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_thread_prep(const struct lu_env *env, struct lustre_scrub *scrub,
276 uuid_t uuid, u64 start)
278 struct scrub_file *sf = &scrub->os_file;
279 u32 flags = scrub->os_start_flags;
280 bool drop_dryrun = false;
284 CDEBUG(D_LFSCK, "%s: OI scrub prep, flags = 0x%x\n",
285 scrub->os_name, flags);
287 down_write(&scrub->os_rwsem);
288 if (flags & SS_SET_FAILOUT)
289 sf->sf_param |= SP_FAILOUT;
290 else if (flags & SS_CLEAR_FAILOUT)
291 sf->sf_param &= ~SP_FAILOUT;
293 if (flags & SS_SET_DRYRUN) {
294 sf->sf_param |= SP_DRYRUN;
295 } else if (flags & SS_CLEAR_DRYRUN && sf->sf_param & SP_DRYRUN) {
296 sf->sf_param &= ~SP_DRYRUN;
300 if (flags & SS_RESET)
301 scrub_file_reset(scrub, uuid, 0);
303 spin_lock(&scrub->os_lock);
304 scrub->os_partial_scan = 0;
305 if (flags & SS_AUTO_FULL) {
306 scrub->os_full_speed = 1;
307 sf->sf_flags |= SF_AUTO;
308 } else if (flags & SS_AUTO_PARTIAL) {
309 scrub->os_full_speed = 0;
310 scrub->os_partial_scan = 1;
311 sf->sf_flags |= SF_AUTO;
312 } else if (sf->sf_flags & (SF_RECREATED | SF_INCONSISTENT |
314 scrub->os_full_speed = 1;
316 scrub->os_full_speed = 0;
319 scrub->os_in_prior = 0;
320 scrub->os_waiting = 0;
321 scrub->os_paused = 0;
322 scrub->os_in_join = 0;
323 scrub->os_full_scrub = 0;
324 spin_unlock(&scrub->os_lock);
325 scrub->os_new_checked = 0;
326 if (drop_dryrun && sf->sf_pos_first_inconsistent != 0)
327 sf->sf_pos_latest_start = sf->sf_pos_first_inconsistent;
328 else if (sf->sf_pos_last_checkpoint != 0)
329 sf->sf_pos_latest_start = sf->sf_pos_last_checkpoint + 1;
331 sf->sf_pos_latest_start = start;
333 scrub->os_pos_current = sf->sf_pos_latest_start;
334 sf->sf_status = SS_SCANNING;
335 sf->sf_time_latest_start = ktime_get_real_seconds();
336 sf->sf_time_last_checkpoint = sf->sf_time_latest_start;
337 sf->sf_pos_last_checkpoint = sf->sf_pos_latest_start - 1;
338 rc = scrub_file_store(env, scrub);
340 spin_lock(&scrub->os_lock);
341 scrub->os_running = 1;
342 spin_unlock(&scrub->os_lock);
345 up_write(&scrub->os_rwsem);
349 EXPORT_SYMBOL(scrub_thread_prep);
351 int scrub_thread_post(const struct lu_env *env, struct lustre_scrub *scrub,
354 struct scrub_file *sf = &scrub->os_file;
358 CDEBUG(D_LFSCK, "%s: OI scrub post with result = %d\n",
359 scrub->os_name, result);
361 down_write(&scrub->os_rwsem);
362 spin_lock(&scrub->os_lock);
363 scrub->os_running = 0;
364 spin_unlock(&scrub->os_lock);
365 if (scrub->os_new_checked > 0) {
366 sf->sf_items_checked += scrub->os_new_checked;
367 scrub->os_new_checked = 0;
368 sf->sf_pos_last_checkpoint = scrub->os_pos_current;
370 sf->sf_time_last_checkpoint = ktime_get_real_seconds();
372 sf->sf_status = SS_COMPLETED;
373 if (!(sf->sf_param & SP_DRYRUN)) {
374 memset(sf->sf_oi_bitmap, 0, SCRUB_OI_BITMAP_SIZE);
375 sf->sf_flags &= ~(SF_RECREATED | SF_INCONSISTENT |
376 SF_UPGRADE | SF_AUTO);
378 sf->sf_time_last_complete = sf->sf_time_last_checkpoint;
379 sf->sf_success_count++;
380 } else if (result == 0) {
381 if (scrub->os_paused)
382 sf->sf_status = SS_PAUSED;
384 sf->sf_status = SS_STOPPED;
386 sf->sf_status = SS_FAILED;
388 sf->sf_run_time += ktime_get_seconds() -
389 scrub->os_time_last_checkpoint;
391 rc = scrub_file_store(env, scrub);
392 up_write(&scrub->os_rwsem);
394 RETURN(rc < 0 ? rc : result);
396 EXPORT_SYMBOL(scrub_thread_post);
398 int scrub_start(int (*threadfn)(void *data), struct lustre_scrub *scrub,
399 void *data, __u32 flags)
401 struct task_struct *task;
408 if (scrub->os_file.sf_status == SS_COMPLETED) {
409 if (!(flags & SS_SET_FAILOUT))
410 flags |= SS_CLEAR_FAILOUT;
412 if (!(flags & SS_SET_DRYRUN))
413 flags |= SS_CLEAR_DRYRUN;
418 task = kthread_create(threadfn, data, "OI_scrub");
421 CERROR("%s: cannot start iteration thread: rc = %d\n",
425 spin_lock(&scrub->os_lock);
426 if (scrub->os_task) {
428 spin_unlock(&scrub->os_lock);
432 scrub->os_start_flags = flags;
433 scrub->os_task = task;
434 wake_up_process(task);
435 spin_unlock(&scrub->os_lock);
436 wait_var_event(scrub, scrub->os_running || !scrub->os_task);
440 EXPORT_SYMBOL(scrub_start);
442 void scrub_stop(struct lustre_scrub *scrub)
444 struct task_struct *task;
446 spin_lock(&scrub->os_lock);
447 scrub->os_running = 0;
448 spin_unlock(&scrub->os_lock);
449 task = xchg(&scrub->os_task, NULL);
453 EXPORT_SYMBOL(scrub_stop);
455 const char *const scrub_status_names[] = {
466 const char *const scrub_flags_names[] = {
474 const char *const scrub_param_names[] = {
480 static void scrub_bits_dump(struct seq_file *m, int bits,
481 const char *const names[],
487 seq_printf(m, "%s:%c", prefix, bits != 0 ? ' ' : '\n');
489 for (i = 0, flag = 1; bits != 0; i++, flag = BIT(i)) {
492 seq_printf(m, "%s%c", names[i],
493 bits != 0 ? ',' : '\n');
498 static void scrub_time_dump(struct seq_file *m, time64_t time,
502 seq_printf(m, "%s: %llu seconds\n", prefix,
503 ktime_get_real_seconds() - time);
505 seq_printf(m, "%s: N/A\n", prefix);
508 static void scrub_pos_dump(struct seq_file *m, __u64 pos, const char *prefix)
511 seq_printf(m, "%s: %llu\n", prefix, pos);
513 seq_printf(m, "%s: N/A\n", prefix);
516 void scrub_dump(struct seq_file *m, struct lustre_scrub *scrub)
518 struct scrub_file *sf = &scrub->os_file;
522 down_read(&scrub->os_rwsem);
523 seq_printf(m, "name: OI_scrub\n"
527 sf->sf_magic, (int)sf->sf_oi_count,
528 scrub_status_names[sf->sf_status]);
530 scrub_bits_dump(m, sf->sf_flags, scrub_flags_names, "flags");
532 scrub_bits_dump(m, sf->sf_param, scrub_param_names, "param");
534 scrub_time_dump(m, sf->sf_time_last_complete,
535 "time_since_last_completed");
537 scrub_time_dump(m, sf->sf_time_latest_start,
538 "time_since_latest_start");
540 scrub_time_dump(m, sf->sf_time_last_checkpoint,
541 "time_since_last_checkpoint");
543 scrub_pos_dump(m, sf->sf_pos_latest_start,
544 "latest_start_position");
546 scrub_pos_dump(m, sf->sf_pos_last_checkpoint,
547 "last_checkpoint_position");
549 scrub_pos_dump(m, sf->sf_pos_first_inconsistent,
550 "first_failure_position");
552 checked = sf->sf_items_checked + scrub->os_new_checked;
553 seq_printf(m, "checked: %llu\n"
559 "success_count: %u\n",
561 sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
562 sf->sf_items_updated, sf->sf_items_failed,
563 sf->sf_param & SP_DRYRUN ? "inconsistent" : "updated",
564 sf->sf_items_updated_prior, sf->sf_items_noscrub,
565 sf->sf_items_igif, sf->sf_success_count);
568 if (scrub->os_running) {
569 s64 new_checked = scrub->os_new_checked;
573 /* Since the time resolution is in seconds for new system
574 * or small devices it ismore likely that duration will be
575 * zero which will lead to inaccurate results.
577 duration = ktime_get_seconds() -
578 scrub->os_time_last_checkpoint;
580 new_checked = div_s64(new_checked, duration);
582 rtime = sf->sf_run_time + duration;
584 speed = div_s64(speed, rtime);
586 seq_printf(m, "run_time: %lld seconds\n"
587 "average_speed: %lld objects/sec\n"
588 "real_time_speed: %lld objects/sec\n"
589 "current_position: %llu\n"
590 "scrub_in_prior: %s\n"
591 "scrub_full_speed: %s\n"
592 "partial_scan: %s\n",
593 rtime, speed, new_checked,
594 scrub->os_pos_current,
595 scrub->os_in_prior ? "yes" : "no",
596 scrub->os_full_speed ? "yes" : "no",
597 scrub->os_partial_scan ? "yes" : "no");
599 if (sf->sf_run_time != 0)
600 speed = div_s64(speed, sf->sf_run_time);
601 seq_printf(m, "run_time: %d seconds\n"
602 "average_speed: %lld objects/sec\n"
603 "real_time_speed: N/A\n"
604 "current_position: N/A\n",
605 sf->sf_run_time, speed);
608 up_read(&scrub->os_rwsem);
610 EXPORT_SYMBOL(scrub_dump);
612 int lustre_liru_new(struct list_head *head, const struct lu_fid *pfid,
613 const struct lu_fid *cfid, __u64 child,
614 const char *name, int namelen)
616 struct lustre_index_restore_unit *liru;
617 int len = sizeof(*liru) + namelen + 1;
619 OBD_ALLOC(liru, len);
623 INIT_LIST_HEAD(&liru->liru_link);
624 liru->liru_pfid = *pfid;
625 liru->liru_cfid = *cfid;
626 liru->liru_clid = child;
627 liru->liru_len = len;
628 memcpy(liru->liru_name, name, namelen);
629 liru->liru_name[namelen] = 0;
630 list_add_tail(&liru->liru_link, head);
634 EXPORT_SYMBOL(lustre_liru_new);
636 int lustre_index_register(struct dt_device *dev, const char *devname,
637 struct list_head *head, spinlock_t *lock, int *guard,
638 const struct lu_fid *fid,
639 __u32 keysize, __u32 recsize)
641 struct lustre_index_backup_unit *libu, *pos;
645 if (dev->dd_rdonly || *guard)
652 INIT_LIST_HEAD(&libu->libu_link);
653 libu->libu_keysize = keysize;
654 libu->libu_recsize = recsize;
655 libu->libu_fid = *fid;
658 if (unlikely(*guard)) {
665 list_for_each_entry_reverse(pos, head, libu_link) {
666 rc = lu_fid_cmp(&pos->libu_fid, fid);
668 list_add(&libu->libu_link, &pos->libu_link);
675 /* Registered already. But the former registered one
676 * has different keysize/recsize. It may because that
677 * the former values are from disk and corrupted, then
678 * replace it with new values. */
679 if (unlikely(keysize != pos->libu_keysize ||
680 recsize != pos->libu_recsize)) {
681 CWARN("%s: the index "DFID" has registered "
682 "with %u/%u, may be invalid, replace "
684 devname, PFID(fid), pos->libu_keysize,
685 pos->libu_recsize, keysize, recsize);
687 pos->libu_keysize = keysize;
688 pos->libu_recsize = recsize;
700 list_add(&libu->libu_link, head);
705 EXPORT_SYMBOL(lustre_index_register);
707 static void lustre_index_degister(struct list_head *head, spinlock_t *lock,
708 const struct lu_fid *fid)
710 struct lustre_index_backup_unit *libu;
714 list_for_each_entry_reverse(libu, head, libu_link) {
715 rc = lu_fid_cmp(&libu->libu_fid, fid);
716 /* NOT registered. */
721 list_del(&libu->libu_link);
732 lustre_index_backup_make_header(struct lustre_index_backup_header *header,
733 __u32 keysize, __u32 recsize,
734 const struct lu_fid *fid, __u32 count)
736 memset(header, 0, sizeof(*header));
737 header->libh_magic = cpu_to_le32(INDEX_BACKUP_MAGIC_V1);
738 header->libh_count = cpu_to_le32(count);
739 header->libh_keysize = cpu_to_le32(keysize);
740 header->libh_recsize = cpu_to_le32(recsize);
741 fid_cpu_to_le(&header->libh_owner, fid);
744 static int lustre_index_backup_body(const struct lu_env *env,
745 struct dt_object *obj, loff_t *pos,
746 void *buf, int bufsize)
748 struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
750 struct lu_buf lbuf = {
757 th = dt_trans_create(env, dev);
761 rc = dt_declare_record_write(env, obj, &lbuf, *pos, th);
765 rc = dt_trans_start_local(env, dev, th);
769 rc = dt_record_write(env, obj, &lbuf, pos, th);
774 dt_trans_stop(env, dev, th);
778 static int lustre_index_backup_header(const struct lu_env *env,
779 struct dt_object *obj,
780 const struct lu_fid *tgt_fid,
781 __u32 keysize, __u32 recsize,
782 void *buf, int bufsize, int count)
784 struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
785 struct lustre_index_backup_header *header = buf;
786 struct lu_attr *la = buf;
788 struct lu_buf lbuf = {
790 .lb_len = sizeof(*header)
792 loff_t size = sizeof(*header) + (keysize + recsize) * count;
798 LASSERT(sizeof(*la) <= bufsize);
799 LASSERT(sizeof(*header) <= bufsize);
801 rc = dt_attr_get(env, obj, la);
805 if (la->la_size > size)
808 lustre_index_backup_make_header(header, keysize, recsize,
810 th = dt_trans_create(env, dev);
814 rc = dt_declare_record_write(env, obj, &lbuf, pos, th);
819 rc = dt_declare_punch(env, obj, size, OBD_OBJECT_EOF, th);
824 rc = dt_trans_start_local(env, dev, th);
828 rc = dt_record_write(env, obj, &lbuf, &pos, th);
830 rc = dt_punch(env, obj, size, OBD_OBJECT_EOF, th);
835 dt_trans_stop(env, dev, th);
839 static int lustre_index_update_lma(const struct lu_env *env,
840 struct dt_object *obj,
841 void *buf, int bufsize)
843 struct dt_device *dev = lu2dt_dev(obj->do_lu.lo_dev);
844 struct lustre_mdt_attrs *lma = buf;
845 struct lu_buf lbuf = {
847 .lb_len = sizeof(struct lustre_ost_attrs)
850 int fl = LU_XATTR_REPLACE;
854 LASSERT(bufsize >= lbuf.lb_len);
856 rc = dt_xattr_get(env, obj, &lbuf, XATTR_NAME_LMA);
857 if (unlikely(rc == -ENODATA)) {
858 fl = LU_XATTR_CREATE;
859 lustre_lma_init(lma, lu_object_fid(&obj->do_lu),
862 } else if (rc < sizeof(*lma)) {
863 RETURN(rc < 0 ? rc : -EFAULT);
865 lustre_lma_swab(lma);
866 if (lma->lma_compat & LMAC_IDX_BACKUP)
869 lma->lma_compat |= LMAC_IDX_BACKUP;
872 lustre_lma_swab(lma);
874 th = dt_trans_create(env, dev);
878 rc = dt_declare_xattr_set(env, obj, &lbuf, XATTR_NAME_LMA, fl, th);
882 rc = dt_trans_start_local(env, dev, th);
886 rc = dt_xattr_set(env, obj, &lbuf, XATTR_NAME_LMA, fl, th);
891 dt_trans_stop(env, dev, th);
895 static int lustre_index_backup_one(const struct lu_env *env,
896 struct local_oid_storage *los,
897 struct dt_object *parent,
898 struct lustre_index_backup_unit *libu,
899 char *buf, int bufsize)
901 struct dt_device *dev = scrub_obj2dev(parent);
902 struct dt_object *tgt_obj = NULL;
903 struct dt_object *bak_obj = NULL;
904 const struct dt_it_ops *iops;
906 loff_t pos = sizeof(struct lustre_index_backup_header);
912 tgt_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
913 &libu->libu_fid, NULL));
914 if (IS_ERR_OR_NULL(tgt_obj))
915 GOTO(out, rc = tgt_obj ? PTR_ERR(tgt_obj) : -ENOENT);
917 if (!dt_object_exists(tgt_obj))
920 if (!tgt_obj->do_index_ops) {
921 struct dt_index_features feat;
923 feat.dif_flags = DT_IND_UPDATE;
924 feat.dif_keysize_min = libu->libu_keysize;
925 feat.dif_keysize_max = libu->libu_keysize;
926 feat.dif_recsize_min = libu->libu_recsize;
927 feat.dif_recsize_max = libu->libu_recsize;
928 feat.dif_ptrsize = 4;
929 rc = tgt_obj->do_ops->do_index_try(env, tgt_obj, &feat);
934 lustre_fid2lbx(buf, &libu->libu_fid, bufsize);
935 bak_obj = local_file_find_or_create(env, los, parent, buf,
936 S_IFREG | S_IRUGO | S_IWUSR);
937 if (IS_ERR_OR_NULL(bak_obj))
938 GOTO(out, rc = bak_obj ? PTR_ERR(bak_obj) : -ENOENT);
940 iops = &tgt_obj->do_index_ops->dio_it;
941 di = iops->init(env, tgt_obj, 0);
943 GOTO(out, rc = PTR_ERR(di));
945 rc = iops->load(env, di, 0);
947 rc = iops->next(env, di);
955 key = iops->key(env, di);
956 memcpy(&buf[size], key, libu->libu_keysize);
957 size += libu->libu_keysize;
959 rc = iops->rec(env, di, rec, 0);
963 size += libu->libu_recsize;
965 if (size + libu->libu_keysize + libu->libu_recsize > bufsize) {
966 rc = lustre_index_backup_body(env, bak_obj, &pos,
974 rc = iops->next(env, di);
977 if (rc >= 0 && size > 0)
978 rc = lustre_index_backup_body(env, bak_obj, &pos, buf, size);
983 rc = lustre_index_backup_header(env, bak_obj, &libu->libu_fid,
984 libu->libu_keysize, libu->libu_recsize,
985 buf, bufsize, count);
987 rc = lustre_index_update_lma(env, tgt_obj, buf, bufsize);
989 if (!rc && OBD_FAIL_CHECK(OBD_FAIL_OSD_INDEX_CRASH)) {
990 LASSERT(bufsize >= 512);
994 lustre_index_backup_body(env, tgt_obj, &pos, buf, 512);
1000 iops->fini(env, di);
1002 if (!IS_ERR_OR_NULL(tgt_obj))
1003 dt_object_put_nocache(env, tgt_obj);
1004 if (!IS_ERR_OR_NULL(bak_obj))
1005 dt_object_put_nocache(env, bak_obj);
1009 void lustre_index_backup(const struct lu_env *env, struct dt_device *dev,
1010 const char *devname, struct list_head *head,
1011 spinlock_t *lock, int *guard, bool backup)
1013 struct lustre_index_backup_unit *libu;
1014 struct local_oid_storage *los = NULL;
1015 struct dt_object *parent = NULL;
1021 if (dev->dd_rdonly || *guard)
1028 if (list_empty(head))
1031 /* Handle kinds of failures during mount process. */
1032 if (!dev->dd_lu_dev.ld_site || !dev->dd_lu_dev.ld_site->ls_top_dev)
1036 OBD_ALLOC_LARGE(buf, INDEX_BACKUP_BUFSIZE);
1042 lu_local_obj_fid(&fid, INDEX_BACKUP_OID);
1043 parent = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1045 if (IS_ERR_OR_NULL(parent)) {
1046 CERROR("%s: failed to locate backup dir: rc = %ld\n",
1047 devname, parent ? PTR_ERR(parent) : -ENOENT);
1052 lu_local_name_obj_fid(&fid, 1);
1053 rc = local_oid_storage_init(env, dev, &fid, &los);
1055 CERROR("%s: failed to init local storage: rc = %d\n",
1063 while (!list_empty(head)) {
1064 libu = list_entry(head->next,
1065 struct lustre_index_backup_unit, libu_link);
1066 list_del_init(&libu->libu_link);
1070 rc = lustre_index_backup_one(env, los, parent, libu,
1071 buf, INDEX_BACKUP_BUFSIZE);
1072 CDEBUG(D_WARNING, "%s: backup index "DFID": rc = %d\n",
1073 devname, PFID(&libu->libu_fid), rc);
1082 local_oid_storage_fini(env, los);
1084 dt_object_put_nocache(env, parent);
1086 OBD_FREE_LARGE(buf, INDEX_BACKUP_BUFSIZE);
1090 EXPORT_SYMBOL(lustre_index_backup);
1092 int lustre_index_restore(const struct lu_env *env, struct dt_device *dev,
1093 const struct lu_fid *parent_fid,
1094 const struct lu_fid *tgt_fid,
1095 const struct lu_fid *bak_fid, const char *name,
1096 struct list_head *head, spinlock_t *lock,
1097 char *buf, int bufsize)
1099 struct dt_object *parent_obj = NULL;
1100 struct dt_object *tgt_obj = NULL;
1101 struct dt_object *bak_obj = NULL;
1102 struct lustre_index_backup_header *header;
1103 struct dt_index_features *feat;
1104 struct dt_object_format *dof;
1107 struct lu_object_conf conf;
1108 struct dt_insert_rec ent;
1117 bool registered = false;
1120 LASSERT(bufsize >= sizeof(*la) + sizeof(*dof) +
1121 sizeof(*feat) + sizeof(*header));
1123 memset(buf, 0, bufsize);
1124 la = (struct lu_attr *)buf;
1125 dof = (void *)la + sizeof(*la);
1126 feat = (void *)dof + sizeof(*dof);
1127 header = (void *)feat + sizeof(*feat);
1128 lbuf.lb_buf = header;
1129 lbuf.lb_len = sizeof(*header);
1131 tgt_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1133 if (IS_ERR_OR_NULL(tgt_obj))
1134 GOTO(out, rc = tgt_obj ? PTR_ERR(tgt_obj) : -ENOENT);
1136 bak_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1138 if (IS_ERR_OR_NULL(bak_obj))
1139 GOTO(out, rc = bak_obj ? PTR_ERR(bak_obj) : -ENOENT);
1141 if (!dt_object_exists(bak_obj))
1142 GOTO(out, rc = -ENOENT);
1144 parent_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1146 if (IS_ERR_OR_NULL(parent_obj))
1147 GOTO(out, rc = parent_obj ? PTR_ERR(parent_obj) : -ENOENT);
1149 LASSERT(dt_object_exists(parent_obj));
1151 if (unlikely(!dt_try_as_dir(env, parent_obj)))
1152 GOTO(out, rc = -ENOTDIR);
1154 rc = dt_attr_get(env, tgt_obj, la);
1158 rc = dt_record_read(env, bak_obj, &lbuf, &pos);
1162 if (le32_to_cpu(header->libh_magic) != INDEX_BACKUP_MAGIC_V1)
1163 GOTO(out, rc = -EINVAL);
1165 fid_le_to_cpu(&tfid, &header->libh_owner);
1166 if (unlikely(!lu_fid_eq(tgt_fid, &tfid)))
1167 GOTO(out, rc = -EINVAL);
1169 keysize = le32_to_cpu(header->libh_keysize);
1170 recsize = le32_to_cpu(header->libh_recsize);
1171 pairsize = keysize + recsize;
1173 memset(feat, 0, sizeof(*feat));
1174 feat->dif_flags = DT_IND_UPDATE;
1175 feat->dif_keysize_min = feat->dif_keysize_max = keysize;
1176 feat->dif_recsize_min = feat->dif_recsize_max = recsize;
1177 feat->dif_ptrsize = 4;
1179 /* T1: remove old name entry and destroy old index. */
1180 th = dt_trans_create(env, dev);
1182 GOTO(out, rc = PTR_ERR(th));
1184 rc = dt_declare_delete(env, parent_obj,
1185 (const struct dt_key *)name, th);
1189 rc = dt_declare_ref_del(env, tgt_obj, th);
1193 rc = dt_declare_destroy(env, tgt_obj, th);
1197 rc = dt_trans_start_local(env, dev, th);
1201 rc = dt_delete(env, parent_obj, (const struct dt_key *)name, th);
1205 dt_write_lock(env, tgt_obj, 0);
1206 rc = dt_ref_del(env, tgt_obj, th);
1208 if (S_ISDIR(tgt_obj->do_lu.lo_header->loh_attr))
1209 dt_ref_del(env, tgt_obj, th);
1210 rc = dt_destroy(env, tgt_obj, th);
1212 dt_write_unlock(env, tgt_obj);
1213 dt_trans_stop(env, dev, th);
1217 la->la_valid = LA_MODE | LA_UID | LA_GID;
1218 conf.loc_flags = LOC_F_NEW;
1219 dof->u.dof_idx.di_feat = feat;
1220 dof->dof_type = DFT_INDEX;
1221 ent.rec_type = S_IFREG;
1222 ent.rec_fid = tgt_fid;
1224 /* Drop cache before re-create it. */
1225 dt_object_put_nocache(env, tgt_obj);
1226 tgt_obj = lu2dt(lu_object_find_slice(env, &dev->dd_lu_dev,
1228 if (IS_ERR_OR_NULL(tgt_obj))
1229 GOTO(out, rc = tgt_obj ? PTR_ERR(tgt_obj) : -ENOENT);
1231 LASSERT(!dt_object_exists(tgt_obj));
1233 /* T2: create new index and insert new name entry. */
1234 th = dt_trans_create(env, dev);
1236 GOTO(out, rc = PTR_ERR(th));
1238 rc = dt_declare_create(env, tgt_obj, la, NULL, dof, th);
1242 rc = dt_declare_insert(env, parent_obj, (const struct dt_rec *)&ent,
1243 (const struct dt_key *)name, th);
1247 rc = dt_trans_start_local(env, dev, th);
1251 dt_write_lock(env, tgt_obj, 0);
1252 rc = dt_create(env, tgt_obj, la, NULL, dof, th);
1253 dt_write_unlock(env, tgt_obj);
1257 rc = dt_insert(env, parent_obj, (const struct dt_rec *)&ent,
1258 (const struct dt_key *)name, th);
1259 dt_trans_stop(env, dev, th);
1260 /* Some index name may has been inserted by OSD
1261 * automatically when create the index object. */
1262 if (unlikely(rc == -EEXIST))
1267 /* The new index will register via index_try. */
1268 rc = tgt_obj->do_ops->do_index_try(env, tgt_obj, feat);
1273 count = le32_to_cpu(header->libh_count);
1274 while (!rc && count > 0) {
1275 int size = pairsize * count;
1279 if (size > bufsize) {
1280 items = bufsize / pairsize;
1281 size = pairsize * items;
1286 rc = dt_record_read(env, bak_obj, &lbuf, &pos);
1287 for (i = 0; i < items && !rc; i++) {
1288 void *key = &buf[i * pairsize];
1289 void *rec = &buf[i * pairsize + keysize];
1291 /* Tn: restore the records. */
1292 th = dt_trans_create(env, dev);
1294 GOTO(out, rc = -ENOMEM);
1296 rc = dt_declare_insert(env, tgt_obj, rec, key, th);
1300 rc = dt_trans_start_local(env, dev, th);
1304 rc = dt_insert(env, tgt_obj, rec, key, th);
1305 if (unlikely(rc == -EEXIST))
1308 dt_trans_stop(env, dev, th);
1317 dt_trans_stop(env, dev, th);
1318 if (rc && registered)
1319 /* Degister the index to avoid overwriting the backup. */
1320 lustre_index_degister(head, lock, tgt_fid);
1323 if (!IS_ERR_OR_NULL(tgt_obj))
1324 dt_object_put_nocache(env, tgt_obj);
1325 if (!IS_ERR_OR_NULL(bak_obj))
1326 dt_object_put_nocache(env, bak_obj);
1327 if (!IS_ERR_OR_NULL(parent_obj))
1328 dt_object_put_nocache(env, parent_obj);
1331 EXPORT_SYMBOL(lustre_index_restore);