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 lfsck_assistant_object *lso,
612 struct lu_dirent *ent, __u16 type)
614 struct lfsck_component *com;
617 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
618 rc = com->lc_ops->lfsck_exec_dir(env, com, lso, ent, type);
625 static int lfsck_master_dir_engine(const struct lu_env *env,
626 struct lfsck_instance *lfsck);
628 static int lfsck_post(const struct lu_env *env, struct lfsck_instance *lfsck,
631 struct lfsck_component *com;
632 struct lfsck_component *next;
635 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
636 lfsck_close_dir(env, lfsck, result);
638 while (thread_is_running(&lfsck->li_thread) && rc > 0 &&
639 !list_empty(&lfsck->li_list_lmv)) {
640 struct lfsck_lmv_unit *llu;
642 spin_lock(&lfsck->li_lock);
643 llu = list_entry(lfsck->li_list_lmv.next,
644 struct lfsck_lmv_unit, llu_link);
645 list_del_init(&llu->llu_link);
646 spin_unlock(&lfsck->li_lock);
648 lfsck->li_lmv = &llu->llu_lmv;
649 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
650 rc = lfsck_open_dir(env, lfsck, 0);
652 rc = lfsck_master_dir_engine(env, lfsck);
653 lfsck_close_dir(env, lfsck, result);
659 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
660 rc = com->lc_ops->lfsck_post(env, com, result, false);
662 CDEBUG(D_LFSCK, "%s: lfsck_post at the component %u: "
663 "rc = %d\n", lfsck_lfsck2name(lfsck),
664 (__u32)com->lc_type, rc);
667 lfsck->li_time_last_checkpoint = cfs_time_current();
668 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
669 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
671 /* Ignore some component post failure to make other can go ahead. */
675 static int lfsck_double_scan(const struct lu_env *env,
676 struct lfsck_instance *lfsck)
678 struct lfsck_component *com;
679 struct lfsck_component *next;
680 struct l_wait_info lwi = { 0 };
684 list_for_each_entry(com, &lfsck->li_list_double_scan, lc_link) {
685 rc = com->lc_ops->lfsck_double_scan(env, com);
690 l_wait_event(lfsck->li_thread.t_ctl_waitq,
691 atomic_read(&lfsck->li_double_scan_count) == 0,
694 if (lfsck->li_status != LS_PAUSED &&
695 lfsck->li_status != LS_CO_PAUSED) {
696 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
698 spin_lock(&lfsck->li_lock);
699 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
700 spin_unlock(&lfsck->li_lock);
704 return rc1 != 0 ? rc1 : rc;
707 static void lfsck_quit(const struct lu_env *env, struct lfsck_instance *lfsck)
709 struct lfsck_component *com;
710 struct lfsck_component *next;
712 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
714 if (com->lc_ops->lfsck_quit != NULL)
715 com->lc_ops->lfsck_quit(env, com);
717 spin_lock(&lfsck->li_lock);
718 list_del_init(&com->lc_link_dir);
719 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
720 spin_unlock(&lfsck->li_lock);
723 list_for_each_entry_safe(com, next, &lfsck->li_list_double_scan,
725 if (com->lc_ops->lfsck_quit != NULL)
726 com->lc_ops->lfsck_quit(env, com);
728 spin_lock(&lfsck->li_lock);
729 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
730 spin_unlock(&lfsck->li_lock);
736 static int lfsck_master_dir_engine(const struct lu_env *env,
737 struct lfsck_instance *lfsck)
739 struct lfsck_thread_info *info = lfsck_env_info(env);
740 struct dt_object *dir = lfsck->li_obj_dir;
741 const struct dt_it_ops *iops = &dir->do_index_ops->dio_it;
742 struct dt_it *di = lfsck->li_di_dir;
743 struct lu_dirent *ent =
744 (struct lu_dirent *)info->lti_key;
745 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
746 struct ptlrpc_thread *thread = &lfsck->li_thread;
747 struct lfsck_assistant_object *lso = NULL;
753 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
754 unlikely(!thread_is_running(thread))) {
755 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
756 "parent "DFID", cookie "LPX64"\n",
757 lfsck_lfsck2name(lfsck),
758 PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
763 lfsck->li_new_scanned++;
764 rc = iops->rec(env, di, (struct dt_rec *)ent,
767 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
771 CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
772 "parent "DFID", cookie "LPX64": rc = %d\n",
773 lfsck_lfsck2name(lfsck),
774 PFID(lfsck_dto2fid(dir)),
775 lfsck->li_cookie_dir, rc);
776 lfsck_fail(env, lfsck, true);
777 if (bk->lb_param & LPF_FAILOUT)
783 if (ent->lde_attrs & LUDA_IGNORE &&
784 strcmp(ent->lde_name, dotdot) != 0)
787 /* skip dot entry. */
788 if (ent->lde_namelen == 1 && ent->lde_name[0] == '.')
792 lso = lfsck_assistant_object_init(env,
793 lfsck_dto2fid(dir), NULL,
794 lfsck->li_pos_current.lp_oit_cookie, true);
796 if (bk->lb_param & LPF_FAILOUT)
797 RETURN(PTR_ERR(lso));
804 /* The type in the @ent structure may has been overwritten,
805 * so we need to pass the @type parameter independently. */
806 rc = lfsck_exec_dir(env, lfsck, lso, ent, type);
807 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
811 rc = lfsck_checkpoint(env, lfsck);
812 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
816 lfsck_control_speed(lfsck);
817 if (unlikely(!thread_is_running(thread))) {
818 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
819 "parent "DFID", cookie "LPX64"\n",
820 lfsck_lfsck2name(lfsck),
821 PFID(lfsck_dto2fid(dir)),
822 lfsck->li_cookie_dir);
826 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
827 spin_lock(&lfsck->li_lock);
828 thread_set_flags(thread, SVC_STOPPING);
829 spin_unlock(&lfsck->li_lock);
830 GOTO(out, rc = -EINVAL);
833 rc = iops->next(env, di);
836 if (rc > 0 && !lfsck->li_oit_over)
837 lfsck_close_dir(env, lfsck, rc);
843 lfsck_assistant_object_put(env, lso);
849 * Object-table based iteration engine.
851 * Object-table based iteration is the basic linear engine to scan all the
852 * objects on current device in turn. For each object, it calls all the
853 * registered LFSCK component(s)' API to perform related consistency
856 * It flushes related LFSCK trace files to disk via making checkpoint
857 * periodically. Then if the server crashed or the LFSCK is paused, the
858 * LFSCK can resume from the latest checkpoint.
860 * It also controls the whole LFSCK speed via lfsck_control_speed() to
861 * avoid the server to become overload.
863 * \param[in] env pointer to the thread context
864 * \param[in] lfsck pointer to the lfsck instance
866 * \retval positive number if all objects have been scanned
867 * \retval 0 if the iteration is stopped or paused
868 * \retval negative error number on failure
870 static int lfsck_master_oit_engine(const struct lu_env *env,
871 struct lfsck_instance *lfsck)
873 struct lfsck_thread_info *info = lfsck_env_info(env);
874 const struct dt_it_ops *iops =
875 &lfsck->li_obj_oit->do_index_ops->dio_it;
876 struct dt_it *di = lfsck->li_di_oit;
877 struct lu_fid *fid = &info->lti_fid;
878 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
879 struct ptlrpc_thread *thread = &lfsck->li_thread;
880 struct seq_server_site *ss = lfsck_dev_site(lfsck);
881 __u32 idx = lfsck_dev_idx(lfsck);
885 if (unlikely(ss == NULL))
889 struct dt_object *target;
890 bool update_lma = false;
892 if (lfsck->li_di_dir != NULL) {
893 rc = lfsck_master_dir_engine(env, lfsck);
898 if (unlikely(lfsck->li_oit_over))
901 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
902 unlikely(!thread_is_running(thread))) {
903 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
905 lfsck_lfsck2name(lfsck), iops->store(env, di));
910 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
913 lfsck->li_current_oit_processed = 1;
915 if (!list_empty(&lfsck->li_list_lmv)) {
916 struct lfsck_lmv_unit *llu;
918 spin_lock(&lfsck->li_lock);
919 llu = list_entry(lfsck->li_list_lmv.next,
920 struct lfsck_lmv_unit, llu_link);
921 list_del_init(&llu->llu_link);
922 spin_unlock(&lfsck->li_lock);
924 lfsck->li_lmv = &llu->llu_lmv;
925 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
926 rc = lfsck_open_dir(env, lfsck, 0);
928 rc = lfsck_master_dir_engine(env, lfsck);
934 lfsck->li_new_scanned++;
935 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
936 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
938 CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
939 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
940 lfsck_fail(env, lfsck, true);
941 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
947 if (unlikely(!fid_is_sane(fid))) {
948 CDEBUG(D_LFSCK, "%s: OIT scan find invalid FID "DFID
950 lfsck_lfsck2name(lfsck), PFID(fid));
954 if (fid_is_idif(fid)) {
955 __u32 idx1 = fid_idif_ost_idx(fid);
957 LASSERT(!lfsck->li_master);
959 /* It is an old format device, update the LMA. */
961 struct ost_id *oi = &info->lti_oi;
963 fid_to_ostid(fid, oi);
964 ostid_to_fid(fid, oi, idx);
967 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
968 !fid_is_last_id(fid) &&
969 !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
971 /* If the FID/object is only used locally and invisible
972 * to external nodes, then LFSCK will not handle it.
974 * dot_lustre sequence has been handled specially. */
977 struct lu_seq_range *range = &info->lti_range;
979 if (lfsck->li_master)
980 fld_range_set_mdt(range);
982 fld_range_set_ost(range);
983 rc = fld_local_lookup(env, ss->ss_server_fld,
984 fid_seq(fid), range);
985 if (rc != 0 || range->lsr_index != idx) {
986 /* Remote object will be handled by the LFSCK
987 * instance on the MDT where the remote object
988 * really resides on. */
994 target = lfsck_object_find_bottom(env, lfsck, fid);
995 if (IS_ERR(target)) {
996 CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
997 DFID", cookie "LPU64": rc = %d\n",
998 lfsck_lfsck2name(lfsck), PFID(fid),
999 iops->store(env, di), rc);
1000 lfsck_fail(env, lfsck, true);
1001 if (bk->lb_param & LPF_FAILOUT)
1002 RETURN(PTR_ERR(target));
1007 if (dt_object_exists(target)) {
1009 rc = lfsck_update_lma(env, lfsck, target);
1011 CDEBUG(D_LFSCK, "%s: fail to update "
1012 "LMA for "DFID": rc = %d\n",
1013 lfsck_lfsck2name(lfsck),
1014 PFID(lfsck_dto2fid(target)), rc);
1017 rc = lfsck_exec_oit(env, lfsck, target);
1019 lfsck_object_put(env, target);
1020 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1024 rc = lfsck_checkpoint(env, lfsck);
1025 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1029 lfsck_control_speed(lfsck);
1031 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
1032 spin_lock(&lfsck->li_lock);
1033 thread_set_flags(thread, SVC_STOPPING);
1034 spin_unlock(&lfsck->li_lock);
1038 rc = iops->next(env, di);
1039 if (unlikely(rc > 0))
1040 lfsck->li_oit_over = 1;
1041 else if (likely(rc == 0))
1042 lfsck->li_current_oit_processed = 0;
1044 if (unlikely(!thread_is_running(thread))) {
1045 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1046 "cookie "LPU64"\n", lfsck_lfsck2name(lfsck),
1047 iops->store(env, di));
1050 } while (rc == 0 || lfsck->li_di_dir != NULL);
1055 int lfsck_master_engine(void *args)
1057 struct lfsck_thread_args *lta = args;
1058 struct lu_env *env = <a->lta_env;
1059 struct lfsck_instance *lfsck = lta->lta_lfsck;
1060 struct ptlrpc_thread *thread = &lfsck->li_thread;
1061 struct dt_object *oit_obj = lfsck->li_obj_oit;
1062 const struct dt_it_ops *oit_iops = &oit_obj->do_index_ops->dio_it;
1063 struct dt_it *oit_di;
1064 struct l_wait_info lwi = { 0 };
1068 if (lfsck->li_master &&
1069 (!list_empty(&lfsck->li_list_scan) ||
1070 !list_empty(&lfsck->li_list_double_scan))) {
1071 rc = lfsck_verify_lpf(env, lfsck);
1072 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1073 * fatal, because the .lustre/lost+found/ maybe not accessed
1074 * by the LFSCK if it does not add orphans or others to such
1075 * directory. So go ahead until hit failure when really uses
1078 CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1079 ".lustre/lost+found/, go ahead: rc = %d\n",
1080 lfsck_lfsck2name(lfsck), rc);
1083 oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit);
1084 if (IS_ERR(oit_di)) {
1085 rc = PTR_ERR(oit_di);
1086 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1087 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1089 GOTO(fini_args, rc);
1092 spin_lock(&lfsck->li_lock);
1093 lfsck->li_di_oit = oit_di;
1094 spin_unlock(&lfsck->li_lock);
1095 rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1099 CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1100 "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1101 ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1102 lfsck->li_pos_checkpoint.lp_oit_cookie,
1103 lfsck->li_pos_checkpoint.lp_dir_cookie,
1104 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1107 spin_lock(&lfsck->li_lock);
1108 thread_set_flags(thread, SVC_RUNNING);
1109 spin_unlock(&lfsck->li_lock);
1110 wake_up_all(&thread->t_ctl_waitq);
1112 l_wait_event(thread->t_ctl_waitq,
1113 lfsck->li_start_unplug ||
1114 !thread_is_running(thread),
1116 if (!thread_is_running(thread))
1117 GOTO(fini_oit, rc = 0);
1119 if (!list_empty(&lfsck->li_list_scan) ||
1120 list_empty(&lfsck->li_list_double_scan))
1121 rc = lfsck_master_oit_engine(env, lfsck);
1125 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
1126 CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1127 "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1128 ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1129 lfsck->li_pos_checkpoint.lp_oit_cookie,
1130 lfsck->li_pos_checkpoint.lp_dir_cookie,
1131 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1134 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1135 rc = lfsck_post(env, lfsck, rc);
1137 lfsck_close_dir(env, lfsck, rc);
1140 lfsck_di_oit_put(env, lfsck);
1141 oit_iops->fini(env, oit_di);
1143 if (!list_empty(&lfsck->li_list_double_scan))
1144 rc = lfsck_double_scan(env, lfsck);
1148 lfsck_quit(env, lfsck);
1151 /* XXX: Purge the pinned objects in the future. */
1154 spin_lock(&lfsck->li_lock);
1155 thread_set_flags(thread, SVC_STOPPED);
1156 spin_unlock(&lfsck->li_lock);
1157 wake_up_all(&thread->t_ctl_waitq);
1158 lfsck_thread_args_fini(lta);
1162 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1166 spin_lock(&lad->lad_lock);
1167 if (list_empty(&lad->lad_req_list))
1169 spin_unlock(&lad->lad_lock);
1175 * Query the LFSCK status from the instatnces on remote servers.
1177 * The LFSCK assistant thread queries the LFSCK instances on other
1178 * servers (MDT/OST) about their status, such as whether they have
1179 * finished the phase1/phase2 scanning or not, and so on.
1181 * \param[in] env pointer to the thread context
1182 * \param[in] com pointer to the lfsck component
1184 * \retval 0 for success
1185 * \retval negative error number on failure
1187 static int lfsck_assistant_query_others(const struct lu_env *env,
1188 struct lfsck_component *com)
1190 struct lfsck_thread_info *info = lfsck_env_info(env);
1191 struct lfsck_request *lr = &info->lti_lr;
1192 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1193 struct lfsck_instance *lfsck = com->lc_lfsck;
1194 struct lfsck_assistant_data *lad = com->lc_data;
1195 struct ptlrpc_request_set *set;
1196 struct lfsck_tgt_descs *ltds;
1197 struct lfsck_tgt_desc *ltd;
1198 struct list_head *phase_head;
1203 set = ptlrpc_prep_set();
1207 lad->lad_touch_gen++;
1208 memset(lr, 0, sizeof(*lr));
1209 lr->lr_event = LE_QUERY;
1210 lr->lr_active = com->lc_type;
1211 laia->laia_com = com;
1213 laia->laia_shared = 0;
1215 if (!list_empty(&lad->lad_mdt_phase1_list)) {
1216 ltds = &lfsck->li_mdt_descs;
1218 phase_head = &lad->lad_mdt_phase1_list;
1219 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1224 ltds = &lfsck->li_ost_descs;
1225 lr->lr_flags = LEF_TO_OST;
1226 phase_head = &lad->lad_ost_phase1_list;
1229 laia->laia_ltds = ltds;
1230 spin_lock(<ds->ltd_lock);
1231 while (!list_empty(phase_head)) {
1232 struct list_head *phase_list;
1235 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1236 ltd = list_entry(phase_head->next,
1237 struct lfsck_tgt_desc,
1238 ltd_layout_phase_list);
1239 phase_list = <d->ltd_layout_phase_list;
1240 gen = <d->ltd_layout_gen;
1242 ltd = list_entry(phase_head->next,
1243 struct lfsck_tgt_desc,
1244 ltd_namespace_phase_list);
1245 phase_list = <d->ltd_namespace_phase_list;
1246 gen = <d->ltd_namespace_gen;
1249 if (*gen == lad->lad_touch_gen)
1252 *gen = lad->lad_touch_gen;
1253 list_move_tail(phase_list, phase_head);
1254 atomic_inc(<d->ltd_ref);
1255 laia->laia_ltd = ltd;
1256 spin_unlock(<ds->ltd_lock);
1257 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1258 lfsck_async_interpret_common,
1261 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1262 "%s %x for %s: rc = %d\n",
1263 lfsck_lfsck2name(lfsck),
1264 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1265 ltd->ltd_index, lad->lad_name, rc);
1269 spin_lock(<ds->ltd_lock);
1271 spin_unlock(<ds->ltd_lock);
1273 rc = ptlrpc_set_wait(set);
1275 ptlrpc_set_destroy(set);
1279 if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1280 list_empty(&lad->lad_mdt_phase1_list))
1284 ptlrpc_set_destroy(set);
1286 RETURN(rc1 != 0 ? rc1 : rc);
1290 * Notify the LFSCK event to the instances on remote servers.
1292 * The LFSCK assistant thread notifies the LFSCK instances on other
1293 * servers (MDT/OST) about some events, such as start new scanning,
1294 * stop the scanning, this LFSCK instance will exit, and so on.
1296 * \param[in] env pointer to the thread context
1297 * \param[in] com pointer to the lfsck component
1298 * \param[in] lr pointer to the LFSCK event request
1300 * \retval 0 for success
1301 * \retval negative error number on failure
1303 static int lfsck_assistant_notify_others(const struct lu_env *env,
1304 struct lfsck_component *com,
1305 struct lfsck_request *lr)
1307 struct lfsck_thread_info *info = lfsck_env_info(env);
1308 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1309 struct lfsck_instance *lfsck = com->lc_lfsck;
1310 struct lfsck_assistant_data *lad = com->lc_data;
1311 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1312 struct ptlrpc_request_set *set;
1313 struct lfsck_tgt_descs *ltds;
1314 struct lfsck_tgt_desc *ltd;
1315 struct lfsck_tgt_desc *next;
1321 set = ptlrpc_prep_set();
1325 lr->lr_index = lfsck_dev_idx(lfsck);
1326 lr->lr_active = com->lc_type;
1327 laia->laia_com = com;
1329 laia->laia_shared = 0;
1331 switch (lr->lr_event) {
1333 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1336 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1337 lr->lr_speed = bk->lb_speed_limit;
1338 lr->lr_version = bk->lb_version;
1339 lr->lr_param |= bk->lb_param;
1340 lr->lr_async_windows = bk->lb_async_windows;
1341 lr->lr_flags = LEF_TO_OST;
1343 /* Notify OSTs firstly, then handle other MDTs if needed. */
1344 ltds = &lfsck->li_ost_descs;
1345 laia->laia_ltds = ltds;
1346 down_read(<ds->ltd_rw_sem);
1347 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1348 ltd = lfsck_tgt_get(ltds, idx);
1349 LASSERT(ltd != NULL);
1351 laia->laia_ltd = ltd;
1352 ltd->ltd_layout_done = 0;
1353 ltd->ltd_synced_failures = 0;
1354 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1355 lfsck_async_interpret_common,
1356 laia, LFSCK_NOTIFY);
1358 lfsck_lad_set_bitmap(env, com, idx);
1359 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1360 "notify OST %x for %s start: rc = %d\n",
1361 lfsck_lfsck2name(lfsck), idx,
1366 up_read(<ds->ltd_rw_sem);
1369 rc = ptlrpc_set_wait(set);
1371 ptlrpc_set_destroy(set);
1376 if (!(bk->lb_param & LPF_ALL_TGT))
1379 /* link other MDT targets locallly. */
1380 ltds = &lfsck->li_mdt_descs;
1381 spin_lock(<ds->ltd_lock);
1382 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1383 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1384 ltd = LTD_TGT(ltds, idx);
1385 LASSERT(ltd != NULL);
1387 if (!list_empty(<d->ltd_layout_list))
1390 list_add_tail(<d->ltd_layout_list,
1391 &lad->lad_mdt_list);
1392 list_add_tail(<d->ltd_layout_phase_list,
1393 &lad->lad_mdt_phase1_list);
1396 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1397 ltd = LTD_TGT(ltds, idx);
1398 LASSERT(ltd != NULL);
1400 if (!list_empty(<d->ltd_namespace_list))
1403 list_add_tail(<d->ltd_namespace_list,
1404 &lad->lad_mdt_list);
1405 list_add_tail(<d->ltd_namespace_phase_list,
1406 &lad->lad_mdt_phase1_list);
1409 spin_unlock(<ds->ltd_lock);
1412 case LE_PHASE2_DONE:
1413 case LE_PEER_EXIT: {
1414 struct list_head *phase_head;
1416 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1417 if (bk->lb_param & LPF_ALL_TGT) {
1418 phase_head = &lad->lad_mdt_list;
1419 ltds = &lfsck->li_mdt_descs;
1420 if (lr->lr_event == LE_STOP) {
1421 /* unlink other MDT targets locallly. */
1422 spin_lock(<ds->ltd_lock);
1423 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1424 list_for_each_entry_safe(ltd, next,
1425 phase_head, ltd_layout_list) {
1427 <d->ltd_layout_phase_list);
1429 <d->ltd_layout_list);
1432 list_for_each_entry_safe(ltd, next,
1434 ltd_namespace_list) {
1436 <d->ltd_namespace_phase_list);
1438 <d->ltd_namespace_list);
1441 spin_unlock(<ds->ltd_lock);
1443 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1446 lr->lr_flags |= LEF_TO_OST;
1447 phase_head = &lad->lad_ost_list;
1448 ltds = &lfsck->li_ost_descs;
1450 lr->lr_flags &= ~LEF_TO_OST;
1452 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1455 lr->lr_flags |= LEF_TO_OST;
1456 phase_head = &lad->lad_ost_list;
1457 ltds = &lfsck->li_ost_descs;
1461 laia->laia_ltds = ltds;
1462 spin_lock(<ds->ltd_lock);
1463 while (!list_empty(phase_head)) {
1464 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1465 ltd = list_entry(phase_head->next,
1466 struct lfsck_tgt_desc,
1468 if (!list_empty(<d->ltd_layout_phase_list))
1470 <d->ltd_layout_phase_list);
1471 list_del_init(<d->ltd_layout_list);
1473 ltd = list_entry(phase_head->next,
1474 struct lfsck_tgt_desc,
1475 ltd_namespace_list);
1476 if (!list_empty(<d->ltd_namespace_phase_list))
1478 <d->ltd_namespace_phase_list);
1479 list_del_init(<d->ltd_namespace_list);
1481 atomic_inc(<d->ltd_ref);
1482 laia->laia_ltd = ltd;
1483 spin_unlock(<ds->ltd_lock);
1484 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1485 lfsck_async_interpret_common,
1486 laia, LFSCK_NOTIFY);
1488 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1489 "notify %s %x for %s stop/phase2_done/"
1490 "peer_exit: rc = %d\n",
1491 lfsck_lfsck2name(lfsck),
1492 (lr->lr_flags & LEF_TO_OST) ?
1493 "OST" : "MDT", ltd->ltd_index,
1497 spin_lock(<ds->ltd_lock);
1499 spin_unlock(<ds->ltd_lock);
1501 rc = ptlrpc_set_wait(set);
1503 ptlrpc_set_destroy(set);
1507 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1508 !(lr->lr_flags & LEF_TO_OST)) {
1509 lr->lr_flags |= LEF_TO_OST;
1510 phase_head = &lad->lad_ost_list;
1511 ltds = &lfsck->li_ost_descs;
1516 case LE_PHASE1_DONE:
1517 lad->lad_ops->la_sync_failures(env, com, lr);
1518 lad->lad_touch_gen++;
1519 ltds = &lfsck->li_mdt_descs;
1520 laia->laia_ltds = ltds;
1521 spin_lock(<ds->ltd_lock);
1522 while (!list_empty(&lad->lad_mdt_list)) {
1523 struct list_head *list;
1526 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1527 ltd = list_entry(lad->lad_mdt_list.next,
1528 struct lfsck_tgt_desc,
1530 list = <d->ltd_layout_list;
1531 gen = <d->ltd_layout_gen;
1533 struct lfsck_namespace *ns = com->lc_file_ram;
1535 ltd = list_entry(lad->lad_mdt_list.next,
1536 struct lfsck_tgt_desc,
1537 ltd_namespace_list);
1538 list = <d->ltd_namespace_list;
1539 gen = <d->ltd_namespace_gen;
1540 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1543 if (*gen == lad->lad_touch_gen)
1546 *gen = lad->lad_touch_gen;
1547 list_move_tail(list, &lad->lad_mdt_list);
1548 if (ltd->ltd_synced_failures)
1551 atomic_inc(<d->ltd_ref);
1552 laia->laia_ltd = ltd;
1553 spin_unlock(<ds->ltd_lock);
1554 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1555 lfsck_async_interpret_common,
1556 laia, LFSCK_NOTIFY);
1558 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1559 "notify MDT %x for %s phase1 done: "
1560 "rc = %d\n", lfsck_lfsck2name(lfsck),
1561 ltd->ltd_index, lad->lad_name, rc);
1564 spin_lock(<ds->ltd_lock);
1566 spin_unlock(<ds->ltd_lock);
1569 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1570 "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1575 rc1 = ptlrpc_set_wait(set);
1576 ptlrpc_set_destroy(set);
1578 RETURN(rc != 0 ? rc : rc1);
1582 * The LFSCK assistant thread is triggered by the LFSCK main engine.
1583 * They co-work together as an asynchronous pipeline: the LFSCK main
1584 * engine scans the system and pre-fetches the objects, attributes,
1585 * or name entries, etc, and pushes them into the pipeline as input
1586 * requests for the LFSCK assistant thread; on the other end of the
1587 * pipeline, the LFSCK assistant thread performs the real check and
1588 * repair for every request from the main engine.
1590 * Generally, the assistant engine may be blocked when check/repair
1591 * something, so the LFSCK main engine will run some faster. On the
1592 * other hand, the LFSCK main engine will drive multiple assistant
1593 * threads in parallel, means for each LFSCK component on the master
1594 * (such as layout LFSCK, namespace LFSCK), there is an independent
1595 * LFSCK assistant thread. So under such 1:N multiple asynchronous
1596 * pipelines mode, the whole LFSCK performance will be much better
1597 * than check/repair everything by the LFSCK main engine itself.
1599 int lfsck_assistant_engine(void *args)
1601 struct lfsck_thread_args *lta = args;
1602 struct lu_env *env = <a->lta_env;
1603 struct lfsck_component *com = lta->lta_com;
1604 struct lfsck_instance *lfsck = lta->lta_lfsck;
1605 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1606 struct lfsck_position *pos = &com->lc_pos_start;
1607 struct lfsck_thread_info *info = lfsck_env_info(env);
1608 struct lfsck_request *lr = &info->lti_lr;
1609 struct lfsck_assistant_data *lad = com->lc_data;
1610 struct ptlrpc_thread *mthread = &lfsck->li_thread;
1611 struct ptlrpc_thread *athread = &lad->lad_thread;
1612 struct lfsck_assistant_operations *lao = lad->lad_ops;
1613 struct lfsck_assistant_req *lar;
1614 struct l_wait_info lwi = { 0 };
1620 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1621 lfsck_lfsck2name(lfsck), lad->lad_name);
1623 memset(lr, 0, sizeof(*lr));
1624 lr->lr_event = LE_START;
1625 if (pos->lp_oit_cookie <= 1)
1626 lr->lr_param = LPF_RESET;
1627 rc = lfsck_assistant_notify_others(env, com, lr);
1629 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1630 "to start %s: rc = %d\n",
1631 lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1635 spin_lock(&lad->lad_lock);
1636 thread_set_flags(athread, SVC_RUNNING);
1637 spin_unlock(&lad->lad_lock);
1638 wake_up_all(&mthread->t_ctl_waitq);
1641 while (!list_empty(&lad->lad_req_list)) {
1642 bool wakeup = false;
1644 if (unlikely(lad->lad_exit ||
1645 !thread_is_running(mthread)))
1646 GOTO(cleanup1, rc = lad->lad_post_result);
1648 lar = list_entry(lad->lad_req_list.next,
1649 struct lfsck_assistant_req,
1651 /* Only the lfsck_assistant_engine thread itself can
1652 * remove the "lar" from the head of the list, LFSCK
1653 * engine thread only inserts other new "lar" at the
1654 * end of the list. So it is safe to handle current
1655 * "lar" without the spin_lock. */
1656 rc = lao->la_handler_p1(env, com, lar);
1657 spin_lock(&lad->lad_lock);
1658 list_del_init(&lar->lar_list);
1659 lad->lad_prefetched--;
1660 /* Wake up the main engine thread only when the list
1661 * is empty or half of the prefetched items have been
1662 * handled to avoid too frequent thread schedule. */
1663 if (lad->lad_prefetched <= (bk->lb_async_windows / 2))
1665 spin_unlock(&lad->lad_lock);
1667 wake_up_all(&mthread->t_ctl_waitq);
1669 lao->la_req_fini(env, lar);
1670 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1674 l_wait_event(athread->t_ctl_waitq,
1675 !lfsck_assistant_req_empty(lad) ||
1678 lad->lad_to_double_scan,
1681 if (unlikely(lad->lad_exit))
1682 GOTO(cleanup1, rc = lad->lad_post_result);
1684 if (!list_empty(&lad->lad_req_list))
1687 if (lad->lad_to_post) {
1688 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1689 lfsck_lfsck2name(lfsck), lad->lad_name);
1691 if (unlikely(lad->lad_exit))
1692 GOTO(cleanup1, rc = lad->lad_post_result);
1694 lad->lad_to_post = 0;
1695 LASSERT(lad->lad_post_result > 0);
1697 memset(lr, 0, sizeof(*lr));
1698 lr->lr_event = LE_PHASE1_DONE;
1699 lr->lr_status = lad->lad_post_result;
1700 rc = lfsck_assistant_notify_others(env, com, lr);
1702 CDEBUG(D_LFSCK, "%s: LFSCK assistant notified "
1703 "others for %s post: rc = %d\n",
1704 lfsck_lfsck2name(lfsck),
1707 /* Wakeup the master engine to go ahead. */
1708 wake_up_all(&mthread->t_ctl_waitq);
1711 if (lad->lad_to_double_scan) {
1712 lad->lad_to_double_scan = 0;
1713 atomic_inc(&lfsck->li_double_scan_count);
1714 lad->lad_in_double_scan = 1;
1715 wake_up_all(&mthread->t_ctl_waitq);
1717 com->lc_new_checked = 0;
1718 com->lc_new_scanned = 0;
1719 com->lc_time_last_checkpoint = cfs_time_current();
1720 com->lc_time_next_checkpoint =
1721 com->lc_time_last_checkpoint +
1722 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1724 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1725 "the second-stage scaning\n",
1726 lfsck_lfsck2name(lfsck));
1728 /* Flush async updates before handling orphan. */
1729 rc2 = dt_sync(env, lfsck->li_next);
1731 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1732 "scan start, synced: rc = %d\n",
1733 lfsck_lfsck2name(lfsck), rc2);
1735 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1736 GOTO(cleanup2, rc = 0);
1738 while (lad->lad_in_double_scan) {
1739 rc = lfsck_assistant_query_others(env, com);
1740 if (lfsck_phase2_next_ready(lad))
1746 /* Pull LFSCK status on related targets once
1747 * per 30 seconds if we are not notified. */
1748 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1749 cfs_time_seconds(1),
1751 rc = l_wait_event(athread->t_ctl_waitq,
1752 lfsck_phase2_next_ready(lad) ||
1754 !thread_is_running(mthread),
1757 if (unlikely(lad->lad_exit ||
1758 !thread_is_running(mthread)))
1759 GOTO(cleanup2, rc = 0);
1761 if (rc == -ETIMEDOUT)
1768 rc = lao->la_handler_p2(env, com);
1772 if (unlikely(lad->lad_exit ||
1773 !thread_is_running(mthread)))
1774 GOTO(cleanup2, rc = 0);
1780 /* Cleanup the unfinished requests. */
1781 spin_lock(&lad->lad_lock);
1783 lad->lad_assistant_status = rc;
1785 if (lad->lad_exit && lad->lad_post_result <= 0)
1786 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1788 while (!list_empty(&lad->lad_req_list)) {
1789 lar = list_entry(lad->lad_req_list.next,
1790 struct lfsck_assistant_req,
1792 list_del_init(&lar->lar_list);
1793 lad->lad_prefetched--;
1794 spin_unlock(&lad->lad_lock);
1795 lao->la_req_fini(env, lar);
1796 spin_lock(&lad->lad_lock);
1798 spin_unlock(&lad->lad_lock);
1800 LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1801 lad->lad_prefetched);
1804 memset(lr, 0, sizeof(*lr));
1806 lr->lr_event = LE_PHASE2_DONE;
1808 } else if (rc == 0) {
1809 if (lfsck->li_flags & LPF_ALL_TGT) {
1810 lr->lr_event = LE_STOP;
1811 lr->lr_status = LS_STOPPED;
1813 lr->lr_event = LE_PEER_EXIT;
1814 switch (lfsck->li_status) {
1817 lr->lr_status = LS_CO_PAUSED;
1821 lr->lr_status = LS_CO_STOPPED;
1824 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1825 "status: rc = %d\n",
1826 lfsck_lfsck2name(lfsck),
1828 lr->lr_status = LS_CO_FAILED;
1833 if (lfsck->li_flags & LPF_ALL_TGT) {
1834 lr->lr_event = LE_STOP;
1835 lr->lr_status = LS_FAILED;
1837 lr->lr_event = LE_PEER_EXIT;
1838 lr->lr_status = LS_CO_FAILED;
1842 rc1 = lfsck_assistant_notify_others(env, com, lr);
1844 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1845 "others for %s quit: rc = %d\n",
1846 lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1850 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1851 lfsck_lfsck2name(lfsck));
1853 /* Flush async updates before exit. */
1854 rc2 = dt_sync(env, lfsck->li_next);
1856 CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1857 lfsck_lfsck2name(lfsck), rc2);
1859 /* Under force exit case, some requests may be just freed without
1860 * verification, those objects should be re-handled when next run.
1861 * So not update the on-disk trace file under such case. */
1862 if (lad->lad_in_double_scan) {
1864 rc1 = lao->la_double_scan_result(env, com, rc);
1866 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1867 "finished: rc = %d\n",
1868 lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1872 if (lad->lad_in_double_scan)
1873 atomic_dec(&lfsck->li_double_scan_count);
1875 spin_lock(&lad->lad_lock);
1876 lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1877 thread_set_flags(athread, SVC_STOPPED);
1878 wake_up_all(&mthread->t_ctl_waitq);
1879 spin_unlock(&lad->lad_lock);
1881 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1882 lfsck_lfsck2name(lfsck), lad->lad_name,
1883 lad->lad_assistant_status);
1885 lfsck_thread_args_fini(lta);