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(cfs_time_beforeq(cfs_time_current(),
399 lfsck->li_time_next_checkpoint)))
402 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
403 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
404 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
409 lfsck->li_time_last_checkpoint = cfs_time_current();
410 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
411 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
412 return rc1 != 0 ? rc1 : rc;
415 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
416 struct lfsck_start_param *lsp)
418 struct dt_object *obj = NULL;
419 struct lfsck_component *com;
420 struct lfsck_component *next;
421 struct lfsck_position *pos = NULL;
422 const struct dt_it_ops *iops =
423 &lfsck->li_obj_oit->do_index_ops->dio_it;
427 LASSERT(lfsck->li_obj_dir == NULL);
428 LASSERT(lfsck->li_di_dir == NULL);
430 lfsck->li_current_oit_processed = 0;
431 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
432 com->lc_new_checked = 0;
433 rc = com->lc_ops->lfsck_prep(env, com, lsp);
438 (!lfsck_pos_is_zero(&com->lc_pos_start) &&
439 lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
440 pos = &com->lc_pos_start;
443 /* Init otable-based iterator. */
445 rc = iops->load(env, lfsck->li_di_oit, 0);
446 if (rc > 0 || unlikely(rc == -ENODATA)) {
447 lfsck->li_oit_over = 1;
454 rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
455 if (rc > 0 || unlikely(rc == -ENODATA))
456 lfsck->li_oit_over = 1;
460 if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
463 /* Find the directory for namespace-based traverse. */
464 obj = lfsck_object_find_bottom(env, lfsck, &pos->lp_dir_parent);
466 RETURN(PTR_ERR(obj));
468 /* Remote directory will be scanned by the LFSCK instance
469 * on the MDT where the remote object really resides on. */
470 if (!dt_object_exists(obj) || dt_object_remote(obj) ||
471 unlikely(!S_ISDIR(lfsck_object_type(obj))))
474 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
476 /* For the master MDT-object of a striped directory,
477 * reset the iteration from the directory beginning. */
478 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
479 pos->lp_dir_cookie = 0;
481 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
483 /* The end of the directory. */
491 lfsck_object_put(env, obj);
494 lfsck_close_dir(env, lfsck, rc);
495 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
497 com->lc_ops->lfsck_post(env, com, rc, true);
504 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
505 lfsck->li_pos_current = lfsck->li_pos_checkpoint;
506 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
507 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
512 lfsck->li_time_last_checkpoint = cfs_time_current();
513 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
514 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
518 static int lfsck_exec_oit(const struct lu_env *env,
519 struct lfsck_instance *lfsck, struct dt_object *obj)
521 struct lfsck_component *com;
525 LASSERT(lfsck->li_obj_dir == NULL);
527 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
528 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
533 rc = lfsck_needs_scan_dir(env, lfsck, obj);
537 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
539 rc = lfsck_open_dir(env, lfsck, 0);
545 lfsck_fail(env, lfsck, false);
548 lfsck_close_dir(env, lfsck, rc);
550 return rc > 0 ? 0 : rc;
553 static int lfsck_exec_dir(const struct lu_env *env,
554 struct lfsck_instance *lfsck,
555 struct lfsck_assistant_object *lso,
556 struct lu_dirent *ent, __u16 type)
558 struct lfsck_component *com;
561 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
562 rc = com->lc_ops->lfsck_exec_dir(env, com, lso, ent, type);
569 static int lfsck_master_dir_engine(const struct lu_env *env,
570 struct lfsck_instance *lfsck);
572 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
575 struct lfsck_component *com;
576 struct lfsck_component *next;
579 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
580 lfsck_close_dir(env, lfsck, result);
582 while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
583 !list_empty(&lfsck->li_list_lmv)) {
584 struct lfsck_lmv_unit *llu;
586 spin_lock(&lfsck->li_lock);
587 llu = list_entry(lfsck->li_list_lmv.next,
588 struct lfsck_lmv_unit, llu_link);
589 list_del_init(&llu->llu_link);
590 spin_unlock(&lfsck->li_lock);
592 lfsck->li_lmv = &llu->llu_lmv;
593 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
594 rc = lfsck_open_dir(env, lfsck, 0);
596 rc = lfsck_master_dir_engine(env, lfsck);
597 lfsck_close_dir(env, lfsck, result);
603 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
604 rc = com->lc_ops->lfsck_post(env, com, result, false);
606 CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
607 "rc = %d\n", lfsck_lfsck2name(lfsck),
608 (__u32)com->lc_type, rc);
611 lfsck->li_time_last_checkpoint = cfs_time_current();
612 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
613 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
615 /* Ignore some component post failure to make other can go ahead. */
619 static int lfsck_double_scan(const struct lu_env *env,
620 struct lfsck_instance *lfsck)
622 struct lfsck_component *com;
623 struct lfsck_component *next;
624 struct l_wait_info lwi = { 0 };
628 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
629 rc = com->lc_ops->lfsck_double_scan(env, com);
634 l_wait_event(lfsck->li_thread.t_ctl_waitq,
635 atomic_read(&lfsck->li_double_scan_count) == 0,
638 if (lfsck->li_status != LS_PAUSED &&
639 lfsck->li_status != LS_CO_PAUSED) {
640 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
642 spin_lock(&lfsck->li_lock);
643 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
644 spin_unlock(&lfsck->li_lock);
648 return rc1 != 0 ? rc1 : rc;
651 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
653 struct lfsck_component *com;
654 struct lfsck_component *next;
656 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
658 if (com->lc_ops->lfsck_quit != NULL)
659 com->lc_ops->lfsck_quit(env, com);
661 spin_lock(&lfsck->li_lock);
662 list_del_init(&com->lc_link_dir);
663 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
664 spin_unlock(&lfsck->li_lock);
667 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
669 if (com->lc_ops->lfsck_quit != NULL)
670 com->lc_ops->lfsck_quit(env, com);
672 spin_lock(&lfsck->li_lock);
673 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
674 spin_unlock(&lfsck->li_lock);
680 static int lfsck_master_dir_engine(const struct lu_env *env,
681 struct lfsck_instance *lfsck)
683 struct lfsck_thread_info *info = lfsck_env_info(env);
684 struct dt_object *dir = lfsck->li_obj_dir;
685 const struct dt_it_ops *iops = &dir->do_index_ops->dio_it;
686 struct dt_it *di = lfsck->li_di_dir;
687 struct lu_dirent *ent =
688 (struct lu_dirent *)info->lti_key;
689 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
690 struct ptlrpc_thread *thread = &lfsck->li_thread;
691 struct lfsck_assistant_object *lso = NULL;
697 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
698 unlikely(!thread_is_running(thread))) {
699 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
700 "parent "DFID", cookie %#llx\n",
701 lfsck_lfsck2name(lfsck),
702 PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
707 lfsck->li_new_scanned++;
708 rc = iops->rec(env, di, (struct dt_rec *)ent,
711 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
715 CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
716 "parent "DFID", cookie %#llx: rc = %d\n",
717 lfsck_lfsck2name(lfsck),
718 PFID(lfsck_dto2fid(dir)),
719 lfsck->li_cookie_dir, rc);
720 lfsck_fail(env, lfsck, true);
721 if (bk->lb_param & LPF_FAILOUT)
727 if (ent->lde_attrs & LUDA_IGNORE)
730 /* skip dot entry. */
731 if (ent->lde_namelen == 1 && ent->lde_name[0] == '.')
735 lso = lfsck_assistant_object_init(env,
736 lfsck_dto2fid(dir), NULL,
737 lfsck->li_pos_current.lp_oit_cookie, true);
739 if (bk->lb_param & LPF_FAILOUT)
740 RETURN(PTR_ERR(lso));
747 /* The type in the @ent structure may has been overwritten,
748 * so we need to pass the @type parameter independently. */
749 rc = lfsck_exec_dir(env, lfsck, lso, ent, type);
750 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
754 rc = lfsck_checkpoint(env, lfsck);
755 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
759 lfsck_control_speed(lfsck);
760 if (unlikely(!thread_is_running(thread))) {
761 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
762 "parent "DFID", cookie %#llx\n",
763 lfsck_lfsck2name(lfsck),
764 PFID(lfsck_dto2fid(dir)),
765 lfsck->li_cookie_dir);
769 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
770 spin_lock(&lfsck->li_lock);
771 thread_set_flags(thread, SVC_STOPPING);
772 spin_unlock(&lfsck->li_lock);
773 GOTO(out, rc = -EINVAL);
776 rc = iops->next(env, di);
778 CDEBUG(D_LFSCK, "%s dir engine fail to locate next "
779 "for the directory "DFID": rc = %d\n",
780 lfsck_lfsck2name(lfsck),
781 PFID(&lfsck->li_pos_current.lp_dir_parent), rc);
784 if (rc > 0 && !lfsck->li_oit_over)
785 lfsck_close_dir(env, lfsck, rc);
791 lfsck_assistant_object_put(env, lso);
797 * Object-table based iteration engine.
799 * Object-table based iteration is the basic linear engine to scan all the
800 * objects on current device in turn. For each object, it calls all the
801 * registered LFSCK component(s)' API to perform related consistency
804 * It flushes related LFSCK trace files to disk via making checkpoint
805 * periodically. Then if the server crashed or the LFSCK is paused, the
806 * LFSCK can resume from the latest checkpoint.
808 * It also controls the whole LFSCK speed via lfsck_control_speed() to
809 * avoid the server to become overload.
811 * \param[in] env pointer to the thread context
812 * \param[in] lfsck pointer to the lfsck instance
814 * \retval positive number if all objects have been scanned
815 * \retval 0 if the iteration is stopped or paused
816 * \retval negative error number on failure
818 static int lfsck_master_oit_engine(const struct lu_env *env,
819 struct lfsck_instance *lfsck)
821 struct lfsck_thread_info *info = lfsck_env_info(env);
822 const struct dt_it_ops *iops =
823 &lfsck->li_obj_oit->do_index_ops->dio_it;
824 struct dt_it *di = lfsck->li_di_oit;
825 struct lu_fid *fid = &info->lti_fid;
826 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
827 struct ptlrpc_thread *thread = &lfsck->li_thread;
828 struct seq_server_site *ss = lfsck_dev_site(lfsck);
829 __u32 idx = lfsck_dev_idx(lfsck);
833 if (unlikely(ss == NULL))
837 struct dt_object *target;
839 if (lfsck->li_di_dir != NULL) {
840 rc = lfsck_master_dir_engine(env, lfsck);
845 if (unlikely(lfsck->li_oit_over))
848 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
849 unlikely(!thread_is_running(thread))) {
850 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
852 lfsck_lfsck2name(lfsck), iops->store(env, di));
857 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
860 lfsck->li_current_oit_processed = 1;
862 if (!list_empty(&lfsck->li_list_lmv)) {
863 struct lfsck_lmv_unit *llu;
865 spin_lock(&lfsck->li_lock);
866 llu = list_entry(lfsck->li_list_lmv.next,
867 struct lfsck_lmv_unit, llu_link);
868 list_del_init(&llu->llu_link);
869 spin_unlock(&lfsck->li_lock);
871 lfsck->li_lmv = &llu->llu_lmv;
872 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
873 rc = lfsck_open_dir(env, lfsck, 0);
875 rc = lfsck_master_dir_engine(env, lfsck);
881 lfsck->li_new_scanned++;
882 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
883 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
885 CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
886 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
887 lfsck_fail(env, lfsck, true);
888 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
894 if (unlikely(!fid_is_sane(fid))) {
895 CDEBUG(D_LFSCK, "%s: OIT scan find invalid FID "DFID
897 lfsck_lfsck2name(lfsck), PFID(fid));
901 if (fid_is_idif(fid)) {
902 __u32 idx1 = fid_idif_ost_idx(fid);
904 LASSERT(!lfsck->li_master);
907 struct ost_id *oi = &info->lti_oi;
909 if (unlikely(idx1 != 0)) {
910 CDEBUG(D_LFSCK, "%s: invalid IDIF "DFID
911 ", not match device index %u\n",
912 lfsck_lfsck2name(lfsck),
918 /* rebuild the IDIF with index to
919 * avoid double instances for the
921 fid_to_ostid(fid, oi);
922 ostid_to_fid(fid, oi, idx);
924 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
925 !fid_is_last_id(fid) &&
926 !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
928 /* If the FID/object is only used locally and invisible
929 * to external nodes, then LFSCK will not handle it.
931 * dot_lustre sequence has been handled specially. */
934 struct lu_seq_range *range = &info->lti_range;
936 if (lfsck->li_master)
937 fld_range_set_mdt(range);
939 fld_range_set_ost(range);
940 rc = fld_local_lookup(env, ss->ss_server_fld,
941 fid_seq(fid), range);
942 if (rc != 0 || range->lsr_index != idx) {
943 /* Remote object will be handled by the LFSCK
944 * instance on the MDT where the remote object
945 * really resides on. */
951 target = lfsck_object_find_bottom(env, lfsck, fid);
952 if (IS_ERR(target)) {
953 CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
954 DFID", cookie %llu: rc = %d\n",
955 lfsck_lfsck2name(lfsck), PFID(fid),
956 iops->store(env, di), rc);
957 lfsck_fail(env, lfsck, true);
958 if (bk->lb_param & LPF_FAILOUT)
959 RETURN(PTR_ERR(target));
964 if (dt_object_exists(target))
965 rc = lfsck_exec_oit(env, lfsck, target);
967 lfsck_object_put(env, target);
968 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
972 rc = lfsck_checkpoint(env, lfsck);
973 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
977 lfsck_control_speed(lfsck);
979 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
980 spin_lock(&lfsck->li_lock);
981 thread_set_flags(thread, SVC_STOPPING);
982 spin_unlock(&lfsck->li_lock);
986 rc = iops->next(env, di);
987 if (unlikely(rc > 0))
988 lfsck->li_oit_over = 1;
989 else if (likely(rc == 0))
990 lfsck->li_current_oit_processed = 0;
992 CDEBUG(D_LFSCK, "%s oit engine fail to locate next at "
993 "%llu: rc = %d\n", lfsck_lfsck2name(lfsck),
994 iops->store(env, di), rc);
996 if (unlikely(!thread_is_running(thread))) {
997 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
998 "cookie %llu\n", lfsck_lfsck2name(lfsck),
999 iops->store(env, di));
1002 } while (rc == 0 || lfsck->li_di_dir != NULL);
1007 int lfsck_master_engine(void *args)
1009 struct lfsck_thread_args *lta = args;
1010 struct lu_env *env = <a->lta_env;
1011 struct lfsck_instance *lfsck = lta->lta_lfsck;
1012 struct ptlrpc_thread *thread = &lfsck->li_thread;
1013 struct dt_object *oit_obj = lfsck->li_obj_oit;
1014 const struct dt_it_ops *oit_iops = &oit_obj->do_index_ops->dio_it;
1015 struct dt_it *oit_di;
1016 struct l_wait_info lwi = { 0 };
1020 /* There will be some objects verification during the LFSCK start,
1021 * such as the subsequent lfsck_verify_lpf(). Trigger low layer OI
1022 * OI scrub before that to handle the potential inconsistence. */
1023 oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit);
1024 if (IS_ERR(oit_di)) {
1025 rc = PTR_ERR(oit_di);
1026 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1027 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1029 GOTO(fini_args, rc);
1032 if (lfsck->li_master &&
1033 (!list_empty(&lfsck->li_list_scan) ||
1034 !list_empty(&lfsck->li_list_double_scan))) {
1035 rc = lfsck_verify_lpf(env, lfsck);
1036 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1037 * fatal, because the .lustre/lost+found/ maybe not accessed
1038 * by the LFSCK if it does not add orphans or others to such
1039 * directory. So go ahead until hit failure when really uses
1042 CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1043 ".lustre/lost+found/, go ahead: rc = %d\n",
1044 lfsck_lfsck2name(lfsck), rc);
1047 spin_lock(&lfsck->li_lock);
1048 lfsck->li_di_oit = oit_di;
1049 spin_unlock(&lfsck->li_lock);
1050 rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1054 CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1055 "oit_cookie = %llu, dir_cookie = %#llx, parent = "DFID
1056 ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1057 lfsck->li_pos_checkpoint.lp_oit_cookie,
1058 lfsck->li_pos_checkpoint.lp_dir_cookie,
1059 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1062 spin_lock(&lfsck->li_lock);
1063 thread_set_flags(thread, SVC_RUNNING);
1064 spin_unlock(&lfsck->li_lock);
1065 wake_up_all(&thread->t_ctl_waitq);
1067 l_wait_event(thread->t_ctl_waitq,
1068 lfsck->li_start_unplug ||
1069 !thread_is_running(thread),
1071 if (!thread_is_running(thread))
1072 GOTO(fini_oit, rc = 0);
1074 if (!list_empty(&lfsck->li_list_scan) ||
1075 list_empty(&lfsck->li_list_double_scan))
1076 rc = lfsck_master_oit_engine(env, lfsck);
1080 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
1081 CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1082 "oit_cookie = %llu, dir_cookie = %#llx, parent = "DFID
1083 ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1084 lfsck->li_pos_checkpoint.lp_oit_cookie,
1085 lfsck->li_pos_checkpoint.lp_dir_cookie,
1086 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1089 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1090 rc = lfsck_post(env, lfsck, rc);
1092 lfsck_close_dir(env, lfsck, rc);
1095 lfsck_di_oit_put(env, lfsck);
1096 oit_iops->fini(env, oit_di);
1098 if (!list_empty(&lfsck->li_list_double_scan))
1099 rc = lfsck_double_scan(env, lfsck);
1103 lfsck_quit(env, lfsck);
1106 /* XXX: Purge the pinned objects in the future. */
1109 spin_lock(&lfsck->li_lock);
1110 thread_set_flags(thread, SVC_STOPPED);
1111 spin_unlock(&lfsck->li_lock);
1112 wake_up_all(&thread->t_ctl_waitq);
1113 lfsck_thread_args_fini(lta);
1117 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1121 spin_lock(&lad->lad_lock);
1122 if (list_empty(&lad->lad_req_list))
1124 spin_unlock(&lad->lad_lock);
1130 * Query the LFSCK status from the instatnces on remote servers.
1132 * The LFSCK assistant thread queries the LFSCK instances on other
1133 * servers (MDT/OST) about their status, such as whether they have
1134 * finished the phase1/phase2 scanning or not, and so on.
1136 * \param[in] env pointer to the thread context
1137 * \param[in] com pointer to the lfsck component
1139 * \retval 0 for success
1140 * \retval negative error number on failure
1142 static int lfsck_assistant_query_others(const struct lu_env *env,
1143 struct lfsck_component *com)
1145 struct lfsck_thread_info *info = lfsck_env_info(env);
1146 struct lfsck_request *lr = &info->lti_lr;
1147 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1148 struct lfsck_instance *lfsck = com->lc_lfsck;
1149 struct lfsck_assistant_data *lad = com->lc_data;
1150 struct ptlrpc_request_set *set;
1151 struct lfsck_tgt_descs *ltds;
1152 struct lfsck_tgt_desc *ltd;
1153 struct list_head *phase_head;
1158 set = ptlrpc_prep_set();
1162 lad->lad_touch_gen++;
1163 memset(lr, 0, sizeof(*lr));
1164 lr->lr_event = LE_QUERY;
1165 lr->lr_active = com->lc_type;
1167 memset(laia, 0, sizeof(*laia));
1168 laia->laia_com = com;
1171 if (!list_empty(&lad->lad_mdt_phase1_list)) {
1172 ltds = &lfsck->li_mdt_descs;
1174 phase_head = &lad->lad_mdt_phase1_list;
1175 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1180 ltds = &lfsck->li_ost_descs;
1181 lr->lr_flags = LEF_TO_OST;
1182 phase_head = &lad->lad_ost_phase1_list;
1185 laia->laia_ltds = ltds;
1186 spin_lock(<ds->ltd_lock);
1187 while (!list_empty(phase_head)) {
1188 struct list_head *phase_list;
1191 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1192 ltd = list_entry(phase_head->next,
1193 struct lfsck_tgt_desc,
1194 ltd_layout_phase_list);
1195 phase_list = <d->ltd_layout_phase_list;
1196 gen = <d->ltd_layout_gen;
1198 ltd = list_entry(phase_head->next,
1199 struct lfsck_tgt_desc,
1200 ltd_namespace_phase_list);
1201 phase_list = <d->ltd_namespace_phase_list;
1202 gen = <d->ltd_namespace_gen;
1205 if (*gen == lad->lad_touch_gen)
1208 *gen = lad->lad_touch_gen;
1209 list_move_tail(phase_list, phase_head);
1210 atomic_inc(<d->ltd_ref);
1211 laia->laia_ltd = ltd;
1212 spin_unlock(<ds->ltd_lock);
1213 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1214 lfsck_async_interpret_common,
1217 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1218 "%s %x for %s: rc = %d\n",
1219 lfsck_lfsck2name(lfsck),
1220 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1221 ltd->ltd_index, lad->lad_name, rc);
1225 spin_lock(<ds->ltd_lock);
1227 spin_unlock(<ds->ltd_lock);
1229 rc = ptlrpc_set_wait(set);
1231 ptlrpc_set_destroy(set);
1235 if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1236 list_empty(&lad->lad_mdt_phase1_list))
1240 ptlrpc_set_destroy(set);
1242 RETURN(rc1 != 0 ? rc1 : rc);
1246 * Notify the LFSCK event to the instances on remote servers.
1248 * The LFSCK assistant thread notifies the LFSCK instances on other
1249 * servers (MDT/OST) about some events, such as start new scanning,
1250 * stop the scanning, this LFSCK instance will exit, and so on.
1252 * \param[in] env pointer to the thread context
1253 * \param[in] com pointer to the lfsck component
1254 * \param[in] lr pointer to the LFSCK event request
1256 * \retval 0 for success
1257 * \retval negative error number on failure
1259 static int lfsck_assistant_notify_others(const struct lu_env *env,
1260 struct lfsck_component *com,
1261 struct lfsck_request *lr)
1263 struct lfsck_thread_info *info = lfsck_env_info(env);
1264 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1265 struct lfsck_instance *lfsck = com->lc_lfsck;
1266 struct lfsck_assistant_data *lad = com->lc_data;
1267 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1268 struct ptlrpc_request_set *set;
1269 struct lfsck_tgt_descs *ltds;
1270 struct lfsck_tgt_desc *ltd;
1271 struct lfsck_tgt_desc *next;
1277 set = ptlrpc_prep_set();
1281 lr->lr_index = lfsck_dev_idx(lfsck);
1282 lr->lr_active = com->lc_type;
1284 memset(laia, 0, sizeof(*laia));
1285 laia->laia_com = com;
1288 switch (lr->lr_event) {
1290 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1293 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1294 lr->lr_speed = bk->lb_speed_limit;
1295 lr->lr_version = bk->lb_version;
1296 lr->lr_param |= bk->lb_param;
1297 lr->lr_async_windows = bk->lb_async_windows;
1298 lr->lr_flags = LEF_TO_OST;
1300 /* Notify OSTs firstly, then handle other MDTs if needed. */
1301 ltds = &lfsck->li_ost_descs;
1302 laia->laia_ltds = ltds;
1303 down_read(<ds->ltd_rw_sem);
1304 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1305 ltd = lfsck_tgt_get(ltds, idx);
1306 LASSERT(ltd != NULL);
1308 laia->laia_ltd = ltd;
1309 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1310 lfsck_async_interpret_common,
1311 laia, LFSCK_NOTIFY);
1313 lfsck_lad_set_bitmap(env, com, idx);
1314 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1315 "notify OST %x for %s start: rc = %d\n",
1316 lfsck_lfsck2name(lfsck), idx,
1321 up_read(<ds->ltd_rw_sem);
1324 rc = ptlrpc_set_wait(set);
1326 ptlrpc_set_destroy(set);
1331 if (!(bk->lb_param & LPF_ALL_TGT))
1334 /* link other MDT targets locallly. */
1335 ltds = &lfsck->li_mdt_descs;
1336 spin_lock(<ds->ltd_lock);
1337 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1338 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1339 ltd = lfsck_ltd2tgt(ltds, idx);
1340 LASSERT(ltd != NULL);
1342 if (!list_empty(<d->ltd_layout_list))
1345 list_add_tail(<d->ltd_layout_list,
1346 &lad->lad_mdt_list);
1347 list_add_tail(<d->ltd_layout_phase_list,
1348 &lad->lad_mdt_phase1_list);
1351 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1352 ltd = lfsck_ltd2tgt(ltds, idx);
1353 LASSERT(ltd != NULL);
1355 if (!list_empty(<d->ltd_namespace_list))
1358 list_add_tail(<d->ltd_namespace_list,
1359 &lad->lad_mdt_list);
1360 list_add_tail(<d->ltd_namespace_phase_list,
1361 &lad->lad_mdt_phase1_list);
1364 spin_unlock(<ds->ltd_lock);
1367 case LE_PHASE2_DONE:
1368 case LE_PEER_EXIT: {
1369 struct list_head *phase_head;
1371 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1372 if (bk->lb_param & LPF_ALL_TGT) {
1373 phase_head = &lad->lad_mdt_list;
1374 ltds = &lfsck->li_mdt_descs;
1375 if (lr->lr_event == LE_STOP) {
1376 /* unlink other MDT targets locallly. */
1377 spin_lock(<ds->ltd_lock);
1378 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1379 list_for_each_entry_safe(ltd, next,
1380 phase_head, ltd_layout_list) {
1382 <d->ltd_layout_phase_list);
1384 <d->ltd_layout_list);
1387 list_for_each_entry_safe(ltd, next,
1389 ltd_namespace_list) {
1391 <d->ltd_namespace_phase_list);
1393 <d->ltd_namespace_list);
1396 spin_unlock(<ds->ltd_lock);
1398 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1401 lr->lr_flags |= LEF_TO_OST;
1402 phase_head = &lad->lad_ost_list;
1403 ltds = &lfsck->li_ost_descs;
1405 lr->lr_flags &= ~LEF_TO_OST;
1407 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1410 lr->lr_flags |= LEF_TO_OST;
1411 phase_head = &lad->lad_ost_list;
1412 ltds = &lfsck->li_ost_descs;
1416 laia->laia_ltds = ltds;
1417 spin_lock(<ds->ltd_lock);
1418 while (!list_empty(phase_head)) {
1419 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1420 ltd = list_entry(phase_head->next,
1421 struct lfsck_tgt_desc,
1423 if (!list_empty(<d->ltd_layout_phase_list))
1425 <d->ltd_layout_phase_list);
1426 list_del_init(<d->ltd_layout_list);
1428 ltd = list_entry(phase_head->next,
1429 struct lfsck_tgt_desc,
1430 ltd_namespace_list);
1431 if (!list_empty(<d->ltd_namespace_phase_list))
1433 <d->ltd_namespace_phase_list);
1434 list_del_init(<d->ltd_namespace_list);
1436 atomic_inc(<d->ltd_ref);
1437 laia->laia_ltd = ltd;
1438 spin_unlock(<ds->ltd_lock);
1439 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1440 lfsck_async_interpret_common,
1441 laia, LFSCK_NOTIFY);
1443 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1444 "notify %s %x for %s stop/phase2_done/"
1445 "peer_exit: rc = %d\n",
1446 lfsck_lfsck2name(lfsck),
1447 (lr->lr_flags & LEF_TO_OST) ?
1448 "OST" : "MDT", ltd->ltd_index,
1452 spin_lock(<ds->ltd_lock);
1454 spin_unlock(<ds->ltd_lock);
1456 rc = ptlrpc_set_wait(set);
1458 ptlrpc_set_destroy(set);
1462 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1463 !(lr->lr_flags & LEF_TO_OST)) {
1464 lr->lr_flags |= LEF_TO_OST;
1465 phase_head = &lad->lad_ost_list;
1466 ltds = &lfsck->li_ost_descs;
1471 case LE_PHASE1_DONE:
1472 lad->lad_ops->la_sync_failures(env, com, lr);
1473 lad->lad_touch_gen++;
1474 ltds = &lfsck->li_mdt_descs;
1475 laia->laia_ltds = ltds;
1476 spin_lock(<ds->ltd_lock);
1477 while (!list_empty(&lad->lad_mdt_list)) {
1478 struct list_head *list;
1481 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1482 ltd = list_entry(lad->lad_mdt_list.next,
1483 struct lfsck_tgt_desc,
1485 list = <d->ltd_layout_list;
1486 gen = <d->ltd_layout_gen;
1488 struct lfsck_namespace *ns = com->lc_file_ram;
1490 ltd = list_entry(lad->lad_mdt_list.next,
1491 struct lfsck_tgt_desc,
1492 ltd_namespace_list);
1493 list = <d->ltd_namespace_list;
1494 gen = <d->ltd_namespace_gen;
1495 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1498 if (*gen == lad->lad_touch_gen)
1501 *gen = lad->lad_touch_gen;
1502 list_move_tail(list, &lad->lad_mdt_list);
1503 if (ltd->ltd_synced_failures)
1506 atomic_inc(<d->ltd_ref);
1507 laia->laia_ltd = ltd;
1508 spin_unlock(<ds->ltd_lock);
1509 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1510 lfsck_async_interpret_common,
1511 laia, LFSCK_NOTIFY);
1513 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1514 "notify MDT %x for %s phase1 done: "
1515 "rc = %d\n", lfsck_lfsck2name(lfsck),
1516 ltd->ltd_index, lad->lad_name, rc);
1519 spin_lock(<ds->ltd_lock);
1521 spin_unlock(<ds->ltd_lock);
1524 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1525 "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1530 rc1 = ptlrpc_set_wait(set);
1531 ptlrpc_set_destroy(set);
1533 RETURN(rc != 0 ? rc : rc1);
1537 * The LFSCK assistant thread is triggered by the LFSCK main engine.
1538 * They co-work together as an asynchronous pipeline: the LFSCK main
1539 * engine scans the system and pre-fetches the objects, attributes,
1540 * or name entries, etc, and pushes them into the pipeline as input
1541 * requests for the LFSCK assistant thread; on the other end of the
1542 * pipeline, the LFSCK assistant thread performs the real check and
1543 * repair for every request from the main engine.
1545 * Generally, the assistant engine may be blocked when check/repair
1546 * something, so the LFSCK main engine will run some faster. On the
1547 * other hand, the LFSCK main engine will drive multiple assistant
1548 * threads in parallel, means for each LFSCK component on the master
1549 * (such as layout LFSCK, namespace LFSCK), there is an independent
1550 * LFSCK assistant thread. So under such 1:N multiple asynchronous
1551 * pipelines mode, the whole LFSCK performance will be much better
1552 * than check/repair everything by the LFSCK main engine itself.
1554 int lfsck_assistant_engine(void *args)
1556 struct lfsck_thread_args *lta = args;
1557 struct lu_env *env = <a->lta_env;
1558 struct lfsck_component *com = lta->lta_com;
1559 struct lfsck_instance *lfsck = lta->lta_lfsck;
1560 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1561 struct lfsck_position *pos = &com->lc_pos_start;
1562 struct lfsck_thread_info *info = lfsck_env_info(env);
1563 struct lfsck_request *lr = &info->lti_lr;
1564 struct lfsck_assistant_data *lad = com->lc_data;
1565 struct ptlrpc_thread *mthread = &lfsck->li_thread;
1566 struct ptlrpc_thread *athread = &lad->lad_thread;
1567 struct lfsck_assistant_operations *lao = lad->lad_ops;
1568 struct lfsck_assistant_req *lar;
1569 struct l_wait_info lwi = { 0 };
1575 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1576 lfsck_lfsck2name(lfsck), lad->lad_name);
1578 memset(lr, 0, sizeof(*lr));
1579 lr->lr_event = LE_START;
1580 if (pos->lp_oit_cookie <= 1)
1581 lr->lr_param = LPF_RESET;
1582 rc = lfsck_assistant_notify_others(env, com, lr);
1584 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1585 "to start %s: rc = %d\n",
1586 lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1590 spin_lock(&lad->lad_lock);
1591 lad->lad_task = current;
1592 thread_set_flags(athread, SVC_RUNNING);
1593 spin_unlock(&lad->lad_lock);
1594 wake_up_all(&mthread->t_ctl_waitq);
1597 while (!list_empty(&lad->lad_req_list)) {
1598 bool wakeup = false;
1600 if (unlikely(lad->lad_exit ||
1601 !thread_is_running(mthread)))
1602 GOTO(cleanup, rc = lad->lad_post_result);
1604 lar = list_entry(lad->lad_req_list.next,
1605 struct lfsck_assistant_req,
1607 /* Only the lfsck_assistant_engine thread itself can
1608 * remove the "lar" from the head of the list, LFSCK
1609 * engine thread only inserts other new "lar" at the
1610 * end of the list. So it is safe to handle current
1611 * "lar" without the spin_lock. */
1612 rc = lao->la_handler_p1(env, com, lar);
1613 spin_lock(&lad->lad_lock);
1614 list_del_init(&lar->lar_list);
1615 lad->lad_prefetched--;
1616 /* Wake up the main engine thread only when the list
1617 * is empty or half of the prefetched items have been
1618 * handled to avoid too frequent thread schedule. */
1619 if (lad->lad_prefetched <= (bk->lb_async_windows / 2))
1621 spin_unlock(&lad->lad_lock);
1623 wake_up_all(&mthread->t_ctl_waitq);
1625 lao->la_req_fini(env, lar);
1626 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1630 l_wait_event(athread->t_ctl_waitq,
1631 !lfsck_assistant_req_empty(lad) ||
1634 lad->lad_to_double_scan,
1637 if (unlikely(lad->lad_exit))
1638 GOTO(cleanup, rc = lad->lad_post_result);
1640 if (!list_empty(&lad->lad_req_list))
1643 if (lad->lad_to_post) {
1644 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1645 lfsck_lfsck2name(lfsck), lad->lad_name);
1647 if (unlikely(lad->lad_exit))
1648 GOTO(cleanup, rc = lad->lad_post_result);
1650 lad->lad_to_post = 0;
1651 LASSERT(lad->lad_post_result > 0);
1653 /* Wakeup the master engine to go ahead. */
1654 wake_up_all(&mthread->t_ctl_waitq);
1656 memset(lr, 0, sizeof(*lr));
1657 lr->lr_event = LE_PHASE1_DONE;
1658 lr->lr_status = lad->lad_post_result;
1659 rc = lfsck_assistant_notify_others(env, com, lr);
1661 CDEBUG(D_LFSCK, "%s: LFSCK assistant notified "
1662 "others for %s post: rc = %d\n",
1663 lfsck_lfsck2name(lfsck),
1667 if (lad->lad_to_double_scan) {
1668 lad->lad_to_double_scan = 0;
1669 atomic_inc(&lfsck->li_double_scan_count);
1670 lad->lad_in_double_scan = 1;
1671 wake_up_all(&mthread->t_ctl_waitq);
1673 com->lc_new_checked = 0;
1674 com->lc_new_scanned = 0;
1675 com->lc_time_last_checkpoint = cfs_time_current();
1676 com->lc_time_next_checkpoint =
1677 com->lc_time_last_checkpoint +
1678 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1680 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1681 "the second-stage scaning\n",
1682 lfsck_lfsck2name(lfsck));
1684 /* Flush async updates before handling orphan. */
1685 rc2 = dt_sync(env, lfsck->li_next);
1687 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1688 "scan start, synced: rc = %d\n",
1689 lfsck_lfsck2name(lfsck), rc2);
1691 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1692 GOTO(cleanup, rc = 0);
1694 while (lad->lad_in_double_scan) {
1695 rc = lfsck_assistant_query_others(env, com);
1696 if (lfsck_phase2_next_ready(lad))
1702 /* Pull LFSCK status on related targets once
1703 * per 30 seconds if we are not notified. */
1704 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1705 cfs_time_seconds(1),
1707 rc = l_wait_event(athread->t_ctl_waitq,
1708 lfsck_phase2_next_ready(lad) ||
1710 !thread_is_running(mthread),
1713 if (unlikely(lad->lad_exit ||
1714 !thread_is_running(mthread)))
1715 GOTO(cleanup, rc = 0);
1717 if (rc == -ETIMEDOUT)
1724 rc = lao->la_handler_p2(env, com);
1728 if (unlikely(lad->lad_exit ||
1729 !thread_is_running(mthread)))
1730 GOTO(cleanup, rc = 0);
1736 /* Cleanup the unfinished requests. */
1737 spin_lock(&lad->lad_lock);
1739 lad->lad_assistant_status = rc;
1741 if (lad->lad_exit && lad->lad_post_result <= 0)
1742 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1744 thread_set_flags(athread, SVC_STOPPING);
1745 while (!list_empty(&lad->lad_req_list)) {
1746 lar = list_entry(lad->lad_req_list.next,
1747 struct lfsck_assistant_req,
1749 list_del_init(&lar->lar_list);
1750 lad->lad_prefetched--;
1751 spin_unlock(&lad->lad_lock);
1752 lao->la_req_fini(env, lar);
1753 spin_lock(&lad->lad_lock);
1755 spin_unlock(&lad->lad_lock);
1757 LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1758 lad->lad_prefetched);
1760 memset(lr, 0, sizeof(*lr));
1762 lr->lr_event = LE_PHASE2_DONE;
1764 } else if (rc == 0) {
1765 if (lfsck->li_flags & LPF_ALL_TGT) {
1766 lr->lr_event = LE_STOP;
1767 lr->lr_status = LS_STOPPED;
1769 lr->lr_event = LE_PEER_EXIT;
1770 switch (lfsck->li_status) {
1773 lr->lr_status = LS_CO_PAUSED;
1777 lr->lr_status = LS_CO_STOPPED;
1780 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1781 "status: rc = %d\n",
1782 lfsck_lfsck2name(lfsck),
1784 lr->lr_status = LS_CO_FAILED;
1789 if (lfsck->li_flags & LPF_ALL_TGT) {
1790 lr->lr_event = LE_STOP;
1791 lr->lr_status = LS_FAILED;
1793 lr->lr_event = LE_PEER_EXIT;
1794 lr->lr_status = LS_CO_FAILED;
1798 rc1 = lfsck_assistant_notify_others(env, com, lr);
1800 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1801 "others for %s quit: rc = %d\n",
1802 lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1806 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1807 lfsck_lfsck2name(lfsck));
1809 /* Flush async updates before exit. */
1810 rc2 = dt_sync(env, lfsck->li_next);
1812 CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1813 lfsck_lfsck2name(lfsck), rc2);
1815 /* Under force exit case, some requests may be just freed without
1816 * verification, those objects should be re-handled when next run.
1817 * So not update the on-disk trace file under such case. */
1818 if (lad->lad_in_double_scan) {
1820 rc1 = lao->la_double_scan_result(env, com, rc);
1822 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1823 "finished: rc = %d\n",
1824 lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1828 if (lad->lad_in_double_scan)
1829 atomic_dec(&lfsck->li_double_scan_count);
1831 spin_lock(&lad->lad_lock);
1832 lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1833 thread_set_flags(athread, SVC_STOPPED);
1834 wake_up_all(&mthread->t_ctl_waitq);
1835 lad->lad_task = NULL;
1836 spin_unlock(&lad->lad_lock);
1838 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1839 lfsck_lfsck2name(lfsck), lad->lad_name,
1840 lad->lad_assistant_status);
1842 lfsck_thread_args_fini(lta);