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) 2012, 2013, 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. */
237 /* normal FID on this target (locally) must be for the
238 * client-side visiable object. */
239 if (fid_is_norm(fid))
243 obj = lfsck_object_find(env, lfsck, fid);
248 if (!dt_object_exists(obj))
252 dt_read_lock(env, obj, MOR_TGT_CHILD);
253 if (unlikely(lfsck_is_dead_obj(obj))) {
254 dt_read_unlock(env, obj);
259 rc = dt_xattr_get(env, obj,
260 lfsck_buf_get(env, NULL, 0), XATTR_NAME_LINK,
262 dt_read_unlock(env, obj);
266 if (rc < 0 && rc != -ENODATA)
269 rc = lfsck_parent_fid(env, obj, fid);
271 lfsck_object_put(env, obj);
277 if (!fid_is_sane(fid))
282 if (depth > 0 && obj != NULL)
283 lfsck_object_put(env, obj);
288 static int lfsck_load_stripe_lmv(const struct lu_env *env,
289 struct lfsck_instance *lfsck,
290 struct dt_object *obj)
292 struct lmv_mds_md_v1 *lmv = &lfsck_env_info(env)->lti_lmv;
293 struct lfsck_lmv *llmv;
297 LASSERT(lfsck->li_obj_dir == NULL);
298 LASSERT(lfsck->li_lmv == NULL);
300 rc = lfsck_read_stripe_lmv(env, obj, lmv);
301 if (rc == -ENODATA) {
302 lfsck->li_obj_dir = lfsck_object_get(obj);
314 if (lmv->lmv_magic == LMV_MAGIC) {
315 struct lfsck_slave_lmv_rec *lslr;
318 llmv->ll_lmv_master = 1;
319 if (lmv->lmv_stripe_count < 1)
320 stripes = LFSCK_LMV_DEF_STRIPES;
321 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
322 stripes = LFSCK_LMV_MAX_STRIPES;
324 stripes = lmv->lmv_stripe_count;
326 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
333 /* Find the object against the bottom device. */
334 obj = lfsck_object_find_by_dev(env, lfsck->li_bottom,
337 OBD_FREE_LARGE(lslr, sizeof(*lslr) * stripes);
340 RETURN(PTR_ERR(obj));
343 llmv->ll_stripes_allocated = stripes;
344 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
345 llmv->ll_lslr = lslr;
346 lfsck->li_obj_dir = obj;
348 llmv->ll_lmv_slave = 1;
349 lfsck->li_obj_dir = lfsck_object_get(obj);
353 atomic_set(&llmv->ll_ref, 1);
354 lfsck->li_lmv = llmv;
359 /* LFSCK wrap functions */
361 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
364 struct lfsck_component *com;
366 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
367 com->lc_ops->lfsck_fail(env, com, new_checked);
371 static void lfsck_close_dir(const struct lu_env *env,
372 struct lfsck_instance *lfsck,
375 struct lfsck_component *com;
378 if (lfsck->li_lmv != NULL) {
379 lfsck->li_lmv->ll_exit_value = result;
380 if (lfsck->li_obj_dir != NULL) {
381 list_for_each_entry(com, &lfsck->li_list_dir,
383 com->lc_ops->lfsck_close_dir(env, com);
387 lfsck_lmv_put(env, lfsck->li_lmv);
388 lfsck->li_lmv = NULL;
391 if (lfsck->li_di_dir != NULL) {
392 const struct dt_it_ops *dir_iops =
393 &lfsck->li_obj_dir->do_index_ops->dio_it;
394 struct dt_it *dir_di = lfsck->li_di_dir;
396 lfsck_di_dir_put(env, lfsck);
397 dir_iops->fini(env, dir_di);
400 if (lfsck->li_obj_dir != NULL) {
401 struct dt_object *dir_obj = lfsck->li_obj_dir;
403 lfsck->li_obj_dir = NULL;
404 lfsck_object_put(env, dir_obj);
410 static int lfsck_open_dir(const struct lu_env *env,
411 struct lfsck_instance *lfsck, __u64 cookie)
413 struct dt_object *obj = lfsck->li_obj_dir;
414 struct dt_it *di = lfsck->li_di_dir;
415 struct lfsck_component *com;
416 const struct dt_it_ops *iops;
420 LASSERT(obj != NULL);
423 if (unlikely(!dt_try_as_dir(env, obj)))
424 GOTO(out, rc = -ENOTDIR);
426 list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
427 rc = com->lc_ops->lfsck_open_dir(env, com);
432 iops = &obj->do_index_ops->dio_it;
433 di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA);
435 GOTO(out, rc = PTR_ERR(di));
437 rc = iops->load(env, di, cookie);
438 if (rc == 0 || (rc > 0 && cookie > 0))
439 rc = iops->next(env, di);
447 lfsck->li_cookie_dir = iops->store(env, di);
448 spin_lock(&lfsck->li_lock);
449 lfsck->li_di_dir = di;
450 spin_unlock(&lfsck->li_lock);
457 lfsck_close_dir(env, lfsck, rc);
462 static int lfsck_checkpoint(const struct lu_env *env,
463 struct lfsck_instance *lfsck)
465 struct lfsck_component *com;
469 if (likely(cfs_time_beforeq(cfs_time_current(),
470 lfsck->li_time_next_checkpoint)))
473 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
474 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
475 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
480 lfsck->li_time_last_checkpoint = cfs_time_current();
481 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
482 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
483 return rc1 != 0 ? rc1 : rc;
486 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
487 struct lfsck_start_param *lsp)
489 struct dt_object *obj = NULL;
490 struct lfsck_component *com;
491 struct lfsck_component *next;
492 struct lfsck_position *pos = NULL;
493 const struct dt_it_ops *iops =
494 &lfsck->li_obj_oit->do_index_ops->dio_it;
498 LASSERT(lfsck->li_obj_dir == NULL);
499 LASSERT(lfsck->li_di_dir == NULL);
501 lfsck->li_current_oit_processed = 0;
502 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
503 com->lc_new_checked = 0;
504 rc = com->lc_ops->lfsck_prep(env, com, lsp);
509 (!lfsck_pos_is_zero(&com->lc_pos_start) &&
510 lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
511 pos = &com->lc_pos_start;
514 /* Init otable-based iterator. */
516 rc = iops->load(env, lfsck->li_di_oit, 0);
518 lfsck->li_oit_over = 1;
525 rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
529 lfsck->li_oit_over = 1;
531 if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
534 /* Find the directory for namespace-based traverse. */
535 obj = lfsck_object_find(env, lfsck, &pos->lp_dir_parent);
537 RETURN(PTR_ERR(obj));
539 /* Remote directory will be scanned by the LFSCK instance
540 * on the MDT where the remote object really resides on. */
541 if (!dt_object_exists(obj) || dt_object_remote(obj) ||
542 unlikely(!S_ISDIR(lfsck_object_type(obj))))
545 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
547 /* For the master MDT-object of a striped directory,
548 * reset the iteration from the directory beginning. */
549 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
550 pos->lp_dir_cookie = 0;
552 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
559 lfsck_object_put(env, obj);
562 lfsck_close_dir(env, lfsck, rc);
563 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
565 com->lc_ops->lfsck_post(env, com, rc, true);
572 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
573 lfsck->li_pos_current = lfsck->li_pos_checkpoint;
574 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
575 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
580 lfsck->li_time_last_checkpoint = cfs_time_current();
581 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
582 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
586 static int lfsck_exec_oit(const struct lu_env *env,
587 struct lfsck_instance *lfsck, struct dt_object *obj)
589 struct lfsck_component *com;
593 LASSERT(lfsck->li_obj_dir == NULL);
595 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
596 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
601 rc = lfsck_needs_scan_dir(env, lfsck, obj);
605 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
607 rc = lfsck_open_dir(env, lfsck, 0);
613 lfsck_fail(env, lfsck, false);
616 lfsck_close_dir(env, lfsck, rc);
618 return rc > 0 ? 0 : rc;
621 static int lfsck_exec_dir(const struct lu_env *env,
622 struct lfsck_instance *lfsck,
623 struct lu_dirent *ent, __u16 type)
625 struct lfsck_component *com;
628 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
629 rc = com->lc_ops->lfsck_exec_dir(env, com, ent, type);
636 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
639 struct lfsck_component *com;
640 struct lfsck_component *next;
644 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
645 lfsck_close_dir(env, lfsck, result);
646 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
647 rc = com->lc_ops->lfsck_post(env, com, result, false);
652 lfsck->li_time_last_checkpoint = cfs_time_current();
653 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
654 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
656 /* Ignore some component post failure to make other can go ahead. */
660 static int lfsck_double_scan(const struct lu_env *env,
661 struct lfsck_instance *lfsck)
663 struct lfsck_component *com;
664 struct lfsck_component *next;
665 struct l_wait_info lwi = { 0 };
669 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
670 rc = com->lc_ops->lfsck_double_scan(env, com);
675 l_wait_event(lfsck->li_thread.t_ctl_waitq,
676 atomic_read(&lfsck->li_double_scan_count) == 0,
679 if (lfsck->li_status != LS_PAUSED &&
680 lfsck->li_status != LS_CO_PAUSED) {
681 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
683 spin_lock(&lfsck->li_lock);
684 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
685 spin_unlock(&lfsck->li_lock);
689 return rc1 != 0 ? rc1 : rc;
692 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
694 struct lfsck_component *com;
695 struct lfsck_component *next;
697 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
699 if (com->lc_ops->lfsck_quit != NULL)
700 com->lc_ops->lfsck_quit(env, com);
702 spin_lock(&lfsck->li_lock);
703 list_del_init(&com->lc_link_dir);
704 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
705 spin_unlock(&lfsck->li_lock);
708 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
710 if (com->lc_ops->lfsck_quit != NULL)
711 com->lc_ops->lfsck_quit(env, com);
713 spin_lock(&lfsck->li_lock);
714 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
715 spin_unlock(&lfsck->li_lock);
721 static int lfsck_master_dir_engine(const struct lu_env *env,
722 struct lfsck_instance *lfsck)
724 struct lfsck_thread_info *info = lfsck_env_info(env);
725 struct dt_object *dir = lfsck->li_obj_dir;
726 const struct dt_it_ops *iops = &dir->do_index_ops->dio_it;
727 struct dt_it *di = lfsck->li_di_dir;
728 struct lu_dirent *ent =
729 (struct lu_dirent *)info->lti_key;
730 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
731 struct ptlrpc_thread *thread = &lfsck->li_thread;
737 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY2) &&
739 struct l_wait_info lwi;
741 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
743 l_wait_event(thread->t_ctl_waitq,
744 !thread_is_running(thread),
747 if (unlikely(!thread_is_running(thread))) {
748 CDEBUG(D_LFSCK, "%s: scan dir exit for engine "
749 "stop, parent "DFID", cookie "LPX64"\n",
750 lfsck_lfsck2name(lfsck),
751 PFID(lfsck_dto2fid(dir)),
752 lfsck->li_cookie_dir);
757 lfsck->li_new_scanned++;
758 rc = iops->rec(env, di, (struct dt_rec *)ent,
761 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
765 CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
766 "parent "DFID", cookie "LPX64": rc = %d\n",
767 lfsck_lfsck2name(lfsck),
768 PFID(lfsck_dto2fid(dir)),
769 lfsck->li_cookie_dir, rc);
770 lfsck_fail(env, lfsck, true);
771 if (bk->lb_param & LPF_FAILOUT)
777 if (ent->lde_attrs & LUDA_IGNORE &&
778 strcmp(ent->lde_name, dotdot) != 0)
781 /* The type in the @ent structure may has been overwritten,
782 * so we need to pass the @type parameter independently. */
783 rc = lfsck_exec_dir(env, lfsck, ent, type);
784 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
788 rc = lfsck_checkpoint(env, lfsck);
789 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
793 lfsck_control_speed(lfsck);
794 if (unlikely(!thread_is_running(thread))) {
795 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
796 "parent "DFID", cookie "LPX64"\n",
797 lfsck_lfsck2name(lfsck),
798 PFID(lfsck_dto2fid(dir)),
799 lfsck->li_cookie_dir);
803 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
804 spin_lock(&lfsck->li_lock);
805 thread_set_flags(thread, SVC_STOPPING);
806 spin_unlock(&lfsck->li_lock);
810 rc = iops->next(env, di);
813 if (rc > 0 && !lfsck->li_oit_over)
814 lfsck_close_dir(env, lfsck, rc);
820 * Object-table based iteration engine.
822 * Object-table based iteration is the basic linear engine to scan all the
823 * objects on current device in turn. For each object, it calls all the
824 * registered LFSCK component(s)' API to perform related consistency
827 * It flushes related LFSCK trace files to disk via making checkpoint
828 * periodically. Then if the server crashed or the LFSCK is paused, the
829 * LFSCK can resume from the latest checkpoint.
831 * It also controls the whole LFSCK speed via lfsck_control_speed() to
832 * avoid the server to become overload.
834 * \param[in] env pointer to the thread context
835 * \param[in] lfsck pointer to the lfsck instance
837 * \retval positive number if all objects have been scanned
838 * \retval 0 if the iteration is stopped or paused
839 * \retval negative error number on failure
841 static int lfsck_master_oit_engine(const struct lu_env *env,
842 struct lfsck_instance *lfsck)
844 struct lfsck_thread_info *info = lfsck_env_info(env);
845 const struct dt_it_ops *iops =
846 &lfsck->li_obj_oit->do_index_ops->dio_it;
847 struct dt_it *di = lfsck->li_di_oit;
848 struct lu_fid *fid = &info->lti_fid;
849 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
850 struct ptlrpc_thread *thread = &lfsck->li_thread;
851 struct dt_device *dev = lfsck->li_bottom;
852 struct seq_server_site *ss = lu_site2seq(dev->dd_lu_dev.ld_site);
853 __u32 idx = lfsck_dev_idx(dev);
857 if (unlikely(ss == NULL))
861 struct dt_object *target;
862 bool update_lma = false;
864 if (lfsck->li_di_dir != NULL) {
865 rc = lfsck_master_dir_engine(env, lfsck);
870 if (unlikely(lfsck->li_oit_over))
873 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY1) &&
875 struct l_wait_info lwi;
877 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
879 l_wait_event(thread->t_ctl_waitq,
880 !thread_is_running(thread),
883 if (unlikely(!thread_is_running(thread))) {
884 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine "
885 "stop, cookie "LPU64"\n",
886 lfsck_lfsck2name(lfsck),
887 iops->store(env, di));
892 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
895 lfsck->li_current_oit_processed = 1;
896 lfsck->li_new_scanned++;
897 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
898 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
900 CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
901 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
902 lfsck_fail(env, lfsck, true);
903 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
909 if (fid_is_idif(fid)) {
910 __u32 idx1 = fid_idif_ost_idx(fid);
912 LASSERT(!lfsck->li_master);
914 /* It is an old format device, update the LMA. */
916 struct ost_id *oi = &info->lti_oi;
918 fid_to_ostid(fid, oi);
919 ostid_to_fid(fid, oi, idx);
922 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
923 !fid_is_last_id(fid) &&
924 !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
926 /* If the FID/object is only used locally and invisible
927 * to external nodes, then LFSCK will not handle it.
929 * dot_lustre sequence has been handled specially. */
932 struct lu_seq_range *range = &info->lti_range;
934 if (lfsck->li_master)
935 fld_range_set_mdt(range);
937 fld_range_set_ost(range);
938 rc = fld_local_lookup(env, ss->ss_server_fld,
939 fid_seq(fid), range);
940 if (rc != 0 || range->lsr_index != idx) {
941 /* Remote object will be handled by the LFSCK
942 * instance on the MDT where the remote object
943 * really resides on. */
949 target = lfsck_object_find(env, lfsck, fid);
950 if (IS_ERR(target)) {
951 CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
952 DFID", cookie "LPU64": rc = %d\n",
953 lfsck_lfsck2name(lfsck), PFID(fid),
954 iops->store(env, di), rc);
955 lfsck_fail(env, lfsck, true);
956 if (bk->lb_param & LPF_FAILOUT)
957 RETURN(PTR_ERR(target));
962 if (dt_object_exists(target)) {
964 rc = lfsck_update_lma(env, lfsck, target);
966 CDEBUG(D_LFSCK, "%s: fail to update "
967 "LMA for "DFID": rc = %d\n",
968 lfsck_lfsck2name(lfsck),
969 PFID(lfsck_dto2fid(target)), rc);
972 rc = lfsck_exec_oit(env, lfsck, target);
974 lfsck_object_put(env, target);
975 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
979 rc = lfsck_checkpoint(env, lfsck);
980 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
984 lfsck_control_speed(lfsck);
986 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
987 spin_lock(&lfsck->li_lock);
988 thread_set_flags(thread, SVC_STOPPING);
989 spin_unlock(&lfsck->li_lock);
993 rc = iops->next(env, di);
994 if (unlikely(rc > 0))
995 lfsck->li_oit_over = 1;
996 else if (likely(rc == 0))
997 lfsck->li_current_oit_processed = 0;
999 if (unlikely(!thread_is_running(thread))) {
1000 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1001 "cookie "LPU64"\n", lfsck_lfsck2name(lfsck),
1002 iops->store(env, di));
1005 } while (rc == 0 || lfsck->li_di_dir != NULL);
1010 int lfsck_master_engine(void *args)
1012 struct lfsck_thread_args *lta = args;
1013 struct lu_env *env = <a->lta_env;
1014 struct lfsck_instance *lfsck = lta->lta_lfsck;
1015 struct ptlrpc_thread *thread = &lfsck->li_thread;
1016 struct dt_object *oit_obj = lfsck->li_obj_oit;
1017 const struct dt_it_ops *oit_iops = &oit_obj->do_index_ops->dio_it;
1018 struct dt_it *oit_di;
1019 struct l_wait_info lwi = { 0 };
1023 if (lfsck->li_master &&
1024 (!list_empty(&lfsck->li_list_scan) ||
1025 !list_empty(&lfsck->li_list_double_scan))) {
1026 rc = lfsck_verify_lpf(env, lfsck);
1027 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1028 * fatal, because the .lustre/lost+found/ maybe not accessed
1029 * by the LFSCK if it does not add orphans or others to such
1030 * directory. So go ahead until hit failure when really uses
1033 CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1034 ".lustre/lost+found/, go ahead: rc = %d\n",
1035 lfsck_lfsck2name(lfsck), rc);
1038 oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit, BYPASS_CAPA);
1039 if (IS_ERR(oit_di)) {
1040 rc = PTR_ERR(oit_di);
1041 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1042 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1044 GOTO(fini_args, 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 = "LPU64", dir_cookie = "LPX64", 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 CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1081 "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1082 ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1083 lfsck->li_pos_checkpoint.lp_oit_cookie,
1084 lfsck->li_pos_checkpoint.lp_dir_cookie,
1085 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1088 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1089 rc = lfsck_post(env, lfsck, rc);
1091 lfsck_close_dir(env, lfsck, rc);
1094 lfsck_di_oit_put(env, lfsck);
1095 oit_iops->fini(env, oit_di);
1097 if (!list_empty(&lfsck->li_list_double_scan))
1098 rc = lfsck_double_scan(env, lfsck);
1102 lfsck_quit(env, lfsck);
1105 /* XXX: Purge the pinned objects in the future. */
1108 spin_lock(&lfsck->li_lock);
1109 thread_set_flags(thread, SVC_STOPPED);
1110 spin_unlock(&lfsck->li_lock);
1111 wake_up_all(&thread->t_ctl_waitq);
1112 lfsck_thread_args_fini(lta);
1116 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1120 spin_lock(&lad->lad_lock);
1121 if (list_empty(&lad->lad_req_list))
1123 spin_unlock(&lad->lad_lock);
1129 * Query the LFSCK status from the instatnces on remote servers.
1131 * The LFSCK assistant thread queries the LFSCK instances on other
1132 * servers (MDT/OST) about their status, such as whether they have
1133 * finished the phase1/phase2 scanning or not, and so on.
1135 * \param[in] env pointer to the thread context
1136 * \param[in] com pointer to the lfsck component
1138 * \retval 0 for success
1139 * \retval negative error number on failure
1141 static int lfsck_assistant_query_others(const struct lu_env *env,
1142 struct lfsck_component *com)
1144 struct lfsck_thread_info *info = lfsck_env_info(env);
1145 struct lfsck_request *lr = &info->lti_lr;
1146 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1147 struct lfsck_instance *lfsck = com->lc_lfsck;
1148 struct lfsck_assistant_data *lad = com->lc_data;
1149 struct ptlrpc_request_set *set;
1150 struct lfsck_tgt_descs *ltds;
1151 struct lfsck_tgt_desc *ltd;
1152 struct list_head *phase_head;
1157 set = ptlrpc_prep_set();
1161 lad->lad_touch_gen++;
1162 memset(lr, 0, sizeof(*lr));
1163 lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1164 lr->lr_event = LE_QUERY;
1165 lr->lr_active = com->lc_type;
1166 laia->laia_com = com;
1168 laia->laia_shared = 0;
1170 if (!list_empty(&lad->lad_mdt_phase1_list)) {
1171 ltds = &lfsck->li_mdt_descs;
1173 phase_head = &lad->lad_mdt_phase1_list;
1174 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1179 ltds = &lfsck->li_ost_descs;
1180 lr->lr_flags = LEF_TO_OST;
1181 phase_head = &lad->lad_ost_phase1_list;
1184 laia->laia_ltds = ltds;
1185 spin_lock(<ds->ltd_lock);
1186 while (!list_empty(phase_head)) {
1187 struct list_head *phase_list;
1190 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1191 ltd = list_entry(phase_head->next,
1192 struct lfsck_tgt_desc,
1193 ltd_layout_phase_list);
1194 phase_list = <d->ltd_layout_phase_list;
1195 gen = <d->ltd_layout_gen;
1197 ltd = list_entry(phase_head->next,
1198 struct lfsck_tgt_desc,
1199 ltd_namespace_phase_list);
1200 phase_list = <d->ltd_namespace_phase_list;
1201 gen = <d->ltd_namespace_gen;
1204 if (*gen == lad->lad_touch_gen)
1207 *gen = lad->lad_touch_gen;
1208 list_move_tail(phase_list, phase_head);
1209 atomic_inc(<d->ltd_ref);
1210 laia->laia_ltd = ltd;
1211 spin_unlock(<ds->ltd_lock);
1212 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1213 lfsck_async_interpret_common,
1216 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1217 "%s %x for %s: rc = %d\n",
1218 lfsck_lfsck2name(lfsck),
1219 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1220 ltd->ltd_index, lad->lad_name, rc);
1224 spin_lock(<ds->ltd_lock);
1226 spin_unlock(<ds->ltd_lock);
1228 rc = ptlrpc_set_wait(set);
1230 ptlrpc_set_destroy(set);
1234 if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1235 list_empty(&lad->lad_mdt_phase1_list))
1239 ptlrpc_set_destroy(set);
1241 RETURN(rc1 != 0 ? rc1 : rc);
1245 * Notify the LFSCK event to the instatnces on remote servers.
1247 * The LFSCK assistant thread notifies the LFSCK instances on other
1248 * servers (MDT/OST) about some events, such as start new scanning,
1249 * stop the scanning, this LFSCK instance will exit, and so on.
1251 * \param[in] env pointer to the thread context
1252 * \param[in] com pointer to the lfsck component
1253 * \param[in] lr pointer to the LFSCK event request
1255 * \retval 0 for success
1256 * \retval negative error number on failure
1258 static int lfsck_assistant_notify_others(const struct lu_env *env,
1259 struct lfsck_component *com,
1260 struct lfsck_request *lr)
1262 struct lfsck_thread_info *info = lfsck_env_info(env);
1263 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1264 struct lfsck_instance *lfsck = com->lc_lfsck;
1265 struct lfsck_assistant_data *lad = com->lc_data;
1266 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1267 struct ptlrpc_request_set *set;
1268 struct lfsck_tgt_descs *ltds;
1269 struct lfsck_tgt_desc *ltd;
1270 struct lfsck_tgt_desc *next;
1276 set = ptlrpc_prep_set();
1280 lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1281 lr->lr_active = com->lc_type;
1282 laia->laia_com = com;
1284 laia->laia_shared = 0;
1286 switch (lr->lr_event) {
1288 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1291 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN |
1292 LSV_ASYNC_WINDOWS | LSV_CREATE_OSTOBJ;
1293 lr->lr_speed = bk->lb_speed_limit;
1294 lr->lr_version = bk->lb_version;
1295 lr->lr_param |= bk->lb_param;
1296 lr->lr_async_windows = bk->lb_async_windows;
1297 lr->lr_flags = LEF_TO_OST;
1299 /* Notify OSTs firstly, then handle other MDTs if needed. */
1300 ltds = &lfsck->li_ost_descs;
1301 laia->laia_ltds = ltds;
1302 down_read(<ds->ltd_rw_sem);
1303 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1304 ltd = lfsck_tgt_get(ltds, idx);
1305 LASSERT(ltd != NULL);
1307 laia->laia_ltd = ltd;
1308 ltd->ltd_layout_done = 0;
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 = LTD_TGT(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 = LTD_TGT(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_namespace_failed)
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 };
1574 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1575 lfsck_lfsck2name(lfsck), lad->lad_name);
1577 memset(lr, 0, sizeof(*lr));
1578 lr->lr_event = LE_START;
1579 if (pos->lp_oit_cookie <= 1)
1580 lr->lr_param = LPF_RESET;
1581 rc = lfsck_assistant_notify_others(env, com, lr);
1583 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1584 "to start %s: rc = %d\n",
1585 lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1589 spin_lock(&lad->lad_lock);
1590 thread_set_flags(athread, SVC_RUNNING);
1591 spin_unlock(&lad->lad_lock);
1592 wake_up_all(&mthread->t_ctl_waitq);
1595 while (!list_empty(&lad->lad_req_list)) {
1596 bool wakeup = false;
1598 if (unlikely(lad->lad_exit ||
1599 !thread_is_running(mthread)))
1600 GOTO(cleanup1, rc = lad->lad_post_result);
1602 lar = list_entry(lad->lad_req_list.next,
1603 struct lfsck_assistant_req,
1605 /* Only the lfsck_assistant_engine thread itself can
1606 * remove the "lar" from the head of the list, LFSCK
1607 * engine thread only inserts other new "lar" at the
1608 * end of the list. So it is safe to handle current
1609 * "lar" without the spin_lock. */
1610 rc = lao->la_handler_p1(env, com, lar);
1611 spin_lock(&lad->lad_lock);
1612 list_del_init(&lar->lar_list);
1613 lad->lad_prefetched--;
1614 /* Wake up the main engine thread only when the list
1615 * is empty or half of the prefetched items have been
1616 * handled to avoid too frequent thread schedule. */
1617 if (lad->lad_prefetched == 0 ||
1618 (bk->lb_async_windows != 0 &&
1619 bk->lb_async_windows / 2 ==
1620 lad->lad_prefetched))
1622 spin_unlock(&lad->lad_lock);
1624 wake_up_all(&mthread->t_ctl_waitq);
1626 lao->la_req_fini(env, lar);
1627 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1631 l_wait_event(athread->t_ctl_waitq,
1632 !lfsck_assistant_req_empty(lad) ||
1635 lad->lad_to_double_scan,
1638 if (unlikely(lad->lad_exit))
1639 GOTO(cleanup1, rc = lad->lad_post_result);
1641 if (!list_empty(&lad->lad_req_list))
1644 if (lad->lad_to_post) {
1645 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1646 lfsck_lfsck2name(lfsck), lad->lad_name);
1648 if (unlikely(lad->lad_exit))
1649 GOTO(cleanup1, rc = lad->lad_post_result);
1651 lad->lad_to_post = 0;
1652 LASSERT(lad->lad_post_result > 0);
1654 memset(lr, 0, sizeof(*lr));
1655 lr->lr_event = LE_PHASE1_DONE;
1656 lr->lr_status = lad->lad_post_result;
1657 rc = lfsck_assistant_notify_others(env, com, lr);
1659 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to "
1660 "notify others for %s post: rc = %d\n",
1661 lfsck_lfsck2name(lfsck),
1664 /* Wakeup the master engine to go ahead. */
1665 wake_up_all(&mthread->t_ctl_waitq);
1668 if (lad->lad_to_double_scan) {
1669 lad->lad_to_double_scan = 0;
1670 atomic_inc(&lfsck->li_double_scan_count);
1671 lad->lad_in_double_scan = 1;
1672 wake_up_all(&mthread->t_ctl_waitq);
1674 com->lc_new_checked = 0;
1675 com->lc_new_scanned = 0;
1676 com->lc_time_last_checkpoint = cfs_time_current();
1677 com->lc_time_next_checkpoint =
1678 com->lc_time_last_checkpoint +
1679 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1681 /* Flush async updates before handling orphan. */
1682 dt_sync(env, lfsck->li_next);
1684 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1685 "scan start\n", lfsck_lfsck2name(lfsck));
1687 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1688 GOTO(cleanup2, rc = 0);
1690 while (lad->lad_in_double_scan) {
1691 rc = lfsck_assistant_query_others(env, com);
1692 if (lfsck_phase2_next_ready(lad))
1698 /* Pull LFSCK status on related targets once
1699 * per 30 seconds if we are not notified. */
1700 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1701 cfs_time_seconds(1),
1703 rc = l_wait_event(athread->t_ctl_waitq,
1704 lfsck_phase2_next_ready(lad) ||
1706 !thread_is_running(mthread),
1709 if (unlikely(lad->lad_exit ||
1710 !thread_is_running(mthread)))
1711 GOTO(cleanup2, rc = 0);
1713 if (rc == -ETIMEDOUT)
1720 rc = lao->la_handler_p2(env, com);
1724 if (unlikely(lad->lad_exit ||
1725 !thread_is_running(mthread)))
1726 GOTO(cleanup2, rc = 0);
1732 /* Cleanup the unfinished requests. */
1733 spin_lock(&lad->lad_lock);
1735 lad->lad_assistant_status = rc;
1737 if (lad->lad_exit && lad->lad_post_result <= 0)
1738 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1740 while (!list_empty(&lad->lad_req_list)) {
1741 lar = list_entry(lad->lad_req_list.next,
1742 struct lfsck_assistant_req,
1744 list_del_init(&lar->lar_list);
1745 lad->lad_prefetched--;
1746 spin_unlock(&lad->lad_lock);
1747 lao->la_req_fini(env, lar);
1748 spin_lock(&lad->lad_lock);
1750 spin_unlock(&lad->lad_lock);
1752 LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1753 lad->lad_prefetched);
1756 memset(lr, 0, sizeof(*lr));
1758 lr->lr_event = LE_PHASE2_DONE;
1760 } else if (rc == 0) {
1761 if (lfsck->li_flags & LPF_ALL_TGT) {
1762 lr->lr_event = LE_STOP;
1763 lr->lr_status = LS_STOPPED;
1765 lr->lr_event = LE_PEER_EXIT;
1766 switch (lfsck->li_status) {
1769 lr->lr_status = LS_CO_PAUSED;
1773 lr->lr_status = LS_CO_STOPPED;
1776 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1777 "status: rc = %d\n",
1778 lfsck_lfsck2name(lfsck),
1780 lr->lr_status = LS_CO_FAILED;
1785 if (lfsck->li_flags & LPF_ALL_TGT) {
1786 lr->lr_event = LE_STOP;
1787 lr->lr_status = LS_FAILED;
1789 lr->lr_event = LE_PEER_EXIT;
1790 lr->lr_status = LS_CO_FAILED;
1794 rc1 = lfsck_assistant_notify_others(env, com, lr);
1796 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1797 "others for %s quit: rc = %d\n",
1798 lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1802 /* Flush async updates before exit. */
1803 dt_sync(env, lfsck->li_next);
1805 /* Under force exit case, some requests may be just freed without
1806 * verification, those objects should be re-handled when next run.
1807 * So not update the on-disk trace file under such case. */
1808 if (lad->lad_in_double_scan) {
1810 rc1 = lao->la_double_scan_result(env, com, rc);
1812 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1813 "finished: rc = %d\n",
1814 lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1818 if (lad->lad_in_double_scan)
1819 atomic_dec(&lfsck->li_double_scan_count);
1821 spin_lock(&lad->lad_lock);
1822 lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1823 thread_set_flags(athread, SVC_STOPPED);
1824 wake_up_all(&mthread->t_ctl_waitq);
1825 spin_unlock(&lad->lad_lock);
1827 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1828 lfsck_lfsck2name(lfsck), lad->lad_name,
1829 lad->lad_assistant_status);
1831 lfsck_thread_args_fini(lta);