize, count);
864 rc = lustre_index_update_lma(env, tgt_obj, buf, bufsize);
866 if (!rc && OBD_FAIL_CHECK(OBD_FAIL_OSD_INDEX_CRASH)) {
867 LASSERT(bufsize >= 512);
871 lustre_index_backup_body(env, tgt_obj, &pos, buf, 512);
879 if (!IS_ERR_OR_NULL(tgt_obj))
880 dt_object_put_nocache(env, tgt_obj);
881 if (!IS_ERR_OR_NULL(bak_obj))
882 dt_object_put_nocache(env, bak_obj);
886 void lustre_index_backup(const struct lu_env *env, struct dt_device *dev,
887 const char *devname, struct list_head *head,
888 spinlock_t *lock, int *guard, bool backup)
890 struct lustre_index_backup_unit *libu;
891 struct local_oid_storage *los = NULL;
892 struct dt_object *parent = NULL;
898 if (dev->dd_rdonly || *guard)
905 if (list_empty(head))
908 /* Handle kinds of failures during mount process. */
909 if (!dev->dd_lu_dev.ld_site || !dev->dd_lu_dev.ld_site->ls_top_dev)
913 OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
919 lu_local_obj_fid(&fid, INDEX_BACKUP_OID);
920 parent = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
922 if (IS_ERR_OR_NULL(parent)) {
923 CERROR("%s: failed to locate backup dir: rc = %ld\n",
924 devname, parent ? PTR_ERR(parent) : -ENOENT);
929 lu_local_name_obj_fid(&fid, 1);
930 rc = local_oid_storage_init(env, dev, &fid, &los);
932 CERROR("%s: failed to init local storage: rc = %d\n",
940 while (!list_empty(head)) {
941 libu = list_entry(head->next,
942 struct lustre_index_backup_unit, libu_link);
943 list_del_init(&libu->libu_link);
947 rc = lustre_index_backup_one(env, los, parent, libu,
948 buf, INDEX_BACKUP_BUFSIZE);
949 CDEBUG(D_WARNING, "%s: backup index "DFID": rc = %d\n",
950 devname, PFID(&libu->libu_fid), rc);
959 local_oid_storage_fini(env, los);
961 dt_object_put_nocache(env, parent);
963 OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
967 EXPORT_SYMBOL(lustre_index_backup);
969 int lustre_index_restore(const struct lu_env *env, struct dt_device *dev,
970 const struct lu_fid *parent_fid,
971 const struct lu_fid *tgt_fid,
972 const struct lu_fid *bak_fid, const char *name,
973 struct list_head *head, spinlock_t *lock,
974 char *buf, int bufsize)
976 struct dt_object *parent_obj = NULL;
977 struct dt_object *tgt_obj = NULL;
978 struct dt_object *bak_obj = NULL;
979 struct lustre_index_backup_header *header;
980 struct dt_index_features *feat;
981 struct dt_object_format *dof;
984 struct lu_object_conf conf;
985 struct dt_insert_rec ent;
994 bool registered = false;
997 LASSERT(bufsize >= sizeof(*la) + sizeof(*dof) +
998 sizeof(*feat) + sizeof(*header));
1000 memset(buf, 0, bufsize);
1001 la = (struct lu_attr *)buf;
1002 dof = (void *)la + sizeof(*la);
1003 feat = (void *)dof + sizeof(*dof);
1004 header = (void *)feat + sizeof(*feat);
1005 lbuf.lb_buf = header;
1006 lbuf.lb_len = sizeof(*header);
1008 tgt_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1010 if (IS_ERR_OR_NULL(tgt_obj))
1011 GOTO(out, rc = tgt_obj ? PTR_ERR(tgt_obj) : -ENOENT);
1013 bak_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1015 if (IS_ERR_OR_NULL(bak_obj))
1016 GOTO(out, rc = bak_obj ? PTR_ERR(bak_obj) : -ENOENT);
1018 if (!dt_object_exists(bak_obj))
1019 GOTO(out, rc = -ENOENT);
1021 parent_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1023 if (IS_ERR_OR_NULL(parent_obj))
1024 GOTO(out, rc = parent_obj ? PTR_ERR(parent_obj) : -ENOENT);
1026 LASSERT(dt_object_exists(parent_obj));
1028 if (unlikely(!dt_try_as_dir(env, parent_obj)))
1029 GOTO(out, rc = -ENOTDIR);
1031 rc = dt_attr_get(env, tgt_obj, la);
1035 rc = dt_record_read(env, bak_obj, &lbuf, &pos);
1039 if (le32_to_cpu(header->libh_magic) != INDEX_BACKUP_MAGIC_V1)
1040 GOTO(out, rc = -EINVAL);
1042 fid_le_to_cpu(&tfid, &header->libh_owner);
1043 if (unlikely(!lu_fid_eq(tgt_fid, &tfid)))
1044 GOTO(out, rc = -EINVAL);
1046 keysize = le32_to_cpu(header->libh_keysize);
1047 recsize = le32_to_cpu(header->libh_recsize);
1048 pairsize = keysize + recsize;
1050 memset(feat, 0, sizeof(*feat));
1051 feat->dif_flags = DT_IND_UPDATE;
1052 feat->dif_keysize_min = feat->dif_keysize_max = keysize;
1053 feat->dif_recsize_min = feat->dif_recsize_max = recsize;
1054 feat->dif_ptrsize = 4;
1056 /* T1: remove old name entry and destroy old index. */
1057 th = dt_trans_create(env, dev);
1059 GOTO(out, rc = PTR_ERR(th));
1061 rc = dt_declare_delete(env, parent_obj,
1062 (const struct dt_key *)name, th);
1066 rc = dt_declare_ref_del(env, tgt_obj, th);
1070 rc = dt_declare_destroy(env, tgt_obj, th);
1074 rc = dt_trans_start_local(env, dev, th);
1078 rc = dt_delete(env, parent_obj, (const struct dt_key *)name, th);
1082 dt_write_lock(env, tgt_obj, 0);
1083 rc = dt_ref_del(env, tgt_obj, th);
1085 if (S_ISDIR(tgt_obj->do_lu.lo_header->loh_attr))
1086 dt_ref_del(env, tgt_obj, th);
1087 rc = dt_destroy(env, tgt_obj, th);
1089 dt_write_unlock(env, tgt_obj);
1090 dt_trans_stop(env, dev, th);
1094 la->la_valid = LA_MODE | LA_UID | LA_GID;
1095 conf.loc_flags = LOC_F_NEW;
1096 dof->u.dof_idx.di_feat = feat;
1097 dof->dof_type = DFT_INDEX;
1098 ent.rec_type = S_IFREG;
1099 ent.rec_fid = tgt_fid;
1101 /* Drop cache before re-create it. */
1102 dt_object_put_nocache(env, tgt_obj);
1103 tgt_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1105 if (IS_ERR_OR_NULL(tgt_obj))
1106 GOTO(out, rc = tgt_obj ? PTR_ERR(tgt_obj) : -ENOENT);
1108 LASSERT(!dt_object_exists(tgt_obj));
1110 /* T2: create new index and insert new name entry. */
1111 th = dt_trans_create(env, dev);
1113 GOTO(out, rc = PTR_ERR(th));
1115 rc = dt_declare_create(env, tgt_obj, la, NULL, dof, th);
1119 rc = dt_declare_insert(env, parent_obj, (const struct dt_rec *)&ent,
1120 (const struct dt_key *)name, th);
1124 rc = dt_trans_start_local(env, dev, th);
1128 dt_write_lock(env, tgt_obj, 0);
1129 rc = dt_create(env, tgt_obj, la, NULL, dof, th);
1130 dt_write_unlock(env, tgt_obj);
1134 rc = dt_insert(env, parent_obj, (const struct dt_rec *)&ent,
1135 (const struct dt_key *)name, th);
1136 dt_trans_stop(env, dev, th);
1137 /* Some index name may has been inserted by OSD
1138 * automatically when create the index object. */
1139 if (unlikely(rc == -EEXIST))
1144 /* The new index will register via index_try. */
1145 rc = tgt_obj->do_ops->do_index_try(env, tgt_obj, feat);
1150 count = le32_to_cpu(header->libh_count);
1151 while (!rc && count > 0) {
1152 int size = pairsize * count;
1156 if (size > bufsize) {
1157 items = bufsize / pairsize;
1158 size = pairsize * items;
1163 rc = dt_record_read(env, bak_obj, &lbuf, &pos);
1164 for (i = 0; i < items && !rc; i++) {
1165 void *key = &buf[i * pairsize];
1166 void *rec = &buf[i * pairsize + keysize];
1168 /* Tn: restore the records. */
1169 th = dt_trans_create(env, dev);
1171 GOTO(out, rc = -ENOMEM);
1173 rc = dt_declare_insert(env, tgt_obj, rec, key, th);
1177 rc = dt_trans_start_local(env, dev, th);
1181 rc = dt_insert(env, tgt_obj, rec, key, th);
1182 if (unlikely(rc == -EEXIST))
1185 dt_trans_stop(env, dev, th);
1194 dt_trans_stop(env, dev, th);
1195 if (rc && registered)
1196 /* Degister the index to avoid overwriting the backup. */
1197 lustre_index_degister(head, lock, tgt_fid);
1200 if (!IS_ERR_OR_NULL(tgt_obj))
1201 dt_object_put_nocache(env, tgt_obj);
1202 if (!IS_ERR_OR_NULL(bak_obj))
1203 dt_object_put_nocache(env, bak_obj);
1204 if (!IS_ERR_OR_NULL(parent_obj))
1205 dt_object_put_nocache(env, parent_obj);
1208 EXPORT_SYMBOL(lustre_index_restore);