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, 2014, 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_update_lma(const struct lu_env *env,
97 struct lfsck_instance *lfsck, struct dt_object *obj)
99 struct lfsck_thread_info *info = lfsck_env_info(env);
100 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
101 struct dt_device *dev = lfsck_obj2dev(obj);
102 struct lustre_mdt_attrs *lma = &info->lti_lma;
109 if (bk->lb_param & LPF_DRYRUN)
112 buf = lfsck_buf_get(env, info->lti_lma_old, LMA_OLD_SIZE);
113 rc = dt_xattr_get(env, obj, buf, XATTR_NAME_LMA);
118 fl = LU_XATTR_CREATE;
119 lustre_lma_init(lma, lfsck_dto2fid(obj), LMAC_FID_ON_OST, 0);
121 if (rc != LMA_OLD_SIZE && rc != sizeof(struct lustre_mdt_attrs))
124 fl = LU_XATTR_REPLACE;
125 lustre_lma_swab(lma);
126 lustre_lma_init(lma, lfsck_dto2fid(obj),
127 lma->lma_compat | LMAC_FID_ON_OST,
130 lustre_lma_swab(lma);
132 th = dt_trans_create(env, dev);
136 buf = lfsck_buf_get(env, lma, sizeof(*lma));
137 rc = dt_declare_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th);
141 rc = dt_trans_start_local(env, dev, th);
145 rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th);
150 dt_trans_stop(env, dev, th);
154 static int lfsck_parent_fid(const struct lu_env *env, struct dt_object *obj,
157 if (unlikely(!S_ISDIR(lfsck_object_type(obj)) ||
158 !dt_try_as_dir(env, obj)))
161 return dt_lookup(env, obj, (struct dt_rec *)fid,
162 (const struct dt_key *)"..");
166 * Check whether needs to scan the directory or not.
168 * 1) If we are not doing namespace LFSCK, or the given @obj is not directory,
169 * then needs not to scan the @obj. Otherwise,
170 * 2) Global /ROOT needs to be scanned, backend root needs not to be scanned.
171 * 3) If the @obj is neither IGIF nor normal FID (including .lustre and its
172 * sub-directories that have been scanned when the LFSCK engine start),
173 * then needs not to be scanned.
174 * 4) If it is a remote object, then scanning the object will be done on the
175 * MDT on which the object really resides.
176 * 5) If the local object has normal FID, then needs to be scanned. Otherwise,
177 * 6) If the object has linkEA, then needs to be scanned. Otherwise,
178 * 7) If none of the previous conditions are true, we need to check the parent
179 * directories whether this subdirectory is in a tree that should be scanned.
180 * Set the parent as current @obj, repeat 2)-7).
182 * \param[in] env pointer to the thread context
183 * \param[in] lfsck pointer to the lfsck instance
184 * \param[in] obj pointer to the object to be checked
186 * \retval positive number if the directory needs to be scanned
187 * \retval 0 if the directory needs NOT to be scanned
188 * \retval negative error number on failure
190 static int lfsck_needs_scan_dir(const struct lu_env *env,
191 struct lfsck_instance *lfsck,
192 struct dt_object *obj)
194 struct lfsck_thread_info *info = lfsck_env_info(env);
195 struct lu_fid *fid = &info->lti_fid;
196 struct lu_seq_range *range = &info->lti_range;
197 struct seq_server_site *ss = lfsck_dev_site(lfsck);
198 __u32 idx = lfsck_dev_idx(lfsck);
202 if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
207 *fid = *lfsck_dto2fid(obj);
209 /* Global /ROOT is visible. */
210 if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
213 /* Backend root is invisible. */
214 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
217 if (!fid_is_norm(fid) && !fid_is_igif(fid))
220 fld_range_set_mdt(range);
221 rc = fld_local_lookup(env, ss->ss_server_fld,
222 fid_seq(fid), range);
223 if (rc != 0 || range->lsr_index != idx)
224 /* Current FID should NOT be for the input parameter
225 * @obj, because the lfsck_master_oit_engine() has
226 * filtered out agent object. So current FID is for
227 * the ancestor of the original input parameter @obj.
228 * So the ancestor is a remote directory. The input
229 * parameter @obj is local directory, and should be
230 * scanned under such case. */
233 /* normal FID on this target (locally) must be for the
234 * client-side visiable object. */
235 if (fid_is_norm(fid))
239 obj = lfsck_object_find_bottom(env, lfsck, fid);
244 if (!dt_object_exists(obj))
248 dt_read_lock(env, obj, MOR_TGT_CHILD);
249 if (unlikely(lfsck_is_dead_obj(obj))) {
250 dt_read_unlock(env, obj);
255 rc = dt_xattr_get(env, obj,
256 lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK);
257 dt_read_unlock(env, obj);
261 if (rc < 0 && rc != -ENODATA)
264 rc = lfsck_parent_fid(env, obj, fid);
266 lfsck_object_put(env, obj);
272 if (!fid_is_sane(fid))
277 if (depth > 0 && obj != NULL)
278 lfsck_object_put(env, obj);
283 static int lfsck_load_stripe_lmv(const struct lu_env *env,
284 struct lfsck_instance *lfsck,
285 struct dt_object *obj)
287 struct lmv_mds_md_v1 *lmv = &lfsck_env_info(env)->lti_lmv;
288 struct lfsck_lmv *llmv;
292 LASSERT(lfsck->li_obj_dir == NULL);
293 LASSERT(lfsck->li_lmv == NULL);
295 rc = lfsck_read_stripe_lmv(env, obj, lmv);
296 if (rc == -ENODATA) {
297 lfsck->li_obj_dir = lfsck_object_get(obj);
309 if (lmv->lmv_magic == LMV_MAGIC) {
310 struct lfsck_slave_lmv_rec *lslr;
313 llmv->ll_lmv_master = 1;
314 if (lmv->lmv_stripe_count < 1)
315 stripes = LFSCK_LMV_DEF_STRIPES;
316 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
317 stripes = LFSCK_LMV_MAX_STRIPES;
319 stripes = lmv->lmv_stripe_count;
321 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
328 llmv->ll_stripes_allocated = stripes;
329 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
330 llmv->ll_lslr = lslr;
332 llmv->ll_lmv_slave = 1;
335 lfsck->li_obj_dir = lfsck_object_get(obj);
337 atomic_set(&llmv->ll_ref, 1);
338 lfsck->li_lmv = llmv;
343 /* LFSCK wrap functions */
345 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
348 struct lfsck_component *com;
350 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
351 com->lc_ops->lfsck_fail(env, com, new_checked);
355 void lfsck_close_dir(const struct lu_env *env,
356 struct lfsck_instance *lfsck, int result)
358 struct lfsck_component *com;
361 if (lfsck->li_lmv != NULL) {
362 lfsck->li_lmv->ll_exit_value = result;
363 if (lfsck->li_obj_dir != NULL) {
364 list_for_each_entry(com, &lfsck->li_list_dir,
366 com->lc_ops->lfsck_close_dir(env, com);
370 lfsck_lmv_put(env, lfsck->li_lmv);
371 lfsck->li_lmv = NULL;
374 if (lfsck->li_di_dir != NULL) {
375 const struct dt_it_ops *dir_iops;
376 struct dt_it *dir_di = lfsck->li_di_dir;
378 LASSERT(lfsck->li_obj_dir != NULL);
380 dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
381 lfsck_di_dir_put(env, lfsck);
382 dir_iops->fini(env, dir_di);
385 if (lfsck->li_obj_dir != NULL) {
386 struct dt_object *dir_obj = lfsck->li_obj_dir;
388 lfsck->li_obj_dir = NULL;
389 lfsck_object_put(env, dir_obj);
395 int lfsck_open_dir(const struct lu_env *env,
396 struct lfsck_instance *lfsck, __u64 cookie)
398 struct dt_object *obj = lfsck->li_obj_dir;
399 struct dt_it *di = lfsck->li_di_dir;
400 struct lfsck_component *com;
401 const struct dt_it_ops *iops;
405 LASSERT(obj != NULL);
408 if (unlikely(!dt_try_as_dir(env, obj)))
409 GOTO(out, rc = -ENOTDIR);
411 list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
412 rc = com->lc_ops->lfsck_open_dir(env, com);
417 iops = &obj->do_index_ops->dio_it;
418 di = iops->init(env, obj, lfsck->li_args_dir);
420 GOTO(out, rc = PTR_ERR(di));
422 rc = iops->load(env, di, cookie);
423 if (rc == 0 || (rc > 0 && cookie > 0))
424 rc = iops->next(env, di);
432 lfsck->li_cookie_dir = iops->store(env, di);
433 spin_lock(&lfsck->li_lock);
434 lfsck->li_di_dir = di;
435 spin_unlock(&lfsck->li_lock);
442 lfsck_close_dir(env, lfsck, rc);
447 static int lfsck_checkpoint(const struct lu_env *env,
448 struct lfsck_instance *lfsck)
450 struct lfsck_component *com;
454 if (likely(cfs_time_beforeq(cfs_time_current(),
455 lfsck->li_time_next_checkpoint)))
458 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
459 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
460 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
465 lfsck->li_time_last_checkpoint = cfs_time_current();
466 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
467 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
468 return rc1 != 0 ? rc1 : rc;
471 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
472 struct lfsck_start_param *lsp)
474 struct dt_object *obj = NULL;
475 struct lfsck_component *com;
476 struct lfsck_component *next;
477 struct lfsck_position *pos = NULL;
478 const struct dt_it_ops *iops =
479 &lfsck->li_obj_oit->do_index_ops->dio_it;
483 LASSERT(lfsck->li_obj_dir == NULL);
484 LASSERT(lfsck->li_di_dir == NULL);
486 lfsck->li_current_oit_processed = 0;
487 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
488 com->lc_new_checked = 0;
489 rc = com->lc_ops->lfsck_prep(env, com, lsp);
494 (!lfsck_pos_is_zero(&com->lc_pos_start) &&
495 lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
496 pos = &com->lc_pos_start;
499 /* Init otable-based iterator. */
501 rc = iops->load(env, lfsck->li_di_oit, 0);
503 lfsck->li_oit_over = 1;
510 rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
514 lfsck->li_oit_over = 1;
516 if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
519 /* Find the directory for namespace-based traverse. */
520 obj = lfsck_object_find_bottom(env, lfsck, &pos->lp_dir_parent);
522 RETURN(PTR_ERR(obj));
524 /* Remote directory will be scanned by the LFSCK instance
525 * on the MDT where the remote object really resides on. */
526 if (!dt_object_exists(obj) || dt_object_remote(obj) ||
527 unlikely(!S_ISDIR(lfsck_object_type(obj))))
530 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
532 /* For the master MDT-object of a striped directory,
533 * reset the iteration from the directory beginning. */
534 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
535 pos->lp_dir_cookie = 0;
537 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
539 /* The end of the directory. */
547 lfsck_object_put(env, obj);
550 lfsck_close_dir(env, lfsck, rc);
551 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
553 com->lc_ops->lfsck_post(env, com, rc, true);
560 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
561 lfsck->li_pos_current = lfsck->li_pos_checkpoint;
562 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
563 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
568 lfsck->li_time_last_checkpoint = cfs_time_current();
569 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
570 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
574 static int lfsck_exec_oit(const struct lu_env *env,
575 struct lfsck_instance *lfsck, struct dt_object *obj)
577 struct lfsck_component *com;
581 LASSERT(lfsck->li_obj_dir == NULL);
583 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
584 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
589 rc = lfsck_needs_scan_dir(env, lfsck, obj);
593 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
595 rc = lfsck_open_dir(env, lfsck, 0);
601 lfsck_fail(env, lfsck, false);
604 lfsck_close_dir(env, lfsck, rc);
606 return rc > 0 ? 0 : rc;
609 static int lfsck_exec_dir(const struct lu_env *env,
610 struct lfsck_instance *lfsck,
611 struct lu_dirent *ent, __u16 type)
613 struct lfsck_component *com;
616 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
617 rc = com->lc_ops->lfsck_exec_dir(env, com, ent, type);
624 static int lfsck_master_dir_engine(const struct lu_env *env,
625 struct lfsck_instance *lfsck);
627 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
630 struct lfsck_component *com;
631 struct lfsck_component *next;
634 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
635 lfsck_close_dir(env, lfsck, result);
637 while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
638 !list_empty(&lfsck->li_list_lmv)) {
639 struct lfsck_lmv_unit *llu;
641 spin_lock(&lfsck->li_lock);
642 llu = list_entry(lfsck->li_list_lmv.next,
643 struct lfsck_lmv_unit, llu_link);
644 list_del_init(&llu->llu_link);
645 spin_unlock(&lfsck->li_lock);
647 lfsck->li_lmv = &llu->llu_lmv;
648 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
649 rc = lfsck_open_dir(env, lfsck, 0);
651 rc = lfsck_master_dir_engine(env, lfsck);
652 lfsck_close_dir(env, lfsck, result);
658 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
659 rc = com->lc_ops->lfsck_post(env, com, result, false);
661 CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
662 "rc = %d\n", lfsck_lfsck2name(lfsck),
663 (__u32)com->lc_type, rc);
666 lfsck->li_time_last_checkpoint = cfs_time_current();
667 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
668 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
670 /* Ignore some component post failure to make other can go ahead. */
674 static int lfsck_double_scan(const struct lu_env *env,
675 struct lfsck_instance *lfsck)
677 struct lfsck_component *com;
678 struct lfsck_component *next;
679 struct l_wait_info lwi = { 0 };
683 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
684 rc = com->lc_ops->lfsck_double_scan(env, com);
689 l_wait_event(lfsck->li_thread.t_ctl_waitq,
690 atomic_read(&lfsck->li_double_scan_count) == 0,
693 if (lfsck->li_status != LS_PAUSED &&
694 lfsck->li_status != LS_CO_PAUSED) {
695 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
697 spin_lock(&lfsck->li_lock);
698 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
699 spin_unlock(&lfsck->li_lock);
703 return rc1 != 0 ? rc1 : rc;
706 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
708 struct lfsck_component *com;
709 struct lfsck_component *next;
711 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
713 if (com->lc_ops->lfsck_quit != NULL)
714 com->lc_ops->lfsck_quit(env, com);
716 spin_lock(&lfsck->li_lock);
717 list_del_init(&com->lc_link_dir);
718 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
719 spin_unlock(&lfsck->li_lock);
722 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
724 if (com->lc_ops->lfsck_quit != NULL)
725 com->lc_ops->lfsck_quit(env, com);
727 spin_lock(&lfsck->li_lock);
728 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
729 spin_unlock(&lfsck->li_lock);
735 static int lfsck_master_dir_engine(const struct lu_env *env,
736 struct lfsck_instance *lfsck)
738 struct lfsck_thread_info *info = lfsck_env_info(env);
739 struct dt_object *dir = lfsck->li_obj_dir;
740 const struct dt_it_ops *iops = &dir->do_index_ops->dio_it;
741 struct dt_it *di = lfsck->li_di_dir;
742 struct lu_dirent *ent =
743 (struct lu_dirent *)info->lti_key;
744 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
745 struct ptlrpc_thread *thread = &lfsck->li_thread;
751 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
752 unlikely(!thread_is_running(thread))) {
753 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
754 "parent "DFID", cookie "LPX64"\n",
755 lfsck_lfsck2name(lfsck),
756 PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
761 lfsck->li_new_scanned++;
762 rc = iops->rec(env, di, (struct dt_rec *)ent,
765 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
769 CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
770 "parent "DFID", cookie "LPX64": rc = %d\n",
771 lfsck_lfsck2name(lfsck),
772 PFID(lfsck_dto2fid(dir)),
773 lfsck->li_cookie_dir, rc);
774 lfsck_fail(env, lfsck, true);
775 if (bk->lb_param & LPF_FAILOUT)
781 if (ent->lde_attrs & LUDA_IGNORE &&
782 strcmp(ent->lde_name, dotdot) != 0)
785 /* skip dot entry. */
786 if (ent->lde_namelen == 1 && ent->lde_name[0] == '.')
789 /* The type in the @ent structure may has been overwritten,
790 * so we need to pass the @type parameter independently. */
791 rc = lfsck_exec_dir(env, lfsck, ent, type);
792 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
796 rc = lfsck_checkpoint(env, lfsck);
797 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
801 lfsck_control_speed(lfsck);
802 if (unlikely(!thread_is_running(thread))) {
803 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
804 "parent "DFID", cookie "LPX64"\n",
805 lfsck_lfsck2name(lfsck),
806 PFID(lfsck_dto2fid(dir)),
807 lfsck->li_cookie_dir);
811 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
812 spin_lock(&lfsck->li_lock);
813 thread_set_flags(thread, SVC_STOPPING);
814 spin_unlock(&lfsck->li_lock);
818 rc = iops->next(env, di);
821 if (rc > 0 && !lfsck->li_oit_over)
822 lfsck_close_dir(env, lfsck, rc);
828 * Object-table based iteration engine.
830 * Object-table based iteration is the basic linear engine to scan all the
831 * objects on current device in turn. For each object, it calls all the
832 * registered LFSCK component(s)' API to perform related consistency
835 * It flushes related LFSCK trace files to disk via making checkpoint
836 * periodically. Then if the server crashed or the LFSCK is paused, the
837 * LFSCK can resume from the latest checkpoint.
839 * It also controls the whole LFSCK speed via lfsck_control_speed() to
840 * avoid the server to become overload.
842 * \param[in] env pointer to the thread context
843 * \param[in] lfsck pointer to the lfsck instance
845 * \retval positive number if all objects have been scanned
846 * \retval 0 if the iteration is stopped or paused
847 * \retval negative error number on failure
849 static int lfsck_master_oit_engine(const struct lu_env *env,
850 struct lfsck_instance *lfsck)
852 struct lfsck_thread_info *info = lfsck_env_info(env);
853 const struct dt_it_ops *iops =
854 &lfsck->li_obj_oit->do_index_ops->dio_it;
855 struct dt_it *di = lfsck->li_di_oit;
856 struct lu_fid *fid = &info->lti_fid;
857 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
858 struct ptlrpc_thread *thread = &lfsck->li_thread;
859 struct seq_server_site *ss = lfsck_dev_site(lfsck);
860 __u32 idx = lfsck_dev_idx(lfsck);
864 if (unlikely(ss == NULL))
868 struct dt_object *target;
869 bool update_lma = false;
871 if (lfsck->li_di_dir != NULL) {
872 rc = lfsck_master_dir_engine(env, lfsck);
877 if (unlikely(lfsck->li_oit_over))
880 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
881 unlikely(!thread_is_running(thread))) {
882 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
884 lfsck_lfsck2name(lfsck), iops->store(env, di));
889 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
892 lfsck->li_current_oit_processed = 1;
894 if (!list_empty(&lfsck->li_list_lmv)) {
895 struct lfsck_lmv_unit *llu;
897 spin_lock(&lfsck->li_lock);
898 llu = list_entry(lfsck->li_list_lmv.next,
899 struct lfsck_lmv_unit, llu_link);
900 list_del_init(&llu->llu_link);
901 spin_unlock(&lfsck->li_lock);
903 lfsck->li_lmv = &llu->llu_lmv;
904 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
905 rc = lfsck_open_dir(env, lfsck, 0);
907 rc = lfsck_master_dir_engine(env, lfsck);
913 lfsck->li_new_scanned++;
914 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
915 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
917 CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
918 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
919 lfsck_fail(env, lfsck, true);
920 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
926 if (unlikely(!fid_is_sane(fid))) {
927 CDEBUG(D_LFSCK, "%s: OIT scan find invalid FID "DFID
929 lfsck_lfsck2name(lfsck), PFID(fid));
933 if (fid_is_idif(fid)) {
934 __u32 idx1 = fid_idif_ost_idx(fid);
936 LASSERT(!lfsck->li_master);
938 /* It is an old format device, update the LMA. */
940 struct ost_id *oi = &info->lti_oi;
942 fid_to_ostid(fid, oi);
943 ostid_to_fid(fid, oi, idx);
946 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
947 !fid_is_last_id(fid) &&
948 !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
950 /* If the FID/object is only used locally and invisible
951 * to external nodes, then LFSCK will not handle it.
953 * dot_lustre sequence has been handled specially. */
956 struct lu_seq_range *range = &info->lti_range;
958 if (lfsck->li_master)
959 fld_range_set_mdt(range);
961 fld_range_set_ost(range);
962 rc = fld_local_lookup(env, ss->ss_server_fld,
963 fid_seq(fid), range);
964 if (rc != 0 || range->lsr_index != idx) {
965 /* Remote object will be handled by the LFSCK
966 * instance on the MDT where the remote object
967 * really resides on. */
973 target = lfsck_object_find_bottom(env, lfsck, fid);
974 if (IS_ERR(target)) {
975 CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
976 DFID", cookie "LPU64": rc = %d\n",
977 lfsck_lfsck2name(lfsck), PFID(fid),
978 iops->store(env, di), rc);
979 lfsck_fail(env, lfsck, true);
980 if (bk->lb_param & LPF_FAILOUT)
981 RETURN(PTR_ERR(target));
986 if (dt_object_exists(target)) {
988 rc = lfsck_update_lma(env, lfsck, target);
990 CDEBUG(D_LFSCK, "%s: fail to update "
991 "LMA for "DFID": rc = %d\n",
992 lfsck_lfsck2name(lfsck),
993 PFID(lfsck_dto2fid(target)), rc);
996 rc = lfsck_exec_oit(env, lfsck, target);
998 lfsck_object_put(env, target);
999 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1003 rc = lfsck_checkpoint(env, lfsck);
1004 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1008 lfsck_control_speed(lfsck);
1010 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
1011 spin_lock(&lfsck->li_lock);
1012 thread_set_flags(thread, SVC_STOPPING);
1013 spin_unlock(&lfsck->li_lock);
1017 rc = iops->next(env, di);
1018 if (unlikely(rc > 0))
1019 lfsck->li_oit_over = 1;
1020 else if (likely(rc == 0))
1021 lfsck->li_current_oit_processed = 0;
1023 if (unlikely(!thread_is_running(thread))) {
1024 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1025 "cookie "LPU64"\n", lfsck_lfsck2name(lfsck),
1026 iops->store(env, di));
1029 } while (rc == 0 || lfsck->li_di_dir != NULL);
1034 int lfsck_master_engine(void *args)
1036 struct lfsck_thread_args *lta = args;
1037 struct lu_env *env = <a->lta_env;
1038 struct lfsck_instance *lfsck = lta->lta_lfsck;
1039 struct ptlrpc_thread *thread = &lfsck->li_thread;
1040 struct dt_object *oit_obj = lfsck->li_obj_oit;
1041 const struct dt_it_ops *oit_iops = &oit_obj->do_index_ops->dio_it;
1042 struct dt_it *oit_di;
1043 struct l_wait_info lwi = { 0 };
1047 if (lfsck->li_master &&
1048 (!list_empty(&lfsck->li_list_scan) ||
1049 !list_empty(&lfsck->li_list_double_scan))) {
1050 rc = lfsck_verify_lpf(env, lfsck);
1051 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1052 * fatal, because the .lustre/lost+found/ maybe not accessed
1053 * by the LFSCK if it does not add orphans or others to such
1054 * directory. So go ahead until hit failure when really uses
1057 CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1058 ".lustre/lost+found/, go ahead: rc = %d\n",
1059 lfsck_lfsck2name(lfsck), rc);
1062 oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit);
1063 if (IS_ERR(oit_di)) {
1064 rc = PTR_ERR(oit_di);
1065 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1066 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1068 GOTO(fini_args, rc);
1071 spin_lock(&lfsck->li_lock);
1072 lfsck->li_di_oit = oit_di;
1073 spin_unlock(&lfsck->li_lock);
1074 rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1078 CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1079 "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1080 ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1081 lfsck->li_pos_checkpoint.lp_oit_cookie,
1082 lfsck->li_pos_checkpoint.lp_dir_cookie,
1083 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1086 spin_lock(&lfsck->li_lock);
1087 thread_set_flags(thread, SVC_RUNNING);
1088 spin_unlock(&lfsck->li_lock);
1089 wake_up_all(&thread->t_ctl_waitq);
1091 l_wait_event(thread->t_ctl_waitq,
1092 lfsck->li_start_unplug ||
1093 !thread_is_running(thread),
1095 if (!thread_is_running(thread))
1096 GOTO(fini_oit, rc = 0);
1098 if (!list_empty(&lfsck->li_list_scan) ||
1099 list_empty(&lfsck->li_list_double_scan))
1100 rc = lfsck_master_oit_engine(env, lfsck);
1104 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
1105 CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1106 "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1107 ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1108 lfsck->li_pos_checkpoint.lp_oit_cookie,
1109 lfsck->li_pos_checkpoint.lp_dir_cookie,
1110 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1113 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1114 rc = lfsck_post(env, lfsck, rc);
1116 lfsck_close_dir(env, lfsck, rc);
1119 lfsck_di_oit_put(env, lfsck);
1120 oit_iops->fini(env, oit_di);
1122 if (!list_empty(&lfsck->li_list_double_scan))
1123 rc = lfsck_double_scan(env, lfsck);
1127 lfsck_quit(env, lfsck);
1130 /* XXX: Purge the pinned objects in the future. */
1133 spin_lock(&lfsck->li_lock);
1134 thread_set_flags(thread, SVC_STOPPED);
1135 spin_unlock(&lfsck->li_lock);
1136 wake_up_all(&thread->t_ctl_waitq);
1137 lfsck_thread_args_fini(lta);
1141 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1145 spin_lock(&lad->lad_lock);
1146 if (list_empty(&lad->lad_req_list))
1148 spin_unlock(&lad->lad_lock);
1154 * Query the LFSCK status from the instatnces on remote servers.
1156 * The LFSCK assistant thread queries the LFSCK instances on other
1157 * servers (MDT/OST) about their status, such as whether they have
1158 * finished the phase1/phase2 scanning or not, and so on.
1160 * \param[in] env pointer to the thread context
1161 * \param[in] com pointer to the lfsck component
1163 * \retval 0 for success
1164 * \retval negative error number on failure
1166 static int lfsck_assistant_query_others(const struct lu_env *env,
1167 struct lfsck_component *com)
1169 struct lfsck_thread_info *info = lfsck_env_info(env);
1170 struct lfsck_request *lr = &info->lti_lr;
1171 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1172 struct lfsck_instance *lfsck = com->lc_lfsck;
1173 struct lfsck_assistant_data *lad = com->lc_data;
1174 struct ptlrpc_request_set *set;
1175 struct lfsck_tgt_descs *ltds;
1176 struct lfsck_tgt_desc *ltd;
1177 struct list_head *phase_head;
1182 set = ptlrpc_prep_set();
1186 lad->lad_touch_gen++;
1187 memset(lr, 0, sizeof(*lr));
1188 lr->lr_event = LE_QUERY;
1189 lr->lr_active = com->lc_type;
1190 laia->laia_com = com;
1192 laia->laia_shared = 0;
1194 if (!list_empty(&lad->lad_mdt_phase1_list)) {
1195 ltds = &lfsck->li_mdt_descs;
1197 phase_head = &lad->lad_mdt_phase1_list;
1198 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1203 ltds = &lfsck->li_ost_descs;
1204 lr->lr_flags = LEF_TO_OST;
1205 phase_head = &lad->lad_ost_phase1_list;
1208 laia->laia_ltds = ltds;
1209 spin_lock(<ds->ltd_lock);
1210 while (!list_empty(phase_head)) {
1211 struct list_head *phase_list;
1214 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1215 ltd = list_entry(phase_head->next,
1216 struct lfsck_tgt_desc,
1217 ltd_layout_phase_list);
1218 phase_list = <d->ltd_layout_phase_list;
1219 gen = <d->ltd_layout_gen;
1221 ltd = list_entry(phase_head->next,
1222 struct lfsck_tgt_desc,
1223 ltd_namespace_phase_list);
1224 phase_list = <d->ltd_namespace_phase_list;
1225 gen = <d->ltd_namespace_gen;
1228 if (*gen == lad->lad_touch_gen)
1231 *gen = lad->lad_touch_gen;
1232 list_move_tail(phase_list, phase_head);
1233 atomic_inc(<d->ltd_ref);
1234 laia->laia_ltd = ltd;
1235 spin_unlock(<ds->ltd_lock);
1236 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1237 lfsck_async_interpret_common,
1240 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1241 "%s %x for %s: rc = %d\n",
1242 lfsck_lfsck2name(lfsck),
1243 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1244 ltd->ltd_index, lad->lad_name, rc);
1248 spin_lock(<ds->ltd_lock);
1250 spin_unlock(<ds->ltd_lock);
1252 rc = ptlrpc_set_wait(set);
1254 ptlrpc_set_destroy(set);
1258 if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1259 list_empty(&lad->lad_mdt_phase1_list))
1263 ptlrpc_set_destroy(set);
1265 RETURN(rc1 != 0 ? rc1 : rc);
1269 * Notify the LFSCK event to the instances on remote servers.
1271 * The LFSCK assistant thread notifies the LFSCK instances on other
1272 * servers (MDT/OST) about some events, such as start new scanning,
1273 * stop the scanning, this LFSCK instance will exit, and so on.
1275 * \param[in] env pointer to the thread context
1276 * \param[in] com pointer to the lfsck component
1277 * \param[in] lr pointer to the LFSCK event request
1279 * \retval 0 for success
1280 * \retval negative error number on failure
1282 static int lfsck_assistant_notify_others(const struct lu_env *env,
1283 struct lfsck_component *com,
1284 struct lfsck_request *lr)
1286 struct lfsck_thread_info *info = lfsck_env_info(env);
1287 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1288 struct lfsck_instance *lfsck = com->lc_lfsck;
1289 struct lfsck_assistant_data *lad = com->lc_data;
1290 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1291 struct ptlrpc_request_set *set;
1292 struct lfsck_tgt_descs *ltds;
1293 struct lfsck_tgt_desc *ltd;
1294 struct lfsck_tgt_desc *next;
1300 set = ptlrpc_prep_set();
1304 lr->lr_index = lfsck_dev_idx(lfsck);
1305 lr->lr_active = com->lc_type;
1306 laia->laia_com = com;
1308 laia->laia_shared = 0;
1310 switch (lr->lr_event) {
1312 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1315 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1316 lr->lr_speed = bk->lb_speed_limit;
1317 lr->lr_version = bk->lb_version;
1318 lr->lr_param |= bk->lb_param;
1319 lr->lr_async_windows = bk->lb_async_windows;
1320 lr->lr_flags = LEF_TO_OST;
1322 /* Notify OSTs firstly, then handle other MDTs if needed. */
1323 ltds = &lfsck->li_ost_descs;
1324 laia->laia_ltds = ltds;
1325 down_read(<ds->ltd_rw_sem);
1326 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1327 ltd = lfsck_tgt_get(ltds, idx);
1328 LASSERT(ltd != NULL);
1330 laia->laia_ltd = ltd;
1331 ltd->ltd_layout_done = 0;
1332 ltd->ltd_synced_failures = 0;
1333 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1334 lfsck_async_interpret_common,
1335 laia, LFSCK_NOTIFY);
1337 lfsck_lad_set_bitmap(env, com, idx);
1338 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1339 "notify OST %x for %s start: rc = %d\n",
1340 lfsck_lfsck2name(lfsck), idx,
1345 up_read(<ds->ltd_rw_sem);
1348 rc = ptlrpc_set_wait(set);
1350 ptlrpc_set_destroy(set);
1355 if (!(bk->lb_param & LPF_ALL_TGT))
1358 /* link other MDT targets locallly. */
1359 ltds = &lfsck->li_mdt_descs;
1360 spin_lock(<ds->ltd_lock);
1361 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1362 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1363 ltd = LTD_TGT(ltds, idx);
1364 LASSERT(ltd != NULL);
1366 if (!list_empty(<d->ltd_layout_list))
1369 list_add_tail(<d->ltd_layout_list,
1370 &lad->lad_mdt_list);
1371 list_add_tail(<d->ltd_layout_phase_list,
1372 &lad->lad_mdt_phase1_list);
1375 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1376 ltd = LTD_TGT(ltds, idx);
1377 LASSERT(ltd != NULL);
1379 if (!list_empty(<d->ltd_namespace_list))
1382 list_add_tail(<d->ltd_namespace_list,
1383 &lad->lad_mdt_list);
1384 list_add_tail(<d->ltd_namespace_phase_list,
1385 &lad->lad_mdt_phase1_list);
1388 spin_unlock(<ds->ltd_lock);
1391 case LE_PHASE2_DONE:
1392 case LE_PEER_EXIT: {
1393 struct list_head *phase_head;
1395 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1396 if (bk->lb_param & LPF_ALL_TGT) {
1397 phase_head = &lad->lad_mdt_list;
1398 ltds = &lfsck->li_mdt_descs;
1399 if (lr->lr_event == LE_STOP) {
1400 /* unlink other MDT targets locallly. */
1401 spin_lock(<ds->ltd_lock);
1402 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1403 list_for_each_entry_safe(ltd, next,
1404 phase_head, ltd_layout_list) {
1406 <d->ltd_layout_phase_list);
1408 <d->ltd_layout_list);
1411 list_for_each_entry_safe(ltd, next,
1413 ltd_namespace_list) {
1415 <d->ltd_namespace_phase_list);
1417 <d->ltd_namespace_list);
1420 spin_unlock(<ds->ltd_lock);
1422 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1425 lr->lr_flags |= LEF_TO_OST;
1426 phase_head = &lad->lad_ost_list;
1427 ltds = &lfsck->li_ost_descs;
1429 lr->lr_flags &= ~LEF_TO_OST;
1431 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1434 lr->lr_flags |= LEF_TO_OST;
1435 phase_head = &lad->lad_ost_list;
1436 ltds = &lfsck->li_ost_descs;
1440 laia->laia_ltds = ltds;
1441 spin_lock(<ds->ltd_lock);
1442 while (!list_empty(phase_head)) {
1443 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1444 ltd = list_entry(phase_head->next,
1445 struct lfsck_tgt_desc,
1447 if (!list_empty(<d->ltd_layout_phase_list))
1449 <d->ltd_layout_phase_list);
1450 list_del_init(<d->ltd_layout_list);
1452 ltd = list_entry(phase_head->next,
1453 struct lfsck_tgt_desc,
1454 ltd_namespace_list);
1455 if (!list_empty(<d->ltd_namespace_phase_list))
1457 <d->ltd_namespace_phase_list);
1458 list_del_init(<d->ltd_namespace_list);
1460 atomic_inc(<d->ltd_ref);
1461 laia->laia_ltd = ltd;
1462 spin_unlock(<ds->ltd_lock);
1463 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1464 lfsck_async_interpret_common,
1465 laia, LFSCK_NOTIFY);
1467 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1468 "notify %s %x for %s stop/phase2_done/"
1469 "peer_exit: rc = %d\n",
1470 lfsck_lfsck2name(lfsck),
1471 (lr->lr_flags & LEF_TO_OST) ?
1472 "OST" : "MDT", ltd->ltd_index,
1476 spin_lock(<ds->ltd_lock);
1478 spin_unlock(<ds->ltd_lock);
1480 rc = ptlrpc_set_wait(set);
1482 ptlrpc_set_destroy(set);
1486 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1487 !(lr->lr_flags & LEF_TO_OST)) {
1488 lr->lr_flags |= LEF_TO_OST;
1489 phase_head = &lad->lad_ost_list;
1490 ltds = &lfsck->li_ost_descs;
1495 case LE_PHASE1_DONE:
1496 lad->lad_ops->la_sync_failures(env, com, lr);
1497 lad->lad_touch_gen++;
1498 ltds = &lfsck->li_mdt_descs;
1499 laia->laia_ltds = ltds;
1500 spin_lock(<ds->ltd_lock);
1501 while (!list_empty(&lad->lad_mdt_list)) {
1502 struct list_head *list;
1505 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1506 ltd = list_entry(lad->lad_mdt_list.next,
1507 struct lfsck_tgt_desc,
1509 list = <d->ltd_layout_list;
1510 gen = <d->ltd_layout_gen;
1512 struct lfsck_namespace *ns = com->lc_file_ram;
1514 ltd = list_entry(lad->lad_mdt_list.next,
1515 struct lfsck_tgt_desc,
1516 ltd_namespace_list);
1517 list = <d->ltd_namespace_list;
1518 gen = <d->ltd_namespace_gen;
1519 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1522 if (*gen == lad->lad_touch_gen)
1525 *gen = lad->lad_touch_gen;
1526 list_move_tail(list, &lad->lad_mdt_list);
1527 if (ltd->ltd_synced_failures)
1530 atomic_inc(<d->ltd_ref);
1531 laia->laia_ltd = ltd;
1532 spin_unlock(<ds->ltd_lock);
1533 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1534 lfsck_async_interpret_common,
1535 laia, LFSCK_NOTIFY);
1537 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1538 "notify MDT %x for %s phase1 done: "
1539 "rc = %d\n", lfsck_lfsck2name(lfsck),
1540 ltd->ltd_index, lad->lad_name, rc);
1543 spin_lock(<ds->ltd_lock);
1545 spin_unlock(<ds->ltd_lock);
1548 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1549 "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1554 rc1 = ptlrpc_set_wait(set);
1555 ptlrpc_set_destroy(set);
1557 RETURN(rc != 0 ? rc : rc1);
1561 * The LFSCK assistant thread is triggered by the LFSCK main engine.
1562 * They co-work together as an asynchronous pipeline: the LFSCK main
1563 * engine scans the system and pre-fetches the objects, attributes,
1564 * or name entries, etc, and pushes them into the pipeline as input
1565 * requests for the LFSCK assistant thread; on the other end of the
1566 * pipeline, the LFSCK assistant thread performs the real check and
1567 * repair for every request from the main engine.
1569 * Generally, the assistant engine may be blocked when check/repair
1570 * something, so the LFSCK main engine will run some faster. On the
1571 * other hand, the LFSCK main engine will drive multiple assistant
1572 * threads in parallel, means for each LFSCK component on the master
1573 * (such as layout LFSCK, namespace LFSCK), there is an independent
1574 * LFSCK assistant thread. So under such 1:N multiple asynchronous
1575 * pipelines mode, the whole LFSCK performance will be much better
1576 * than check/repair everything by the LFSCK main engine itself.
1578 int lfsck_assistant_engine(void *args)
1580 struct lfsck_thread_args *lta = args;
1581 struct lu_env *env = <a->lta_env;
1582 struct lfsck_component *com = lta->lta_com;
1583 struct lfsck_instance *lfsck = lta->lta_lfsck;
1584 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1585 struct lfsck_position *pos = &com->lc_pos_start;
1586 struct lfsck_thread_info *info = lfsck_env_info(env);
1587 struct lfsck_request *lr = &info->lti_lr;
1588 struct lfsck_assistant_data *lad = com->lc_data;
1589 struct ptlrpc_thread *mthread = &lfsck->li_thread;
1590 struct ptlrpc_thread *athread = &lad->lad_thread;
1591 struct lfsck_assistant_operations *lao = lad->lad_ops;
1592 struct lfsck_assistant_req *lar;
1593 struct l_wait_info lwi = { 0 };
1599 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1600 lfsck_lfsck2name(lfsck), lad->lad_name);
1602 memset(lr, 0, sizeof(*lr));
1603 lr->lr_event = LE_START;
1604 if (pos->lp_oit_cookie <= 1)
1605 lr->lr_param = LPF_RESET;
1606 rc = lfsck_assistant_notify_others(env, com, lr);
1608 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1609 "to start %s: rc = %d\n",
1610 lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1614 spin_lock(&lad->lad_lock);
1615 thread_set_flags(athread, SVC_RUNNING);
1616 spin_unlock(&lad->lad_lock);
1617 wake_up_all(&mthread->t_ctl_waitq);
1620 while (!list_empty(&lad->lad_req_list)) {
1621 bool wakeup = false;
1623 if (unlikely(lad->lad_exit ||
1624 !thread_is_running(mthread)))
1625 GOTO(cleanup1, rc = lad->lad_post_result);
1627 lar = list_entry(lad->lad_req_list.next,
1628 struct lfsck_assistant_req,
1630 /* Only the lfsck_assistant_engine thread itself can
1631 * remove the "lar" from the head of the list, LFSCK
1632 * engine thread only inserts other new "lar" at the
1633 * end of the list. So it is safe to handle current
1634 * "lar" without the spin_lock. */
1635 rc = lao->la_handler_p1(env, com, lar);
1636 spin_lock(&lad->lad_lock);
1637 list_del_init(&lar->lar_list);
1638 lad->lad_prefetched--;
1639 /* Wake up the main engine thread only when the list
1640 * is empty or half of the prefetched items have been
1641 * handled to avoid too frequent thread schedule. */
1642 if (lad->lad_prefetched <= (bk->lb_async_windows / 2))
1644 spin_unlock(&lad->lad_lock);
1646 wake_up_all(&mthread->t_ctl_waitq);
1648 lao->la_req_fini(env, lar);
1649 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1653 l_wait_event(athread->t_ctl_waitq,
1654 !lfsck_assistant_req_empty(lad) ||
1657 lad->lad_to_double_scan,
1660 if (unlikely(lad->lad_exit))
1661 GOTO(cleanup1, rc = lad->lad_post_result);
1663 if (!list_empty(&lad->lad_req_list))
1666 if (lad->lad_to_post) {
1667 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1668 lfsck_lfsck2name(lfsck), lad->lad_name);
1670 if (unlikely(lad->lad_exit))
1671 GOTO(cleanup1, rc = lad->lad_post_result);
1673 lad->lad_to_post = 0;
1674 LASSERT(lad->lad_post_result > 0);
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),
1686 /* Wakeup the master engine to go ahead. */
1687 wake_up_all(&mthread->t_ctl_waitq);
1690 if (lad->lad_to_double_scan) {
1691 lad->lad_to_double_scan = 0;
1692 atomic_inc(&lfsck->li_double_scan_count);
1693 lad->lad_in_double_scan = 1;
1694 wake_up_all(&mthread->t_ctl_waitq);
1696 com->lc_new_checked = 0;
1697 com->lc_new_scanned = 0;
1698 com->lc_time_last_checkpoint = cfs_time_current();
1699 com->lc_time_next_checkpoint =
1700 com->lc_time_last_checkpoint +
1701 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1703 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1704 "the second-stage scaning\n",
1705 lfsck_lfsck2name(lfsck));
1707 /* Flush async updates before handling orphan. */
1708 rc2 = dt_sync(env, lfsck->li_next);
1710 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1711 "scan start, synced: rc = %d\n",
1712 lfsck_lfsck2name(lfsck), rc2);
1714 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1715 GOTO(cleanup2, rc = 0);
1717 while (lad->lad_in_double_scan) {
1718 rc = lfsck_assistant_query_others(env, com);
1719 if (lfsck_phase2_next_ready(lad))
1725 /* Pull LFSCK status on related targets once
1726 * per 30 seconds if we are not notified. */
1727 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1728 cfs_time_seconds(1),
1730 rc = l_wait_event(athread->t_ctl_waitq,
1731 lfsck_phase2_next_ready(lad) ||
1733 !thread_is_running(mthread),
1736 if (unlikely(lad->lad_exit ||
1737 !thread_is_running(mthread)))
1738 GOTO(cleanup2, rc = 0);
1740 if (rc == -ETIMEDOUT)
1747 rc = lao->la_handler_p2(env, com);
1751 if (unlikely(lad->lad_exit ||
1752 !thread_is_running(mthread)))
1753 GOTO(cleanup2, rc = 0);
1759 /* Cleanup the unfinished requests. */
1760 spin_lock(&lad->lad_lock);
1762 lad->lad_assistant_status = rc;
1764 if (lad->lad_exit && lad->lad_post_result <= 0)
1765 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1767 while (!list_empty(&lad->lad_req_list)) {
1768 lar = list_entry(lad->lad_req_list.next,
1769 struct lfsck_assistant_req,
1771 list_del_init(&lar->lar_list);
1772 lad->lad_prefetched--;
1773 spin_unlock(&lad->lad_lock);
1774 lao->la_req_fini(env, lar);
1775 spin_lock(&lad->lad_lock);
1777 spin_unlock(&lad->lad_lock);
1779 LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1780 lad->lad_prefetched);
1783 memset(lr, 0, sizeof(*lr));
1785 lr->lr_event = LE_PHASE2_DONE;
1787 } else if (rc == 0) {
1788 if (lfsck->li_flags & LPF_ALL_TGT) {
1789 lr->lr_event = LE_STOP;
1790 lr->lr_status = LS_STOPPED;
1792 lr->lr_event = LE_PEER_EXIT;
1793 switch (lfsck->li_status) {
1796 lr->lr_status = LS_CO_PAUSED;
1800 lr->lr_status = LS_CO_STOPPED;
1803 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1804 "status: rc = %d\n",
1805 lfsck_lfsck2name(lfsck),
1807 lr->lr_status = LS_CO_FAILED;
1812 if (lfsck->li_flags & LPF_ALL_TGT) {
1813 lr->lr_event = LE_STOP;
1814 lr->lr_status = LS_FAILED;
1816 lr->lr_event = LE_PEER_EXIT;
1817 lr->lr_status = LS_CO_FAILED;
1821 rc1 = lfsck_assistant_notify_others(env, com, lr);
1823 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1824 "others for %s quit: rc = %d\n",
1825 lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1829 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1830 lfsck_lfsck2name(lfsck));
1832 /* Flush async updates before exit. */
1833 rc2 = dt_sync(env, lfsck->li_next);
1835 CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1836 lfsck_lfsck2name(lfsck), rc2);
1838 /* Under force exit case, some requests may be just freed without
1839 * verification, those objects should be re-handled when next run.
1840 * So not update the on-disk trace file under such case. */
1841 if (lad->lad_in_double_scan) {
1843 rc1 = lao->la_double_scan_result(env, com, rc);
1845 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1846 "finished: rc = %d\n",
1847 lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1851 if (lad->lad_in_double_scan)
1852 atomic_dec(&lfsck->li_double_scan_count);
1854 spin_lock(&lad->lad_lock);
1855 lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1856 thread_set_flags(athread, SVC_STOPPED);
1857 wake_up_all(&mthread->t_ctl_waitq);
1858 spin_unlock(&lad->lad_lock);
1860 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1861 lfsck_lfsck2name(lfsck), lad->lad_name,
1862 lad->lad_assistant_status);
1864 lfsck_thread_args_fini(lta);