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) 2013, 2016, Intel Corporation.
26 * lustre/lfsck/lfsck_engine.c
28 * Author: Fan, Yong <fan.yong@intel.com>
31 #define DEBUG_SUBSYSTEM S_LFSCK
33 #include <lu_object.h>
34 #include <dt_object.h>
35 #include <lustre_net.h>
36 #include <lustre_fid.h>
37 #include <obd_support.h>
38 #include <lustre_lib.h>
40 #include "lfsck_internal.h"
42 int lfsck_unpack_ent(struct lu_dirent *ent, __u64 *cookie, __u16 *type)
45 int align = sizeof(*lt) - 1;
48 fid_le_to_cpu(&ent->lde_fid, &ent->lde_fid);
49 *cookie = le64_to_cpu(ent->lde_hash);
50 ent->lde_reclen = le16_to_cpu(ent->lde_reclen);
51 ent->lde_namelen = le16_to_cpu(ent->lde_namelen);
52 ent->lde_attrs = le32_to_cpu(ent->lde_attrs);
54 if (unlikely(!(ent->lde_attrs & LUDA_TYPE)))
57 len = (ent->lde_namelen + align) & ~align;
58 lt = (struct luda_type *)(ent->lde_name + len);
59 *type = le16_to_cpu(lt->lt_type);
61 /* Make sure the name is terminated with '\0'. The data (object type)
62 * after ent::lde_name maybe broken, but we have stored such data in
63 * the output parameter @type as above. */
64 ent->lde_name[ent->lde_namelen] = '\0';
69 static void lfsck_di_oit_put(const struct lu_env *env, struct lfsck_instance *lfsck)
71 const struct dt_it_ops *iops;
74 spin_lock(&lfsck->li_lock);
75 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
76 di = lfsck->li_di_oit;
77 lfsck->li_di_oit = NULL;
78 spin_unlock(&lfsck->li_lock);
82 static void lfsck_di_dir_put(const struct lu_env *env, struct lfsck_instance *lfsck)
84 const struct dt_it_ops *iops;
87 spin_lock(&lfsck->li_lock);
88 iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
89 di = lfsck->li_di_dir;
90 lfsck->li_di_dir = NULL;
91 lfsck->li_cookie_dir = 0;
92 spin_unlock(&lfsck->li_lock);
96 static int lfsck_parent_fid(const struct lu_env *env, struct dt_object *obj,
99 if (unlikely(!S_ISDIR(lfsck_object_type(obj)) ||
100 !dt_try_as_dir(env, obj)))
103 return dt_lookup(env, obj, (struct dt_rec *)fid,
104 (const struct dt_key *)"..");
108 * Check whether needs to scan the directory or not.
110 * 1) If we are not doing namespace LFSCK, or the given @obj is not directory,
111 * then needs not to scan the @obj. Otherwise,
112 * 2) Global /ROOT needs to be scanned, backend root needs not to be scanned.
113 * 3) If the @obj is neither IGIF nor normal FID (including .lustre and its
114 * sub-directories that have been scanned when the LFSCK engine start),
115 * then needs not to be scanned.
116 * 4) If it is a remote object, then scanning the object will be done on the
117 * MDT on which the object really resides.
118 * 5) If the local object has normal FID, then needs to be scanned. Otherwise,
119 * 6) If the object has linkEA, then needs to be scanned. Otherwise,
120 * 7) If none of the previous conditions are true, we need to check the parent
121 * directories whether this subdirectory is in a tree that should be scanned.
122 * Set the parent as current @obj, repeat 2)-7).
124 * \param[in] env pointer to the thread context
125 * \param[in] lfsck pointer to the lfsck instance
126 * \param[in] obj pointer to the object to be checked
128 * \retval positive number if the directory needs to be scanned
129 * \retval 0 if the directory needs NOT to be scanned
130 * \retval negative error number on failure
132 static int lfsck_needs_scan_dir(const struct lu_env *env,
133 struct lfsck_instance *lfsck,
134 struct dt_object *obj)
136 struct lfsck_thread_info *info = lfsck_env_info(env);
137 struct lu_fid *fid = &info->lti_fid;
138 struct lu_seq_range *range = &info->lti_range;
139 struct seq_server_site *ss = lfsck_dev_site(lfsck);
140 __u32 idx = lfsck_dev_idx(lfsck);
144 if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
149 *fid = *lfsck_dto2fid(obj);
151 /* Global /ROOT is visible. */
152 if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
155 /* Backend root is invisible. */
156 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
159 if (!fid_is_norm(fid) && !fid_is_igif(fid))
162 fld_range_set_mdt(range);
163 rc = fld_local_lookup(env, ss->ss_server_fld,
164 fid_seq(fid), range);
165 if (rc != 0 || range->lsr_index != idx)
166 /* Current FID should NOT be for the input parameter
167 * @obj, because the lfsck_master_oit_engine() has
168 * filtered out agent object. So current FID is for
169 * the ancestor of the original input parameter @obj.
170 * So the ancestor is a remote directory. The input
171 * parameter @obj is local directory, and should be
172 * scanned under such case. */
175 /* normal FID on this target (locally) must be for the
176 * client-side visiable object. */
177 if (fid_is_norm(fid))
181 obj = lfsck_object_find_bottom(env, lfsck, fid);
186 if (!dt_object_exists(obj))
190 dt_read_lock(env, obj, MOR_TGT_CHILD);
191 if (unlikely(lfsck_is_dead_obj(obj))) {
192 dt_read_unlock(env, obj);
197 rc = dt_xattr_get(env, obj,
198 lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK);
199 dt_read_unlock(env, obj);
203 if (rc < 0 && rc != -ENODATA)
206 rc = lfsck_parent_fid(env, obj, fid);
208 lfsck_object_put(env, obj);
214 if (!fid_is_sane(fid))
219 if (depth > 0 && obj != NULL)
220 lfsck_object_put(env, obj);
225 static int lfsck_load_stripe_lmv(const struct lu_env *env,
226 struct lfsck_instance *lfsck,
227 struct dt_object *obj)
229 struct lmv_mds_md_v1 *lmv = &lfsck_env_info(env)->lti_lmv;
230 struct lfsck_lmv *llmv;
234 LASSERT(lfsck->li_obj_dir == NULL);
235 LASSERT(lfsck->li_lmv == NULL);
237 rc = lfsck_read_stripe_lmv(env, obj, lmv);
238 if (rc == -ENODATA) {
239 lfsck->li_obj_dir = lfsck_object_get(obj);
251 if (lmv->lmv_magic == LMV_MAGIC) {
252 struct lfsck_slave_lmv_rec *lslr;
255 llmv->ll_lmv_master = 1;
256 if (lmv->lmv_stripe_count < 1)
257 stripes = LFSCK_LMV_DEF_STRIPES;
258 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
259 stripes = LFSCK_LMV_MAX_STRIPES;
261 stripes = lmv->lmv_stripe_count;
263 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
270 llmv->ll_stripes_allocated = stripes;
271 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
272 llmv->ll_lslr = lslr;
274 llmv->ll_lmv_slave = 1;
277 lfsck->li_obj_dir = lfsck_object_get(obj);
279 atomic_set(&llmv->ll_ref, 1);
280 lfsck->li_lmv = llmv;
285 /* LFSCK wrap functions */
287 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
290 struct lfsck_component *com;
292 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
293 com->lc_ops->lfsck_fail(env, com, new_checked);
297 void lfsck_close_dir(const struct lu_env *env,
298 struct lfsck_instance *lfsck, int result)
300 struct lfsck_component *com;
303 if (lfsck->li_lmv != NULL) {
304 lfsck->li_lmv->ll_exit_value = result;
305 if (lfsck->li_obj_dir != NULL) {
306 list_for_each_entry(com, &lfsck->li_list_dir,
308 com->lc_ops->lfsck_close_dir(env, com);
312 lfsck_lmv_put(env, lfsck->li_lmv);
313 lfsck->li_lmv = NULL;
316 if (lfsck->li_di_dir != NULL) {
317 const struct dt_it_ops *dir_iops;
318 struct dt_it *dir_di = lfsck->li_di_dir;
320 LASSERT(lfsck->li_obj_dir != NULL);
322 dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
323 lfsck_di_dir_put(env, lfsck);
324 dir_iops->fini(env, dir_di);
327 if (lfsck->li_obj_dir != NULL) {
328 struct dt_object *dir_obj = lfsck->li_obj_dir;
330 lfsck->li_obj_dir = NULL;
331 lfsck_object_put(env, dir_obj);
337 int lfsck_open_dir(const struct lu_env *env,
338 struct lfsck_instance *lfsck, __u64 cookie)
340 struct dt_object *obj = lfsck->li_obj_dir;
341 struct dt_it *di = lfsck->li_di_dir;
342 struct lfsck_component *com;
343 const struct dt_it_ops *iops;
347 LASSERT(obj != NULL);
350 if (unlikely(!dt_try_as_dir(env, obj)))
351 GOTO(out, rc = -ENOTDIR);
353 list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
354 rc = com->lc_ops->lfsck_open_dir(env, com);
359 iops = &obj->do_index_ops->dio_it;
360 di = iops->init(env, obj, lfsck->li_args_dir);
362 GOTO(out, rc = PTR_ERR(di));
364 rc = iops->load(env, di, cookie);
367 else if (rc == 0 || (rc > 0 && cookie > 0))
368 rc = iops->next(env, di);
376 lfsck->li_cookie_dir = iops->store(env, di);
377 spin_lock(&lfsck->li_lock);
378 lfsck->li_di_dir = di;
379 spin_unlock(&lfsck->li_lock);
386 lfsck_close_dir(env, lfsck, rc);
391 static int lfsck_checkpoint(const struct lu_env *env,
392 struct lfsck_instance *lfsck)
394 struct lfsck_component *com;
398 if (likely(ktime_get_seconds() <= lfsck->li_time_next_checkpoint))
401 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
402 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
403 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
408 lfsck->li_time_last_checkpoint = ktime_get_seconds();
409 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
410 LFSCK_CHECKPOINT_INTERVAL;
411 return rc1 != 0 ? rc1 : rc;
414 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
415 struct lfsck_start_param *lsp)
417 struct dt_object *obj = NULL;
418 struct lfsck_component *com;
419 struct lfsck_component *next;
420 struct lfsck_position *pos = NULL;
421 const struct dt_it_ops *iops =
422 &lfsck->li_obj_oit->do_index_ops->dio_it;
426 LASSERT(lfsck->li_obj_dir == NULL);
427 LASSERT(lfsck->li_di_dir == NULL);
429 lfsck->li_current_oit_processed = 0;
430 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
431 com->lc_new_checked = 0;
432 rc = com->lc_ops->lfsck_prep(env, com, lsp);
437 (!lfsck_pos_is_zero(&com->lc_pos_start) &&
438 lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
439 pos = &com->lc_pos_start;
442 /* Init otable-based iterator. */
444 rc = iops->load(env, lfsck->li_di_oit, 0);
445 if (rc > 0 || unlikely(rc == -ENODATA)) {
446 lfsck->li_oit_over = 1;
453 rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
454 if (rc > 0 || unlikely(rc == -ENODATA))
455 lfsck->li_oit_over = 1;
459 if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
462 /* Find the directory for namespace-based traverse. */
463 obj = lfsck_object_find_bottom(env, lfsck, &pos->lp_dir_parent);
465 RETURN(PTR_ERR(obj));
467 /* Remote directory will be scanned by the LFSCK instance
468 * on the MDT where the remote object really resides on. */
469 if (!dt_object_exists(obj) || dt_object_remote(obj) ||
470 unlikely(!S_ISDIR(lfsck_object_type(obj))))
473 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
475 /* For the master MDT-object of a striped directory,
476 * reset the iteration from the directory beginning. */
477 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
478 pos->lp_dir_cookie = 0;
480 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
482 /* The end of the directory. */
490 lfsck_object_put(env, obj);
493 lfsck_close_dir(env, lfsck, rc);
494 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
496 com->lc_ops->lfsck_post(env, com, rc, true);
503 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
504 lfsck->li_pos_current = lfsck->li_pos_checkpoint;
505 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
506 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
511 lfsck->li_time_last_checkpoint = ktime_get_seconds();
512 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
513 LFSCK_CHECKPOINT_INTERVAL;
517 static int lfsck_exec_oit(const struct lu_env *env,
518 struct lfsck_instance *lfsck, struct dt_object *obj)
520 struct lfsck_component *com;
524 LASSERT(lfsck->li_obj_dir == NULL);
526 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
527 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
532 rc = lfsck_needs_scan_dir(env, lfsck, obj);
536 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
538 rc = lfsck_open_dir(env, lfsck, 0);
544 lfsck_fail(env, lfsck, false);
547 lfsck_close_dir(env, lfsck, rc);
549 return rc > 0 ? 0 : rc;
552 static int lfsck_exec_dir(const struct lu_env *env,
553 struct lfsck_instance *lfsck,
554 struct lfsck_assistant_object *lso,
555 struct lu_dirent *ent, __u16 type)
557 struct lfsck_component *com;
560 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
561 rc = com->lc_ops->lfsck_exec_dir(env, com, lso, ent, type);
568 static int lfsck_master_dir_engine(const struct lu_env *env,
569 struct lfsck_instance *lfsck);
571 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
574 struct lfsck_component *com;
575 struct lfsck_component *next;
578 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
579 lfsck_close_dir(env, lfsck, result);
581 while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
582 !list_empty(&lfsck->li_list_lmv)) {
583 struct lfsck_lmv_unit *llu;
585 spin_lock(&lfsck->li_lock);
586 llu = list_entry(lfsck->li_list_lmv.next,
587 struct lfsck_lmv_unit, llu_link);
588 list_del_init(&llu->llu_link);
589 spin_unlock(&lfsck->li_lock);
591 lfsck->li_lmv = &llu->llu_lmv;
592 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
593 rc = lfsck_open_dir(env, lfsck, 0);
595 rc = lfsck_master_dir_engine(env, lfsck);
596 lfsck_close_dir(env, lfsck, result);
602 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
603 rc = com->lc_ops->lfsck_post(env, com, result, false);
605 CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
606 "rc = %d\n", lfsck_lfsck2name(lfsck),
607 (__u32)com->lc_type, rc);
610 lfsck->li_time_last_checkpoint = ktime_get_seconds();
611 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
612 LFSCK_CHECKPOINT_INTERVAL;
614 /* Ignore some component post failure to make other can go ahead. */
618 static int lfsck_double_scan(const struct lu_env *env,
619 struct lfsck_instance *lfsck)
621 struct lfsck_component *com;
622 struct lfsck_component *next;
623 struct l_wait_info lwi = { 0 };
627 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
628 rc = com->lc_ops->lfsck_double_scan(env, com);
633 l_wait_event(lfsck->li_thread.t_ctl_waitq,
634 atomic_read(&lfsck->li_double_scan_count) == 0,
637 if (lfsck->li_status != LS_PAUSED &&
638 lfsck->li_status != LS_CO_PAUSED) {
639 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
641 spin_lock(&lfsck->li_lock);
642 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
643 spin_unlock(&lfsck->li_lock);
647 return rc1 != 0 ? rc1 : rc;
650 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
652 struct lfsck_component *com;
653 struct lfsck_component *next;
655 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
657 if (com->lc_ops->lfsck_quit != NULL)
658 com->lc_ops->lfsck_quit(env, com);
660 spin_lock(&lfsck->li_lock);
661 list_del_init(&com->lc_link_dir);
662 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
663 spin_unlock(&lfsck->li_lock);
666 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
668 if (com->lc_ops->lfsck_quit != NULL)
669 com->lc_ops->lfsck_quit(env, com);
671 spin_lock(&lfsck->li_lock);
672 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
673 spin_unlock(&lfsck->li_lock);
679 static int lfsck_master_dir_engine(const struct lu_env *env,
680 struct lfsck_instance *lfsck)
682 struct lfsck_thread_info *info = lfsck_env_info(env);
683 struct dt_object *dir = lfsck->li_obj_dir;
684 const struct dt_it_ops *iops = &dir->do_index_ops->dio_it;
685 struct dt_it *di = lfsck->li_di_dir;
686 struct lu_dirent *ent =
687 (struct lu_dirent *)info->lti_key;
688 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
689 struct ptlrpc_thread *thread = &lfsck->li_thread;
690 struct lfsck_assistant_object *lso = NULL;
696 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
697 unlikely(!thread_is_running(thread))) {
698 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
699 "parent "DFID", cookie %#llx\n",
700 lfsck_lfsck2name(lfsck),
701 PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
706 lfsck->li_new_scanned++;
707 rc = iops->rec(env, di, (struct dt_rec *)ent,
710 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
714 CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
715 "parent "DFID", cookie %#llx: rc = %d\n",
716 lfsck_lfsck2name(lfsck),
717 PFID(lfsck_dto2fid(dir)),
718 lfsck->li_cookie_dir, rc);
719 lfsck_fail(env, lfsck, true);
720 if (bk->lb_param & LPF_FAILOUT)
726 if (ent->lde_attrs & LUDA_IGNORE)
729 /* skip dot entry. */
730 if (ent->lde_namelen == 1 && ent->lde_name[0] == '.')
734 lso = lfsck_assistant_object_init(env,
735 lfsck_dto2fid(dir), NULL,
736 lfsck->li_pos_current.lp_oit_cookie, true);
738 if (bk->lb_param & LPF_FAILOUT)
739 RETURN(PTR_ERR(lso));
746 /* The type in the @ent structure may has been overwritten,
747 * so we need to pass the @type parameter independently. */
748 rc = lfsck_exec_dir(env, lfsck, lso, ent, type);
749 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
753 rc = lfsck_checkpoint(env, lfsck);
754 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
758 lfsck_control_speed(lfsck);
759 if (unlikely(!thread_is_running(thread))) {
760 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
761 "parent "DFID", cookie %#llx\n",
762 lfsck_lfsck2name(lfsck),
763 PFID(lfsck_dto2fid(dir)),
764 lfsck->li_cookie_dir);
768 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
769 spin_lock(&lfsck->li_lock);
770 thread_set_flags(thread, SVC_STOPPING);
771 spin_unlock(&lfsck->li_lock);
772 GOTO(out, rc = -EINVAL);
775 rc = iops->next(env, di);
777 CDEBUG(D_LFSCK, "%s dir engine fail to locate next "
778 "for the directory "DFID": rc = %d\n",
779 lfsck_lfsck2name(lfsck),
780 PFID(&lfsck->li_pos_current.lp_dir_parent), rc);
783 if (rc > 0 && !lfsck->li_oit_over)
784 lfsck_close_dir(env, lfsck, rc);
790 lfsck_assistant_object_put(env, lso);
796 * Object-table based iteration engine.
798 * Object-table based iteration is the basic linear engine to scan all the
799 * objects on current device in turn. For each object, it calls all the
800 * registered LFSCK component(s)' API to perform related consistency
803 * It flushes related LFSCK trace files to disk via making checkpoint
804 * periodically. Then if the server crashed or the LFSCK is paused, the
805 * LFSCK can resume from the latest checkpoint.
807 * It also controls the whole LFSCK speed via lfsck_control_speed() to
808 * avoid the server to become overload.
810 * \param[in] env pointer to the thread context
811 * \param[in] lfsck pointer to the lfsck instance
813 * \retval positive number if all objects have been scanned
814 * \retval 0 if the iteration is stopped or paused
815 * \retval negative error number on failure
817 static int lfsck_master_oit_engine(const struct lu_env *env,
818 struct lfsck_instance *lfsck)
820 struct lfsck_thread_info *info = lfsck_env_info(env);
821 const struct dt_it_ops *iops =
822 &lfsck->li_obj_oit->do_index_ops->dio_it;
823 struct dt_it *di = lfsck->li_di_oit;
824 struct lu_fid *fid = &info->lti_fid;
825 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
826 struct ptlrpc_thread *thread = &lfsck->li_thread;
827 struct seq_server_site *ss = lfsck_dev_site(lfsck);
828 __u32 idx = lfsck_dev_idx(lfsck);
832 if (unlikely(ss == NULL))
836 struct dt_object *target;
838 if (lfsck->li_di_dir != NULL) {
839 rc = lfsck_master_dir_engine(env, lfsck);
844 if (unlikely(lfsck->li_oit_over))
847 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
848 unlikely(!thread_is_running(thread))) {
849 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
851 lfsck_lfsck2name(lfsck), iops->store(env, di));
856 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
859 lfsck->li_current_oit_processed = 1;
861 if (!list_empty(&lfsck->li_list_lmv)) {
862 struct lfsck_lmv_unit *llu;
864 spin_lock(&lfsck->li_lock);
865 llu = list_entry(lfsck->li_list_lmv.next,
866 struct lfsck_lmv_unit, llu_link);
867 list_del_init(&llu->llu_link);
868 spin_unlock(&lfsck->li_lock);
870 lfsck->li_lmv = &llu->llu_lmv;
871 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
872 rc = lfsck_open_dir(env, lfsck, 0);
874 rc = lfsck_master_dir_engine(env, lfsck);
880 lfsck->li_new_scanned++;
881 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
882 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
884 CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
885 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
886 lfsck_fail(env, lfsck, true);
887 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
893 if (unlikely(!fid_is_sane(fid))) {
894 CDEBUG(D_LFSCK, "%s: OIT scan find invalid FID "DFID
896 lfsck_lfsck2name(lfsck), PFID(fid));
900 if (fid_is_idif(fid)) {
901 __u32 idx1 = fid_idif_ost_idx(fid);
903 LASSERT(!lfsck->li_master);
906 struct ost_id *oi = &info->lti_oi;
908 if (unlikely(idx1 != 0)) {
909 CDEBUG(D_LFSCK, "%s: invalid IDIF "DFID
910 ", not match device index %u\n",
911 lfsck_lfsck2name(lfsck),
917 /* rebuild the IDIF with index to
918 * avoid double instances for the
920 fid_to_ostid(fid, oi);
921 ostid_to_fid(fid, oi, idx);
923 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
924 !fid_is_last_id(fid) &&
925 !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
927 /* If the FID/object is only used locally and invisible
928 * to external nodes, then LFSCK will not handle it.
930 * dot_lustre sequence has been handled specially. */
933 struct lu_seq_range *range = &info->lti_range;
935 if (lfsck->li_master)
936 fld_range_set_mdt(range);
938 fld_range_set_ost(range);
939 rc = fld_local_lookup(env, ss->ss_server_fld,
940 fid_seq(fid), range);
941 if (rc != 0 || range->lsr_index != idx) {
942 /* Remote object will be handled by the LFSCK
943 * instance on the MDT where the remote object
944 * really resides on. */
950 target = lfsck_object_find_bottom(env, lfsck, fid);
951 if (IS_ERR(target)) {
952 CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
953 DFID", cookie %llu: rc = %d\n",
954 lfsck_lfsck2name(lfsck), PFID(fid),
955 iops->store(env, di), rc);
956 lfsck_fail(env, lfsck, true);
957 if (bk->lb_param & LPF_FAILOUT)
958 RETURN(PTR_ERR(target));
963 if (dt_object_exists(target)) {
964 struct lu_attr la = { .la_valid = 0 };
966 rc = dt_attr_get(env, target, &la);
967 if (likely(!rc && (!(la.la_valid & LA_FLAGS) ||
968 !(la.la_flags & LUSTRE_ORPHAN_FL))))
969 rc = lfsck_exec_oit(env, lfsck, target);
972 "%s: orphan "DFID", %llx/%x: rc = %d\n",
973 lfsck_lfsck2name(lfsck), PFID(fid),
974 la.la_valid, la.la_flags, rc);
977 lfsck_object_put(env, target);
978 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
982 rc = lfsck_checkpoint(env, lfsck);
983 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
987 lfsck_control_speed(lfsck);
989 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
990 spin_lock(&lfsck->li_lock);
991 thread_set_flags(thread, SVC_STOPPING);
992 spin_unlock(&lfsck->li_lock);
996 rc = iops->next(env, di);
997 if (unlikely(rc > 0))
998 lfsck->li_oit_over = 1;
999 else if (likely(rc == 0))
1000 lfsck->li_current_oit_processed = 0;
1002 CDEBUG(D_LFSCK, "%s oit engine fail to locate next at "
1003 "%llu: rc = %d\n", lfsck_lfsck2name(lfsck),
1004 iops->store(env, di), rc);
1006 if (unlikely(!thread_is_running(thread))) {
1007 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1008 "cookie %llu\n", lfsck_lfsck2name(lfsck),
1009 iops->store(env, di));
1012 } while (rc == 0 || lfsck->li_di_dir != NULL);
1017 int lfsck_master_engine(void *args)
1019 struct lfsck_thread_args *lta = args;
1020 struct lu_env *env = <a->lta_env;
1021 struct lfsck_instance *lfsck = lta->lta_lfsck;
1022 struct ptlrpc_thread *thread = &lfsck->li_thread;
1023 struct dt_object *oit_obj = lfsck->li_obj_oit;
1024 const struct dt_it_ops *oit_iops = &oit_obj->do_index_ops->dio_it;
1025 struct dt_it *oit_di;
1026 struct l_wait_info lwi = { 0 };
1030 spin_lock(&lfsck->li_lock);
1031 lfsck->li_task = current;
1032 spin_unlock(&lfsck->li_lock);
1034 /* There will be some objects verification during the LFSCK start,
1035 * such as the subsequent lfsck_verify_lpf(). Trigger low layer OI
1036 * OI scrub before that to handle the potential inconsistence. */
1037 oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit);
1038 if (IS_ERR(oit_di)) {
1039 rc = PTR_ERR(oit_di);
1040 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1041 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1043 GOTO(fini_args, rc);
1046 if (lfsck->li_master &&
1047 (!list_empty(&lfsck->li_list_scan) ||
1048 !list_empty(&lfsck->li_list_double_scan))) {
1049 rc = lfsck_verify_lpf(env, lfsck);
1050 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1051 * fatal, because the .lustre/lost+found/ maybe not accessed
1052 * by the LFSCK if it does not add orphans or others to such
1053 * directory. So go ahead until hit failure when really uses
1056 CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1057 ".lustre/lost+found/, go ahead: rc = %d\n",
1058 lfsck_lfsck2name(lfsck), rc);
1061 spin_lock(&lfsck->li_lock);
1062 lfsck->li_di_oit = oit_di;
1063 spin_unlock(&lfsck->li_lock);
1064 rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1068 CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1069 "oit_cookie = %llu, dir_cookie = %#llx, parent = "DFID
1070 ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1071 lfsck->li_pos_checkpoint.lp_oit_cookie,
1072 lfsck->li_pos_checkpoint.lp_dir_cookie,
1073 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1076 spin_lock(&lfsck->li_lock);
1077 if (unlikely(!thread_is_starting(thread))) {
1078 spin_unlock(&lfsck->li_lock);
1079 GOTO(fini_oit, rc = 0);
1082 thread_set_flags(thread, SVC_RUNNING);
1083 spin_unlock(&lfsck->li_lock);
1084 wake_up_all(&thread->t_ctl_waitq);
1086 l_wait_event(thread->t_ctl_waitq,
1087 lfsck->li_start_unplug ||
1088 !thread_is_running(thread),
1090 if (!thread_is_running(thread))
1091 GOTO(fini_oit, rc = 0);
1093 if (!list_empty(&lfsck->li_list_scan) ||
1094 list_empty(&lfsck->li_list_double_scan))
1095 rc = lfsck_master_oit_engine(env, lfsck);
1099 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
1100 CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1101 "oit_cookie = %llu, dir_cookie = %#llx, parent = "DFID
1102 ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1103 lfsck->li_pos_checkpoint.lp_oit_cookie,
1104 lfsck->li_pos_checkpoint.lp_dir_cookie,
1105 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1108 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1109 rc = lfsck_post(env, lfsck, rc);
1111 lfsck_close_dir(env, lfsck, rc);
1114 lfsck_di_oit_put(env, lfsck);
1115 oit_iops->fini(env, oit_di);
1117 if (!list_empty(&lfsck->li_list_double_scan))
1118 rc = lfsck_double_scan(env, lfsck);
1122 lfsck_quit(env, lfsck);
1125 /* XXX: Purge the pinned objects in the future. */
1128 spin_lock(&lfsck->li_lock);
1129 thread_set_flags(thread, SVC_STOPPED);
1130 lfsck->li_task = NULL;
1131 spin_unlock(&lfsck->li_lock);
1132 wake_up_all(&thread->t_ctl_waitq);
1133 lfsck_thread_args_fini(lta);
1137 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1141 spin_lock(&lad->lad_lock);
1142 if (list_empty(&lad->lad_req_list))
1144 spin_unlock(&lad->lad_lock);
1150 * Query the LFSCK status from the instatnces on remote servers.
1152 * The LFSCK assistant thread queries the LFSCK instances on other
1153 * servers (MDT/OST) about their status, such as whether they have
1154 * finished the phase1/phase2 scanning or not, and so on.
1156 * \param[in] env pointer to the thread context
1157 * \param[in] com pointer to the lfsck component
1159 * \retval 0 for success
1160 * \retval negative error number on failure
1162 static int lfsck_assistant_query_others(const struct lu_env *env,
1163 struct lfsck_component *com)
1165 struct lfsck_thread_info *info = lfsck_env_info(env);
1166 struct lfsck_request *lr = &info->lti_lr;
1167 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1168 struct lfsck_instance *lfsck = com->lc_lfsck;
1169 struct lfsck_assistant_data *lad = com->lc_data;
1170 struct ptlrpc_request_set *set;
1171 struct lfsck_tgt_descs *ltds;
1172 struct lfsck_tgt_desc *ltd;
1173 struct list_head *phase_head;
1178 set = ptlrpc_prep_set();
1182 lad->lad_touch_gen++;
1183 memset(lr, 0, sizeof(*lr));
1184 lr->lr_event = LE_QUERY;
1185 lr->lr_active = com->lc_type;
1187 memset(laia, 0, sizeof(*laia));
1188 laia->laia_com = com;
1191 if (!list_empty(&lad->lad_mdt_phase1_list)) {
1192 ltds = &lfsck->li_mdt_descs;
1194 phase_head = &lad->lad_mdt_phase1_list;
1195 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1200 ltds = &lfsck->li_ost_descs;
1201 lr->lr_flags = LEF_TO_OST;
1202 phase_head = &lad->lad_ost_phase1_list;
1205 laia->laia_ltds = ltds;
1206 spin_lock(<ds->ltd_lock);
1207 while (!list_empty(phase_head)) {
1208 struct list_head *phase_list;
1211 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1212 ltd = list_entry(phase_head->next,
1213 struct lfsck_tgt_desc,
1214 ltd_layout_phase_list);
1215 phase_list = <d->ltd_layout_phase_list;
1216 gen = <d->ltd_layout_gen;
1218 ltd = list_entry(phase_head->next,
1219 struct lfsck_tgt_desc,
1220 ltd_namespace_phase_list);
1221 phase_list = <d->ltd_namespace_phase_list;
1222 gen = <d->ltd_namespace_gen;
1225 if (*gen == lad->lad_touch_gen)
1228 *gen = lad->lad_touch_gen;
1229 list_move_tail(phase_list, phase_head);
1230 atomic_inc(<d->ltd_ref);
1231 laia->laia_ltd = ltd;
1232 spin_unlock(<ds->ltd_lock);
1233 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1234 lfsck_async_interpret_common,
1237 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1238 "%s %x for %s: rc = %d\n",
1239 lfsck_lfsck2name(lfsck),
1240 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1241 ltd->ltd_index, lad->lad_name, rc);
1245 spin_lock(<ds->ltd_lock);
1247 spin_unlock(<ds->ltd_lock);
1249 rc = ptlrpc_set_wait(env, set);
1251 ptlrpc_set_destroy(set);
1255 if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1256 list_empty(&lad->lad_mdt_phase1_list))
1260 ptlrpc_set_destroy(set);
1262 RETURN(rc1 != 0 ? rc1 : rc);
1266 * Notify the LFSCK event to the instances on remote servers.
1268 * The LFSCK assistant thread notifies the LFSCK instances on other
1269 * servers (MDT/OST) about some events, such as start new scanning,
1270 * stop the scanning, this LFSCK instance will exit, and so on.
1272 * \param[in] env pointer to the thread context
1273 * \param[in] com pointer to the lfsck component
1274 * \param[in] lr pointer to the LFSCK event request
1276 * \retval 0 for success
1277 * \retval negative error number on failure
1279 static int lfsck_assistant_notify_others(const struct lu_env *env,
1280 struct lfsck_component *com,
1281 struct lfsck_request *lr)
1283 struct lfsck_thread_info *info = lfsck_env_info(env);
1284 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1285 struct lfsck_instance *lfsck = com->lc_lfsck;
1286 struct lfsck_assistant_data *lad = com->lc_data;
1287 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1288 struct ptlrpc_request_set *set;
1289 struct lfsck_tgt_descs *ltds;
1290 struct lfsck_tgt_desc *ltd;
1291 struct lfsck_tgt_desc *next;
1297 set = ptlrpc_prep_set();
1301 lr->lr_index = lfsck_dev_idx(lfsck);
1302 lr->lr_active = com->lc_type;
1304 memset(laia, 0, sizeof(*laia));
1305 laia->laia_com = com;
1308 switch (lr->lr_event) {
1310 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1313 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1314 lr->lr_speed = bk->lb_speed_limit;
1315 lr->lr_version = bk->lb_version;
1316 lr->lr_param |= bk->lb_param;
1317 lr->lr_async_windows = bk->lb_async_windows;
1318 lr->lr_flags = LEF_TO_OST;
1320 /* Notify OSTs firstly, then handle other MDTs if needed. */
1321 ltds = &lfsck->li_ost_descs;
1322 laia->laia_ltds = ltds;
1323 down_read(<ds->ltd_rw_sem);
1324 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1325 ltd = lfsck_tgt_get(ltds, idx);
1326 LASSERT(ltd != NULL);
1328 laia->laia_ltd = ltd;
1329 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1330 lfsck_async_interpret_common,
1331 laia, LFSCK_NOTIFY);
1333 lfsck_lad_set_bitmap(env, com, idx);
1334 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1335 "notify OST %x for %s start: rc = %d\n",
1336 lfsck_lfsck2name(lfsck), idx,
1341 up_read(<ds->ltd_rw_sem);
1344 rc = ptlrpc_set_wait(env, set);
1346 ptlrpc_set_destroy(set);
1351 if (!(bk->lb_param & LPF_ALL_TGT))
1354 /* link other MDT targets locallly. */
1355 ltds = &lfsck->li_mdt_descs;
1356 spin_lock(<ds->ltd_lock);
1357 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1358 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1359 ltd = lfsck_ltd2tgt(ltds, idx);
1360 LASSERT(ltd != NULL);
1362 if (!list_empty(<d->ltd_layout_list))
1365 list_add_tail(<d->ltd_layout_list,
1366 &lad->lad_mdt_list);
1367 list_add_tail(<d->ltd_layout_phase_list,
1368 &lad->lad_mdt_phase1_list);
1371 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1372 ltd = lfsck_ltd2tgt(ltds, idx);
1373 LASSERT(ltd != NULL);
1375 if (!list_empty(<d->ltd_namespace_list))
1378 list_add_tail(<d->ltd_namespace_list,
1379 &lad->lad_mdt_list);
1380 list_add_tail(<d->ltd_namespace_phase_list,
1381 &lad->lad_mdt_phase1_list);
1384 spin_unlock(<ds->ltd_lock);
1387 case LE_PHASE2_DONE:
1388 case LE_PEER_EXIT: {
1389 struct list_head *phase_head;
1391 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1392 if (bk->lb_param & LPF_ALL_TGT) {
1393 phase_head = &lad->lad_mdt_list;
1394 ltds = &lfsck->li_mdt_descs;
1395 if (lr->lr_event == LE_STOP) {
1396 /* unlink other MDT targets locallly. */
1397 spin_lock(<ds->ltd_lock);
1398 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1399 list_for_each_entry_safe(ltd, next,
1400 phase_head, ltd_layout_list) {
1402 <d->ltd_layout_phase_list);
1404 <d->ltd_layout_list);
1407 list_for_each_entry_safe(ltd, next,
1409 ltd_namespace_list) {
1411 <d->ltd_namespace_phase_list);
1413 <d->ltd_namespace_list);
1416 spin_unlock(<ds->ltd_lock);
1418 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1421 lr->lr_flags |= LEF_TO_OST;
1422 phase_head = &lad->lad_ost_list;
1423 ltds = &lfsck->li_ost_descs;
1425 lr->lr_flags &= ~LEF_TO_OST;
1427 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1430 lr->lr_flags |= LEF_TO_OST;
1431 phase_head = &lad->lad_ost_list;
1432 ltds = &lfsck->li_ost_descs;
1436 laia->laia_ltds = ltds;
1437 spin_lock(<ds->ltd_lock);
1438 while (!list_empty(phase_head)) {
1439 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1440 ltd = list_entry(phase_head->next,
1441 struct lfsck_tgt_desc,
1443 if (!list_empty(<d->ltd_layout_phase_list))
1445 <d->ltd_layout_phase_list);
1446 list_del_init(<d->ltd_layout_list);
1448 ltd = list_entry(phase_head->next,
1449 struct lfsck_tgt_desc,
1450 ltd_namespace_list);
1451 if (!list_empty(<d->ltd_namespace_phase_list))
1453 <d->ltd_namespace_phase_list);
1454 list_del_init(<d->ltd_namespace_list);
1456 atomic_inc(<d->ltd_ref);
1457 laia->laia_ltd = ltd;
1458 spin_unlock(<ds->ltd_lock);
1459 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1460 lfsck_async_interpret_common,
1461 laia, LFSCK_NOTIFY);
1463 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1464 "notify %s %x for %s stop/phase2_done/"
1465 "peer_exit: rc = %d\n",
1466 lfsck_lfsck2name(lfsck),
1467 (lr->lr_flags & LEF_TO_OST) ?
1468 "OST" : "MDT", ltd->ltd_index,
1472 spin_lock(<ds->ltd_lock);
1474 spin_unlock(<ds->ltd_lock);
1476 rc = ptlrpc_set_wait(env, set);
1478 ptlrpc_set_destroy(set);
1482 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1483 !(lr->lr_flags & LEF_TO_OST)) {
1484 lr->lr_flags |= LEF_TO_OST;
1485 phase_head = &lad->lad_ost_list;
1486 ltds = &lfsck->li_ost_descs;
1491 case LE_PHASE1_DONE:
1492 lad->lad_ops->la_sync_failures(env, com, lr);
1493 lad->lad_touch_gen++;
1494 ltds = &lfsck->li_mdt_descs;
1495 laia->laia_ltds = ltds;
1496 spin_lock(<ds->ltd_lock);
1497 while (!list_empty(&lad->lad_mdt_list)) {
1498 struct list_head *list;
1501 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1502 ltd = list_entry(lad->lad_mdt_list.next,
1503 struct lfsck_tgt_desc,
1505 list = <d->ltd_layout_list;
1506 gen = <d->ltd_layout_gen;
1508 struct lfsck_namespace *ns = com->lc_file_ram;
1510 ltd = list_entry(lad->lad_mdt_list.next,
1511 struct lfsck_tgt_desc,
1512 ltd_namespace_list);
1513 list = <d->ltd_namespace_list;
1514 gen = <d->ltd_namespace_gen;
1515 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1518 if (*gen == lad->lad_touch_gen)
1521 *gen = lad->lad_touch_gen;
1522 list_move_tail(list, &lad->lad_mdt_list);
1523 if (ltd->ltd_synced_failures)
1526 atomic_inc(<d->ltd_ref);
1527 laia->laia_ltd = ltd;
1528 spin_unlock(<ds->ltd_lock);
1529 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1530 lfsck_async_interpret_common,
1531 laia, LFSCK_NOTIFY);
1533 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1534 "notify MDT %x for %s phase1 done: "
1535 "rc = %d\n", lfsck_lfsck2name(lfsck),
1536 ltd->ltd_index, lad->lad_name, rc);
1539 spin_lock(<ds->ltd_lock);
1541 spin_unlock(<ds->ltd_lock);
1544 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1545 "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1550 rc1 = ptlrpc_set_wait(env, set);
1551 ptlrpc_set_destroy(set);
1553 RETURN(rc != 0 ? rc : rc1);
1557 * The LFSCK assistant thread is triggered by the LFSCK main engine.
1558 * They co-work together as an asynchronous pipeline: the LFSCK main
1559 * engine scans the system and pre-fetches the objects, attributes,
1560 * or name entries, etc, and pushes them into the pipeline as input
1561 * requests for the LFSCK assistant thread; on the other end of the
1562 * pipeline, the LFSCK assistant thread performs the real check and
1563 * repair for every request from the main engine.
1565 * Generally, the assistant engine may be blocked when check/repair
1566 * something, so the LFSCK main engine will run some faster. On the
1567 * other hand, the LFSCK main engine will drive multiple assistant
1568 * threads in parallel, means for each LFSCK component on the master
1569 * (such as layout LFSCK, namespace LFSCK), there is an independent
1570 * LFSCK assistant thread. So under such 1:N multiple asynchronous
1571 * pipelines mode, the whole LFSCK performance will be much better
1572 * than check/repair everything by the LFSCK main engine itself.
1574 int lfsck_assistant_engine(void *args)
1576 struct lfsck_thread_args *lta = args;
1577 struct lu_env *env = <a->lta_env;
1578 struct lfsck_component *com = lta->lta_com;
1579 struct lfsck_instance *lfsck = lta->lta_lfsck;
1580 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1581 struct lfsck_position *pos = &com->lc_pos_start;
1582 struct lfsck_thread_info *info = lfsck_env_info(env);
1583 struct lfsck_request *lr = &info->lti_lr;
1584 struct lfsck_assistant_data *lad = com->lc_data;
1585 struct ptlrpc_thread *mthread = &lfsck->li_thread;
1586 struct ptlrpc_thread *athread = &lad->lad_thread;
1587 struct lfsck_assistant_operations *lao = lad->lad_ops;
1588 struct lfsck_assistant_req *lar;
1589 struct l_wait_info lwi = { 0 };
1595 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1596 lfsck_lfsck2name(lfsck), lad->lad_name);
1598 memset(lr, 0, sizeof(*lr));
1599 lr->lr_event = LE_START;
1600 if (pos->lp_oit_cookie <= 1)
1601 lr->lr_param = LPF_RESET;
1602 rc = lfsck_assistant_notify_others(env, com, lr);
1604 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1605 "to start %s: rc = %d\n",
1606 lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1610 spin_lock(&lad->lad_lock);
1611 lad->lad_task = current;
1612 thread_set_flags(athread, SVC_RUNNING);
1613 spin_unlock(&lad->lad_lock);
1614 wake_up_all(&mthread->t_ctl_waitq);
1617 while (!list_empty(&lad->lad_req_list)) {
1618 bool wakeup = false;
1620 if (unlikely(lad->lad_exit ||
1621 !thread_is_running(mthread)))
1622 GOTO(cleanup, rc = lad->lad_post_result);
1624 lar = list_entry(lad->lad_req_list.next,
1625 struct lfsck_assistant_req,
1627 /* Only the lfsck_assistant_engine thread itself can
1628 * remove the "lar" from the head of the list, LFSCK
1629 * engine thread only inserts other new "lar" at the
1630 * end of the list. So it is safe to handle current
1631 * "lar" without the spin_lock. */
1632 rc = lao->la_handler_p1(env, com, lar);
1633 spin_lock(&lad->lad_lock);
1634 list_del_init(&lar->lar_list);
1635 lad->lad_prefetched--;
1636 /* Wake up the main engine thread only when the list
1637 * is empty or half of the prefetched items have been
1638 * handled to avoid too frequent thread schedule. */
1639 if (lad->lad_prefetched <= (bk->lb_async_windows / 2))
1641 spin_unlock(&lad->lad_lock);
1643 wake_up_all(&mthread->t_ctl_waitq);
1645 lao->la_req_fini(env, lar);
1646 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1650 l_wait_event(athread->t_ctl_waitq,
1651 !lfsck_assistant_req_empty(lad) ||
1654 lad->lad_to_double_scan,
1657 if (unlikely(lad->lad_exit))
1658 GOTO(cleanup, rc = lad->lad_post_result);
1660 if (!list_empty(&lad->lad_req_list))
1663 if (lad->lad_to_post) {
1664 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1665 lfsck_lfsck2name(lfsck), lad->lad_name);
1667 if (unlikely(lad->lad_exit))
1668 GOTO(cleanup, rc = lad->lad_post_result);
1670 lad->lad_to_post = 0;
1671 LASSERT(lad->lad_post_result > 0);
1673 /* Wakeup the master engine to go ahead. */
1674 wake_up_all(&mthread->t_ctl_waitq);
1676 memset(lr, 0, sizeof(*lr));
1677 lr->lr_event = LE_PHASE1_DONE;
1678 lr->lr_status = lad->lad_post_result;
1679 rc = lfsck_assistant_notify_others(env, com, lr);
1681 CDEBUG(D_LFSCK, "%s: LFSCK assistant notified "
1682 "others for %s post: rc = %d\n",
1683 lfsck_lfsck2name(lfsck),
1687 if (lad->lad_to_double_scan) {
1688 lad->lad_to_double_scan = 0;
1689 atomic_inc(&lfsck->li_double_scan_count);
1690 lad->lad_in_double_scan = 1;
1691 wake_up_all(&mthread->t_ctl_waitq);
1693 com->lc_new_checked = 0;
1694 com->lc_new_scanned = 0;
1695 com->lc_time_last_checkpoint = ktime_get_seconds();
1696 com->lc_time_next_checkpoint =
1697 com->lc_time_last_checkpoint +
1698 LFSCK_CHECKPOINT_INTERVAL;
1700 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1701 "the second-stage scaning\n",
1702 lfsck_lfsck2name(lfsck));
1704 /* Flush async updates before handling orphan. */
1705 rc2 = dt_sync(env, lfsck->li_next);
1707 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1708 "scan start, synced: rc = %d\n",
1709 lfsck_lfsck2name(lfsck), rc2);
1711 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1712 GOTO(cleanup, rc = 0);
1714 while (lad->lad_in_double_scan) {
1715 rc = lfsck_assistant_query_others(env, com);
1716 if (lfsck_phase2_next_ready(lad))
1722 /* Pull LFSCK status on related targets once
1723 * per 30 seconds if we are not notified. */
1724 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1725 cfs_time_seconds(1),
1727 rc = l_wait_event(athread->t_ctl_waitq,
1728 lfsck_phase2_next_ready(lad) ||
1730 !thread_is_running(mthread),
1733 if (unlikely(lad->lad_exit ||
1734 !thread_is_running(mthread)))
1735 GOTO(cleanup, rc = 0);
1737 if (rc == -ETIMEDOUT)
1744 rc = lao->la_handler_p2(env, com);
1748 if (unlikely(lad->lad_exit ||
1749 !thread_is_running(mthread)))
1750 GOTO(cleanup, rc = 0);
1756 /* Cleanup the unfinished requests. */
1757 spin_lock(&lad->lad_lock);
1759 lad->lad_assistant_status = rc;
1761 if (lad->lad_exit && lad->lad_post_result <= 0)
1762 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1764 thread_set_flags(athread, SVC_STOPPING);
1765 while (!list_empty(&lad->lad_req_list)) {
1766 lar = list_entry(lad->lad_req_list.next,
1767 struct lfsck_assistant_req,
1769 list_del_init(&lar->lar_list);
1770 lad->lad_prefetched--;
1771 spin_unlock(&lad->lad_lock);
1772 lao->la_req_fini(env, lar);
1773 spin_lock(&lad->lad_lock);
1775 spin_unlock(&lad->lad_lock);
1777 memset(lr, 0, sizeof(*lr));
1779 lr->lr_event = LE_PHASE2_DONE;
1781 } else if (rc == 0) {
1782 if (lfsck->li_flags & LPF_ALL_TGT) {
1783 lr->lr_event = LE_STOP;
1784 lr->lr_status = LS_STOPPED;
1786 lr->lr_event = LE_PEER_EXIT;
1787 switch (lfsck->li_status) {
1790 lr->lr_status = LS_CO_PAUSED;
1794 lr->lr_status = LS_CO_STOPPED;
1797 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1798 "status: rc = %d\n",
1799 lfsck_lfsck2name(lfsck),
1801 lr->lr_status = LS_CO_FAILED;
1806 if (lfsck->li_flags & LPF_ALL_TGT) {
1807 lr->lr_event = LE_STOP;
1808 lr->lr_status = LS_FAILED;
1810 lr->lr_event = LE_PEER_EXIT;
1811 lr->lr_status = LS_CO_FAILED;
1815 rc1 = lfsck_assistant_notify_others(env, com, lr);
1817 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1818 "others for %s quit: rc = %d\n",
1819 lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1823 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1824 lfsck_lfsck2name(lfsck));
1826 /* Flush async updates before exit. */
1827 rc2 = dt_sync(env, lfsck->li_next);
1829 CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1830 lfsck_lfsck2name(lfsck), rc2);
1832 /* Under force exit case, some requests may be just freed without
1833 * verification, those objects should be re-handled when next run.
1834 * So not update the on-disk trace file under such case. */
1835 if (lad->lad_in_double_scan) {
1837 rc1 = lao->la_double_scan_result(env, com, rc);
1839 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1840 "finished: rc = %d\n",
1841 lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1845 if (lad->lad_in_double_scan)
1846 atomic_dec(&lfsck->li_double_scan_count);
1848 spin_lock(&lad->lad_lock);
1849 lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1850 thread_set_flags(athread, SVC_STOPPED);
1851 lad->lad_task = NULL;
1852 spin_unlock(&lad->lad_lock);
1854 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1855 lfsck_lfsck2name(lfsck), lad->lad_name,
1856 lad->lad_assistant_status);
1858 lfsck_thread_args_fini(lta);
1859 wake_up_all(&mthread->t_ctl_waitq);