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 *dt = lfsck->li_bottom;
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, BYPASS_CAPA);
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, dt);
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(env, dt, th);
145 rc = dt_xattr_set(env, obj, buf, XATTR_NAME_LMA, fl, th, BYPASS_CAPA);
150 dt_trans_stop(env, dt, 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 *)"..", BYPASS_CAPA);
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 dt_device *dev = lfsck->li_bottom;
198 struct seq_server_site *ss = lu_site2seq(dev->dd_lu_dev.ld_site);
199 __u32 idx = lfsck_dev_idx(dev);
203 if (list_empty(&lfsck->li_list_dir) || !S_ISDIR(lfsck_object_type(obj)))
208 *fid = *lfsck_dto2fid(obj);
210 /* Global /ROOT is visible. */
211 if (unlikely(lu_fid_eq(fid, &lfsck->li_global_root_fid)))
214 /* Backend root is invisible. */
215 if (unlikely(lu_fid_eq(fid, &lfsck->li_local_root_fid)))
218 if (!fid_is_norm(fid) && !fid_is_igif(fid))
221 fld_range_set_mdt(range);
222 rc = fld_local_lookup(env, ss->ss_server_fld,
223 fid_seq(fid), range);
224 if (rc != 0 || range->lsr_index != idx)
225 /* Current FID should NOT be for the input parameter
226 * @obj, because the lfsck_master_oit_engine() has
227 * filtered out agent object. So current FID is for
228 * the ancestor of the original input parameter @obj.
229 * So the ancestor is a remote directory. The input
230 * parameter @obj is local directory, and should be
231 * scanned under such case. */
234 /* normal FID on this target (locally) must be for the
235 * client-side visiable object. */
236 if (fid_is_norm(fid))
240 obj = lfsck_object_find(env, lfsck, fid);
245 if (!dt_object_exists(obj))
249 dt_read_lock(env, obj, MOR_TGT_CHILD);
250 if (unlikely(lfsck_is_dead_obj(obj))) {
251 dt_read_unlock(env, obj);
256 rc = dt_xattr_get(env, obj,
257 lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK,
259 dt_read_unlock(env, obj);
263 if (rc < 0 && rc != -ENODATA)
266 rc = lfsck_parent_fid(env, obj, fid);
268 lfsck_object_put(env, obj);
274 if (!fid_is_sane(fid))
279 if (depth > 0 && obj != NULL)
280 lfsck_object_put(env, obj);
285 static int lfsck_load_stripe_lmv(const struct lu_env *env,
286 struct lfsck_instance *lfsck,
287 struct dt_object *obj)
289 struct lmv_mds_md_v1 *lmv = &lfsck_env_info(env)->lti_lmv;
290 struct lfsck_lmv *llmv;
294 LASSERT(lfsck->li_obj_dir == NULL);
295 LASSERT(lfsck->li_lmv == NULL);
297 rc = lfsck_read_stripe_lmv(env, obj, lmv);
298 if (rc == -ENODATA) {
299 lfsck->li_obj_dir = lfsck_object_get(obj);
311 if (lmv->lmv_magic == LMV_MAGIC) {
312 struct lfsck_slave_lmv_rec *lslr;
315 llmv->ll_lmv_master = 1;
316 if (lmv->lmv_stripe_count < 1)
317 stripes = LFSCK_LMV_DEF_STRIPES;
318 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
319 stripes = LFSCK_LMV_MAX_STRIPES;
321 stripes = lmv->lmv_stripe_count;
323 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
330 /* Find the object against the bottom device. */
331 obj = lfsck_object_find_by_dev(env, lfsck->li_bottom,
334 OBD_FREE_LARGE(lslr, sizeof(*lslr) * stripes);
337 RETURN(PTR_ERR(obj));
340 llmv->ll_stripes_allocated = stripes;
341 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
342 llmv->ll_lslr = lslr;
343 lfsck->li_obj_dir = obj;
345 llmv->ll_lmv_slave = 1;
346 lfsck->li_obj_dir = lfsck_object_get(obj);
350 atomic_set(&llmv->ll_ref, 1);
351 lfsck->li_lmv = llmv;
356 /* LFSCK wrap functions */
358 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
361 struct lfsck_component *com;
363 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
364 com->lc_ops->lfsck_fail(env, com, new_checked);
368 void lfsck_close_dir(const struct lu_env *env,
369 struct lfsck_instance *lfsck, int result)
371 struct lfsck_component *com;
374 if (lfsck->li_lmv != NULL) {
375 lfsck->li_lmv->ll_exit_value = result;
376 if (lfsck->li_obj_dir != NULL) {
377 list_for_each_entry(com, &lfsck->li_list_dir,
379 com->lc_ops->lfsck_close_dir(env, com);
383 lfsck_lmv_put(env, lfsck->li_lmv);
384 lfsck->li_lmv = NULL;
387 if (lfsck->li_di_dir != NULL) {
388 const struct dt_it_ops *dir_iops;
389 struct dt_it *dir_di = lfsck->li_di_dir;
391 LASSERT(lfsck->li_obj_dir != NULL);
393 dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
394 lfsck_di_dir_put(env, lfsck);
395 dir_iops->fini(env, dir_di);
398 if (lfsck->li_obj_dir != NULL) {
399 struct dt_object *dir_obj = lfsck->li_obj_dir;
401 lfsck->li_obj_dir = NULL;
402 lfsck_object_put(env, dir_obj);
408 int lfsck_open_dir(const struct lu_env *env,
409 struct lfsck_instance *lfsck, __u64 cookie)
411 struct dt_object *obj = lfsck->li_obj_dir;
412 struct dt_it *di = lfsck->li_di_dir;
413 struct lfsck_component *com;
414 const struct dt_it_ops *iops;
418 LASSERT(obj != NULL);
421 if (unlikely(!dt_try_as_dir(env, obj)))
422 GOTO(out, rc = -ENOTDIR);
424 list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
425 rc = com->lc_ops->lfsck_open_dir(env, com);
430 iops = &obj->do_index_ops->dio_it;
431 di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA);
433 GOTO(out, rc = PTR_ERR(di));
435 rc = iops->load(env, di, cookie);
436 if (rc == 0 || (rc > 0 && cookie > 0))
437 rc = iops->next(env, di);
445 lfsck->li_cookie_dir = iops->store(env, di);
446 spin_lock(&lfsck->li_lock);
447 lfsck->li_di_dir = di;
448 spin_unlock(&lfsck->li_lock);
455 lfsck_close_dir(env, lfsck, rc);
460 static int lfsck_checkpoint(const struct lu_env *env,
461 struct lfsck_instance *lfsck)
463 struct lfsck_component *com;
467 if (likely(cfs_time_beforeq(cfs_time_current(),
468 lfsck->li_time_next_checkpoint)))
471 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
472 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
473 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
478 lfsck->li_time_last_checkpoint = cfs_time_current();
479 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
480 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
481 return rc1 != 0 ? rc1 : rc;
484 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
485 struct lfsck_start_param *lsp)
487 struct dt_object *obj = NULL;
488 struct lfsck_component *com;
489 struct lfsck_component *next;
490 struct lfsck_position *pos = NULL;
491 const struct dt_it_ops *iops =
492 &lfsck->li_obj_oit->do_index_ops->dio_it;
496 LASSERT(lfsck->li_obj_dir == NULL);
497 LASSERT(lfsck->li_di_dir == NULL);
499 lfsck->li_current_oit_processed = 0;
500 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
501 com->lc_new_checked = 0;
502 rc = com->lc_ops->lfsck_prep(env, com, lsp);
507 (!lfsck_pos_is_zero(&com->lc_pos_start) &&
508 lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
509 pos = &com->lc_pos_start;
512 /* Init otable-based iterator. */
514 rc = iops->load(env, lfsck->li_di_oit, 0);
516 lfsck->li_oit_over = 1;
523 rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
527 lfsck->li_oit_over = 1;
529 if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
532 /* Find the directory for namespace-based traverse. */
533 obj = lfsck_object_find(env, lfsck, &pos->lp_dir_parent);
535 RETURN(PTR_ERR(obj));
537 /* Remote directory will be scanned by the LFSCK instance
538 * on the MDT where the remote object really resides on. */
539 if (!dt_object_exists(obj) || dt_object_remote(obj) ||
540 unlikely(!S_ISDIR(lfsck_object_type(obj))))
543 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
545 /* For the master MDT-object of a striped directory,
546 * reset the iteration from the directory beginning. */
547 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
548 pos->lp_dir_cookie = 0;
550 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
552 /* The end of the directory. */
560 lfsck_object_put(env, obj);
563 lfsck_close_dir(env, lfsck, rc);
564 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
566 com->lc_ops->lfsck_post(env, com, rc, true);
573 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
574 lfsck->li_pos_current = lfsck->li_pos_checkpoint;
575 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
576 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
581 lfsck->li_time_last_checkpoint = cfs_time_current();
582 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
583 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
587 static int lfsck_exec_oit(const struct lu_env *env,
588 struct lfsck_instance *lfsck, struct dt_object *obj)
590 struct lfsck_component *com;
594 LASSERT(lfsck->li_obj_dir == NULL);
596 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
597 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
602 rc = lfsck_needs_scan_dir(env, lfsck, obj);
606 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
608 rc = lfsck_open_dir(env, lfsck, 0);
614 lfsck_fail(env, lfsck, false);
617 lfsck_close_dir(env, lfsck, rc);
619 return rc > 0 ? 0 : rc;
622 static int lfsck_exec_dir(const struct lu_env *env,
623 struct lfsck_instance *lfsck,
624 struct lu_dirent *ent, __u16 type)
626 struct lfsck_component *com;
629 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
630 rc = com->lc_ops->lfsck_exec_dir(env, com, ent, type);
637 static int lfsck_master_dir_engine(const struct lu_env *env,
638 struct lfsck_instance *lfsck);
640 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
643 struct lfsck_component *com;
644 struct lfsck_component *next;
647 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
648 lfsck_close_dir(env, lfsck, result);
650 while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
651 !list_empty(&lfsck->li_list_lmv)) {
652 struct lfsck_lmv_unit *llu;
654 spin_lock(&lfsck->li_lock);
655 llu = list_entry(lfsck->li_list_lmv.next,
656 struct lfsck_lmv_unit, llu_link);
657 list_del_init(&llu->llu_link);
658 spin_unlock(&lfsck->li_lock);
660 lfsck->li_lmv = &llu->llu_lmv;
661 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
662 rc = lfsck_open_dir(env, lfsck, 0);
664 rc = lfsck_master_dir_engine(env, lfsck);
665 lfsck_close_dir(env, lfsck, result);
671 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
672 rc = com->lc_ops->lfsck_post(env, com, result, false);
674 CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
675 "rc = %d\n", lfsck_lfsck2name(lfsck),
676 (__u32)com->lc_type, rc);
679 lfsck->li_time_last_checkpoint = cfs_time_current();
680 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
681 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
683 /* Ignore some component post failure to make other can go ahead. */
687 static int lfsck_double_scan(const struct lu_env *env,
688 struct lfsck_instance *lfsck)
690 struct lfsck_component *com;
691 struct lfsck_component *next;
692 struct l_wait_info lwi = { 0 };
696 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
697 rc = com->lc_ops->lfsck_double_scan(env, com);
702 l_wait_event(lfsck->li_thread.t_ctl_waitq,
703 atomic_read(&lfsck->li_double_scan_count) == 0,
706 if (lfsck->li_status != LS_PAUSED &&
707 lfsck->li_status != LS_CO_PAUSED) {
708 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
710 spin_lock(&lfsck->li_lock);
711 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
712 spin_unlock(&lfsck->li_lock);
716 return rc1 != 0 ? rc1 : rc;
719 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
721 struct lfsck_component *com;
722 struct lfsck_component *next;
724 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
726 if (com->lc_ops->lfsck_quit != NULL)
727 com->lc_ops->lfsck_quit(env, com);
729 spin_lock(&lfsck->li_lock);
730 list_del_init(&com->lc_link_dir);
731 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
732 spin_unlock(&lfsck->li_lock);
735 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
737 if (com->lc_ops->lfsck_quit != NULL)
738 com->lc_ops->lfsck_quit(env, com);
740 spin_lock(&lfsck->li_lock);
741 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
742 spin_unlock(&lfsck->li_lock);
748 static int lfsck_master_dir_engine(const struct lu_env *env,
749 struct lfsck_instance *lfsck)
751 struct lfsck_thread_info *info = lfsck_env_info(env);
752 struct dt_object *dir = lfsck->li_obj_dir;
753 const struct dt_it_ops *iops = &dir->do_index_ops->dio_it;
754 struct dt_it *di = lfsck->li_di_dir;
755 struct lu_dirent *ent =
756 (struct lu_dirent *)info->lti_key;
757 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
758 struct ptlrpc_thread *thread = &lfsck->li_thread;
764 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
765 unlikely(!thread_is_running(thread))) {
766 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
767 "parent "DFID", cookie "LPX64"\n",
768 lfsck_lfsck2name(lfsck),
769 PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
774 lfsck->li_new_scanned++;
775 rc = iops->rec(env, di, (struct dt_rec *)ent,
778 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
782 CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
783 "parent "DFID", cookie "LPX64": rc = %d\n",
784 lfsck_lfsck2name(lfsck),
785 PFID(lfsck_dto2fid(dir)),
786 lfsck->li_cookie_dir, rc);
787 lfsck_fail(env, lfsck, true);
788 if (bk->lb_param & LPF_FAILOUT)
794 if (ent->lde_attrs & LUDA_IGNORE &&
795 strcmp(ent->lde_name, dotdot) != 0)
798 /* The type in the @ent structure may has been overwritten,
799 * so we need to pass the @type parameter independently. */
800 rc = lfsck_exec_dir(env, lfsck, ent, type);
801 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
805 rc = lfsck_checkpoint(env, lfsck);
806 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
810 lfsck_control_speed(lfsck);
811 if (unlikely(!thread_is_running(thread))) {
812 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
813 "parent "DFID", cookie "LPX64"\n",
814 lfsck_lfsck2name(lfsck),
815 PFID(lfsck_dto2fid(dir)),
816 lfsck->li_cookie_dir);
820 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
821 spin_lock(&lfsck->li_lock);
822 thread_set_flags(thread, SVC_STOPPING);
823 spin_unlock(&lfsck->li_lock);
827 rc = iops->next(env, di);
830 if (rc > 0 && !lfsck->li_oit_over)
831 lfsck_close_dir(env, lfsck, rc);
837 * Object-table based iteration engine.
839 * Object-table based iteration is the basic linear engine to scan all the
840 * objects on current device in turn. For each object, it calls all the
841 * registered LFSCK component(s)' API to perform related consistency
844 * It flushes related LFSCK trace files to disk via making checkpoint
845 * periodically. Then if the server crashed or the LFSCK is paused, the
846 * LFSCK can resume from the latest checkpoint.
848 * It also controls the whole LFSCK speed via lfsck_control_speed() to
849 * avoid the server to become overload.
851 * \param[in] env pointer to the thread context
852 * \param[in] lfsck pointer to the lfsck instance
854 * \retval positive number if all objects have been scanned
855 * \retval 0 if the iteration is stopped or paused
856 * \retval negative error number on failure
858 static int lfsck_master_oit_engine(const struct lu_env *env,
859 struct lfsck_instance *lfsck)
861 struct lfsck_thread_info *info = lfsck_env_info(env);
862 const struct dt_it_ops *iops =
863 &lfsck->li_obj_oit->do_index_ops->dio_it;
864 struct dt_it *di = lfsck->li_di_oit;
865 struct lu_fid *fid = &info->lti_fid;
866 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
867 struct ptlrpc_thread *thread = &lfsck->li_thread;
868 struct dt_device *dev = lfsck->li_bottom;
869 struct seq_server_site *ss = lu_site2seq(dev->dd_lu_dev.ld_site);
870 __u32 idx = lfsck_dev_idx(dev);
874 if (unlikely(ss == NULL))
878 struct dt_object *target;
879 bool update_lma = false;
881 if (lfsck->li_di_dir != NULL) {
882 rc = lfsck_master_dir_engine(env, lfsck);
887 if (unlikely(lfsck->li_oit_over))
890 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
891 unlikely(!thread_is_running(thread))) {
892 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
894 lfsck_lfsck2name(lfsck), iops->store(env, di));
899 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
902 lfsck->li_current_oit_processed = 1;
904 if (!list_empty(&lfsck->li_list_lmv)) {
905 struct lfsck_lmv_unit *llu;
907 spin_lock(&lfsck->li_lock);
908 llu = list_entry(lfsck->li_list_lmv.next,
909 struct lfsck_lmv_unit, llu_link);
910 list_del_init(&llu->llu_link);
911 spin_unlock(&lfsck->li_lock);
913 lfsck->li_lmv = &llu->llu_lmv;
914 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
915 rc = lfsck_open_dir(env, lfsck, 0);
917 rc = lfsck_master_dir_engine(env, lfsck);
923 lfsck->li_new_scanned++;
924 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
925 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
927 CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
928 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
929 lfsck_fail(env, lfsck, true);
930 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
936 if (fid_is_idif(fid)) {
937 __u32 idx1 = fid_idif_ost_idx(fid);
939 LASSERT(!lfsck->li_master);
941 /* It is an old format device, update the LMA. */
943 struct ost_id *oi = &info->lti_oi;
945 fid_to_ostid(fid, oi);
946 ostid_to_fid(fid, oi, idx);
949 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
950 !fid_is_last_id(fid) &&
951 !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
953 /* If the FID/object is only used locally and invisible
954 * to external nodes, then LFSCK will not handle it.
956 * dot_lustre sequence has been handled specially. */
959 struct lu_seq_range *range = &info->lti_range;
961 if (lfsck->li_master)
962 fld_range_set_mdt(range);
964 fld_range_set_ost(range);
965 rc = fld_local_lookup(env, ss->ss_server_fld,
966 fid_seq(fid), range);
967 if (rc != 0 || range->lsr_index != idx) {
968 /* Remote object will be handled by the LFSCK
969 * instance on the MDT where the remote object
970 * really resides on. */
976 target = lfsck_object_find(env, lfsck, fid);
977 if (IS_ERR(target)) {
978 CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
979 DFID", cookie "LPU64": rc = %d\n",
980 lfsck_lfsck2name(lfsck), PFID(fid),
981 iops->store(env, di), rc);
982 lfsck_fail(env, lfsck, true);
983 if (bk->lb_param & LPF_FAILOUT)
984 RETURN(PTR_ERR(target));
989 if (dt_object_exists(target)) {
991 rc = lfsck_update_lma(env, lfsck, target);
993 CDEBUG(D_LFSCK, "%s: fail to update "
994 "LMA for "DFID": rc = %d\n",
995 lfsck_lfsck2name(lfsck),
996 PFID(lfsck_dto2fid(target)), rc);
999 rc = lfsck_exec_oit(env, lfsck, target);
1001 lfsck_object_put(env, target);
1002 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1006 rc = lfsck_checkpoint(env, lfsck);
1007 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1011 lfsck_control_speed(lfsck);
1013 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
1014 spin_lock(&lfsck->li_lock);
1015 thread_set_flags(thread, SVC_STOPPING);
1016 spin_unlock(&lfsck->li_lock);
1020 rc = iops->next(env, di);
1021 if (unlikely(rc > 0))
1022 lfsck->li_oit_over = 1;
1023 else if (likely(rc == 0))
1024 lfsck->li_current_oit_processed = 0;
1026 if (unlikely(!thread_is_running(thread))) {
1027 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1028 "cookie "LPU64"\n", lfsck_lfsck2name(lfsck),
1029 iops->store(env, di));
1032 } while (rc == 0 || lfsck->li_di_dir != NULL);
1037 int lfsck_master_engine(void *args)
1039 struct lfsck_thread_args *lta = args;
1040 struct lu_env *env = <a->lta_env;
1041 struct lfsck_instance *lfsck = lta->lta_lfsck;
1042 struct ptlrpc_thread *thread = &lfsck->li_thread;
1043 struct dt_object *oit_obj = lfsck->li_obj_oit;
1044 const struct dt_it_ops *oit_iops = &oit_obj->do_index_ops->dio_it;
1045 struct dt_it *oit_di;
1046 struct l_wait_info lwi = { 0 };
1050 if (lfsck->li_master &&
1051 (!list_empty(&lfsck->li_list_scan) ||
1052 !list_empty(&lfsck->li_list_double_scan))) {
1053 rc = lfsck_verify_lpf(env, lfsck);
1054 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1055 * fatal, because the .lustre/lost+found/ maybe not accessed
1056 * by the LFSCK if it does not add orphans or others to such
1057 * directory. So go ahead until hit failure when really uses
1060 CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1061 ".lustre/lost+found/, go ahead: rc = %d\n",
1062 lfsck_lfsck2name(lfsck), rc);
1065 oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit, BYPASS_CAPA);
1066 if (IS_ERR(oit_di)) {
1067 rc = PTR_ERR(oit_di);
1068 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1069 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1071 GOTO(fini_args, rc);
1074 spin_lock(&lfsck->li_lock);
1075 lfsck->li_di_oit = oit_di;
1076 spin_unlock(&lfsck->li_lock);
1077 rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1081 CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1082 "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1083 ", pid = %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 spin_lock(&lfsck->li_lock);
1090 thread_set_flags(thread, SVC_RUNNING);
1091 spin_unlock(&lfsck->li_lock);
1092 wake_up_all(&thread->t_ctl_waitq);
1094 l_wait_event(thread->t_ctl_waitq,
1095 lfsck->li_start_unplug ||
1096 !thread_is_running(thread),
1098 if (!thread_is_running(thread))
1099 GOTO(fini_oit, rc = 0);
1101 if (!list_empty(&lfsck->li_list_scan) ||
1102 list_empty(&lfsck->li_list_double_scan))
1103 rc = lfsck_master_oit_engine(env, lfsck);
1107 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
1108 CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1109 "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1110 ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1111 lfsck->li_pos_checkpoint.lp_oit_cookie,
1112 lfsck->li_pos_checkpoint.lp_dir_cookie,
1113 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1116 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1117 rc = lfsck_post(env, lfsck, rc);
1119 lfsck_close_dir(env, lfsck, rc);
1122 lfsck_di_oit_put(env, lfsck);
1123 oit_iops->fini(env, oit_di);
1125 if (!list_empty(&lfsck->li_list_double_scan))
1126 rc = lfsck_double_scan(env, lfsck);
1130 lfsck_quit(env, lfsck);
1133 /* XXX: Purge the pinned objects in the future. */
1136 spin_lock(&lfsck->li_lock);
1137 thread_set_flags(thread, SVC_STOPPED);
1138 spin_unlock(&lfsck->li_lock);
1139 wake_up_all(&thread->t_ctl_waitq);
1140 lfsck_thread_args_fini(lta);
1144 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1148 spin_lock(&lad->lad_lock);
1149 if (list_empty(&lad->lad_req_list))
1151 spin_unlock(&lad->lad_lock);
1157 * Query the LFSCK status from the instatnces on remote servers.
1159 * The LFSCK assistant thread queries the LFSCK instances on other
1160 * servers (MDT/OST) about their status, such as whether they have
1161 * finished the phase1/phase2 scanning or not, and so on.
1163 * \param[in] env pointer to the thread context
1164 * \param[in] com pointer to the lfsck component
1166 * \retval 0 for success
1167 * \retval negative error number on failure
1169 static int lfsck_assistant_query_others(const struct lu_env *env,
1170 struct lfsck_component *com)
1172 struct lfsck_thread_info *info = lfsck_env_info(env);
1173 struct lfsck_request *lr = &info->lti_lr;
1174 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1175 struct lfsck_instance *lfsck = com->lc_lfsck;
1176 struct lfsck_assistant_data *lad = com->lc_data;
1177 struct ptlrpc_request_set *set;
1178 struct lfsck_tgt_descs *ltds;
1179 struct lfsck_tgt_desc *ltd;
1180 struct list_head *phase_head;
1185 set = ptlrpc_prep_set();
1189 lad->lad_touch_gen++;
1190 memset(lr, 0, sizeof(*lr));
1191 lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1192 lr->lr_event = LE_QUERY;
1193 lr->lr_active = com->lc_type;
1194 laia->laia_com = com;
1196 laia->laia_shared = 0;
1198 if (!list_empty(&lad->lad_mdt_phase1_list)) {
1199 ltds = &lfsck->li_mdt_descs;
1201 phase_head = &lad->lad_mdt_phase1_list;
1202 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1207 ltds = &lfsck->li_ost_descs;
1208 lr->lr_flags = LEF_TO_OST;
1209 phase_head = &lad->lad_ost_phase1_list;
1212 laia->laia_ltds = ltds;
1213 spin_lock(<ds->ltd_lock);
1214 while (!list_empty(phase_head)) {
1215 struct list_head *phase_list;
1218 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1219 ltd = list_entry(phase_head->next,
1220 struct lfsck_tgt_desc,
1221 ltd_layout_phase_list);
1222 phase_list = <d->ltd_layout_phase_list;
1223 gen = <d->ltd_layout_gen;
1225 ltd = list_entry(phase_head->next,
1226 struct lfsck_tgt_desc,
1227 ltd_namespace_phase_list);
1228 phase_list = <d->ltd_namespace_phase_list;
1229 gen = <d->ltd_namespace_gen;
1232 if (*gen == lad->lad_touch_gen)
1235 *gen = lad->lad_touch_gen;
1236 list_move_tail(phase_list, phase_head);
1237 atomic_inc(<d->ltd_ref);
1238 laia->laia_ltd = ltd;
1239 spin_unlock(<ds->ltd_lock);
1240 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1241 lfsck_async_interpret_common,
1244 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1245 "%s %x for %s: rc = %d\n",
1246 lfsck_lfsck2name(lfsck),
1247 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1248 ltd->ltd_index, lad->lad_name, rc);
1252 spin_lock(<ds->ltd_lock);
1254 spin_unlock(<ds->ltd_lock);
1256 rc = ptlrpc_set_wait(set);
1258 ptlrpc_set_destroy(set);
1262 if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1263 list_empty(&lad->lad_mdt_phase1_list))
1267 ptlrpc_set_destroy(set);
1269 RETURN(rc1 != 0 ? rc1 : rc);
1273 * Notify the LFSCK event to the instances on remote servers.
1275 * The LFSCK assistant thread notifies the LFSCK instances on other
1276 * servers (MDT/OST) about some events, such as start new scanning,
1277 * stop the scanning, this LFSCK instance will exit, and so on.
1279 * \param[in] env pointer to the thread context
1280 * \param[in] com pointer to the lfsck component
1281 * \param[in] lr pointer to the LFSCK event request
1283 * \retval 0 for success
1284 * \retval negative error number on failure
1286 static int lfsck_assistant_notify_others(const struct lu_env *env,
1287 struct lfsck_component *com,
1288 struct lfsck_request *lr)
1290 struct lfsck_thread_info *info = lfsck_env_info(env);
1291 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1292 struct lfsck_instance *lfsck = com->lc_lfsck;
1293 struct lfsck_assistant_data *lad = com->lc_data;
1294 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1295 struct ptlrpc_request_set *set;
1296 struct lfsck_tgt_descs *ltds;
1297 struct lfsck_tgt_desc *ltd;
1298 struct lfsck_tgt_desc *next;
1304 set = ptlrpc_prep_set();
1308 lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1309 lr->lr_active = com->lc_type;
1310 laia->laia_com = com;
1312 laia->laia_shared = 0;
1314 switch (lr->lr_event) {
1316 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1319 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1320 lr->lr_speed = bk->lb_speed_limit;
1321 lr->lr_version = bk->lb_version;
1322 lr->lr_param |= bk->lb_param;
1323 lr->lr_async_windows = bk->lb_async_windows;
1324 lr->lr_flags = LEF_TO_OST;
1326 /* Notify OSTs firstly, then handle other MDTs if needed. */
1327 ltds = &lfsck->li_ost_descs;
1328 laia->laia_ltds = ltds;
1329 down_read(<ds->ltd_rw_sem);
1330 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1331 ltd = lfsck_tgt_get(ltds, idx);
1332 LASSERT(ltd != NULL);
1334 laia->laia_ltd = ltd;
1335 ltd->ltd_layout_done = 0;
1336 ltd->ltd_synced_failures = 0;
1337 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1338 lfsck_async_interpret_common,
1339 laia, LFSCK_NOTIFY);
1341 lfsck_lad_set_bitmap(env, com, idx);
1342 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1343 "notify OST %x for %s start: rc = %d\n",
1344 lfsck_lfsck2name(lfsck), idx,
1349 up_read(<ds->ltd_rw_sem);
1352 rc = ptlrpc_set_wait(set);
1354 ptlrpc_set_destroy(set);
1359 if (!(bk->lb_param & LPF_ALL_TGT))
1362 /* link other MDT targets locallly. */
1363 ltds = &lfsck->li_mdt_descs;
1364 spin_lock(<ds->ltd_lock);
1365 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1366 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1367 ltd = LTD_TGT(ltds, idx);
1368 LASSERT(ltd != NULL);
1370 if (!list_empty(<d->ltd_layout_list))
1373 list_add_tail(<d->ltd_layout_list,
1374 &lad->lad_mdt_list);
1375 list_add_tail(<d->ltd_layout_phase_list,
1376 &lad->lad_mdt_phase1_list);
1379 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1380 ltd = LTD_TGT(ltds, idx);
1381 LASSERT(ltd != NULL);
1383 if (!list_empty(<d->ltd_namespace_list))
1386 list_add_tail(<d->ltd_namespace_list,
1387 &lad->lad_mdt_list);
1388 list_add_tail(<d->ltd_namespace_phase_list,
1389 &lad->lad_mdt_phase1_list);
1392 spin_unlock(<ds->ltd_lock);
1395 case LE_PHASE2_DONE:
1396 case LE_PEER_EXIT: {
1397 struct list_head *phase_head;
1399 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1400 if (bk->lb_param & LPF_ALL_TGT) {
1401 phase_head = &lad->lad_mdt_list;
1402 ltds = &lfsck->li_mdt_descs;
1403 if (lr->lr_event == LE_STOP) {
1404 /* unlink other MDT targets locallly. */
1405 spin_lock(<ds->ltd_lock);
1406 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1407 list_for_each_entry_safe(ltd, next,
1408 phase_head, ltd_layout_list) {
1410 <d->ltd_layout_phase_list);
1412 <d->ltd_layout_list);
1415 list_for_each_entry_safe(ltd, next,
1417 ltd_namespace_list) {
1419 <d->ltd_namespace_phase_list);
1421 <d->ltd_namespace_list);
1424 spin_unlock(<ds->ltd_lock);
1426 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1429 lr->lr_flags |= LEF_TO_OST;
1430 phase_head = &lad->lad_ost_list;
1431 ltds = &lfsck->li_ost_descs;
1433 lr->lr_flags &= ~LEF_TO_OST;
1435 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1438 lr->lr_flags |= LEF_TO_OST;
1439 phase_head = &lad->lad_ost_list;
1440 ltds = &lfsck->li_ost_descs;
1444 laia->laia_ltds = ltds;
1445 spin_lock(<ds->ltd_lock);
1446 while (!list_empty(phase_head)) {
1447 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1448 ltd = list_entry(phase_head->next,
1449 struct lfsck_tgt_desc,
1451 if (!list_empty(<d->ltd_layout_phase_list))
1453 <d->ltd_layout_phase_list);
1454 list_del_init(<d->ltd_layout_list);
1456 ltd = list_entry(phase_head->next,
1457 struct lfsck_tgt_desc,
1458 ltd_namespace_list);
1459 if (!list_empty(<d->ltd_namespace_phase_list))
1461 <d->ltd_namespace_phase_list);
1462 list_del_init(<d->ltd_namespace_list);
1464 atomic_inc(<d->ltd_ref);
1465 laia->laia_ltd = ltd;
1466 spin_unlock(<ds->ltd_lock);
1467 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1468 lfsck_async_interpret_common,
1469 laia, LFSCK_NOTIFY);
1471 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1472 "notify %s %x for %s stop/phase2_done/"
1473 "peer_exit: rc = %d\n",
1474 lfsck_lfsck2name(lfsck),
1475 (lr->lr_flags & LEF_TO_OST) ?
1476 "OST" : "MDT", ltd->ltd_index,
1480 spin_lock(<ds->ltd_lock);
1482 spin_unlock(<ds->ltd_lock);
1484 rc = ptlrpc_set_wait(set);
1486 ptlrpc_set_destroy(set);
1490 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1491 !(lr->lr_flags & LEF_TO_OST)) {
1492 lr->lr_flags |= LEF_TO_OST;
1493 phase_head = &lad->lad_ost_list;
1494 ltds = &lfsck->li_ost_descs;
1499 case LE_PHASE1_DONE:
1500 lad->lad_ops->la_sync_failures(env, com, lr);
1501 lad->lad_touch_gen++;
1502 ltds = &lfsck->li_mdt_descs;
1503 laia->laia_ltds = ltds;
1504 spin_lock(<ds->ltd_lock);
1505 while (!list_empty(&lad->lad_mdt_list)) {
1506 struct list_head *list;
1509 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1510 ltd = list_entry(lad->lad_mdt_list.next,
1511 struct lfsck_tgt_desc,
1513 list = <d->ltd_layout_list;
1514 gen = <d->ltd_layout_gen;
1516 struct lfsck_namespace *ns = com->lc_file_ram;
1518 ltd = list_entry(lad->lad_mdt_list.next,
1519 struct lfsck_tgt_desc,
1520 ltd_namespace_list);
1521 list = <d->ltd_namespace_list;
1522 gen = <d->ltd_namespace_gen;
1523 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1526 if (*gen == lad->lad_touch_gen)
1529 *gen = lad->lad_touch_gen;
1530 list_move_tail(list, &lad->lad_mdt_list);
1531 if (ltd->ltd_synced_failures)
1534 atomic_inc(<d->ltd_ref);
1535 laia->laia_ltd = ltd;
1536 spin_unlock(<ds->ltd_lock);
1537 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1538 lfsck_async_interpret_common,
1539 laia, LFSCK_NOTIFY);
1541 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1542 "notify MDT %x for %s phase1 done: "
1543 "rc = %d\n", lfsck_lfsck2name(lfsck),
1544 ltd->ltd_index, lad->lad_name, rc);
1547 spin_lock(<ds->ltd_lock);
1549 spin_unlock(<ds->ltd_lock);
1552 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1553 "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1558 rc1 = ptlrpc_set_wait(set);
1559 ptlrpc_set_destroy(set);
1561 RETURN(rc != 0 ? rc : rc1);
1565 * The LFSCK assistant thread is triggered by the LFSCK main engine.
1566 * They co-work together as an asynchronous pipeline: the LFSCK main
1567 * engine scans the system and pre-fetches the objects, attributes,
1568 * or name entries, etc, and pushes them into the pipeline as input
1569 * requests for the LFSCK assistant thread; on the other end of the
1570 * pipeline, the LFSCK assistant thread performs the real check and
1571 * repair for every request from the main engine.
1573 * Generally, the assistant engine may be blocked when check/repair
1574 * something, so the LFSCK main engine will run some faster. On the
1575 * other hand, the LFSCK main engine will drive multiple assistant
1576 * threads in parallel, means for each LFSCK component on the master
1577 * (such as layout LFSCK, namespace LFSCK), there is an independent
1578 * LFSCK assistant thread. So under such 1:N multiple asynchronous
1579 * pipelines mode, the whole LFSCK performance will be much better
1580 * than check/repair everything by the LFSCK main engine itself.
1582 int lfsck_assistant_engine(void *args)
1584 struct lfsck_thread_args *lta = args;
1585 struct lu_env *env = <a->lta_env;
1586 struct lfsck_component *com = lta->lta_com;
1587 struct lfsck_instance *lfsck = lta->lta_lfsck;
1588 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1589 struct lfsck_position *pos = &com->lc_pos_start;
1590 struct lfsck_thread_info *info = lfsck_env_info(env);
1591 struct lfsck_request *lr = &info->lti_lr;
1592 struct lfsck_assistant_data *lad = com->lc_data;
1593 struct ptlrpc_thread *mthread = &lfsck->li_thread;
1594 struct ptlrpc_thread *athread = &lad->lad_thread;
1595 struct lfsck_assistant_operations *lao = lad->lad_ops;
1596 struct lfsck_assistant_req *lar;
1597 struct l_wait_info lwi = { 0 };
1603 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1604 lfsck_lfsck2name(lfsck), lad->lad_name);
1606 memset(lr, 0, sizeof(*lr));
1607 lr->lr_event = LE_START;
1608 if (pos->lp_oit_cookie <= 1)
1609 lr->lr_param = LPF_RESET;
1610 rc = lfsck_assistant_notify_others(env, com, lr);
1612 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1613 "to start %s: rc = %d\n",
1614 lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1618 spin_lock(&lad->lad_lock);
1619 thread_set_flags(athread, SVC_RUNNING);
1620 spin_unlock(&lad->lad_lock);
1621 wake_up_all(&mthread->t_ctl_waitq);
1624 while (!list_empty(&lad->lad_req_list)) {
1625 bool wakeup = false;
1627 if (unlikely(lad->lad_exit ||
1628 !thread_is_running(mthread)))
1629 GOTO(cleanup1, rc = lad->lad_post_result);
1631 lar = list_entry(lad->lad_req_list.next,
1632 struct lfsck_assistant_req,
1634 /* Only the lfsck_assistant_engine thread itself can
1635 * remove the "lar" from the head of the list, LFSCK
1636 * engine thread only inserts other new "lar" at the
1637 * end of the list. So it is safe to handle current
1638 * "lar" without the spin_lock. */
1639 rc = lao->la_handler_p1(env, com, lar);
1640 spin_lock(&lad->lad_lock);
1641 list_del_init(&lar->lar_list);
1642 lad->lad_prefetched--;
1643 /* Wake up the main engine thread only when the list
1644 * is empty or half of the prefetched items have been
1645 * handled to avoid too frequent thread schedule. */
1646 if (lad->lad_prefetched == 0 ||
1647 (bk->lb_async_windows != 0 &&
1648 bk->lb_async_windows / 2 ==
1649 lad->lad_prefetched))
1651 spin_unlock(&lad->lad_lock);
1653 wake_up_all(&mthread->t_ctl_waitq);
1655 lao->la_req_fini(env, lar);
1656 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1660 l_wait_event(athread->t_ctl_waitq,
1661 !lfsck_assistant_req_empty(lad) ||
1664 lad->lad_to_double_scan,
1667 if (unlikely(lad->lad_exit))
1668 GOTO(cleanup1, rc = lad->lad_post_result);
1670 if (!list_empty(&lad->lad_req_list))
1673 if (lad->lad_to_post) {
1674 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1675 lfsck_lfsck2name(lfsck), lad->lad_name);
1677 if (unlikely(lad->lad_exit))
1678 GOTO(cleanup1, rc = lad->lad_post_result);
1680 lad->lad_to_post = 0;
1681 LASSERT(lad->lad_post_result > 0);
1683 memset(lr, 0, sizeof(*lr));
1684 lr->lr_event = LE_PHASE1_DONE;
1685 lr->lr_status = lad->lad_post_result;
1686 rc = lfsck_assistant_notify_others(env, com, lr);
1688 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to "
1689 "notify others for %s post: rc = %d\n",
1690 lfsck_lfsck2name(lfsck),
1693 /* Wakeup the master engine to go ahead. */
1694 wake_up_all(&mthread->t_ctl_waitq);
1697 if (lad->lad_to_double_scan) {
1698 lad->lad_to_double_scan = 0;
1699 atomic_inc(&lfsck->li_double_scan_count);
1700 lad->lad_in_double_scan = 1;
1701 wake_up_all(&mthread->t_ctl_waitq);
1703 com->lc_new_checked = 0;
1704 com->lc_new_scanned = 0;
1705 com->lc_time_last_checkpoint = cfs_time_current();
1706 com->lc_time_next_checkpoint =
1707 com->lc_time_last_checkpoint +
1708 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1710 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1711 "the second-stage scaning\n",
1712 lfsck_lfsck2name(lfsck));
1714 /* Flush async updates before handling orphan. */
1715 rc2 = dt_sync(env, lfsck->li_next);
1717 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1718 "scan start, synced: rc = %d\n",
1719 lfsck_lfsck2name(lfsck), rc2);
1721 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1722 GOTO(cleanup2, rc = 0);
1724 while (lad->lad_in_double_scan) {
1725 rc = lfsck_assistant_query_others(env, com);
1726 if (lfsck_phase2_next_ready(lad))
1732 /* Pull LFSCK status on related targets once
1733 * per 30 seconds if we are not notified. */
1734 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1735 cfs_time_seconds(1),
1737 rc = l_wait_event(athread->t_ctl_waitq,
1738 lfsck_phase2_next_ready(lad) ||
1740 !thread_is_running(mthread),
1743 if (unlikely(lad->lad_exit ||
1744 !thread_is_running(mthread)))
1745 GOTO(cleanup2, rc = 0);
1747 if (rc == -ETIMEDOUT)
1754 rc = lao->la_handler_p2(env, com);
1758 if (unlikely(lad->lad_exit ||
1759 !thread_is_running(mthread)))
1760 GOTO(cleanup2, rc = 0);
1766 /* Cleanup the unfinished requests. */
1767 spin_lock(&lad->lad_lock);
1769 lad->lad_assistant_status = rc;
1771 if (lad->lad_exit && lad->lad_post_result <= 0)
1772 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1774 while (!list_empty(&lad->lad_req_list)) {
1775 lar = list_entry(lad->lad_req_list.next,
1776 struct lfsck_assistant_req,
1778 list_del_init(&lar->lar_list);
1779 lad->lad_prefetched--;
1780 spin_unlock(&lad->lad_lock);
1781 lao->la_req_fini(env, lar);
1782 spin_lock(&lad->lad_lock);
1784 spin_unlock(&lad->lad_lock);
1786 LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1787 lad->lad_prefetched);
1790 memset(lr, 0, sizeof(*lr));
1792 lr->lr_event = LE_PHASE2_DONE;
1794 } else if (rc == 0) {
1795 if (lfsck->li_flags & LPF_ALL_TGT) {
1796 lr->lr_event = LE_STOP;
1797 lr->lr_status = LS_STOPPED;
1799 lr->lr_event = LE_PEER_EXIT;
1800 switch (lfsck->li_status) {
1803 lr->lr_status = LS_CO_PAUSED;
1807 lr->lr_status = LS_CO_STOPPED;
1810 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1811 "status: rc = %d\n",
1812 lfsck_lfsck2name(lfsck),
1814 lr->lr_status = LS_CO_FAILED;
1819 if (lfsck->li_flags & LPF_ALL_TGT) {
1820 lr->lr_event = LE_STOP;
1821 lr->lr_status = LS_FAILED;
1823 lr->lr_event = LE_PEER_EXIT;
1824 lr->lr_status = LS_CO_FAILED;
1828 rc1 = lfsck_assistant_notify_others(env, com, lr);
1830 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1831 "others for %s quit: rc = %d\n",
1832 lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1836 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1837 lfsck_lfsck2name(lfsck));
1839 /* Flush async updates before exit. */
1840 rc2 = dt_sync(env, lfsck->li_next);
1842 CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1843 lfsck_lfsck2name(lfsck), rc2);
1845 /* Under force exit case, some requests may be just freed without
1846 * verification, those objects should be re-handled when next run.
1847 * So not update the on-disk trace file under such case. */
1848 if (lad->lad_in_double_scan) {
1850 rc1 = lao->la_double_scan_result(env, com, rc);
1852 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1853 "finished: rc = %d\n",
1854 lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1858 if (lad->lad_in_double_scan)
1859 atomic_dec(&lfsck->li_double_scan_count);
1861 spin_lock(&lad->lad_lock);
1862 lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1863 thread_set_flags(athread, SVC_STOPPED);
1864 wake_up_all(&mthread->t_ctl_waitq);
1865 spin_unlock(&lad->lad_lock);
1867 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1868 lfsck_lfsck2name(lfsck), lad->lad_name,
1869 lad->lad_assistant_status);
1871 lfsck_thread_args_fini(lta);