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, 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, 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, BYPASS_CAPA);
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 *)"..", 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 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,
258 dt_read_unlock(env, obj);
262 if (rc < 0 && rc != -ENODATA)
265 rc = lfsck_parent_fid(env, obj, fid);
267 lfsck_object_put(env, obj);
273 if (!fid_is_sane(fid))
278 if (depth > 0 && obj != NULL)
279 lfsck_object_put(env, obj);
284 static int lfsck_load_stripe_lmv(const struct lu_env *env,
285 struct lfsck_instance *lfsck,
286 struct dt_object *obj)
288 struct lmv_mds_md_v1 *lmv = &lfsck_env_info(env)->lti_lmv;
289 struct lfsck_lmv *llmv;
293 LASSERT(lfsck->li_obj_dir == NULL);
294 LASSERT(lfsck->li_lmv == NULL);
296 rc = lfsck_read_stripe_lmv(env, obj, lmv);
297 if (rc == -ENODATA) {
298 lfsck->li_obj_dir = lfsck_object_get(obj);
310 if (lmv->lmv_magic == LMV_MAGIC) {
311 struct lfsck_slave_lmv_rec *lslr;
314 llmv->ll_lmv_master = 1;
315 if (lmv->lmv_stripe_count < 1)
316 stripes = LFSCK_LMV_DEF_STRIPES;
317 else if (lmv->lmv_stripe_count > LFSCK_LMV_MAX_STRIPES)
318 stripes = LFSCK_LMV_MAX_STRIPES;
320 stripes = lmv->lmv_stripe_count;
322 OBD_ALLOC_LARGE(lslr, sizeof(*lslr) * stripes);
329 llmv->ll_stripes_allocated = stripes;
330 llmv->ll_hash_type = LMV_HASH_TYPE_UNKNOWN;
331 llmv->ll_lslr = lslr;
333 llmv->ll_lmv_slave = 1;
336 lfsck->li_obj_dir = lfsck_object_get(obj);
338 atomic_set(&llmv->ll_ref, 1);
339 lfsck->li_lmv = llmv;
344 /* LFSCK wrap functions */
346 static void lfsck_fail(const struct lu_env *env, struct lfsck_instance *lfsck,
349 struct lfsck_component *com;
351 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
352 com->lc_ops->lfsck_fail(env, com, new_checked);
356 void lfsck_close_dir(const struct lu_env *env,
357 struct lfsck_instance *lfsck, int result)
359 struct lfsck_component *com;
362 if (lfsck->li_lmv != NULL) {
363 lfsck->li_lmv->ll_exit_value = result;
364 if (lfsck->li_obj_dir != NULL) {
365 list_for_each_entry(com, &lfsck->li_list_dir,
367 com->lc_ops->lfsck_close_dir(env, com);
371 lfsck_lmv_put(env, lfsck->li_lmv);
372 lfsck->li_lmv = NULL;
375 if (lfsck->li_di_dir != NULL) {
376 const struct dt_it_ops *dir_iops;
377 struct dt_it *dir_di = lfsck->li_di_dir;
379 LASSERT(lfsck->li_obj_dir != NULL);
381 dir_iops = &lfsck->li_obj_dir->do_index_ops->dio_it;
382 lfsck_di_dir_put(env, lfsck);
383 dir_iops->fini(env, dir_di);
386 if (lfsck->li_obj_dir != NULL) {
387 struct dt_object *dir_obj = lfsck->li_obj_dir;
389 lfsck->li_obj_dir = NULL;
390 lfsck_object_put(env, dir_obj);
396 int lfsck_open_dir(const struct lu_env *env,
397 struct lfsck_instance *lfsck, __u64 cookie)
399 struct dt_object *obj = lfsck->li_obj_dir;
400 struct dt_it *di = lfsck->li_di_dir;
401 struct lfsck_component *com;
402 const struct dt_it_ops *iops;
406 LASSERT(obj != NULL);
409 if (unlikely(!dt_try_as_dir(env, obj)))
410 GOTO(out, rc = -ENOTDIR);
412 list_for_each_entry(com, &lfsck->li_list_dir, lc_link_dir) {
413 rc = com->lc_ops->lfsck_open_dir(env, com);
418 iops = &obj->do_index_ops->dio_it;
419 di = iops->init(env, obj, lfsck->li_args_dir, BYPASS_CAPA);
421 GOTO(out, rc = PTR_ERR(di));
423 rc = iops->load(env, di, cookie);
424 if (rc == 0 || (rc > 0 && cookie > 0))
425 rc = iops->next(env, di);
433 lfsck->li_cookie_dir = iops->store(env, di);
434 spin_lock(&lfsck->li_lock);
435 lfsck->li_di_dir = di;
436 spin_unlock(&lfsck->li_lock);
443 lfsck_close_dir(env, lfsck, rc);
448 static int lfsck_checkpoint(const struct lu_env *env,
449 struct lfsck_instance *lfsck)
451 struct lfsck_component *com;
455 if (likely(cfs_time_beforeq(cfs_time_current(),
456 lfsck->li_time_next_checkpoint)))
459 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
460 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
461 rc = com->lc_ops->lfsck_checkpoint(env, com, false);
466 lfsck->li_time_last_checkpoint = cfs_time_current();
467 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
468 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
469 return rc1 != 0 ? rc1 : rc;
472 static int lfsck_prep(const struct lu_env *env, struct lfsck_instance *lfsck,
473 struct lfsck_start_param *lsp)
475 struct dt_object *obj = NULL;
476 struct lfsck_component *com;
477 struct lfsck_component *next;
478 struct lfsck_position *pos = NULL;
479 const struct dt_it_ops *iops =
480 &lfsck->li_obj_oit->do_index_ops->dio_it;
484 LASSERT(lfsck->li_obj_dir == NULL);
485 LASSERT(lfsck->li_di_dir == NULL);
487 lfsck->li_current_oit_processed = 0;
488 list_for_each_entry_safe(com, next, &lfsck->li_list_scan, lc_link) {
489 com->lc_new_checked = 0;
490 rc = com->lc_ops->lfsck_prep(env, com, lsp);
495 (!lfsck_pos_is_zero(&com->lc_pos_start) &&
496 lfsck_pos_is_eq(pos, &com->lc_pos_start) > 0))
497 pos = &com->lc_pos_start;
500 /* Init otable-based iterator. */
502 rc = iops->load(env, lfsck->li_di_oit, 0);
504 lfsck->li_oit_over = 1;
511 rc = iops->load(env, lfsck->li_di_oit, pos->lp_oit_cookie);
515 lfsck->li_oit_over = 1;
517 if (!lfsck->li_master || fid_is_zero(&pos->lp_dir_parent))
520 /* Find the directory for namespace-based traverse. */
521 obj = lfsck_object_find_bottom(env, lfsck, &pos->lp_dir_parent);
523 RETURN(PTR_ERR(obj));
525 /* Remote directory will be scanned by the LFSCK instance
526 * on the MDT where the remote object really resides on. */
527 if (!dt_object_exists(obj) || dt_object_remote(obj) ||
528 unlikely(!S_ISDIR(lfsck_object_type(obj))))
531 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
533 /* For the master MDT-object of a striped directory,
534 * reset the iteration from the directory beginning. */
535 if (lfsck->li_lmv != NULL && lfsck->li_lmv->ll_lmv_master)
536 pos->lp_dir_cookie = 0;
538 rc = lfsck_open_dir(env, lfsck, pos->lp_dir_cookie);
540 /* The end of the directory. */
548 lfsck_object_put(env, obj);
551 lfsck_close_dir(env, lfsck, rc);
552 list_for_each_entry_safe(com, next, &lfsck->li_list_scan,
554 com->lc_ops->lfsck_post(env, com, rc, true);
561 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, true);
562 lfsck->li_pos_current = lfsck->li_pos_checkpoint;
563 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
564 rc = com->lc_ops->lfsck_checkpoint(env, com, true);
569 lfsck->li_time_last_checkpoint = cfs_time_current();
570 lfsck->li_time_next_checkpoint = lfsck->li_time_last_checkpoint +
571 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
575 static int lfsck_exec_oit(const struct lu_env *env,
576 struct lfsck_instance *lfsck, struct dt_object *obj)
578 struct lfsck_component *com;
582 LASSERT(lfsck->li_obj_dir == NULL);
584 list_for_each_entry(com, &lfsck->li_list_scan, lc_link) {
585 rc = com->lc_ops->lfsck_exec_oit(env, com, obj);
590 rc = lfsck_needs_scan_dir(env, lfsck, obj);
594 rc = lfsck_load_stripe_lmv(env, lfsck, obj);
596 rc = lfsck_open_dir(env, lfsck, 0);
602 lfsck_fail(env, lfsck, false);
605 lfsck_close_dir(env, lfsck, rc);
607 return rc > 0 ? 0 : rc;
610 static int lfsck_exec_dir(const struct lu_env *env,
611 struct lfsck_instance *lfsck,
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, 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;
752 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY2, cfs_fail_val) &&
753 unlikely(!thread_is_running(thread))) {
754 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
755 "parent "DFID", cookie "LPX64"\n",
756 lfsck_lfsck2name(lfsck),
757 PFID(lfsck_dto2fid(dir)), lfsck->li_cookie_dir);
762 lfsck->li_new_scanned++;
763 rc = iops->rec(env, di, (struct dt_rec *)ent,
766 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
770 CDEBUG(D_LFSCK, "%s: scan dir failed at rec(), "
771 "parent "DFID", cookie "LPX64": rc = %d\n",
772 lfsck_lfsck2name(lfsck),
773 PFID(lfsck_dto2fid(dir)),
774 lfsck->li_cookie_dir, rc);
775 lfsck_fail(env, lfsck, true);
776 if (bk->lb_param & LPF_FAILOUT)
782 if (ent->lde_attrs & LUDA_IGNORE &&
783 strcmp(ent->lde_name, dotdot) != 0)
786 /* skip dot entry. */
787 if (ent->lde_namelen == 1 && ent->lde_name[0] == '.')
790 /* The type in the @ent structure may has been overwritten,
791 * so we need to pass the @type parameter independently. */
792 rc = lfsck_exec_dir(env, lfsck, ent, type);
793 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
797 rc = lfsck_checkpoint(env, lfsck);
798 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
802 lfsck_control_speed(lfsck);
803 if (unlikely(!thread_is_running(thread))) {
804 CDEBUG(D_LFSCK, "%s: scan dir exit for engine stop, "
805 "parent "DFID", cookie "LPX64"\n",
806 lfsck_lfsck2name(lfsck),
807 PFID(lfsck_dto2fid(dir)),
808 lfsck->li_cookie_dir);
812 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL2)) {
813 spin_lock(&lfsck->li_lock);
814 thread_set_flags(thread, SVC_STOPPING);
815 spin_unlock(&lfsck->li_lock);
819 rc = iops->next(env, di);
822 if (rc > 0 && !lfsck->li_oit_over)
823 lfsck_close_dir(env, lfsck, rc);
829 * Object-table based iteration engine.
831 * Object-table based iteration is the basic linear engine to scan all the
832 * objects on current device in turn. For each object, it calls all the
833 * registered LFSCK component(s)' API to perform related consistency
836 * It flushes related LFSCK trace files to disk via making checkpoint
837 * periodically. Then if the server crashed or the LFSCK is paused, the
838 * LFSCK can resume from the latest checkpoint.
840 * It also controls the whole LFSCK speed via lfsck_control_speed() to
841 * avoid the server to become overload.
843 * \param[in] env pointer to the thread context
844 * \param[in] lfsck pointer to the lfsck instance
846 * \retval positive number if all objects have been scanned
847 * \retval 0 if the iteration is stopped or paused
848 * \retval negative error number on failure
850 static int lfsck_master_oit_engine(const struct lu_env *env,
851 struct lfsck_instance *lfsck)
853 struct lfsck_thread_info *info = lfsck_env_info(env);
854 const struct dt_it_ops *iops =
855 &lfsck->li_obj_oit->do_index_ops->dio_it;
856 struct dt_it *di = lfsck->li_di_oit;
857 struct lu_fid *fid = &info->lti_fid;
858 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
859 struct ptlrpc_thread *thread = &lfsck->li_thread;
860 struct seq_server_site *ss = lfsck_dev_site(lfsck);
861 __u32 idx = lfsck_dev_idx(lfsck);
865 if (unlikely(ss == NULL))
869 struct dt_object *target;
870 bool update_lma = false;
872 if (lfsck->li_di_dir != NULL) {
873 rc = lfsck_master_dir_engine(env, lfsck);
878 if (unlikely(lfsck->li_oit_over))
881 if (CFS_FAIL_TIMEOUT(OBD_FAIL_LFSCK_DELAY1, cfs_fail_val) &&
882 unlikely(!thread_is_running(thread))) {
883 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
885 lfsck_lfsck2name(lfsck), iops->store(env, di));
890 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
893 lfsck->li_current_oit_processed = 1;
895 if (!list_empty(&lfsck->li_list_lmv)) {
896 struct lfsck_lmv_unit *llu;
898 spin_lock(&lfsck->li_lock);
899 llu = list_entry(lfsck->li_list_lmv.next,
900 struct lfsck_lmv_unit, llu_link);
901 list_del_init(&llu->llu_link);
902 spin_unlock(&lfsck->li_lock);
904 lfsck->li_lmv = &llu->llu_lmv;
905 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
906 rc = lfsck_open_dir(env, lfsck, 0);
908 rc = lfsck_master_dir_engine(env, lfsck);
914 lfsck->li_new_scanned++;
915 lfsck->li_pos_current.lp_oit_cookie = iops->store(env, di);
916 rc = iops->rec(env, di, (struct dt_rec *)fid, 0);
918 CDEBUG(D_LFSCK, "%s: OIT scan failed at rec(): "
919 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
920 lfsck_fail(env, lfsck, true);
921 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
927 if (unlikely(!fid_is_sane(fid))) {
928 CDEBUG(D_LFSCK, "%s: OIT scan find invalid FID "DFID
930 lfsck_lfsck2name(lfsck), PFID(fid));
934 if (fid_is_idif(fid)) {
935 __u32 idx1 = fid_idif_ost_idx(fid);
937 LASSERT(!lfsck->li_master);
939 /* It is an old format device, update the LMA. */
941 struct ost_id *oi = &info->lti_oi;
943 fid_to_ostid(fid, oi);
944 ostid_to_fid(fid, oi, idx);
947 } else if (!fid_is_norm(fid) && !fid_is_igif(fid) &&
948 !fid_is_last_id(fid) &&
949 !lu_fid_eq(fid, &lfsck->li_global_root_fid)) {
951 /* If the FID/object is only used locally and invisible
952 * to external nodes, then LFSCK will not handle it.
954 * dot_lustre sequence has been handled specially. */
957 struct lu_seq_range *range = &info->lti_range;
959 if (lfsck->li_master)
960 fld_range_set_mdt(range);
962 fld_range_set_ost(range);
963 rc = fld_local_lookup(env, ss->ss_server_fld,
964 fid_seq(fid), range);
965 if (rc != 0 || range->lsr_index != idx) {
966 /* Remote object will be handled by the LFSCK
967 * instance on the MDT where the remote object
968 * really resides on. */
974 target = lfsck_object_find_bottom(env, lfsck, fid);
975 if (IS_ERR(target)) {
976 CDEBUG(D_LFSCK, "%s: OIT scan failed at find target "
977 DFID", cookie "LPU64": rc = %d\n",
978 lfsck_lfsck2name(lfsck), PFID(fid),
979 iops->store(env, di), rc);
980 lfsck_fail(env, lfsck, true);
981 if (bk->lb_param & LPF_FAILOUT)
982 RETURN(PTR_ERR(target));
987 if (dt_object_exists(target)) {
989 rc = lfsck_update_lma(env, lfsck, target);
991 CDEBUG(D_LFSCK, "%s: fail to update "
992 "LMA for "DFID": rc = %d\n",
993 lfsck_lfsck2name(lfsck),
994 PFID(lfsck_dto2fid(target)), rc);
997 rc = lfsck_exec_oit(env, lfsck, target);
999 lfsck_object_put(env, target);
1000 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1004 rc = lfsck_checkpoint(env, lfsck);
1005 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
1009 lfsck_control_speed(lfsck);
1011 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_FATAL1)) {
1012 spin_lock(&lfsck->li_lock);
1013 thread_set_flags(thread, SVC_STOPPING);
1014 spin_unlock(&lfsck->li_lock);
1018 rc = iops->next(env, di);
1019 if (unlikely(rc > 0))
1020 lfsck->li_oit_over = 1;
1021 else if (likely(rc == 0))
1022 lfsck->li_current_oit_processed = 0;
1024 if (unlikely(!thread_is_running(thread))) {
1025 CDEBUG(D_LFSCK, "%s: OIT scan exit for engine stop, "
1026 "cookie "LPU64"\n", lfsck_lfsck2name(lfsck),
1027 iops->store(env, di));
1030 } while (rc == 0 || lfsck->li_di_dir != NULL);
1035 int lfsck_master_engine(void *args)
1037 struct lfsck_thread_args *lta = args;
1038 struct lu_env *env = <a->lta_env;
1039 struct lfsck_instance *lfsck = lta->lta_lfsck;
1040 struct ptlrpc_thread *thread = &lfsck->li_thread;
1041 struct dt_object *oit_obj = lfsck->li_obj_oit;
1042 const struct dt_it_ops *oit_iops = &oit_obj->do_index_ops->dio_it;
1043 struct dt_it *oit_di;
1044 struct l_wait_info lwi = { 0 };
1048 if (lfsck->li_master &&
1049 (!list_empty(&lfsck->li_list_scan) ||
1050 !list_empty(&lfsck->li_list_double_scan))) {
1051 rc = lfsck_verify_lpf(env, lfsck);
1052 /* Fail to verify the .lustre/lost+found/MDTxxxx/ may be not
1053 * fatal, because the .lustre/lost+found/ maybe not accessed
1054 * by the LFSCK if it does not add orphans or others to such
1055 * directory. So go ahead until hit failure when really uses
1058 CDEBUG(D_LFSCK, "%s: master engine fail to verify the "
1059 ".lustre/lost+found/, go ahead: rc = %d\n",
1060 lfsck_lfsck2name(lfsck), rc);
1063 oit_di = oit_iops->init(env, oit_obj, lfsck->li_args_oit, BYPASS_CAPA);
1064 if (IS_ERR(oit_di)) {
1065 rc = PTR_ERR(oit_di);
1066 CDEBUG(D_LFSCK, "%s: master engine fail to init iteration: "
1067 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
1069 GOTO(fini_args, rc);
1072 spin_lock(&lfsck->li_lock);
1073 lfsck->li_di_oit = oit_di;
1074 spin_unlock(&lfsck->li_lock);
1075 rc = lfsck_prep(env, lfsck, lta->lta_lsp);
1079 CDEBUG(D_LFSCK, "LFSCK entry: oit_flags = %#x, dir_flags = %#x, "
1080 "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1081 ", pid = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1082 lfsck->li_pos_checkpoint.lp_oit_cookie,
1083 lfsck->li_pos_checkpoint.lp_dir_cookie,
1084 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1087 spin_lock(&lfsck->li_lock);
1088 thread_set_flags(thread, SVC_RUNNING);
1089 spin_unlock(&lfsck->li_lock);
1090 wake_up_all(&thread->t_ctl_waitq);
1092 l_wait_event(thread->t_ctl_waitq,
1093 lfsck->li_start_unplug ||
1094 !thread_is_running(thread),
1096 if (!thread_is_running(thread))
1097 GOTO(fini_oit, rc = 0);
1099 if (!list_empty(&lfsck->li_list_scan) ||
1100 list_empty(&lfsck->li_list_double_scan))
1101 rc = lfsck_master_oit_engine(env, lfsck);
1105 lfsck_pos_fill(env, lfsck, &lfsck->li_pos_checkpoint, false);
1106 CDEBUG(D_LFSCK, "LFSCK exit: oit_flags = %#x, dir_flags = %#x, "
1107 "oit_cookie = "LPU64", dir_cookie = "LPX64", parent = "DFID
1108 ", pid = %d, rc = %d\n", lfsck->li_args_oit, lfsck->li_args_dir,
1109 lfsck->li_pos_checkpoint.lp_oit_cookie,
1110 lfsck->li_pos_checkpoint.lp_dir_cookie,
1111 PFID(&lfsck->li_pos_checkpoint.lp_dir_parent),
1114 if (!OBD_FAIL_CHECK(OBD_FAIL_LFSCK_CRASH))
1115 rc = lfsck_post(env, lfsck, rc);
1117 lfsck_close_dir(env, lfsck, rc);
1120 lfsck_di_oit_put(env, lfsck);
1121 oit_iops->fini(env, oit_di);
1123 if (!list_empty(&lfsck->li_list_double_scan))
1124 rc = lfsck_double_scan(env, lfsck);
1128 lfsck_quit(env, lfsck);
1131 /* XXX: Purge the pinned objects in the future. */
1134 spin_lock(&lfsck->li_lock);
1135 thread_set_flags(thread, SVC_STOPPED);
1136 spin_unlock(&lfsck->li_lock);
1137 wake_up_all(&thread->t_ctl_waitq);
1138 lfsck_thread_args_fini(lta);
1142 static inline bool lfsck_assistant_req_empty(struct lfsck_assistant_data *lad)
1146 spin_lock(&lad->lad_lock);
1147 if (list_empty(&lad->lad_req_list))
1149 spin_unlock(&lad->lad_lock);
1155 * Query the LFSCK status from the instatnces on remote servers.
1157 * The LFSCK assistant thread queries the LFSCK instances on other
1158 * servers (MDT/OST) about their status, such as whether they have
1159 * finished the phase1/phase2 scanning or not, and so on.
1161 * \param[in] env pointer to the thread context
1162 * \param[in] com pointer to the lfsck component
1164 * \retval 0 for success
1165 * \retval negative error number on failure
1167 static int lfsck_assistant_query_others(const struct lu_env *env,
1168 struct lfsck_component *com)
1170 struct lfsck_thread_info *info = lfsck_env_info(env);
1171 struct lfsck_request *lr = &info->lti_lr;
1172 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1173 struct lfsck_instance *lfsck = com->lc_lfsck;
1174 struct lfsck_assistant_data *lad = com->lc_data;
1175 struct ptlrpc_request_set *set;
1176 struct lfsck_tgt_descs *ltds;
1177 struct lfsck_tgt_desc *ltd;
1178 struct list_head *phase_head;
1183 set = ptlrpc_prep_set();
1187 lad->lad_touch_gen++;
1188 memset(lr, 0, sizeof(*lr));
1189 lr->lr_event = LE_QUERY;
1190 lr->lr_active = com->lc_type;
1191 laia->laia_com = com;
1193 laia->laia_shared = 0;
1195 if (!list_empty(&lad->lad_mdt_phase1_list)) {
1196 ltds = &lfsck->li_mdt_descs;
1198 phase_head = &lad->lad_mdt_phase1_list;
1199 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1204 ltds = &lfsck->li_ost_descs;
1205 lr->lr_flags = LEF_TO_OST;
1206 phase_head = &lad->lad_ost_phase1_list;
1209 laia->laia_ltds = ltds;
1210 spin_lock(<ds->ltd_lock);
1211 while (!list_empty(phase_head)) {
1212 struct list_head *phase_list;
1215 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1216 ltd = list_entry(phase_head->next,
1217 struct lfsck_tgt_desc,
1218 ltd_layout_phase_list);
1219 phase_list = <d->ltd_layout_phase_list;
1220 gen = <d->ltd_layout_gen;
1222 ltd = list_entry(phase_head->next,
1223 struct lfsck_tgt_desc,
1224 ltd_namespace_phase_list);
1225 phase_list = <d->ltd_namespace_phase_list;
1226 gen = <d->ltd_namespace_gen;
1229 if (*gen == lad->lad_touch_gen)
1232 *gen = lad->lad_touch_gen;
1233 list_move_tail(phase_list, phase_head);
1234 atomic_inc(<d->ltd_ref);
1235 laia->laia_ltd = ltd;
1236 spin_unlock(<ds->ltd_lock);
1237 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1238 lfsck_async_interpret_common,
1241 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to query "
1242 "%s %x for %s: rc = %d\n",
1243 lfsck_lfsck2name(lfsck),
1244 (lr->lr_flags & LEF_TO_OST) ? "OST" : "MDT",
1245 ltd->ltd_index, lad->lad_name, rc);
1249 spin_lock(<ds->ltd_lock);
1251 spin_unlock(<ds->ltd_lock);
1253 rc = ptlrpc_set_wait(set);
1255 ptlrpc_set_destroy(set);
1259 if (com->lc_type == LFSCK_TYPE_LAYOUT && !(lr->lr_flags & LEF_TO_OST) &&
1260 list_empty(&lad->lad_mdt_phase1_list))
1264 ptlrpc_set_destroy(set);
1266 RETURN(rc1 != 0 ? rc1 : rc);
1270 * Notify the LFSCK event to the instances on remote servers.
1272 * The LFSCK assistant thread notifies the LFSCK instances on other
1273 * servers (MDT/OST) about some events, such as start new scanning,
1274 * stop the scanning, this LFSCK instance will exit, and so on.
1276 * \param[in] env pointer to the thread context
1277 * \param[in] com pointer to the lfsck component
1278 * \param[in] lr pointer to the LFSCK event request
1280 * \retval 0 for success
1281 * \retval negative error number on failure
1283 static int lfsck_assistant_notify_others(const struct lu_env *env,
1284 struct lfsck_component *com,
1285 struct lfsck_request *lr)
1287 struct lfsck_thread_info *info = lfsck_env_info(env);
1288 struct lfsck_async_interpret_args *laia = &info->lti_laia;
1289 struct lfsck_instance *lfsck = com->lc_lfsck;
1290 struct lfsck_assistant_data *lad = com->lc_data;
1291 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1292 struct ptlrpc_request_set *set;
1293 struct lfsck_tgt_descs *ltds;
1294 struct lfsck_tgt_desc *ltd;
1295 struct lfsck_tgt_desc *next;
1301 set = ptlrpc_prep_set();
1305 lr->lr_index = lfsck_dev_idx(lfsck);
1306 lr->lr_active = com->lc_type;
1307 laia->laia_com = com;
1309 laia->laia_shared = 0;
1311 switch (lr->lr_event) {
1313 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1316 lr->lr_valid = LSV_SPEED_LIMIT | LSV_ERROR_HANDLE | LSV_DRYRUN;
1317 lr->lr_speed = bk->lb_speed_limit;
1318 lr->lr_version = bk->lb_version;
1319 lr->lr_param |= bk->lb_param;
1320 lr->lr_async_windows = bk->lb_async_windows;
1321 lr->lr_flags = LEF_TO_OST;
1323 /* Notify OSTs firstly, then handle other MDTs if needed. */
1324 ltds = &lfsck->li_ost_descs;
1325 laia->laia_ltds = ltds;
1326 down_read(<ds->ltd_rw_sem);
1327 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1328 ltd = lfsck_tgt_get(ltds, idx);
1329 LASSERT(ltd != NULL);
1331 laia->laia_ltd = ltd;
1332 ltd->ltd_layout_done = 0;
1333 ltd->ltd_synced_failures = 0;
1334 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1335 lfsck_async_interpret_common,
1336 laia, LFSCK_NOTIFY);
1338 lfsck_lad_set_bitmap(env, com, idx);
1339 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1340 "notify OST %x for %s start: rc = %d\n",
1341 lfsck_lfsck2name(lfsck), idx,
1346 up_read(<ds->ltd_rw_sem);
1349 rc = ptlrpc_set_wait(set);
1351 ptlrpc_set_destroy(set);
1356 if (!(bk->lb_param & LPF_ALL_TGT))
1359 /* link other MDT targets locallly. */
1360 ltds = &lfsck->li_mdt_descs;
1361 spin_lock(<ds->ltd_lock);
1362 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1363 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1364 ltd = LTD_TGT(ltds, idx);
1365 LASSERT(ltd != NULL);
1367 if (!list_empty(<d->ltd_layout_list))
1370 list_add_tail(<d->ltd_layout_list,
1371 &lad->lad_mdt_list);
1372 list_add_tail(<d->ltd_layout_phase_list,
1373 &lad->lad_mdt_phase1_list);
1376 cfs_foreach_bit(ltds->ltd_tgts_bitmap, idx) {
1377 ltd = LTD_TGT(ltds, idx);
1378 LASSERT(ltd != NULL);
1380 if (!list_empty(<d->ltd_namespace_list))
1383 list_add_tail(<d->ltd_namespace_list,
1384 &lad->lad_mdt_list);
1385 list_add_tail(<d->ltd_namespace_phase_list,
1386 &lad->lad_mdt_phase1_list);
1389 spin_unlock(<ds->ltd_lock);
1392 case LE_PHASE2_DONE:
1393 case LE_PEER_EXIT: {
1394 struct list_head *phase_head;
1396 /* Handle other MDTs firstly if needed, then notify the OSTs. */
1397 if (bk->lb_param & LPF_ALL_TGT) {
1398 phase_head = &lad->lad_mdt_list;
1399 ltds = &lfsck->li_mdt_descs;
1400 if (lr->lr_event == LE_STOP) {
1401 /* unlink other MDT targets locallly. */
1402 spin_lock(<ds->ltd_lock);
1403 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1404 list_for_each_entry_safe(ltd, next,
1405 phase_head, ltd_layout_list) {
1407 <d->ltd_layout_phase_list);
1409 <d->ltd_layout_list);
1412 list_for_each_entry_safe(ltd, next,
1414 ltd_namespace_list) {
1416 <d->ltd_namespace_phase_list);
1418 <d->ltd_namespace_list);
1421 spin_unlock(<ds->ltd_lock);
1423 if (com->lc_type != LFSCK_TYPE_LAYOUT)
1426 lr->lr_flags |= LEF_TO_OST;
1427 phase_head = &lad->lad_ost_list;
1428 ltds = &lfsck->li_ost_descs;
1430 lr->lr_flags &= ~LEF_TO_OST;
1432 } else if (com->lc_type != LFSCK_TYPE_LAYOUT) {
1435 lr->lr_flags |= LEF_TO_OST;
1436 phase_head = &lad->lad_ost_list;
1437 ltds = &lfsck->li_ost_descs;
1441 laia->laia_ltds = ltds;
1442 spin_lock(<ds->ltd_lock);
1443 while (!list_empty(phase_head)) {
1444 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1445 ltd = list_entry(phase_head->next,
1446 struct lfsck_tgt_desc,
1448 if (!list_empty(<d->ltd_layout_phase_list))
1450 <d->ltd_layout_phase_list);
1451 list_del_init(<d->ltd_layout_list);
1453 ltd = list_entry(phase_head->next,
1454 struct lfsck_tgt_desc,
1455 ltd_namespace_list);
1456 if (!list_empty(<d->ltd_namespace_phase_list))
1458 <d->ltd_namespace_phase_list);
1459 list_del_init(<d->ltd_namespace_list);
1461 atomic_inc(<d->ltd_ref);
1462 laia->laia_ltd = ltd;
1463 spin_unlock(<ds->ltd_lock);
1464 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1465 lfsck_async_interpret_common,
1466 laia, LFSCK_NOTIFY);
1468 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1469 "notify %s %x for %s stop/phase2_done/"
1470 "peer_exit: rc = %d\n",
1471 lfsck_lfsck2name(lfsck),
1472 (lr->lr_flags & LEF_TO_OST) ?
1473 "OST" : "MDT", ltd->ltd_index,
1477 spin_lock(<ds->ltd_lock);
1479 spin_unlock(<ds->ltd_lock);
1481 rc = ptlrpc_set_wait(set);
1483 ptlrpc_set_destroy(set);
1487 if (com->lc_type == LFSCK_TYPE_LAYOUT &&
1488 !(lr->lr_flags & LEF_TO_OST)) {
1489 lr->lr_flags |= LEF_TO_OST;
1490 phase_head = &lad->lad_ost_list;
1491 ltds = &lfsck->li_ost_descs;
1496 case LE_PHASE1_DONE:
1497 lad->lad_ops->la_sync_failures(env, com, lr);
1498 lad->lad_touch_gen++;
1499 ltds = &lfsck->li_mdt_descs;
1500 laia->laia_ltds = ltds;
1501 spin_lock(<ds->ltd_lock);
1502 while (!list_empty(&lad->lad_mdt_list)) {
1503 struct list_head *list;
1506 if (com->lc_type == LFSCK_TYPE_LAYOUT) {
1507 ltd = list_entry(lad->lad_mdt_list.next,
1508 struct lfsck_tgt_desc,
1510 list = <d->ltd_layout_list;
1511 gen = <d->ltd_layout_gen;
1513 struct lfsck_namespace *ns = com->lc_file_ram;
1515 ltd = list_entry(lad->lad_mdt_list.next,
1516 struct lfsck_tgt_desc,
1517 ltd_namespace_list);
1518 list = <d->ltd_namespace_list;
1519 gen = <d->ltd_namespace_gen;
1520 lr->lr_flags2 = ns->ln_flags & ~LF_INCOMPLETE;
1523 if (*gen == lad->lad_touch_gen)
1526 *gen = lad->lad_touch_gen;
1527 list_move_tail(list, &lad->lad_mdt_list);
1528 if (ltd->ltd_synced_failures)
1531 atomic_inc(<d->ltd_ref);
1532 laia->laia_ltd = ltd;
1533 spin_unlock(<ds->ltd_lock);
1534 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
1535 lfsck_async_interpret_common,
1536 laia, LFSCK_NOTIFY);
1538 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to "
1539 "notify MDT %x for %s phase1 done: "
1540 "rc = %d\n", lfsck_lfsck2name(lfsck),
1541 ltd->ltd_index, lad->lad_name, rc);
1544 spin_lock(<ds->ltd_lock);
1546 spin_unlock(<ds->ltd_lock);
1549 CDEBUG(D_LFSCK, "%s: LFSCK assistant unexpected LFSCK event: "
1550 "rc = %d\n", lfsck_lfsck2name(lfsck), lr->lr_event);
1555 rc1 = ptlrpc_set_wait(set);
1556 ptlrpc_set_destroy(set);
1558 RETURN(rc != 0 ? rc : rc1);
1562 * The LFSCK assistant thread is triggered by the LFSCK main engine.
1563 * They co-work together as an asynchronous pipeline: the LFSCK main
1564 * engine scans the system and pre-fetches the objects, attributes,
1565 * or name entries, etc, and pushes them into the pipeline as input
1566 * requests for the LFSCK assistant thread; on the other end of the
1567 * pipeline, the LFSCK assistant thread performs the real check and
1568 * repair for every request from the main engine.
1570 * Generally, the assistant engine may be blocked when check/repair
1571 * something, so the LFSCK main engine will run some faster. On the
1572 * other hand, the LFSCK main engine will drive multiple assistant
1573 * threads in parallel, means for each LFSCK component on the master
1574 * (such as layout LFSCK, namespace LFSCK), there is an independent
1575 * LFSCK assistant thread. So under such 1:N multiple asynchronous
1576 * pipelines mode, the whole LFSCK performance will be much better
1577 * than check/repair everything by the LFSCK main engine itself.
1579 int lfsck_assistant_engine(void *args)
1581 struct lfsck_thread_args *lta = args;
1582 struct lu_env *env = <a->lta_env;
1583 struct lfsck_component *com = lta->lta_com;
1584 struct lfsck_instance *lfsck = lta->lta_lfsck;
1585 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1586 struct lfsck_position *pos = &com->lc_pos_start;
1587 struct lfsck_thread_info *info = lfsck_env_info(env);
1588 struct lfsck_request *lr = &info->lti_lr;
1589 struct lfsck_assistant_data *lad = com->lc_data;
1590 struct ptlrpc_thread *mthread = &lfsck->li_thread;
1591 struct ptlrpc_thread *athread = &lad->lad_thread;
1592 struct lfsck_assistant_operations *lao = lad->lad_ops;
1593 struct lfsck_assistant_req *lar;
1594 struct l_wait_info lwi = { 0 };
1600 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread start\n",
1601 lfsck_lfsck2name(lfsck), lad->lad_name);
1603 memset(lr, 0, sizeof(*lr));
1604 lr->lr_event = LE_START;
1605 if (pos->lp_oit_cookie <= 1)
1606 lr->lr_param = LPF_RESET;
1607 rc = lfsck_assistant_notify_others(env, com, lr);
1609 CDEBUG(D_LFSCK, "%s: LFSCK assistant fail to notify others "
1610 "to start %s: rc = %d\n",
1611 lfsck_lfsck2name(lfsck), lad->lad_name, rc);
1615 spin_lock(&lad->lad_lock);
1616 thread_set_flags(athread, SVC_RUNNING);
1617 spin_unlock(&lad->lad_lock);
1618 wake_up_all(&mthread->t_ctl_waitq);
1621 while (!list_empty(&lad->lad_req_list)) {
1622 bool wakeup = false;
1624 if (unlikely(lad->lad_exit ||
1625 !thread_is_running(mthread)))
1626 GOTO(cleanup1, rc = lad->lad_post_result);
1628 lar = list_entry(lad->lad_req_list.next,
1629 struct lfsck_assistant_req,
1631 /* Only the lfsck_assistant_engine thread itself can
1632 * remove the "lar" from the head of the list, LFSCK
1633 * engine thread only inserts other new "lar" at the
1634 * end of the list. So it is safe to handle current
1635 * "lar" without the spin_lock. */
1636 rc = lao->la_handler_p1(env, com, lar);
1637 spin_lock(&lad->lad_lock);
1638 list_del_init(&lar->lar_list);
1639 lad->lad_prefetched--;
1640 /* Wake up the main engine thread only when the list
1641 * is empty or half of the prefetched items have been
1642 * handled to avoid too frequent thread schedule. */
1643 if (lad->lad_prefetched <= (bk->lb_async_windows / 2))
1645 spin_unlock(&lad->lad_lock);
1647 wake_up_all(&mthread->t_ctl_waitq);
1649 lao->la_req_fini(env, lar);
1650 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
1654 l_wait_event(athread->t_ctl_waitq,
1655 !lfsck_assistant_req_empty(lad) ||
1658 lad->lad_to_double_scan,
1661 if (unlikely(lad->lad_exit))
1662 GOTO(cleanup1, rc = lad->lad_post_result);
1664 if (!list_empty(&lad->lad_req_list))
1667 if (lad->lad_to_post) {
1668 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread post\n",
1669 lfsck_lfsck2name(lfsck), lad->lad_name);
1671 if (unlikely(lad->lad_exit))
1672 GOTO(cleanup1, rc = lad->lad_post_result);
1674 lad->lad_to_post = 0;
1675 LASSERT(lad->lad_post_result > 0);
1677 memset(lr, 0, sizeof(*lr));
1678 lr->lr_event = LE_PHASE1_DONE;
1679 lr->lr_status = lad->lad_post_result;
1680 rc = lfsck_assistant_notify_others(env, com, lr);
1682 CDEBUG(D_LFSCK, "%s: LFSCK assistant notified "
1683 "others for %s post: rc = %d\n",
1684 lfsck_lfsck2name(lfsck),
1687 /* Wakeup the master engine to go ahead. */
1688 wake_up_all(&mthread->t_ctl_waitq);
1691 if (lad->lad_to_double_scan) {
1692 lad->lad_to_double_scan = 0;
1693 atomic_inc(&lfsck->li_double_scan_count);
1694 lad->lad_in_double_scan = 1;
1695 wake_up_all(&mthread->t_ctl_waitq);
1697 com->lc_new_checked = 0;
1698 com->lc_new_scanned = 0;
1699 com->lc_time_last_checkpoint = cfs_time_current();
1700 com->lc_time_next_checkpoint =
1701 com->lc_time_last_checkpoint +
1702 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
1704 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before "
1705 "the second-stage scaning\n",
1706 lfsck_lfsck2name(lfsck));
1708 /* Flush async updates before handling orphan. */
1709 rc2 = dt_sync(env, lfsck->li_next);
1711 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 "
1712 "scan start, synced: rc = %d\n",
1713 lfsck_lfsck2name(lfsck), rc2);
1715 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_NO_DOUBLESCAN))
1716 GOTO(cleanup2, rc = 0);
1718 while (lad->lad_in_double_scan) {
1719 rc = lfsck_assistant_query_others(env, com);
1720 if (lfsck_phase2_next_ready(lad))
1726 /* Pull LFSCK status on related targets once
1727 * per 30 seconds if we are not notified. */
1728 lwi = LWI_TIMEOUT_INTERVAL(cfs_time_seconds(30),
1729 cfs_time_seconds(1),
1731 rc = l_wait_event(athread->t_ctl_waitq,
1732 lfsck_phase2_next_ready(lad) ||
1734 !thread_is_running(mthread),
1737 if (unlikely(lad->lad_exit ||
1738 !thread_is_running(mthread)))
1739 GOTO(cleanup2, rc = 0);
1741 if (rc == -ETIMEDOUT)
1748 rc = lao->la_handler_p2(env, com);
1752 if (unlikely(lad->lad_exit ||
1753 !thread_is_running(mthread)))
1754 GOTO(cleanup2, rc = 0);
1760 /* Cleanup the unfinished requests. */
1761 spin_lock(&lad->lad_lock);
1763 lad->lad_assistant_status = rc;
1765 if (lad->lad_exit && lad->lad_post_result <= 0)
1766 lao->la_fill_pos(env, com, &lfsck->li_pos_checkpoint);
1768 while (!list_empty(&lad->lad_req_list)) {
1769 lar = list_entry(lad->lad_req_list.next,
1770 struct lfsck_assistant_req,
1772 list_del_init(&lar->lar_list);
1773 lad->lad_prefetched--;
1774 spin_unlock(&lad->lad_lock);
1775 lao->la_req_fini(env, lar);
1776 spin_lock(&lad->lad_lock);
1778 spin_unlock(&lad->lad_lock);
1780 LASSERTF(lad->lad_prefetched == 0, "unmatched prefeteched objs %d\n",
1781 lad->lad_prefetched);
1784 memset(lr, 0, sizeof(*lr));
1786 lr->lr_event = LE_PHASE2_DONE;
1788 } else if (rc == 0) {
1789 if (lfsck->li_flags & LPF_ALL_TGT) {
1790 lr->lr_event = LE_STOP;
1791 lr->lr_status = LS_STOPPED;
1793 lr->lr_event = LE_PEER_EXIT;
1794 switch (lfsck->li_status) {
1797 lr->lr_status = LS_CO_PAUSED;
1801 lr->lr_status = LS_CO_STOPPED;
1804 CDEBUG(D_LFSCK, "%s: LFSCK assistant unknown "
1805 "status: rc = %d\n",
1806 lfsck_lfsck2name(lfsck),
1808 lr->lr_status = LS_CO_FAILED;
1813 if (lfsck->li_flags & LPF_ALL_TGT) {
1814 lr->lr_event = LE_STOP;
1815 lr->lr_status = LS_FAILED;
1817 lr->lr_event = LE_PEER_EXIT;
1818 lr->lr_status = LS_CO_FAILED;
1822 rc1 = lfsck_assistant_notify_others(env, com, lr);
1824 CDEBUG(D_LFSCK, "%s: LFSCK assistant failed to notify "
1825 "others for %s quit: rc = %d\n",
1826 lfsck_lfsck2name(lfsck), lad->lad_name, rc1);
1830 CDEBUG(D_LFSCK, "%s: LFSCK assistant sync before exit\n",
1831 lfsck_lfsck2name(lfsck));
1833 /* Flush async updates before exit. */
1834 rc2 = dt_sync(env, lfsck->li_next);
1836 CDEBUG(D_LFSCK, "%s: LFSCK assistant synced before exit: rc = %d\n",
1837 lfsck_lfsck2name(lfsck), rc2);
1839 /* Under force exit case, some requests may be just freed without
1840 * verification, those objects should be re-handled when next run.
1841 * So not update the on-disk trace file under such case. */
1842 if (lad->lad_in_double_scan) {
1844 rc1 = lao->la_double_scan_result(env, com, rc);
1846 CDEBUG(D_LFSCK, "%s: LFSCK assistant phase2 scan "
1847 "finished: rc = %d\n",
1848 lfsck_lfsck2name(lfsck), rc1 != 0 ? rc1 : rc);
1852 if (lad->lad_in_double_scan)
1853 atomic_dec(&lfsck->li_double_scan_count);
1855 spin_lock(&lad->lad_lock);
1856 lad->lad_assistant_status = (rc1 != 0 ? rc1 : rc);
1857 thread_set_flags(athread, SVC_STOPPED);
1858 wake_up_all(&mthread->t_ctl_waitq);
1859 spin_unlock(&lad->lad_lock);
1861 CDEBUG(D_LFSCK, "%s: %s LFSCK assistant thread exit: rc = %d\n",
1862 lfsck_lfsck2name(lfsck), lad->lad_name,
1863 lad->lad_assistant_status);
1865 lfsck_thread_args_fini(lta);