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_namespace.c
28 * Author: Fan, Yong <fan.yong@intel.com>
31 #define DEBUG_SUBSYSTEM S_LFSCK
33 #include <lustre/lustre_idl.h>
34 #include <lu_object.h>
35 #include <dt_object.h>
36 #include <md_object.h>
37 #include <lustre_fid.h>
38 #include <lustre_lib.h>
39 #include <lustre_net.h>
40 #include <lustre/lustre_user.h>
42 #include "lfsck_internal.h"
44 #define LFSCK_NAMESPACE_MAGIC 0xA0629D03
46 enum lfsck_nameentry_check {
47 LFSCK_NAMEENTRY_DEAD = 1, /* The object has been unlinked. */
48 LFSCK_NAMEENTRY_REMOVED = 2, /* The entry has been removed. */
49 LFSCK_NAMEENTRY_RECREATED = 3, /* The entry has been recreated. */
52 static const char lfsck_namespace_name[] = "lfsck_namespace";
54 static struct lfsck_namespace_req *
55 lfsck_namespace_assistant_req_init(struct lfsck_instance *lfsck,
56 struct lu_dirent *ent, __u16 type)
58 struct lfsck_namespace_req *lnr;
61 size = sizeof(*lnr) + (ent->lde_namelen & ~3) + 4;
64 return ERR_PTR(-ENOMEM);
66 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
67 lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
68 lnr->lnr_lmv = lfsck_lmv_get(lfsck->li_lmv);
69 lnr->lnr_fid = ent->lde_fid;
70 lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
71 lnr->lnr_dir_cookie = ent->lde_hash;
72 lnr->lnr_attr = ent->lde_attrs;
75 lnr->lnr_namelen = ent->lde_namelen;
76 memcpy(lnr->lnr_name, ent->lde_name, ent->lde_namelen);
81 static void lfsck_namespace_assistant_req_fini(const struct lu_env *env,
82 struct lfsck_assistant_req *lar)
84 struct lfsck_namespace_req *lnr =
85 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
87 if (lnr->lnr_lmv != NULL)
88 lfsck_lmv_put(env, lnr->lnr_lmv);
90 lu_object_put(env, &lnr->lnr_obj->do_lu);
91 OBD_FREE(lnr, lnr->lnr_size);
94 static void lfsck_namespace_le_to_cpu(struct lfsck_namespace *dst,
95 struct lfsck_namespace *src)
97 dst->ln_magic = le32_to_cpu(src->ln_magic);
98 dst->ln_status = le32_to_cpu(src->ln_status);
99 dst->ln_flags = le32_to_cpu(src->ln_flags);
100 dst->ln_success_count = le32_to_cpu(src->ln_success_count);
101 dst->ln_run_time_phase1 = le32_to_cpu(src->ln_run_time_phase1);
102 dst->ln_run_time_phase2 = le32_to_cpu(src->ln_run_time_phase2);
103 dst->ln_time_last_complete = le64_to_cpu(src->ln_time_last_complete);
104 dst->ln_time_latest_start = le64_to_cpu(src->ln_time_latest_start);
105 dst->ln_time_last_checkpoint =
106 le64_to_cpu(src->ln_time_last_checkpoint);
107 lfsck_position_le_to_cpu(&dst->ln_pos_latest_start,
108 &src->ln_pos_latest_start);
109 lfsck_position_le_to_cpu(&dst->ln_pos_last_checkpoint,
110 &src->ln_pos_last_checkpoint);
111 lfsck_position_le_to_cpu(&dst->ln_pos_first_inconsistent,
112 &src->ln_pos_first_inconsistent);
113 dst->ln_items_checked = le64_to_cpu(src->ln_items_checked);
114 dst->ln_items_repaired = le64_to_cpu(src->ln_items_repaired);
115 dst->ln_items_failed = le64_to_cpu(src->ln_items_failed);
116 dst->ln_dirs_checked = le64_to_cpu(src->ln_dirs_checked);
117 dst->ln_objs_checked_phase2 = le64_to_cpu(src->ln_objs_checked_phase2);
118 dst->ln_objs_repaired_phase2 =
119 le64_to_cpu(src->ln_objs_repaired_phase2);
120 dst->ln_objs_failed_phase2 = le64_to_cpu(src->ln_objs_failed_phase2);
121 dst->ln_objs_nlink_repaired = le64_to_cpu(src->ln_objs_nlink_repaired);
122 fid_le_to_cpu(&dst->ln_fid_latest_scanned_phase2,
123 &src->ln_fid_latest_scanned_phase2);
124 dst->ln_dirent_repaired = le64_to_cpu(src->ln_dirent_repaired);
125 dst->ln_linkea_repaired = le64_to_cpu(src->ln_linkea_repaired);
126 dst->ln_mul_linked_checked = le64_to_cpu(src->ln_mul_linked_checked);
127 dst->ln_mul_linked_repaired = le64_to_cpu(src->ln_mul_linked_repaired);
128 dst->ln_unknown_inconsistency =
129 le64_to_cpu(src->ln_unknown_inconsistency);
130 dst->ln_unmatched_pairs_repaired =
131 le64_to_cpu(src->ln_unmatched_pairs_repaired);
132 dst->ln_dangling_repaired = le64_to_cpu(src->ln_dangling_repaired);
133 dst->ln_mul_ref_repaired = le64_to_cpu(src->ln_mul_ref_repaired);
134 dst->ln_bad_type_repaired = le64_to_cpu(src->ln_bad_type_repaired);
135 dst->ln_lost_dirent_repaired =
136 le64_to_cpu(src->ln_lost_dirent_repaired);
137 dst->ln_striped_dirs_scanned =
138 le64_to_cpu(src->ln_striped_dirs_scanned);
139 dst->ln_striped_dirs_repaired =
140 le64_to_cpu(src->ln_striped_dirs_repaired);
141 dst->ln_striped_dirs_failed =
142 le64_to_cpu(src->ln_striped_dirs_failed);
143 dst->ln_striped_dirs_disabled =
144 le64_to_cpu(src->ln_striped_dirs_disabled);
145 dst->ln_striped_dirs_skipped =
146 le64_to_cpu(src->ln_striped_dirs_skipped);
147 dst->ln_striped_shards_scanned =
148 le64_to_cpu(src->ln_striped_shards_scanned);
149 dst->ln_striped_shards_repaired =
150 le64_to_cpu(src->ln_striped_shards_repaired);
151 dst->ln_striped_shards_failed =
152 le64_to_cpu(src->ln_striped_shards_failed);
153 dst->ln_striped_shards_skipped =
154 le64_to_cpu(src->ln_striped_shards_skipped);
155 dst->ln_name_hash_repaired = le64_to_cpu(src->ln_name_hash_repaired);
156 dst->ln_local_lpf_scanned = le64_to_cpu(src->ln_local_lpf_scanned);
157 dst->ln_local_lpf_moved = le64_to_cpu(src->ln_local_lpf_moved);
158 dst->ln_local_lpf_skipped = le64_to_cpu(src->ln_local_lpf_skipped);
159 dst->ln_local_lpf_failed = le64_to_cpu(src->ln_local_lpf_failed);
160 dst->ln_bitmap_size = le32_to_cpu(src->ln_bitmap_size);
163 static void lfsck_namespace_cpu_to_le(struct lfsck_namespace *dst,
164 struct lfsck_namespace *src)
166 dst->ln_magic = cpu_to_le32(src->ln_magic);
167 dst->ln_status = cpu_to_le32(src->ln_status);
168 dst->ln_flags = cpu_to_le32(src->ln_flags);
169 dst->ln_success_count = cpu_to_le32(src->ln_success_count);
170 dst->ln_run_time_phase1 = cpu_to_le32(src->ln_run_time_phase1);
171 dst->ln_run_time_phase2 = cpu_to_le32(src->ln_run_time_phase2);
172 dst->ln_time_last_complete = cpu_to_le64(src->ln_time_last_complete);
173 dst->ln_time_latest_start = cpu_to_le64(src->ln_time_latest_start);
174 dst->ln_time_last_checkpoint =
175 cpu_to_le64(src->ln_time_last_checkpoint);
176 lfsck_position_cpu_to_le(&dst->ln_pos_latest_start,
177 &src->ln_pos_latest_start);
178 lfsck_position_cpu_to_le(&dst->ln_pos_last_checkpoint,
179 &src->ln_pos_last_checkpoint);
180 lfsck_position_cpu_to_le(&dst->ln_pos_first_inconsistent,
181 &src->ln_pos_first_inconsistent);
182 dst->ln_items_checked = cpu_to_le64(src->ln_items_checked);
183 dst->ln_items_repaired = cpu_to_le64(src->ln_items_repaired);
184 dst->ln_items_failed = cpu_to_le64(src->ln_items_failed);
185 dst->ln_dirs_checked = cpu_to_le64(src->ln_dirs_checked);
186 dst->ln_objs_checked_phase2 = cpu_to_le64(src->ln_objs_checked_phase2);
187 dst->ln_objs_repaired_phase2 =
188 cpu_to_le64(src->ln_objs_repaired_phase2);
189 dst->ln_objs_failed_phase2 = cpu_to_le64(src->ln_objs_failed_phase2);
190 dst->ln_objs_nlink_repaired = cpu_to_le64(src->ln_objs_nlink_repaired);
191 fid_cpu_to_le(&dst->ln_fid_latest_scanned_phase2,
192 &src->ln_fid_latest_scanned_phase2);
193 dst->ln_dirent_repaired = cpu_to_le64(src->ln_dirent_repaired);
194 dst->ln_linkea_repaired = cpu_to_le64(src->ln_linkea_repaired);
195 dst->ln_mul_linked_checked = cpu_to_le64(src->ln_mul_linked_checked);
196 dst->ln_mul_linked_repaired = cpu_to_le64(src->ln_mul_linked_repaired);
197 dst->ln_unknown_inconsistency =
198 cpu_to_le64(src->ln_unknown_inconsistency);
199 dst->ln_unmatched_pairs_repaired =
200 cpu_to_le64(src->ln_unmatched_pairs_repaired);
201 dst->ln_dangling_repaired = cpu_to_le64(src->ln_dangling_repaired);
202 dst->ln_mul_ref_repaired = cpu_to_le64(src->ln_mul_ref_repaired);
203 dst->ln_bad_type_repaired = cpu_to_le64(src->ln_bad_type_repaired);
204 dst->ln_lost_dirent_repaired =
205 cpu_to_le64(src->ln_lost_dirent_repaired);
206 dst->ln_striped_dirs_scanned =
207 cpu_to_le64(src->ln_striped_dirs_scanned);
208 dst->ln_striped_dirs_repaired =
209 cpu_to_le64(src->ln_striped_dirs_repaired);
210 dst->ln_striped_dirs_failed =
211 cpu_to_le64(src->ln_striped_dirs_failed);
212 dst->ln_striped_dirs_disabled =
213 cpu_to_le64(src->ln_striped_dirs_disabled);
214 dst->ln_striped_dirs_skipped =
215 cpu_to_le64(src->ln_striped_dirs_skipped);
216 dst->ln_striped_shards_scanned =
217 cpu_to_le64(src->ln_striped_shards_scanned);
218 dst->ln_striped_shards_repaired =
219 cpu_to_le64(src->ln_striped_shards_repaired);
220 dst->ln_striped_shards_failed =
221 cpu_to_le64(src->ln_striped_shards_failed);
222 dst->ln_striped_shards_skipped =
223 cpu_to_le64(src->ln_striped_shards_skipped);
224 dst->ln_name_hash_repaired = cpu_to_le64(src->ln_name_hash_repaired);
225 dst->ln_local_lpf_scanned = cpu_to_le64(src->ln_local_lpf_scanned);
226 dst->ln_local_lpf_moved = cpu_to_le64(src->ln_local_lpf_moved);
227 dst->ln_local_lpf_skipped = cpu_to_le64(src->ln_local_lpf_skipped);
228 dst->ln_local_lpf_failed = cpu_to_le64(src->ln_local_lpf_failed);
229 dst->ln_bitmap_size = cpu_to_le32(src->ln_bitmap_size);
232 static void lfsck_namespace_record_failure(const struct lu_env *env,
233 struct lfsck_instance *lfsck,
234 struct lfsck_namespace *ns)
236 struct lfsck_position pos;
238 ns->ln_items_failed++;
239 lfsck_pos_fill(env, lfsck, &pos, false);
240 if (lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent) ||
241 lfsck_pos_is_eq(&pos, &ns->ln_pos_first_inconsistent) < 0) {
242 ns->ln_pos_first_inconsistent = pos;
244 CDEBUG(D_LFSCK, "%s: namespace LFSCK hit first non-repaired "
245 "inconsistency at the pos ["LPU64", "DFID", "LPX64"]\n",
246 lfsck_lfsck2name(lfsck),
247 ns->ln_pos_first_inconsistent.lp_oit_cookie,
248 PFID(&ns->ln_pos_first_inconsistent.lp_dir_parent),
249 ns->ln_pos_first_inconsistent.lp_dir_cookie);
254 * Load the MDT bitmap from the lfsck_namespace trace file.
256 * \param[in] env pointer to the thread context
257 * \param[in] com pointer to the lfsck component
259 * \retval 0 for success
260 * \retval negative error number on failure or data corruption
262 static int lfsck_namespace_load_bitmap(const struct lu_env *env,
263 struct lfsck_component *com)
265 struct dt_object *obj = com->lc_obj;
266 struct lfsck_assistant_data *lad = com->lc_data;
267 struct lfsck_namespace *ns = com->lc_file_ram;
268 cfs_bitmap_t *bitmap = lad->lad_bitmap;
274 if (com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size >
276 nbits = com->lc_lfsck->li_mdt_descs.ltd_tgts_bitmap->size;
278 nbits = ns->ln_bitmap_size;
280 if (unlikely(nbits < BITS_PER_LONG))
281 nbits = BITS_PER_LONG;
283 if (nbits > bitmap->size) {
284 __u32 new_bits = bitmap->size;
285 cfs_bitmap_t *new_bitmap;
287 while (new_bits < nbits)
290 new_bitmap = CFS_ALLOCATE_BITMAP(new_bits);
291 if (new_bitmap == NULL)
294 lad->lad_bitmap = new_bitmap;
295 CFS_FREE_BITMAP(bitmap);
299 if (ns->ln_bitmap_size == 0) {
300 lad->lad_incomplete = 0;
301 CFS_RESET_BITMAP(bitmap);
306 size = (ns->ln_bitmap_size + 7) >> 3;
307 rc = dt_xattr_get(env, obj,
308 lfsck_buf_get(env, bitmap->data, size),
309 XATTR_NAME_LFSCK_BITMAP, BYPASS_CAPA);
311 RETURN(rc >= 0 ? -EINVAL : rc);
313 if (cfs_bitmap_check_empty(bitmap))
314 lad->lad_incomplete = 0;
316 lad->lad_incomplete = 1;
322 * \retval +ve: the lfsck_namespace is broken, the caller should reset it.
323 * \retval 0: succeed.
324 * \retval -ve: failed cases.
326 static int lfsck_namespace_load(const struct lu_env *env,
327 struct lfsck_component *com)
329 int len = com->lc_file_size;
332 rc = dt_xattr_get(env, com->lc_obj,
333 lfsck_buf_get(env, com->lc_file_disk, len),
334 XATTR_NAME_LFSCK_NAMESPACE, BYPASS_CAPA);
336 struct lfsck_namespace *ns = com->lc_file_ram;
338 lfsck_namespace_le_to_cpu(ns,
339 (struct lfsck_namespace *)com->lc_file_disk);
340 if (ns->ln_magic != LFSCK_NAMESPACE_MAGIC) {
341 CDEBUG(D_LFSCK, "%s: invalid lfsck_namespace magic "
342 "%#x != %#x\n", lfsck_lfsck2name(com->lc_lfsck),
343 ns->ln_magic, LFSCK_NAMESPACE_MAGIC);
348 } else if (rc != -ENODATA) {
349 CDEBUG(D_LFSCK, "%s: fail to load lfsck_namespace, "
350 "expected = %d: rc = %d\n",
351 lfsck_lfsck2name(com->lc_lfsck), len, rc);
358 static int lfsck_namespace_store(const struct lu_env *env,
359 struct lfsck_component *com)
361 struct dt_object *obj = com->lc_obj;
362 struct lfsck_instance *lfsck = com->lc_lfsck;
363 struct lfsck_namespace *ns = com->lc_file_ram;
364 struct lfsck_assistant_data *lad = com->lc_data;
365 cfs_bitmap_t *bitmap = NULL;
366 struct thandle *handle;
368 int len = com->lc_file_size;
373 bitmap = lad->lad_bitmap;
374 nbits = bitmap->size;
377 LASSERTF((nbits & 7) == 0, "Invalid nbits %u\n", nbits);
380 ns->ln_bitmap_size = nbits;
381 lfsck_namespace_cpu_to_le((struct lfsck_namespace *)com->lc_file_disk,
383 handle = dt_trans_create(env, lfsck->li_bottom);
385 GOTO(log, rc = PTR_ERR(handle));
387 rc = dt_declare_xattr_set(env, obj,
388 lfsck_buf_get(env, com->lc_file_disk, len),
389 XATTR_NAME_LFSCK_NAMESPACE, 0, handle);
393 if (bitmap != NULL) {
394 rc = dt_declare_xattr_set(env, obj,
395 lfsck_buf_get(env, bitmap->data, nbits >> 3),
396 XATTR_NAME_LFSCK_BITMAP, 0, handle);
401 rc = dt_trans_start_local(env, lfsck->li_bottom, handle);
405 rc = dt_xattr_set(env, obj,
406 lfsck_buf_get(env, com->lc_file_disk, len),
407 XATTR_NAME_LFSCK_NAMESPACE, 0, handle, BYPASS_CAPA);
408 if (rc == 0 && bitmap != NULL)
409 rc = dt_xattr_set(env, obj,
410 lfsck_buf_get(env, bitmap->data, nbits >> 3),
411 XATTR_NAME_LFSCK_BITMAP, 0, handle,
417 dt_trans_stop(env, lfsck->li_bottom, handle);
421 CDEBUG(D_LFSCK, "%s: fail to store lfsck_namespace: rc = %d\n",
422 lfsck_lfsck2name(lfsck), rc);
426 static int lfsck_namespace_init(const struct lu_env *env,
427 struct lfsck_component *com)
429 struct lfsck_namespace *ns = com->lc_file_ram;
432 memset(ns, 0, sizeof(*ns));
433 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
434 ns->ln_status = LS_INIT;
435 down_write(&com->lc_sem);
436 rc = lfsck_namespace_store(env, com);
437 up_write(&com->lc_sem);
442 * Update the namespace LFSCK trace file for the given @fid
444 * \param[in] env pointer to the thread context
445 * \param[in] com pointer to the lfsck component
446 * \param[in] fid the fid which flags to be updated in the lfsck
448 * \param[in] add true if add new flags, otherwise remove flags
450 * \retval 0 for succeed or nothing to be done
451 * \retval negative error number on failure
453 int lfsck_namespace_trace_update(const struct lu_env *env,
454 struct lfsck_component *com,
455 const struct lu_fid *fid,
456 const __u8 flags, bool add)
458 struct lfsck_instance *lfsck = com->lc_lfsck;
459 struct dt_object *obj = com->lc_obj;
460 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
461 struct dt_device *dev = lfsck->li_bottom;
462 struct thandle *th = NULL;
470 down_write(&com->lc_sem);
471 fid_cpu_to_be(key, fid);
472 rc = dt_lookup(env, obj, (struct dt_rec *)&old,
473 (const struct dt_key *)key, BYPASS_CAPA);
476 GOTO(unlock, rc = 0);
480 } else if (rc == 0) {
482 if ((old & flags) == flags)
483 GOTO(unlock, rc = 0);
487 if ((old & flags) == 0)
488 GOTO(unlock, rc = 0);
496 th = dt_trans_create(env, dev);
498 GOTO(log, rc = PTR_ERR(th));
501 rc = dt_declare_delete(env, obj,
502 (const struct dt_key *)key, th);
508 rc = dt_declare_insert(env, obj,
509 (const struct dt_rec *)&new,
510 (const struct dt_key *)key, th);
515 rc = dt_trans_start_local(env, dev, th);
520 rc = dt_delete(env, obj, (const struct dt_key *)key,
527 rc = dt_insert(env, obj, (const struct dt_rec *)&new,
528 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
536 if (th != NULL && !IS_ERR(th))
537 dt_trans_stop(env, dev, th);
539 CDEBUG(D_LFSCK, "%s: namespace LFSCK %s flags for "DFID" in the "
540 "trace file, flags %x, old %x, new %x: rc = %d\n",
541 lfsck_lfsck2name(lfsck), add ? "add" : "del", PFID(fid),
542 (__u32)flags, (__u32)old, (__u32)new, rc);
545 up_write(&com->lc_sem);
550 int lfsck_namespace_check_exist(const struct lu_env *env,
551 struct dt_object *dir,
552 struct dt_object *obj, const char *name)
554 struct lu_fid *fid = &lfsck_env_info(env)->lti_fid;
558 if (unlikely(lfsck_is_dead_obj(obj)))
559 RETURN(LFSCK_NAMEENTRY_DEAD);
561 rc = dt_lookup(env, dir, (struct dt_rec *)fid,
562 (const struct dt_key *)name, BYPASS_CAPA);
564 RETURN(LFSCK_NAMEENTRY_REMOVED);
569 if (!lu_fid_eq(fid, lfsck_dto2fid(obj)))
570 RETURN(LFSCK_NAMEENTRY_RECREATED);
575 static int lfsck_declare_namespace_exec_dir(const struct lu_env *env,
576 struct dt_object *obj,
577 struct thandle *handle)
581 /* For destroying all invalid linkEA entries. */
582 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, handle);
586 /* For insert new linkEA entry. */
587 rc = dt_declare_xattr_set(env, obj,
588 lfsck_buf_get_const(env, NULL, DEFAULT_LINKEA_SIZE),
589 XATTR_NAME_LINK, 0, handle);
593 int __lfsck_links_read(const struct lu_env *env, struct dt_object *obj,
594 struct linkea_data *ldata)
598 if (ldata->ld_buf->lb_buf == NULL)
601 if (!dt_object_exists(obj))
604 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK, BYPASS_CAPA);
606 /* Buf was too small, figure out what we need. */
607 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LINK,
612 lu_buf_realloc(ldata->ld_buf, rc);
613 if (ldata->ld_buf->lb_buf == NULL)
616 rc = dt_xattr_get(env, obj, ldata->ld_buf, XATTR_NAME_LINK,
621 rc = linkea_init(ldata);
627 * Remove linkEA for the given object.
629 * The caller should take the ldlm lock before the calling.
631 * \param[in] env pointer to the thread context
632 * \param[in] com pointer to the lfsck component
633 * \param[in] obj pointer to the dt_object to be handled
635 * \retval 0 for repaired cases
636 * \retval negative error number on failure
638 static int lfsck_namespace_links_remove(const struct lu_env *env,
639 struct lfsck_component *com,
640 struct dt_object *obj)
642 struct lfsck_instance *lfsck = com->lc_lfsck;
643 struct dt_device *dev = lfsck->li_bottom;
644 struct thandle *th = NULL;
648 LASSERT(dt_object_remote(obj) == 0);
650 th = dt_trans_create(env, dev);
652 GOTO(log, rc = PTR_ERR(th));
654 rc = dt_declare_xattr_del(env, obj, XATTR_NAME_LINK, th);
658 rc = dt_trans_start_local(env, dev, th);
662 dt_write_lock(env, obj, 0);
663 if (unlikely(lfsck_is_dead_obj(obj)))
664 GOTO(unlock, rc = -ENOENT);
666 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
667 GOTO(unlock, rc = 0);
669 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, th, BYPASS_CAPA);
674 dt_write_unlock(env, obj);
677 dt_trans_stop(env, dev, th);
680 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove invalid linkEA "
681 "for the object "DFID": rc = %d\n",
682 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
685 struct lfsck_namespace *ns = com->lc_file_ram;
687 ns->ln_flags |= LF_INCONSISTENT;
693 static int lfsck_links_write(const struct lu_env *env, struct dt_object *obj,
694 struct linkea_data *ldata, struct thandle *handle)
696 const struct lu_buf *buf = lfsck_buf_get_const(env,
697 ldata->ld_buf->lb_buf,
698 ldata->ld_leh->leh_len);
700 return dt_xattr_set(env, obj, buf, XATTR_NAME_LINK, 0, handle,
704 static void lfsck_namespace_unpack_linkea_entry(struct linkea_data *ldata,
705 struct lu_name *cname,
709 linkea_entry_unpack(ldata->ld_lee, &ldata->ld_reclen, cname, pfid);
710 /* To guarantee the 'name' is terminated with '0'. */
711 memcpy(buf, cname->ln_name, cname->ln_namelen);
712 buf[cname->ln_namelen] = 0;
713 cname->ln_name = buf;
716 static int lfsck_namespace_filter_linkea_entry(struct linkea_data *ldata,
717 struct lu_name *cname,
721 struct link_ea_entry *oldlee;
725 oldlee = ldata->ld_lee;
726 oldlen = ldata->ld_reclen;
727 linkea_next_entry(ldata);
728 while (ldata->ld_lee != NULL) {
729 ldata->ld_reclen = (ldata->ld_lee->lee_reclen[0] << 8) |
730 ldata->ld_lee->lee_reclen[1];
731 if (unlikely(ldata->ld_reclen == oldlen &&
732 memcmp(ldata->ld_lee, oldlee, oldlen) == 0)) {
737 linkea_del_buf(ldata, cname);
739 linkea_next_entry(ldata);
742 ldata->ld_lee = oldlee;
743 ldata->ld_reclen = oldlen;
749 * Insert orphan into .lustre/lost+found/MDTxxxx/ locally.
751 * Add the specified orphan MDT-object to the .lustre/lost+found/MDTxxxx/
752 * with the given type to generate the name, the detailed rules for name
753 * have been described as following.
755 * The function also generates the linkEA corresponding to the name entry
756 * under the .lustre/lost+found/MDTxxxx/ for the orphan MDT-object.
758 * \param[in] env pointer to the thread context
759 * \param[in] com pointer to the lfsck component
760 * \param[in] orphan pointer to the orphan MDT-object
761 * \param[in] infix additional information for the orphan name, such as
762 * the FID for original
763 * \param[in] type the type for describing why the orphan MDT-object is
764 * created. The rules are as following:
766 * type "D": The MDT-object is a directory, it may knows its parent
767 * but because there is no valid linkEA, the LFSCK cannot
768 * know where to put it back to the namespace.
769 * type "O": The MDT-object has no linkEA, and there is no name
770 * entry that references the MDT-object.
772 * type "S": The orphan MDT-object is a shard of a striped directory
774 * \see lfsck_layout_recreate_parent() for more types.
776 * The orphan name will be like:
777 * ${FID}-${infix}-${type}-${conflict_version}
779 * \param[out] count if some others inserted some linkEA entries by race,
780 * then return the linkEA entries count.
782 * \retval positive number for repaired cases
783 * \retval 0 if needs to repair nothing
784 * \retval negative error number on failure
786 static int lfsck_namespace_insert_orphan(const struct lu_env *env,
787 struct lfsck_component *com,
788 struct dt_object *orphan,
789 const char *infix, const char *type,
792 struct lfsck_thread_info *info = lfsck_env_info(env);
793 struct lu_name *cname = &info->lti_name;
794 struct dt_insert_rec *rec = &info->lti_dt_rec;
795 struct lu_fid *tfid = &info->lti_fid5;
796 struct lu_attr *la = &info->lti_la3;
797 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
798 const struct lu_fid *pfid;
799 struct lfsck_instance *lfsck = com->lc_lfsck;
800 struct dt_device *dev = lfsck->li_bottom;
801 struct dt_object *parent;
802 struct thandle *th = NULL;
803 struct lustre_handle plh = { 0 };
804 struct lustre_handle clh = { 0 };
805 struct linkea_data ldata = { 0 };
806 struct lu_buf linkea_buf;
813 cname->ln_name = NULL;
814 /* Create .lustre/lost+found/MDTxxxx when needed. */
815 if (unlikely(lfsck->li_lpf_obj == NULL)) {
816 rc = lfsck_create_lpf(env, lfsck);
821 parent = lfsck->li_lpf_obj;
822 pfid = lfsck_dto2fid(parent);
824 /* Hold update lock on the parent to prevent others to access. */
825 rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
826 MDS_INODELOCK_UPDATE, LCK_EX);
831 namelen = snprintf(info->lti_key, NAME_MAX, DFID"%s-%s-%d",
832 PFID(cfid), infix, type, idx++);
833 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
834 (const struct dt_key *)info->lti_key,
836 if (rc != 0 && rc != -ENOENT)
839 if (unlikely(rc == 0 && lu_fid_eq(cfid, tfid)))
841 } while (rc == 0 && !exist);
843 cname->ln_name = info->lti_key;
844 cname->ln_namelen = namelen;
845 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
849 rc = linkea_add_buf(&ldata, cname, pfid);
853 rc = lfsck_ibits_lock(env, lfsck, orphan, &clh,
854 MDS_INODELOCK_UPDATE | MDS_INODELOCK_LOOKUP,
859 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
860 ldata.ld_leh->leh_len);
861 th = dt_trans_create(env, dev);
863 GOTO(log, rc = PTR_ERR(th));
865 if (S_ISDIR(lfsck_object_type(orphan))) {
866 rc = dt_declare_delete(env, orphan,
867 (const struct dt_key *)dotdot, th);
871 rec->rec_type = S_IFDIR;
873 rc = dt_declare_insert(env, orphan, (const struct dt_rec *)rec,
874 (const struct dt_key *)dotdot, th);
879 rc = dt_declare_xattr_set(env, orphan, &linkea_buf,
880 XATTR_NAME_LINK, 0, th);
885 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
887 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
888 (const struct dt_key *)cname->ln_name,
893 if (S_ISDIR(rec->rec_type)) {
894 rc = dt_declare_ref_add(env, parent, th);
900 memset(la, 0, sizeof(*la));
901 la->la_ctime = cfs_time_current_sec();
902 la->la_valid = LA_CTIME;
903 rc = dt_declare_attr_set(env, orphan, la, th);
907 rc = dt_trans_start_local(env, dev, th);
911 dt_write_lock(env, orphan, 0);
912 rc = lfsck_links_read(env, orphan, &ldata);
913 if (likely((rc == -ENODATA) || (rc == -EINVAL) ||
914 (rc == 0 && ldata.ld_leh->leh_reccount == 0))) {
915 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
916 GOTO(unlock, rc = 1);
918 if (S_ISDIR(lfsck_object_type(orphan))) {
919 rc = dt_delete(env, orphan,
920 (const struct dt_key *)dotdot, th,
925 rec->rec_type = S_IFDIR;
927 rc = dt_insert(env, orphan, (const struct dt_rec *)rec,
928 (const struct dt_key *)dotdot, th,
934 rc = dt_xattr_set(env, orphan, &linkea_buf, XATTR_NAME_LINK, 0,
937 if (rc == 0 && count != NULL)
938 *count = ldata.ld_leh->leh_reccount;
942 dt_write_unlock(env, orphan);
944 if (rc == 0 && !exist) {
945 rec->rec_type = lfsck_object_type(orphan) & S_IFMT;
947 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
948 (const struct dt_key *)cname->ln_name,
950 if (rc == 0 && S_ISDIR(rec->rec_type)) {
951 dt_write_lock(env, parent, 0);
952 rc = dt_ref_add(env, parent, th);
953 dt_write_unlock(env, parent);
960 rc = dt_attr_set(env, orphan, la, th, BYPASS_CAPA);
962 GOTO(stop, rc = (rc == 0 ? 1 : rc));
965 dt_write_unlock(env, orphan);
968 dt_trans_stop(env, dev, th);
971 lfsck_ibits_unlock(&clh, LCK_EX);
972 lfsck_ibits_unlock(&plh, LCK_EX);
973 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert orphan for the "
974 "object "DFID", name = %s: rc = %d\n",
975 lfsck_lfsck2name(lfsck), PFID(cfid),
976 cname->ln_name != NULL ? cname->ln_name : "<NULL>", rc);
979 struct lfsck_namespace *ns = com->lc_file_ram;
981 ns->ln_flags |= LF_INCONSISTENT;
988 * Add the specified name entry back to namespace.
990 * If there is a linkEA entry that back references a name entry under
991 * some parent directory, but such parent directory does not have the
992 * claimed name entry. On the other hand, the linkEA entries count is
993 * not larger than the MDT-object's hard link count. Under such case,
994 * it is quite possible that the name entry is lost. Then the LFSCK
995 * should add the name entry back to the namespace.
997 * \param[in] env pointer to the thread context
998 * \param[in] com pointer to the lfsck component
999 * \param[in] parent pointer to the directory under which the name entry
1000 * will be inserted into
1001 * \param[in] child pointer to the object referenced by the name entry
1002 * that to be inserted into the parent
1003 * \param[in] name the name for the child in the parent directory
1005 * \retval positive number for repaired cases
1006 * \retval 0 if nothing to be repaired
1007 * \retval negative error number on failure
1009 static int lfsck_namespace_insert_normal(const struct lu_env *env,
1010 struct lfsck_component *com,
1011 struct dt_object *parent,
1012 struct dt_object *child,
1015 struct lfsck_thread_info *info = lfsck_env_info(env);
1016 struct lu_attr *la = &info->lti_la;
1017 struct dt_insert_rec *rec = &info->lti_dt_rec;
1018 struct lfsck_instance *lfsck = com->lc_lfsck;
1019 struct dt_device *dev = lfsck->li_next;
1020 struct thandle *th = NULL;
1021 struct lustre_handle lh = { 0 };
1025 if (unlikely(!dt_try_as_dir(env, parent)))
1026 GOTO(log, rc = -ENOTDIR);
1028 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1031 /* Hold update lock on the parent to prevent others to access. */
1032 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1033 MDS_INODELOCK_UPDATE, LCK_EX);
1037 th = dt_trans_create(env, dev);
1039 GOTO(unlock, rc = PTR_ERR(th));
1041 rec->rec_type = lfsck_object_type(child) & S_IFMT;
1042 rec->rec_fid = lfsck_dto2fid(child);
1043 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1044 (const struct dt_key *)name, th);
1048 if (S_ISDIR(rec->rec_type)) {
1049 rc = dt_declare_ref_add(env, parent, th);
1054 memset(la, 0, sizeof(*la));
1055 la->la_ctime = cfs_time_current_sec();
1056 la->la_valid = LA_CTIME;
1057 rc = dt_declare_attr_set(env, parent, la, th);
1061 rc = dt_declare_attr_set(env, child, la, th);
1065 rc = dt_trans_start_local(env, dev, th);
1069 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1070 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1074 if (S_ISDIR(rec->rec_type)) {
1075 dt_write_lock(env, parent, 0);
1076 rc = dt_ref_add(env, parent, th);
1077 dt_write_unlock(env, parent);
1082 la->la_ctime = cfs_time_current_sec();
1083 rc = dt_attr_set(env, parent, la, th, BYPASS_CAPA);
1087 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
1089 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1092 dt_trans_stop(env, dev, th);
1095 lfsck_ibits_unlock(&lh, LCK_EX);
1098 CDEBUG(D_LFSCK, "%s: namespace LFSCK insert object "DFID" with "
1099 "the name %s and type %o to the parent "DFID": rc = %d\n",
1100 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(child)), name,
1101 lfsck_object_type(child) & S_IFMT,
1102 PFID(lfsck_dto2fid(parent)), rc);
1105 struct lfsck_namespace *ns = com->lc_file_ram;
1107 ns->ln_flags |= LF_INCONSISTENT;
1109 ns->ln_lost_dirent_repaired++;
1116 * Create the specified orphan MDT-object on remote MDT.
1118 * The LFSCK instance on this MDT will send LFSCK RPC to remote MDT to
1119 * ask the remote LFSCK instance to create the specified orphan object
1120 * under .lustre/lost+found/MDTxxxx/ directory with the name:
1121 * ${FID}-P-${conflict_version}.
1123 * \param[in] env pointer to the thread context
1124 * \param[in] com pointer to the lfsck component
1125 * \param[in] orphan pointer to the orphan MDT-object
1126 * \param[in] type the orphan's type to be created
1128 * type "P": The orphan object to be created was a parent directory
1129 * of some MDT-object which linkEA shows that the @orphan
1130 * object is missing.
1132 * \see lfsck_layout_recreate_parent() for more types.
1134 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1136 * \retval positive number for repaired cases
1137 * \retval 0 if needs to repair nothing
1138 * \retval negative error number on failure
1140 static int lfsck_namespace_create_orphan_remote(const struct lu_env *env,
1141 struct lfsck_component *com,
1142 struct dt_object *orphan,
1144 struct lmv_mds_md_v1 *lmv)
1146 struct lfsck_thread_info *info = lfsck_env_info(env);
1147 struct lfsck_request *lr = &info->lti_lr;
1148 struct lu_seq_range *range = &info->lti_range;
1149 const struct lu_fid *fid = lfsck_dto2fid(orphan);
1150 struct lfsck_namespace *ns = com->lc_file_ram;
1151 struct lfsck_instance *lfsck = com->lc_lfsck;
1152 struct seq_server_site *ss =
1153 lu_site2seq(lfsck->li_bottom->dd_lu_dev.ld_site);
1154 struct lfsck_tgt_desc *ltd = NULL;
1155 struct ptlrpc_request *req = NULL;
1159 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1162 fld_range_set_mdt(range);
1163 rc = fld_server_lookup(env, ss->ss_server_fld, fid_seq(fid), range);
1167 ltd = lfsck_tgt_get(&lfsck->li_mdt_descs, range->lsr_index);
1169 ns->ln_flags |= LF_INCOMPLETE;
1171 GOTO(out, rc = -ENODEV);
1174 req = ptlrpc_request_alloc(class_exp2cliimp(ltd->ltd_exp),
1177 GOTO(out, rc = -ENOMEM);
1179 rc = ptlrpc_request_pack(req, LUSTRE_OBD_VERSION, LFSCK_NOTIFY);
1181 ptlrpc_request_free(req);
1186 lr = req_capsule_client_get(&req->rq_pill, &RMF_LFSCK_REQUEST);
1187 memset(lr, 0, sizeof(*lr));
1188 lr->lr_event = LE_CREATE_ORPHAN;
1189 lr->lr_index = lfsck_dev_idx(lfsck->li_bottom);
1190 lr->lr_active = LFSCK_TYPE_NAMESPACE;
1194 lr->lr_hash_type = lmv->lmv_hash_type;
1195 lr->lr_stripe_count = lmv->lmv_stripe_count;
1196 lr->lr_layout_version = lmv->lmv_layout_version;
1197 memcpy(lr->lr_pool_name, lmv->lmv_pool_name,
1198 sizeof(lr->lr_pool_name));
1201 ptlrpc_request_set_replen(req);
1202 rc = ptlrpc_queue_wait(req);
1203 ptlrpc_req_finished(req);
1207 else if (rc == -EEXIST)
1213 CDEBUG(D_LFSCK, "%s: namespace LFSCK create object "
1214 DFID" on the MDT %x remotely: rc = %d\n",
1215 lfsck_lfsck2name(lfsck), PFID(fid),
1216 ltd != NULL ? ltd->ltd_index : -1, rc);
1225 * Create the specified orphan MDT-object locally.
1227 * For the case that the parent MDT-object stored in some MDT-object's
1228 * linkEA entry is lost, the LFSCK will re-create the parent object as
1229 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1230 * with the name ${FID}-P-${conflict_version}.
1232 * \param[in] env pointer to the thread context
1233 * \param[in] com pointer to the lfsck component
1234 * \param[in] orphan pointer to the orphan MDT-object to be created
1235 * \param[in] type the orphan's type to be created
1237 * type "P": The orphan object to be created was a parent directory
1238 * of some MDT-object which linkEA shows that the @orphan
1239 * object is missing.
1241 * \see lfsck_layout_recreate_parent() for more types.
1243 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1245 * \retval positive number for repaired cases
1246 * \retval negative error number on failure
1248 static int lfsck_namespace_create_orphan_local(const struct lu_env *env,
1249 struct lfsck_component *com,
1250 struct dt_object *orphan,
1252 struct lmv_mds_md_v1 *lmv)
1254 struct lfsck_thread_info *info = lfsck_env_info(env);
1255 struct lu_attr *la = &info->lti_la;
1256 struct dt_allocation_hint *hint = &info->lti_hint;
1257 struct dt_object_format *dof = &info->lti_dof;
1258 struct lu_name *cname = &info->lti_name2;
1259 struct dt_insert_rec *rec = &info->lti_dt_rec;
1260 struct lu_fid *tfid = &info->lti_fid;
1261 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
1262 const struct lu_fid *cfid = lfsck_dto2fid(orphan);
1263 const struct lu_fid *pfid;
1264 struct lfsck_instance *lfsck = com->lc_lfsck;
1265 struct dt_device *dev = lfsck->li_bottom;
1266 struct dt_object *parent = NULL;
1267 struct dt_object *child = NULL;
1268 struct thandle *th = NULL;
1269 struct lustre_handle lh = { 0 };
1270 struct linkea_data ldata = { 0 };
1271 struct lu_buf linkea_buf;
1272 struct lu_buf lmv_buf;
1279 LASSERT(!dt_object_exists(orphan));
1280 LASSERT(!dt_object_remote(orphan));
1282 /* @orphan maybe not attached to lfsck->li_bottom */
1283 child = lfsck_object_find_by_dev(env, dev, cfid);
1285 GOTO(log, rc = PTR_ERR(child));
1287 cname->ln_name = NULL;
1288 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1291 /* Create .lustre/lost+found/MDTxxxx when needed. */
1292 if (unlikely(lfsck->li_lpf_obj == NULL)) {
1293 rc = lfsck_create_lpf(env, lfsck);
1298 parent = lfsck->li_lpf_obj;
1299 pfid = lfsck_dto2fid(parent);
1301 /* Hold update lock on the parent to prevent others to access. */
1302 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
1303 MDS_INODELOCK_UPDATE, LCK_EX);
1308 namelen = snprintf(name, 31, DFID"-P-%d",
1310 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1311 (const struct dt_key *)name, BYPASS_CAPA);
1312 if (rc != 0 && rc != -ENOENT)
1316 cname->ln_name = name;
1317 cname->ln_namelen = namelen;
1319 memset(la, 0, sizeof(*la));
1320 la->la_mode = type | (S_ISDIR(type) ? 0700 : 0600);
1321 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
1322 LA_ATIME | LA_MTIME | LA_CTIME;
1324 child->do_ops->do_ah_init(env, hint, parent, child,
1325 la->la_mode & S_IFMT);
1327 memset(dof, 0, sizeof(*dof));
1328 dof->dof_type = dt_mode_to_dft(type);
1330 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
1334 rc = linkea_add_buf(&ldata, cname, pfid);
1338 th = dt_trans_create(env, dev);
1340 GOTO(unlock1, rc = PTR_ERR(th));
1342 rc = dt_declare_create(env, child, la, hint, dof, th);
1343 if (rc == 0 && S_ISDIR(type))
1344 rc = dt_declare_ref_add(env, child, th);
1350 lmv->lmv_magic = LMV_MAGIC;
1351 lmv->lmv_master_mdt_index = lfsck_dev_idx(dev);
1352 lfsck_lmv_header_cpu_to_le(lmv2, lmv);
1353 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
1354 rc = dt_declare_xattr_set(env, child, &lmv_buf,
1355 XATTR_NAME_LMV, 0, th);
1360 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
1361 ldata.ld_leh->leh_len);
1362 rc = dt_declare_xattr_set(env, child, &linkea_buf,
1363 XATTR_NAME_LINK, 0, th);
1367 rec->rec_type = type;
1368 rec->rec_fid = cfid;
1369 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1370 (const struct dt_key *)name, th);
1371 if (rc == 0 && S_ISDIR(type))
1372 rc = dt_declare_ref_add(env, parent, th);
1377 rc = dt_trans_start_local(env, dev, th);
1381 dt_write_lock(env, child, 0);
1382 rc = dt_create(env, child, la, hint, dof, th);
1386 if (S_ISDIR(type)) {
1387 if (unlikely(!dt_try_as_dir(env, child)))
1388 GOTO(unlock2, rc = -ENOTDIR);
1390 rec->rec_type = S_IFDIR;
1391 rec->rec_fid = cfid;
1392 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1393 (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
1397 rec->rec_fid = pfid;
1398 rc = dt_insert(env, child, (const struct dt_rec *)rec,
1399 (const struct dt_key *)dotdot, th,
1404 rc = dt_ref_add(env, child, th);
1410 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV, 0,
1416 rc = dt_xattr_set(env, child, &linkea_buf,
1417 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1418 dt_write_unlock(env, child);
1422 rec->rec_type = type;
1423 rec->rec_fid = cfid;
1424 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1425 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1426 if (rc == 0 && S_ISDIR(type)) {
1427 dt_write_lock(env, parent, 0);
1428 rc = dt_ref_add(env, parent, th);
1429 dt_write_unlock(env, parent);
1432 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1435 dt_write_unlock(env, child);
1438 dt_trans_stop(env, dev, th);
1441 lfsck_ibits_unlock(&lh, LCK_EX);
1444 CDEBUG(D_LFSCK, "%s: namespace LFSCK create orphan locally for "
1445 "the object "DFID", name = %s, type %o: rc = %d\n",
1446 lfsck_lfsck2name(lfsck), PFID(cfid),
1447 cname->ln_name != NULL ? cname->ln_name : "<NULL>", type, rc);
1449 if (child != NULL && !IS_ERR(child))
1450 lfsck_object_put(env, child);
1456 * Create the specified orphan MDT-object.
1458 * For the case that the parent MDT-object stored in some MDT-object's
1459 * linkEA entry is lost, the LFSCK will re-create the parent object as
1460 * an orphan and insert it into .lustre/lost+found/MDTxxxx/ directory
1461 * with the name: ${FID}-P-${conflict_version}.
1463 * \param[in] env pointer to the thread context
1464 * \param[in] com pointer to the lfsck component
1465 * \param[in] orphan pointer to the orphan MDT-object
1467 * type "P": The orphan object to be created was a parent directory
1468 * of some MDT-object which linkEA shows that the @orphan
1469 * object is missing.
1471 * \see lfsck_layout_recreate_parent() for more types.
1473 * \param[in] lmv pointer to master LMV EA that will be set to the orphan
1475 * \retval positive number for repaired cases
1476 * \retval 0 if needs to repair nothing
1477 * \retval negative error number on failure
1479 static int lfsck_namespace_create_orphan(const struct lu_env *env,
1480 struct lfsck_component *com,
1481 struct dt_object *orphan,
1482 struct lmv_mds_md_v1 *lmv)
1484 struct lfsck_namespace *ns = com->lc_file_ram;
1487 if (dt_object_remote(orphan))
1488 rc = lfsck_namespace_create_orphan_remote(env, com, orphan,
1491 rc = lfsck_namespace_create_orphan_local(env, com, orphan,
1495 ns->ln_flags |= LF_INCONSISTENT;
1501 * Remove the specified entry from the linkEA.
1503 * Locate the linkEA entry with the given @cname and @pfid, then
1504 * remove this entry or the other entries those are repeated with
1507 * \param[in] env pointer to the thread context
1508 * \param[in] com pointer to the lfsck component
1509 * \param[in] obj pointer to the dt_object to be handled
1510 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1511 * \param[in] cname the name for the child in the parent directory
1512 * \param[in] pfid the parent directory's FID for the linkEA
1513 * \param[in] next if true, then remove the first found linkEA
1514 * entry, and move the ldata->ld_lee to next entry
1516 * \retval positive number for repaired cases
1517 * \retval 0 if nothing to be repaired
1518 * \retval negative error number on failure
1520 static int lfsck_namespace_shrink_linkea(const struct lu_env *env,
1521 struct lfsck_component *com,
1522 struct dt_object *obj,
1523 struct linkea_data *ldata,
1524 struct lu_name *cname,
1525 struct lu_fid *pfid,
1528 struct lfsck_instance *lfsck = com->lc_lfsck;
1529 struct dt_device *dev = lfsck->li_bottom;
1530 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
1531 struct thandle *th = NULL;
1532 struct lustre_handle lh = { 0 };
1533 struct linkea_data ldata_new = { 0 };
1534 struct lu_buf linkea_buf;
1538 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
1539 MDS_INODELOCK_UPDATE |
1540 MDS_INODELOCK_XATTR, LCK_EX);
1545 linkea_del_buf(ldata, cname);
1547 lfsck_namespace_filter_linkea_entry(ldata, cname, pfid,
1549 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1550 ldata->ld_leh->leh_len);
1553 th = dt_trans_create(env, dev);
1555 GOTO(unlock1, rc = PTR_ERR(th));
1557 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1558 XATTR_NAME_LINK, 0, th);
1562 rc = dt_trans_start_local(env, dev, th);
1566 dt_write_lock(env, obj, 0);
1567 if (unlikely(lfsck_is_dead_obj(obj)))
1568 GOTO(unlock2, rc = -ENOENT);
1570 rc = lfsck_links_read2(env, obj, &ldata_new);
1574 /* The specified linkEA entry has been removed by race. */
1575 rc = linkea_links_find(&ldata_new, cname, pfid);
1577 GOTO(unlock2, rc = 0);
1579 if (bk->lb_param & LPF_DRYRUN)
1580 GOTO(unlock2, rc = 1);
1583 linkea_del_buf(&ldata_new, cname);
1585 lfsck_namespace_filter_linkea_entry(&ldata_new, cname, pfid,
1588 if (linkea_buf.lb_len < ldata_new.ld_leh->leh_len) {
1589 dt_write_unlock(env, obj);
1590 dt_trans_stop(env, dev, th);
1591 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1592 ldata_new.ld_leh->leh_len);
1596 lfsck_buf_init(&linkea_buf, ldata_new.ld_buf->lb_buf,
1597 ldata_new.ld_leh->leh_len);
1598 rc = dt_xattr_set(env, obj, &linkea_buf,
1599 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1601 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
1604 dt_write_unlock(env, obj);
1607 dt_trans_stop(env, dev, th);
1610 lfsck_ibits_unlock(&lh, LCK_EX);
1613 CDEBUG(D_LFSCK, "%s: namespace LFSCK remove %s linkEA entry "
1614 "for the object: "DFID", parent "DFID", name %.*s\n",
1615 lfsck_lfsck2name(lfsck), next ? "invalid" : "redundant",
1616 PFID(lfsck_dto2fid(obj)), PFID(pfid), cname->ln_namelen,
1620 struct lfsck_namespace *ns = com->lc_file_ram;
1622 ns->ln_flags |= LF_INCONSISTENT;
1629 * Conditionally remove the specified entry from the linkEA.
1631 * Take the parent lock firstly, then check whether the specified
1632 * name entry exists or not: if yes, do nothing; otherwise, call
1633 * lfsck_namespace_shrink_linkea() to remove the linkea entry.
1635 * \param[in] env pointer to the thread context
1636 * \param[in] com pointer to the lfsck component
1637 * \param[in] parent pointer to the parent directory
1638 * \param[in] child pointer to the child object that holds the linkEA
1639 * \param[in,out]ldata pointer to the buffer that holds the linkEA
1640 * \param[in] cname the name for the child in the parent directory
1641 * \param[in] pfid the parent directory's FID for the linkEA
1643 * \retval positive number for repaired cases
1644 * \retval 0 if nothing to be repaired
1645 * \retval negative error number on failure
1647 static int lfsck_namespace_shrink_linkea_cond(const struct lu_env *env,
1648 struct lfsck_component *com,
1649 struct dt_object *parent,
1650 struct dt_object *child,
1651 struct linkea_data *ldata,
1652 struct lu_name *cname,
1653 struct lu_fid *pfid)
1655 struct lu_fid *cfid = &lfsck_env_info(env)->lti_fid3;
1656 struct lustre_handle lh = { 0 };
1660 rc = lfsck_ibits_lock(env, com->lc_lfsck, parent, &lh,
1661 MDS_INODELOCK_UPDATE, LCK_EX);
1665 dt_read_lock(env, parent, 0);
1666 if (unlikely(lfsck_is_dead_obj(parent))) {
1667 dt_read_unlock(env, parent);
1668 lfsck_ibits_unlock(&lh, LCK_EX);
1669 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1675 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
1676 (const struct dt_key *)cname->ln_name,
1678 dt_read_unlock(env, parent);
1680 /* It is safe to release the ldlm lock, because when the logic come
1681 * here, we have got all the needed information above whether the
1682 * linkEA entry is valid or not. It is not important that others
1683 * may add new linkEA entry after the ldlm lock released. If other
1684 * has removed the specified linkEA entry by race, then it is OK,
1685 * because the subsequent lfsck_namespace_shrink_linkea() can handle
1687 lfsck_ibits_unlock(&lh, LCK_EX);
1688 if (rc == -ENOENT) {
1689 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata,
1698 /* The LFSCK just found some internal status of cross-MDTs
1699 * create operation. That is normal. */
1700 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
1701 linkea_next_entry(ldata);
1706 rc = lfsck_namespace_shrink_linkea(env, com, child, ldata, cname,
1713 * Conditionally replace name entry in the parent.
1715 * As required, the LFSCK may re-create the lost MDT-object for dangling
1716 * name entry, but such repairing may be wrong because of bad FID in the
1717 * name entry. As the LFSCK processing, the real MDT-object may be found,
1718 * then the LFSCK should check whether the former re-created MDT-object
1719 * has been modified or not, if not, then destroy it and update the name
1720 * entry in the parent to reference the real MDT-object.
1722 * \param[in] env pointer to the thread context
1723 * \param[in] com pointer to the lfsck component
1724 * \param[in] parent pointer to the parent directory
1725 * \param[in] child pointer to the MDT-object that may be the real
1726 * MDT-object corresponding to the name entry in parent
1727 * \param[in] cfid the current FID in the name entry
1728 * \param[in] cname contains the name of the child in the parent directory
1730 * \retval positive number for repaired cases
1731 * \retval 0 if nothing to be repaired
1732 * \retval negative error number on failure
1734 static int lfsck_namespace_replace_cond(const struct lu_env *env,
1735 struct lfsck_component *com,
1736 struct dt_object *parent,
1737 struct dt_object *child,
1738 const struct lu_fid *cfid,
1739 const struct lu_name *cname)
1741 struct lfsck_thread_info *info = lfsck_env_info(env);
1742 struct lu_fid *tfid = &info->lti_fid5;
1743 struct lu_attr *la = &info->lti_la;
1744 struct dt_insert_rec *rec = &info->lti_dt_rec;
1745 struct lfsck_instance *lfsck = com->lc_lfsck;
1746 struct dt_device *dev = lfsck->li_next;
1747 const char *name = cname->ln_name;
1748 struct dt_object *obj = NULL;
1749 struct lustre_handle plh = { 0 };
1750 struct lustre_handle clh = { 0 };
1751 struct linkea_data ldata = { 0 };
1752 struct thandle *th = NULL;
1757 rc = lfsck_ibits_lock(env, lfsck, parent, &plh,
1758 MDS_INODELOCK_UPDATE, LCK_EX);
1762 if (!fid_is_sane(cfid)) {
1767 obj = lfsck_object_find(env, lfsck, cfid);
1770 if (rc == -ENOENT) {
1778 if (!dt_object_exists(obj)) {
1783 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
1784 (const struct dt_key *)name, BYPASS_CAPA);
1785 if (rc == -ENOENT) {
1793 /* Someone changed the name entry, cannot replace it. */
1794 if (!lu_fid_eq(cfid, tfid))
1797 /* lock the object to be destroyed. */
1798 rc = lfsck_ibits_lock(env, lfsck, obj, &clh,
1799 MDS_INODELOCK_UPDATE |
1800 MDS_INODELOCK_XATTR, LCK_EX);
1804 if (unlikely(lfsck_is_dead_obj(obj))) {
1809 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
1813 /* The object has been modified by other(s), or it is not created by
1814 * LFSCK, the two cases are indistinguishable. So cannot replace it. */
1815 if (la->la_ctime != 0)
1818 if (S_ISREG(la->la_mode)) {
1819 rc = dt_xattr_get(env, obj, &LU_BUF_NULL, XATTR_NAME_LOV,
1821 /* If someone has created related OST-object(s),
1823 if ((rc > 0) || (rc < 0 && rc != -ENODATA))
1824 GOTO(log, rc = (rc > 0 ? 0 : rc));
1828 dt_read_lock(env, child, 0);
1829 rc = lfsck_links_read2(env, child, &ldata);
1830 dt_read_unlock(env, child);
1832 /* Someone changed the child, no need to replace. */
1839 rc = linkea_links_find(&ldata, cname, lfsck_dto2fid(parent));
1840 /* Someone moved the child, no need to replace. */
1844 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1847 th = dt_trans_create(env, dev);
1849 GOTO(log, rc = PTR_ERR(th));
1852 rc = dt_declare_destroy(env, obj, th);
1857 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
1861 rec->rec_type = S_IFDIR;
1862 rec->rec_fid = lfsck_dto2fid(child);
1863 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
1864 (const struct dt_key *)name, th);
1868 rc = dt_trans_start(env, dev, th);
1873 rc = dt_destroy(env, obj, th);
1878 /* The old name entry maybe not exist. */
1879 dt_delete(env, parent, (const struct dt_key *)name, th,
1882 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
1883 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
1885 GOTO(stop, rc = (rc == 0 ? 1 : rc));
1888 dt_trans_stop(env, dev, th);
1891 lfsck_ibits_unlock(&clh, LCK_EX);
1892 lfsck_ibits_unlock(&plh, LCK_EX);
1893 if (obj != NULL && !IS_ERR(obj))
1894 lfsck_object_put(env, obj);
1896 CDEBUG(D_LFSCK, "%s: namespace LFSCK conditionally destroy the "
1897 "object "DFID" because of conflict with the object "DFID
1898 " under the parent "DFID" with name %s: rc = %d\n",
1899 lfsck_lfsck2name(lfsck), PFID(cfid),
1900 PFID(lfsck_dto2fid(child)), PFID(lfsck_dto2fid(parent)),
1907 * Overwrite the linkEA for the object with the given ldata.
1909 * The caller should take the ldlm lock before the calling.
1911 * \param[in] env pointer to the thread context
1912 * \param[in] com pointer to the lfsck component
1913 * \param[in] obj pointer to the dt_object to be handled
1914 * \param[in] ldata pointer to the new linkEA data
1916 * \retval positive number for repaired cases
1917 * \retval 0 if nothing to be repaired
1918 * \retval negative error number on failure
1920 int lfsck_namespace_rebuild_linkea(const struct lu_env *env,
1921 struct lfsck_component *com,
1922 struct dt_object *obj,
1923 struct linkea_data *ldata)
1925 struct lfsck_instance *lfsck = com->lc_lfsck;
1926 struct dt_device *dev = lfsck->li_bottom;
1927 struct thandle *th = NULL;
1928 struct lu_buf linkea_buf;
1932 LASSERT(!dt_object_remote(obj));
1934 th = dt_trans_create(env, dev);
1936 GOTO(log, rc = PTR_ERR(th));
1938 lfsck_buf_init(&linkea_buf, ldata->ld_buf->lb_buf,
1939 ldata->ld_leh->leh_len);
1940 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
1941 XATTR_NAME_LINK, 0, th);
1945 rc = dt_trans_start_local(env, dev, th);
1949 dt_write_lock(env, obj, 0);
1950 if (unlikely(lfsck_is_dead_obj(obj)))
1951 GOTO(unlock, rc = 0);
1953 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
1954 GOTO(unlock, rc = 1);
1956 rc = dt_xattr_set(env, obj, &linkea_buf,
1957 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
1959 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
1962 dt_write_unlock(env, obj);
1965 dt_trans_stop(env, dev, th);
1968 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild linkEA for the "
1969 "object "DFID": rc = %d\n",
1970 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)), rc);
1973 struct lfsck_namespace *ns = com->lc_file_ram;
1975 ns->ln_flags |= LF_INCONSISTENT;
1982 * Repair invalid name entry.
1984 * If the name entry contains invalid information, such as bad file type
1985 * or (and) corrupted object FID, then either remove the name entry or
1986 * udpate the name entry with the given (right) information.
1988 * \param[in] env pointer to the thread context
1989 * \param[in] com pointer to the lfsck component
1990 * \param[in] parent pointer to the parent directory
1991 * \param[in] child pointer to the object referenced by the name entry
1992 * \param[in] name the old name of the child under the parent directory
1993 * \param[in] name2 the new name of the child under the parent directory
1994 * \param[in] type the type claimed by the name entry
1995 * \param[in] update update the name entry if true; otherwise, remove it
1996 * \param[in] dec decrease the parent nlink count if true
1998 * \retval positive number for repaired successfully
1999 * \retval 0 if nothing to be repaired
2000 * \retval negative error number on failure
2002 int lfsck_namespace_repair_dirent(const struct lu_env *env,
2003 struct lfsck_component *com,
2004 struct dt_object *parent,
2005 struct dt_object *child,
2006 const char *name, const char *name2,
2007 __u16 type, bool update, bool dec)
2009 struct lfsck_thread_info *info = lfsck_env_info(env);
2010 struct dt_insert_rec *rec = &info->lti_dt_rec;
2011 const struct lu_fid *cfid = lfsck_dto2fid(child);
2012 struct lu_fid *tfid = &info->lti_fid5;
2013 struct lfsck_instance *lfsck = com->lc_lfsck;
2014 struct dt_device *dev = lfsck->li_next;
2015 struct thandle *th = NULL;
2016 struct lustre_handle lh = { 0 };
2020 if (unlikely(!dt_try_as_dir(env, parent)))
2021 GOTO(log, rc = -ENOTDIR);
2023 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
2024 MDS_INODELOCK_UPDATE, LCK_EX);
2028 th = dt_trans_create(env, dev);
2030 GOTO(unlock1, rc = PTR_ERR(th));
2032 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
2037 rec->rec_type = lfsck_object_type(child) & S_IFMT;
2038 rec->rec_fid = cfid;
2039 rc = dt_declare_insert(env, parent,
2040 (const struct dt_rec *)rec,
2041 (const struct dt_key *)name2, th);
2047 rc = dt_declare_ref_del(env, parent, th);
2052 rc = dt_trans_start(env, dev, th);
2056 dt_write_lock(env, parent, 0);
2057 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2058 (const struct dt_key *)name, BYPASS_CAPA);
2059 /* Someone has removed the bad name entry by race. */
2061 GOTO(unlock2, rc = 0);
2066 /* Someone has removed the bad name entry and reused it for other
2067 * object by race. */
2068 if (!lu_fid_eq(tfid, cfid))
2069 GOTO(unlock2, rc = 0);
2071 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2072 GOTO(unlock2, rc = 1);
2074 rc = dt_delete(env, parent, (const struct dt_key *)name, th,
2080 rc = dt_insert(env, parent,
2081 (const struct dt_rec *)rec,
2082 (const struct dt_key *)name2, th,
2089 rc = dt_ref_del(env, parent, th);
2094 GOTO(unlock2, rc = (rc == 0 ? 1 : rc));
2097 dt_write_unlock(env, parent);
2100 dt_trans_stop(env, dev, th);
2102 /* We are not sure whether the child will become orphan or not.
2103 * Record it in the LFSCK trace file for further checking in
2104 * the second-stage scanning. */
2105 if (!update && !dec && rc == 0)
2106 lfsck_namespace_trace_update(env, com, cfid,
2107 LNTF_CHECK_LINKEA, true);
2110 lfsck_ibits_unlock(&lh, LCK_EX);
2113 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found bad name "
2114 "entry for: parent "DFID", child "DFID", name %s, type "
2115 "in name entry %o, type claimed by child %o. repair it "
2116 "by %s with new name2 %s: rc = %d\n", lfsck_lfsck2name(lfsck),
2117 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
2118 name, type, update ? lfsck_object_type(child) : 0,
2119 update ? "updating" : "removing", name2, rc);
2122 struct lfsck_namespace *ns = com->lc_file_ram;
2124 ns->ln_flags |= LF_INCONSISTENT;
2131 * Update the ".." name entry for the given object.
2133 * The object's ".." is corrupted, this function will update the ".." name
2134 * entry with the given pfid, and the linkEA with the given ldata.
2136 * The caller should take the ldlm lock before the calling.
2138 * \param[in] env pointer to the thread context
2139 * \param[in] com pointer to the lfsck component
2140 * \param[in] obj pointer to the dt_object to be handled
2141 * \param[in] pfid the new fid for the object's ".." name entry
2142 * \param[in] cname the name for the @obj in the parent directory
2144 * \retval positive number for repaired cases
2145 * \retval 0 if nothing to be repaired
2146 * \retval negative error number on failure
2148 static int lfsck_namespace_repair_unmatched_pairs(const struct lu_env *env,
2149 struct lfsck_component *com,
2150 struct dt_object *obj,
2151 const struct lu_fid *pfid,
2152 struct lu_name *cname)
2154 struct lfsck_thread_info *info = lfsck_env_info(env);
2155 struct dt_insert_rec *rec = &info->lti_dt_rec;
2156 struct lfsck_instance *lfsck = com->lc_lfsck;
2157 struct dt_device *dev = lfsck->li_bottom;
2158 struct thandle *th = NULL;
2159 struct linkea_data ldata = { 0 };
2160 struct lu_buf linkea_buf;
2164 LASSERT(!dt_object_remote(obj));
2165 LASSERT(S_ISDIR(lfsck_object_type(obj)));
2167 rc = linkea_data_new(&ldata, &info->lti_big_buf);
2171 rc = linkea_add_buf(&ldata, cname, pfid);
2175 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
2176 ldata.ld_leh->leh_len);
2178 th = dt_trans_create(env, dev);
2180 GOTO(log, rc = PTR_ERR(th));
2182 rc = dt_declare_delete(env, obj, (const struct dt_key *)dotdot, th);
2186 rec->rec_type = S_IFDIR;
2187 rec->rec_fid = pfid;
2188 rc = dt_declare_insert(env, obj, (const struct dt_rec *)rec,
2189 (const struct dt_key *)dotdot, th);
2193 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
2194 XATTR_NAME_LINK, 0, th);
2198 rc = dt_trans_start_local(env, dev, th);
2202 dt_write_lock(env, obj, 0);
2203 if (unlikely(lfsck_is_dead_obj(obj)))
2204 GOTO(unlock, rc = 0);
2206 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2207 GOTO(unlock, rc = 1);
2209 /* The old ".." name entry maybe not exist. */
2210 dt_delete(env, obj, (const struct dt_key *)dotdot, th,
2213 rc = dt_insert(env, obj, (const struct dt_rec *)rec,
2214 (const struct dt_key *)dotdot, th, BYPASS_CAPA, 1);
2218 rc = dt_xattr_set(env, obj, &linkea_buf,
2219 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
2221 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2224 dt_write_unlock(env, obj);
2227 dt_trans_stop(env, dev, th);
2230 CDEBUG(D_LFSCK, "%s: namespace LFSCK rebuild dotdot name entry for "
2231 "the object "DFID", new parent "DFID": rc = %d\n",
2232 lfsck_lfsck2name(lfsck), PFID(lfsck_dto2fid(obj)),
2236 struct lfsck_namespace *ns = com->lc_file_ram;
2238 ns->ln_flags |= LF_INCONSISTENT;
2245 * Handle orphan @obj during Double Scan Directory.
2247 * Remove the @obj's current (invalid) linkEA entries, and insert
2248 * it in the directory .lustre/lost+found/MDTxxxx/ with the name:
2249 * ${FID}-${PFID}-D-${conflict_version}
2251 * The caller should take the ldlm lock before the calling.
2253 * \param[in] env pointer to the thread context
2254 * \param[in] com pointer to the lfsck component
2255 * \param[in] obj pointer to the orphan object to be handled
2256 * \param[in] pfid the new fid for the object's ".." name entry
2257 * \param[in,out] lh ldlm lock handler for the given @obj
2258 * \param[out] type to tell the caller what the inconsistency is
2260 * \retval positive number for repaired cases
2261 * \retval 0 if nothing to be repaired
2262 * \retval negative error number on failure
2265 lfsck_namespace_dsd_orphan(const struct lu_env *env,
2266 struct lfsck_component *com,
2267 struct dt_object *obj,
2268 const struct lu_fid *pfid,
2269 struct lustre_handle *lh,
2270 enum lfsck_namespace_inconsistency_type *type)
2272 struct lfsck_thread_info *info = lfsck_env_info(env);
2273 struct lfsck_namespace *ns = com->lc_file_ram;
2277 /* Remove the unrecognized linkEA. */
2278 rc = lfsck_namespace_links_remove(env, com, obj);
2279 lfsck_ibits_unlock(lh, LCK_EX);
2280 if (rc < 0 && rc != -ENODATA)
2283 *type = LNIT_MUL_REF;
2285 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2286 * ever tried to verify some remote MDT-object that resides on this
2287 * MDT, but this MDT failed to respond such request. So means there
2288 * may be some remote name entry on other MDT that references this
2289 * object with another name, so we cannot know whether this linkEA
2290 * is valid or not. So keep it there and maybe resolved when next
2292 if (ns->ln_flags & LF_INCOMPLETE)
2295 /* The unique linkEA is invalid, even if the ".." name entry may be
2296 * valid, we still cannot know via which name entry this directory
2297 * will be referenced. Then handle it as pure orphan. */
2298 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2299 "-"DFID, PFID(pfid));
2300 rc = lfsck_namespace_insert_orphan(env, com, obj,
2301 info->lti_tmpbuf, "D", NULL);
2307 * Double Scan Directory object for single linkEA entry case.
2309 * The given @child has unique linkEA entry. If the linkEA entry is valid,
2310 * then check whether the name is in the namespace or not, if not, add the
2311 * missing name entry back to namespace. If the linkEA entry is invalid,
2312 * then remove it and insert the @child in the .lustre/lost+found/MDTxxxx/
2315 * \param[in] env pointer to the thread context
2316 * \param[in] com pointer to the lfsck component
2317 * \param[in] child pointer to the directory to be double scanned
2318 * \param[in] pfid the FID corresponding to the ".." entry
2319 * \param[in] ldata pointer to the linkEA data for the given @child
2320 * \param[in,out] lh ldlm lock handler for the given @child
2321 * \param[out] type to tell the caller what the inconsistency is
2322 * \param[in] retry if found inconsistency, but the caller does not hold
2323 * ldlm lock on the @child, then set @retry as true
2325 * \retval positive number for repaired cases
2326 * \retval 0 if nothing to be repaired
2327 * \retval negative error number on failure
2330 lfsck_namespace_dsd_single(const struct lu_env *env,
2331 struct lfsck_component *com,
2332 struct dt_object *child,
2333 const struct lu_fid *pfid,
2334 struct linkea_data *ldata,
2335 struct lustre_handle *lh,
2336 enum lfsck_namespace_inconsistency_type *type,
2339 struct lfsck_thread_info *info = lfsck_env_info(env);
2340 struct lu_name *cname = &info->lti_name;
2341 const struct lu_fid *cfid = lfsck_dto2fid(child);
2342 struct lu_fid *tfid = &info->lti_fid3;
2343 struct lfsck_namespace *ns = com->lc_file_ram;
2344 struct lfsck_instance *lfsck = com->lc_lfsck;
2345 struct dt_object *parent = NULL;
2346 struct lmv_mds_md_v1 *lmv;
2350 lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid, info->lti_key);
2351 /* The unique linkEA entry with bad parent will be handled as orphan. */
2352 if (!fid_is_sane(tfid)) {
2353 if (!lustre_handle_is_used(lh) && retry != NULL)
2356 rc = lfsck_namespace_dsd_orphan(env, com, child,
2362 parent = lfsck_object_find_bottom(env, lfsck, tfid);
2364 GOTO(out, rc = PTR_ERR(parent));
2366 /* We trust the unique linkEA entry in spite of whether it matches the
2367 * ".." name entry or not. Because even if the linkEA entry is wrong
2368 * and the ".." name entry is right, we still cannot know via which
2369 * name entry the child will be referenced, since all known entries
2370 * have been verified during the first-stage scanning. */
2371 if (!dt_object_exists(parent)) {
2372 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2373 * has ever tried to verify some remote MDT-object that resides
2374 * on this MDT, but this MDT failed to respond such request. So
2375 * means there may be some remote name entry on other MDT that
2376 * references this object with another name, so we cannot know
2377 * whether this linkEA is valid or not. So keep it there and
2378 * maybe resolved when next LFSCK run. */
2379 if (ns->ln_flags & LF_INCOMPLETE)
2382 if (!lustre_handle_is_used(lh) && retry != NULL) {
2388 lfsck_ibits_unlock(lh, LCK_EX);
2391 lmv = &info->lti_lmv;
2392 rc = lfsck_read_stripe_lmv(env, child, lmv);
2393 if (rc != 0 && rc != -ENODATA)
2396 if (rc == -ENODATA || lmv->lmv_magic != LMV_MAGIC_STRIPE) {
2398 } else if (lfsck_shard_name_to_index(env,
2399 cname->ln_name, cname->ln_namelen,
2400 S_IFDIR, cfid) < 0) {
2401 /* It is an invalid name entry, we
2402 * cannot trust the parent also. */
2403 rc = lfsck_namespace_shrink_linkea(env, com, child,
2404 ldata, cname, tfid, true);
2408 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
2409 "-"DFID, PFID(pfid));
2410 rc = lfsck_namespace_insert_orphan(env, com, child,
2411 info->lti_tmpbuf, "S", NULL);
2416 /* Create the lost parent as an orphan. */
2417 rc = lfsck_namespace_create_orphan(env, com, parent, lmv);
2419 /* Add the missing name entry to the parent. */
2420 rc = lfsck_namespace_insert_normal(env, com, parent,
2421 child, cname->ln_name);
2422 if (unlikely(rc == -EEXIST)) {
2423 /* Unfortunately, someone reused the name
2424 * under the parent by race. So we have
2425 * to remove the linkEA entry from
2426 * current child object. It means that the
2427 * LFSCK cannot recover the system
2428 * totally back to its original status,
2429 * but it is necessary to make the
2430 * current system to be consistent. */
2431 rc = lfsck_namespace_shrink_linkea(env,
2435 snprintf(info->lti_tmpbuf,
2436 sizeof(info->lti_tmpbuf),
2437 "-"DFID, PFID(pfid));
2438 rc = lfsck_namespace_insert_orphan(env,
2439 com, child, info->lti_tmpbuf,
2448 /* The unique linkEA entry with bad parent will be handled as orphan. */
2449 if (unlikely(!dt_try_as_dir(env, parent))) {
2450 if (!lustre_handle_is_used(lh) && retry != NULL)
2453 rc = lfsck_namespace_dsd_orphan(env, com, child,
2459 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2460 (const struct dt_key *)cname->ln_name, BYPASS_CAPA);
2461 if (rc == -ENOENT) {
2462 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT
2463 * has ever tried to verify some remote MDT-object that resides
2464 * on this MDT, but this MDT failed to respond such request. So
2465 * means there may be some remote name entry on other MDT that
2466 * references this object with another name, so we cannot know
2467 * whether this linkEA is valid or not. So keep it there and
2468 * maybe resolved when next LFSCK run. */
2469 if (ns->ln_flags & LF_INCOMPLETE)
2472 if (!lustre_handle_is_used(lh) && retry != NULL) {
2478 lfsck_ibits_unlock(lh, LCK_EX);
2479 rc = lfsck_namespace_check_name(env, parent, child, cname);
2486 /* It is an invalid name entry, drop it. */
2487 if (unlikely(rc > 0)) {
2488 rc = lfsck_namespace_shrink_linkea(env, com, child,
2489 ldata, cname, tfid, true);
2491 snprintf(info->lti_tmpbuf,
2492 sizeof(info->lti_tmpbuf),
2493 "-"DFID, PFID(pfid));
2494 rc = lfsck_namespace_insert_orphan(env, com,
2495 child, info->lti_tmpbuf, "D", NULL);
2501 /* Add the missing name entry back to the namespace. */
2502 rc = lfsck_namespace_insert_normal(env, com, parent, child,
2504 if (unlikely(rc == -ESTALE))
2505 /* It may happen when the remote object has been
2506 * removed, but the local MDT is not aware of that. */
2509 if (unlikely(rc == -EEXIST)) {
2510 /* Unfortunately, someone reused the name under the
2511 * parent by race. So we have to remove the linkEA
2512 * entry from current child object. It means that the
2513 * LFSCK cannot recover the system totally back to
2514 * its original status, but it is necessary to make
2515 * the current system to be consistent.
2517 * It also may be because of the LFSCK found some
2518 * internal status of create operation. Under such
2519 * case, nothing to be done. */
2520 rc = lfsck_namespace_shrink_linkea_cond(env, com,
2521 parent, child, ldata, cname, tfid);
2523 snprintf(info->lti_tmpbuf,
2524 sizeof(info->lti_tmpbuf),
2525 "-"DFID, PFID(pfid));
2526 rc = lfsck_namespace_insert_orphan(env, com,
2527 child, info->lti_tmpbuf, "D", NULL);
2537 if (!lu_fid_eq(tfid, cfid)) {
2538 if (!lustre_handle_is_used(lh) && retry != NULL) {
2544 lfsck_ibits_unlock(lh, LCK_EX);
2545 /* The name entry references another MDT-object that
2546 * may be created by the LFSCK for repairing dangling
2547 * name entry. Try to replace it. */
2548 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2551 rc = lfsck_namespace_dsd_orphan(env, com, child,
2557 if (fid_is_zero(pfid))
2560 /* The ".." name entry is wrong, update it. */
2561 if (!lu_fid_eq(pfid, lfsck_dto2fid(parent))) {
2562 if (!lustre_handle_is_used(lh) && retry != NULL) {
2568 *type = LNIT_UNMATCHED_PAIRS;
2569 rc = lfsck_namespace_repair_unmatched_pairs(env, com, child,
2570 lfsck_dto2fid(parent), cname);
2576 if (parent != NULL && !IS_ERR(parent))
2577 lfsck_object_put(env, parent);
2583 * Double Scan Directory object for multiple linkEA entries case.
2585 * The given @child has multiple linkEA entries. There is at most one linkEA
2586 * entry will be valid, all the others will be removed. Firstly, the function
2587 * will try to find out the linkEA entry for which the name entry exists under
2588 * the given parent (@pfid). If there is no linkEA entry that matches the given
2589 * ".." name entry, then tries to find out the first linkEA entry that both the
2590 * parent and the name entry exist to rebuild a new ".." name entry.
2592 * \param[in] env pointer to the thread context
2593 * \param[in] com pointer to the lfsck component
2594 * \param[in] child pointer to the directory to be double scanned
2595 * \param[in] pfid the FID corresponding to the ".." entry
2596 * \param[in] ldata pointer to the linkEA data for the given @child
2597 * \param[in,out] lh ldlm lock handler for the given @child
2598 * \param[out] type to tell the caller what the inconsistency is
2599 * \param[in] lpf true if the ".." entry is under lost+found/MDTxxxx/
2601 * \retval positive number for repaired cases
2602 * \retval 0 if nothing to be repaired
2603 * \retval negative error number on failure
2606 lfsck_namespace_dsd_multiple(const struct lu_env *env,
2607 struct lfsck_component *com,
2608 struct dt_object *child,
2609 const struct lu_fid *pfid,
2610 struct linkea_data *ldata,
2611 struct lustre_handle *lh,
2612 enum lfsck_namespace_inconsistency_type *type,
2615 struct lfsck_thread_info *info = lfsck_env_info(env);
2616 struct lu_name *cname = &info->lti_name;
2617 const struct lu_fid *cfid = lfsck_dto2fid(child);
2618 struct lu_fid *tfid = &info->lti_fid3;
2619 struct lu_fid *pfid2 = &info->lti_fid4;
2620 struct lfsck_namespace *ns = com->lc_file_ram;
2621 struct lfsck_instance *lfsck = com->lc_lfsck;
2622 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
2623 struct dt_object *parent = NULL;
2624 struct linkea_data ldata_new = { 0 };
2625 int dirent_count = 0;
2626 int linkea_count = 0;
2632 while (ldata->ld_lee != NULL) {
2633 lfsck_namespace_unpack_linkea_entry(ldata, cname, tfid,
2635 /* Drop repeated linkEA entries. */
2636 lfsck_namespace_filter_linkea_entry(ldata, cname, tfid, true);
2637 /* Drop invalid linkEA entry. */
2638 if (!fid_is_sane(tfid)) {
2639 linkea_del_buf(ldata, cname);
2644 /* If current dotdot is the .lustre/lost+found/MDTxxxx/,
2645 * then it is possible that: the directry object has ever
2646 * been lost, but its name entry was there. In the former
2647 * LFSCK run, during the first-stage scanning, the LFSCK
2648 * found the dangling name entry, but it did not recreate
2649 * the lost object, and when moved to the second-stage
2650 * scanning, some children objects of the lost directory
2651 * object were found, then the LFSCK recreated such lost
2652 * directory object as an orphan.
2654 * When the LFSCK runs again, if the dangling name is still
2655 * there, the LFSCK should move the orphan directory object
2656 * back to the normal namespace. */
2657 if (!lpf && !lu_fid_eq(pfid, tfid) && once) {
2658 linkea_next_entry(ldata);
2662 parent = lfsck_object_find_bottom(env, lfsck, tfid);
2664 RETURN(PTR_ERR(parent));
2666 if (!dt_object_exists(parent)) {
2667 lfsck_object_put(env, parent);
2668 if (ldata->ld_leh->leh_reccount > 1) {
2669 /* If it is NOT the last linkEA entry, then
2670 * there is still other chance to make the
2671 * child to be visible via other parent, then
2672 * remove this linkEA entry. */
2673 linkea_del_buf(ldata, cname);
2681 /* The linkEA entry with bad parent will be removed. */
2682 if (unlikely(!dt_try_as_dir(env, parent))) {
2683 lfsck_object_put(env, parent);
2684 linkea_del_buf(ldata, cname);
2689 rc = dt_lookup(env, parent, (struct dt_rec *)tfid,
2690 (const struct dt_key *)cname->ln_name,
2692 *pfid2 = *lfsck_dto2fid(parent);
2693 if (rc == -ENOENT) {
2694 lfsck_object_put(env, parent);
2695 linkea_next_entry(ldata);
2700 lfsck_object_put(env, parent);
2705 if (lu_fid_eq(tfid, cfid)) {
2706 lfsck_object_put(env, parent);
2707 if (!lu_fid_eq(pfid, pfid2)) {
2708 *type = LNIT_UNMATCHED_PAIRS;
2709 rc = lfsck_namespace_repair_unmatched_pairs(env,
2710 com, child, pfid2, cname);
2716 /* It is the most common case that we find the
2717 * name entry corresponding to the linkEA entry
2718 * that matches the ".." name entry. */
2719 rc = linkea_data_new(&ldata_new, &info->lti_big_buf);
2723 rc = linkea_add_buf(&ldata_new, cname, pfid2);
2727 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2732 linkea_del_buf(ldata, cname);
2734 linkea_first_entry(ldata);
2735 /* There may be some invalid dangling name entries under
2736 * other parent directories, remove all of them. */
2737 while (ldata->ld_lee != NULL) {
2738 lfsck_namespace_unpack_linkea_entry(ldata,
2739 cname, tfid, info->lti_key);
2740 if (!fid_is_sane(tfid))
2743 parent = lfsck_object_find_bottom(env, lfsck,
2745 if (IS_ERR(parent)) {
2746 rc = PTR_ERR(parent);
2747 if (rc != -ENOENT &&
2748 bk->lb_param & LPF_FAILOUT)
2754 if (!dt_object_exists(parent)) {
2755 lfsck_object_put(env, parent);
2759 rc = lfsck_namespace_repair_dirent(env, com,
2760 parent, child, cname->ln_name,
2761 cname->ln_name, S_IFDIR, false, true);
2762 lfsck_object_put(env, parent);
2764 if (bk->lb_param & LPF_FAILOUT)
2773 linkea_del_buf(ldata, cname);
2776 ns->ln_dirent_repaired += dirent_count;
2781 lfsck_ibits_unlock(lh, LCK_EX);
2782 /* The name entry references another MDT-object that may be
2783 * created by the LFSCK for repairing dangling name entry.
2784 * Try to replace it. */
2785 rc = lfsck_namespace_replace_cond(env, com, parent, child,
2787 lfsck_object_put(env, parent);
2794 linkea_del_buf(ldata, cname);
2797 linkea_first_entry(ldata);
2798 if (ldata->ld_leh->leh_reccount == 1) {
2799 rc = lfsck_namespace_dsd_single(env, com, child, pfid, ldata,
2802 if (rc == 0 && fid_is_zero(pfid) && linkea_count > 0)
2803 rc = lfsck_namespace_rebuild_linkea(env, com, child,
2809 /* All linkEA entries are invalid and removed, then handle the @child
2811 if (ldata->ld_leh->leh_reccount == 0) {
2812 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, lh,
2818 /* If the dangling name entry for the orphan directory object has
2819 * been remvoed, then just check whether the directory object is
2820 * still under the .lustre/lost+found/MDTxxxx/ or not. */
2826 /* There is no linkEA entry that matches the ".." name entry. Find
2827 * the first linkEA entry that both parent and name entry exist to
2828 * rebuild a new ".." name entry. */
2838 * Repair the object's nlink attribute.
2840 * If all the known name entries have been verified, then the object's hard
2841 * link attribute should match the object's linkEA entries count unless the
2842 * object's has too much hard link to be recorded in the linkEA. Such cases
2843 * should have been marked in the LFSCK trace file. Otherwise, trust the
2844 * linkEA to update the object's nlink attribute.
2846 * \param[in] env pointer to the thread context
2847 * \param[in] com pointer to the lfsck component
2848 * \param[in] obj pointer to the dt_object to be handled
2849 * \param[in,out] nlink pointer to buffer to object's hard lock count before
2850 * and after the repairing
2852 * \retval positive number for repaired cases
2853 * \retval 0 if nothing to be repaired
2854 * \retval negative error number on failure
2856 static int lfsck_namespace_repair_nlink(const struct lu_env *env,
2857 struct lfsck_component *com,
2858 struct dt_object *obj, __u32 *nlink)
2860 struct lfsck_thread_info *info = lfsck_env_info(env);
2861 struct lu_attr *la = &info->lti_la3;
2862 struct lu_fid *tfid = &info->lti_fid3;
2863 struct lfsck_namespace *ns = com->lc_file_ram;
2864 struct lfsck_instance *lfsck = com->lc_lfsck;
2865 struct dt_device *dev = lfsck->li_bottom;
2866 const struct lu_fid *cfid = lfsck_dto2fid(obj);
2867 struct dt_object *child = NULL;
2868 struct thandle *th = NULL;
2869 struct linkea_data ldata = { 0 };
2870 struct lustre_handle lh = { 0 };
2876 LASSERT(!dt_object_remote(obj));
2877 LASSERT(S_ISREG(lfsck_object_type(obj)));
2879 child = lfsck_object_find_by_dev(env, dev, cfid);
2881 GOTO(log, rc = PTR_ERR(child));
2883 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
2884 MDS_INODELOCK_UPDATE |
2885 MDS_INODELOCK_XATTR, LCK_EX);
2889 th = dt_trans_create(env, dev);
2891 GOTO(log, rc = PTR_ERR(th));
2893 la->la_valid = LA_NLINK;
2894 rc = dt_declare_attr_set(env, child, la, th);
2898 rc = dt_trans_start_local(env, dev, th);
2902 dt_write_lock(env, child, 0);
2903 /* If the LFSCK is marked as LF_INCOMPLETE, then means some MDT has
2904 * ever tried to verify some remote MDT-object that resides on this
2905 * MDT, but this MDT failed to respond such request. So means there
2906 * may be some remote name entry on other MDT that references this
2907 * object with another name, so we cannot know whether this linkEA
2908 * is valid or not. So keep it there and maybe resolved when next
2910 if (ns->ln_flags & LF_INCOMPLETE)
2911 GOTO(unlock, rc = 0);
2913 fid_cpu_to_be(tfid, cfid);
2914 rc = dt_lookup(env, com->lc_obj, (struct dt_rec *)&flags,
2915 (const struct dt_key *)tfid, BYPASS_CAPA);
2919 if (flags & LNTF_SKIP_NLINK)
2920 GOTO(unlock, rc = 0);
2922 rc = lfsck_links_read2(env, child, &ldata);
2924 GOTO(unlock, rc = 0);
2929 if (*nlink == ldata.ld_leh->leh_reccount)
2930 GOTO(unlock, rc = 0);
2932 la->la_nlink = *nlink = ldata.ld_leh->leh_reccount;
2933 if (lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN)
2934 GOTO(unlock, rc = 1);
2936 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
2938 GOTO(unlock, rc = (rc == 0 ? 1 : rc));
2941 dt_write_unlock(env, child);
2944 dt_trans_stop(env, dev, th);
2947 lfsck_ibits_unlock(&lh, LCK_EX);
2948 if (child != NULL && !IS_ERR(child))
2949 lfsck_object_put(env, child);
2951 CDEBUG(D_LFSCK, "%s: namespace LFSCK repaired the object "DFID"'s "
2952 "nlink count from %u to %u: rc = %d\n",
2953 lfsck_lfsck2name(lfsck), PFID(cfid), old, *nlink, rc);
2956 ns->ln_flags |= LF_INCONSISTENT;
2962 * Double scan the directory object for namespace LFSCK.
2964 * This function will verify the <parent, child> pairs in the namespace tree:
2965 * the parent references the child via some name entry that should be in the
2966 * child's linkEA entry, the child should back references the parent via its
2969 * The LFSCK will scan every linkEA entry in turn until find out the first
2970 * matched pairs. If found, then all other linkEA entries will be dropped.
2971 * If all the linkEA entries cannot match the ".." name entry, then there
2972 * are serveral possible cases:
2974 * 1) If there is only one linkEA entry, then trust it as long as the PFID
2975 * in the linkEA entry is valid.
2977 * 2) If there are multiple linkEA entries, then try to find the linkEA
2978 * that matches the ".." name entry. If found, then all other entries
2979 * are invalid; otherwise, it is quite possible that the ".." name entry
2980 * is corrupted. Under such case, the LFSCK will rebuild the ".." name
2981 * entry according to the first valid linkEA entry (both the parent and
2982 * the name entry should exist).
2984 * 3) If the directory object has no (valid) linkEA entry, then the
2985 * directory object will be handled as pure orphan and inserted
2986 * in the .lustre/lost+found/MDTxxxx/ with the name:
2987 * ${self_FID}-${PFID}-D-${conflict_version}
2989 * \param[in] env pointer to the thread context
2990 * \param[in] com pointer to the lfsck component
2991 * \param[in] child pointer to the directory object to be handled
2992 * \param[in] flags to indicate the specical checking on the @child
2994 * \retval positive number for repaired cases
2995 * \retval 0 if nothing to be repaired
2996 * \retval negative error number on failure
2998 static int lfsck_namespace_double_scan_dir(const struct lu_env *env,
2999 struct lfsck_component *com,
3000 struct dt_object *child, __u8 flags)
3002 struct lfsck_thread_info *info = lfsck_env_info(env);
3003 const struct lu_fid *cfid = lfsck_dto2fid(child);
3004 struct lu_fid *pfid = &info->lti_fid2;
3005 struct lfsck_namespace *ns = com->lc_file_ram;
3006 struct lfsck_instance *lfsck = com->lc_lfsck;
3007 struct lustre_handle lh = { 0 };
3008 struct linkea_data ldata = { 0 };
3009 bool unknown = false;
3012 enum lfsck_namespace_inconsistency_type type = LNIT_BAD_LINKEA;
3016 LASSERT(!dt_object_remote(child));
3018 if (flags & LNTF_UNCERTAIN_LMV) {
3019 if (flags & LNTF_RECHECK_NAME_HASH) {
3020 rc = lfsck_namespace_scan_shard(env, com, child);
3024 ns->ln_striped_shards_scanned++;
3026 ns->ln_striped_shards_skipped++;
3030 flags &= ~(LNTF_RECHECK_NAME_HASH | LNTF_UNCERTAIN_LMV);
3034 if (flags & (LNTF_CHECK_LINKEA | LNTF_CHECK_PARENT) &&
3035 !(lfsck->li_bookmark_ram.lb_param & LPF_ALL_TGT)) {
3036 CDEBUG(D_LFSCK, "%s: some MDT(s) maybe NOT take part in the"
3037 "the namespace LFSCK, then the LFSCK cannot guarantee"
3038 "all the name entries have been verified in first-stage"
3039 "scanning. So have to skip orphan related handling for"
3040 "the directory object "DFID" with remote name entry\n",
3041 lfsck_lfsck2name(lfsck), PFID(cfid));
3046 if (unlikely(!dt_try_as_dir(env, child)))
3047 GOTO(out, rc = -ENOTDIR);
3049 /* We only take ldlm lock on the @child when required. When the
3050 * logic comes here for the first time, it is always false. */
3054 rc = lfsck_ibits_lock(env, lfsck, child, &lh,
3055 MDS_INODELOCK_UPDATE |
3056 MDS_INODELOCK_XATTR, LCK_EX);
3061 dt_read_lock(env, child, 0);
3062 if (unlikely(lfsck_is_dead_obj(child))) {
3063 dt_read_unlock(env, child);
3068 rc = dt_lookup(env, child, (struct dt_rec *)pfid,
3069 (const struct dt_key *)dotdot, BYPASS_CAPA);
3071 if (rc != -ENOENT && rc != -ENODATA && rc != -EINVAL) {
3072 dt_read_unlock(env, child);
3077 if (!lustre_handle_is_used(&lh)) {
3078 dt_read_unlock(env, child);
3083 } else if (lfsck->li_lpf_obj != NULL &&
3084 lu_fid_eq(pfid, lfsck_dto2fid(lfsck->li_lpf_obj))) {
3086 } else if (unlikely(!fid_is_sane(pfid))) {
3090 rc = lfsck_links_read(env, child, &ldata);
3091 dt_read_unlock(env, child);
3093 if (rc != -ENODATA && rc != -EINVAL)
3096 if (!lustre_handle_is_used(&lh))
3099 if (rc == -EINVAL && !fid_is_zero(pfid)) {
3100 /* Remove the corrupted linkEA. */
3101 rc = lfsck_namespace_links_remove(env, com, child);
3103 /* Here, because of the crashed linkEA, we
3104 * cannot know whether there is some parent
3105 * that references the child directory via
3106 * some name entry or not. So keep it there,
3107 * when the LFSCK run next time, if there is
3108 * some parent that references this object,
3109 * then the LFSCK can rebuild the linkEA;
3110 * otherwise, this object will be handled
3111 * as orphan as above. */
3114 /* 1. If we have neither ".." nor linkEA,
3115 * then it is an orphan.
3117 * 2. If we only have the ".." name entry,
3118 * but no parent references this child
3119 * directory, then handle it as orphan. */
3120 lfsck_ibits_unlock(&lh, LCK_EX);
3121 type = LNIT_MUL_REF;
3123 /* If the LFSCK is marked as LF_INCOMPLETE,
3124 * then means some MDT has ever tried to
3125 * verify some remote MDT-object that resides
3126 * on this MDT, but this MDT failed to respond
3127 * such request. So means there may be some
3128 * remote name entry on other MDT that
3129 * references this object with another name,
3130 * so we cannot know whether this linkEA is
3131 * valid or not. So keep it there and maybe
3132 * resolved when next LFSCK run. */
3133 if (ns->ln_flags & LF_INCOMPLETE)
3136 snprintf(info->lti_tmpbuf, sizeof(info->lti_tmpbuf),
3137 "-"DFID, PFID(pfid));
3138 rc = lfsck_namespace_insert_orphan(env, com, child,
3139 info->lti_tmpbuf, "D", NULL);
3145 linkea_first_entry(&ldata);
3146 /* This is the most common case: the object has unique linkEA entry. */
3147 if (ldata.ld_leh->leh_reccount == 1) {
3148 rc = lfsck_namespace_dsd_single(env, com, child, pfid, &ldata,
3149 &lh, &type, &retry);
3151 LASSERT(!lustre_handle_is_used(&lh));
3160 if (!lustre_handle_is_used(&lh))
3163 if (unlikely(ldata.ld_leh->leh_reccount == 0)) {
3164 rc = lfsck_namespace_dsd_orphan(env, com, child, pfid, &lh,
3170 /* When we come here, the cases usually like that:
3171 * 1) The directory object has a corrupted linkEA entry. During the
3172 * first-stage scanning, the LFSCK cannot know such corruption,
3173 * then it appends the right linkEA entry according to the found
3174 * name entry after the bad one.
3176 * 2) The directory object has a right linkEA entry. During the
3177 * first-stage scanning, the LFSCK finds some bad name entry,
3178 * but the LFSCK cannot aware that at that time, then it adds
3179 * the bad linkEA entry for further processing. */
3180 rc = lfsck_namespace_dsd_multiple(env, com, child, pfid, &ldata,
3186 lfsck_ibits_unlock(&lh, LCK_EX);
3189 case LNIT_BAD_LINKEA:
3190 ns->ln_linkea_repaired++;
3192 case LNIT_UNMATCHED_PAIRS:
3193 ns->ln_unmatched_pairs_repaired++;
3196 ns->ln_mul_ref_repaired++;
3204 ns->ln_unknown_inconsistency++;
3210 * Double scan the MDT-object for namespace LFSCK.
3212 * If the MDT-object contains invalid or repeated linkEA entries, then drop
3213 * those entries from the linkEA; if the linkEA becomes empty or the object
3214 * has no linkEA, then it is an orphan and will be added into the directory
3215 * .lustre/lost+found/MDTxxxx/; if the remote parent is lost, then recreate
3216 * the remote parent; if the name entry corresponding to some linkEA entry
3217 * is lost, then add the name entry back to the namespace.
3219 * \param[in] env pointer to the thread context
3220 * \param[in] com pointer to the lfsck component
3221 * \param[in] child pointer to the dt_object to be handled
3222 * \param[in] flags some hints to indicate how the @child should be handled
3224 * \retval positive number for repaired cases
3225 * \retval 0 if nothing to be repaired
3226 * \retval negative error number on failure
3228 static int lfsck_namespace_double_scan_one(const struct lu_env *env,
3229 struct lfsck_component *com,
3230 struct dt_object *child, __u8 flags)
3232 struct lfsck_thread_info *info = lfsck_env_info(env);
3233 struct lu_attr *la = &info->lti_la;
3234 struct lu_name *cname = &info->lti_name;
3235 struct lu_fid *pfid = &info->lti_fid;
3236 struct lu_fid *cfid = &info->lti_fid2;
3237 struct lfsck_instance *lfsck = com->lc_lfsck;
3238 struct lfsck_namespace *ns = com->lc_file_ram;
3239 struct dt_object *parent = NULL;
3240 struct linkea_data ldata = { 0 };
3241 bool repaired = false;
3246 dt_read_lock(env, child, 0);
3247 if (unlikely(lfsck_is_dead_obj(child))) {
3248 dt_read_unlock(env, child);
3253 if (S_ISDIR(lfsck_object_type(child))) {
3254 dt_read_unlock(env, child);
3255 rc = lfsck_namespace_double_scan_dir(env, com, child, flags);
3260 rc = lfsck_links_read(env, child, &ldata);
3261 dt_read_unlock(env, child);
3265 linkea_first_entry(&ldata);
3266 while (ldata.ld_lee != NULL) {
3267 lfsck_namespace_unpack_linkea_entry(&ldata, cname, pfid,
3269 rc = lfsck_namespace_filter_linkea_entry(&ldata, cname, pfid,
3271 /* Found repeated linkEA entries */
3273 rc = lfsck_namespace_shrink_linkea(env, com, child,
3274 &ldata, cname, pfid, false);
3286 /* Invalid PFID in the linkEA entry. */
3287 if (!fid_is_sane(pfid)) {
3288 rc = lfsck_namespace_shrink_linkea(env, com, child,
3289 &ldata, cname, pfid, true);
3299 parent = lfsck_object_find_bottom(env, lfsck, pfid);
3301 GOTO(out, rc = PTR_ERR(parent));
3303 if (!dt_object_exists(parent)) {
3306 if (ldata.ld_leh->leh_reccount > 1) {
3307 /* If it is NOT the last linkEA entry, then
3308 * there is still other chance to make the
3309 * child to be visible via other parent, then
3310 * remove this linkEA entry. */
3311 rc = lfsck_namespace_shrink_linkea(env, com,
3312 child, &ldata, cname, pfid, true);
3314 /* If the LFSCK is marked as LF_INCOMPLETE,
3315 * then means some MDT has ever tried to
3316 * verify some remote MDT-object that resides
3317 * on this MDT, but this MDT failed to respond
3318 * such request. So means there may be some
3319 * remote name entry on other MDT that
3320 * references this object with another name,
3321 * so we cannot know whether this linkEA is
3322 * valid or not. So keep it there and maybe
3323 * resolved when next LFSCK run. */
3324 if (ns->ln_flags & LF_INCOMPLETE) {
3325 lfsck_object_put(env, parent);
3330 /* Create the lost parent as an orphan. */
3331 rc = lfsck_namespace_create_orphan(env, com,
3334 lfsck_object_put(env, parent);
3342 /* Add the missing name entry to the parent. */
3343 rc = lfsck_namespace_insert_normal(env, com,
3344 parent, child, cname->ln_name);
3345 if (unlikely(rc == -EEXIST))
3346 /* Unfortunately, someone reused the
3347 * name under the parent by race. So we
3348 * have to remove the linkEA entry from
3349 * current child object. It means that
3350 * the LFSCK cannot recover the system
3351 * totally back to its original status,
3352 * but it is necessary to make the
3353 * current system to be consistent. */
3354 rc = lfsck_namespace_shrink_linkea(env,
3358 linkea_next_entry(&ldata);
3361 lfsck_object_put(env, parent);
3371 /* The linkEA entry with bad parent will be removed. */
3372 if (unlikely(!dt_try_as_dir(env, parent))) {
3373 lfsck_object_put(env, parent);
3374 rc = lfsck_namespace_shrink_linkea(env, com, child,
3375 &ldata, cname, pfid, true);
3385 rc = dt_lookup(env, parent, (struct dt_rec *)cfid,
3386 (const struct dt_key *)cname->ln_name,
3388 if (rc != 0 && rc != -ENOENT) {
3389 lfsck_object_put(env, parent);
3395 if (lu_fid_eq(cfid, lfsck_dto2fid(child))) {
3396 /* It is the most common case that we
3397 * find the name entry corresponding
3398 * to the linkEA entry. */
3399 lfsck_object_put(env, parent);
3400 linkea_next_entry(&ldata);
3402 /* The name entry references another
3403 * MDT-object that may be created by
3404 * the LFSCK for repairing dangling
3405 * name entry. Try to replace it. */
3406 rc = lfsck_namespace_replace_cond(env, com,
3407 parent, child, cfid, cname);
3408 lfsck_object_put(env, parent);
3414 linkea_next_entry(&ldata);
3416 rc = lfsck_namespace_shrink_linkea(env,
3430 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3434 /* If there is no name entry in the parent dir and the object
3435 * link count is less than the linkea entries count, then the
3436 * linkea entry should be removed. */
3437 if (ldata.ld_leh->leh_reccount > la->la_nlink) {
3438 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3439 parent, child, &ldata, cname, pfid);
3440 lfsck_object_put(env, parent);
3450 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3451 * MDT has ever tried to verify some remote MDT-object that
3452 * resides on this MDT, but this MDT failed to respond such
3453 * request. So means there may be some remote name entry on
3454 * other MDT that references this object with another name,
3455 * so we cannot know whether this linkEA is valid or not.
3456 * So keep it there and maybe resolved when next LFSCK run. */
3457 if (ns->ln_flags & LF_INCOMPLETE) {
3458 lfsck_object_put(env, parent);
3463 rc = lfsck_namespace_check_name(env, parent, child, cname);
3468 lfsck_object_put(env, parent);
3473 /* It is an invalid name entry, drop it. */
3474 if (unlikely(rc > 0)) {
3475 lfsck_object_put(env, parent);
3476 rc = lfsck_namespace_shrink_linkea(env, com, child,
3477 &ldata, cname, pfid, true);
3487 /* Add the missing name entry back to the namespace. */
3488 rc = lfsck_namespace_insert_normal(env, com, parent, child,
3490 if (unlikely(rc == -ESTALE))
3491 /* It may happen when the remote object has been
3492 * removed, but the local MDT is not aware of that. */
3495 if (unlikely(rc == -EEXIST))
3496 /* Unfortunately, someone reused the name under the
3497 * parent by race. So we have to remove the linkEA
3498 * entry from current child object. It means that the
3499 * LFSCK cannot recover the system totally back to
3500 * its original status, but it is necessary to make
3501 * the current system to be consistent.
3503 * It also may be because of the LFSCK found some
3504 * internal status of create operation. Under such
3505 * case, nothing to be done. */
3506 rc = lfsck_namespace_shrink_linkea_cond(env, com,
3507 parent, child, &ldata, cname, pfid);
3509 linkea_next_entry(&ldata);
3511 lfsck_object_put(env, parent);
3522 if (rc < 0 && rc != -ENODATA)
3526 LASSERT(ldata.ld_leh != NULL);
3528 count = ldata.ld_leh->leh_reccount;
3531 /* If the LFSCK is marked as LF_INCOMPLETE, then means some
3532 * MDT has ever tried to verify some remote MDT-object that
3533 * resides on this MDT, but this MDT failed to respond such
3534 * request. So means there may be some remote name entry on
3535 * other MDT that references this object with another name,
3536 * so we cannot know whether this linkEA is valid or not.
3537 * So keep it there and maybe resolved when next LFSCK run. */
3538 if (count == 0 && !(ns->ln_flags & LF_INCOMPLETE)) {
3539 /* If the child becomes orphan, then insert it into
3540 * the global .lustre/lost+found/MDTxxxx directory. */
3541 rc = lfsck_namespace_insert_orphan(env, com, child, "", "O",
3547 ns->ln_mul_ref_repaired++;
3552 rc = dt_attr_get(env, child, la, BYPASS_CAPA);
3556 if (la->la_nlink != count) {
3557 rc = lfsck_namespace_repair_nlink(env, com, child,
3560 ns->ln_objs_nlink_repaired++;
3566 if (la->la_nlink > 1)
3567 ns->ln_mul_linked_repaired++;
3576 static void lfsck_namespace_dump_statistics(struct seq_file *m,
3577 struct lfsck_namespace *ns,
3578 __u64 checked_phase1,
3579 __u64 checked_phase2,
3583 seq_printf(m, "checked_phase1: "LPU64"\n"
3584 "checked_phase2: "LPU64"\n"
3585 "updated_phase1: "LPU64"\n"
3586 "updated_phase2: "LPU64"\n"
3587 "failed_phase1: "LPU64"\n"
3588 "failed_phase2: "LPU64"\n"
3589 "directories: "LPU64"\n"
3590 "dirent_repaired: "LPU64"\n"
3591 "linkea_repaired: "LPU64"\n"
3592 "nlinks_repaired: "LPU64"\n"
3593 "multiple_linked_checked: "LPU64"\n"
3594 "multiple_linked_repaired: "LPU64"\n"
3595 "unknown_inconsistency: "LPU64"\n"
3596 "unmatched_pairs_repaired: "LPU64"\n"
3597 "dangling_repaired: "LPU64"\n"
3598 "multiple_referenced_repaired: "LPU64"\n"
3599 "bad_file_type_repaired: "LPU64"\n"
3600 "lost_dirent_repaired: "LPU64"\n"
3601 "local_lost_found_scanned: "LPU64"\n"
3602 "local_lost_found_moved: "LPU64"\n"
3603 "local_lost_found_skipped: "LPU64"\n"
3604 "local_lost_found_failed: "LPU64"\n"
3605 "striped_dirs_scanned: "LPU64"\n"
3606 "striped_dirs_repaired: "LPU64"\n"
3607 "striped_dirs_failed: "LPU64"\n"
3608 "striped_dirs_disabled: "LPU64"\n"
3609 "striped_dirs_skipped: "LPU64"\n"
3610 "striped_shards_scanned: "LPU64"\n"
3611 "striped_shards_repaired: "LPU64"\n"
3612 "striped_shards_failed: "LPU64"\n"
3613 "striped_shards_skipped: "LPU64"\n"
3614 "name_hash_repaired: "LPU64"\n"
3615 "success_count: %u\n"
3616 "run_time_phase1: %u seconds\n"
3617 "run_time_phase2: %u seconds\n",
3620 ns->ln_items_repaired,
3621 ns->ln_objs_repaired_phase2,
3622 ns->ln_items_failed,
3623 ns->ln_objs_failed_phase2,
3624 ns->ln_dirs_checked,
3625 ns->ln_dirent_repaired,
3626 ns->ln_linkea_repaired,
3627 ns->ln_objs_nlink_repaired,
3628 ns->ln_mul_linked_checked,
3629 ns->ln_mul_linked_repaired,
3630 ns->ln_unknown_inconsistency,
3631 ns->ln_unmatched_pairs_repaired,
3632 ns->ln_dangling_repaired,
3633 ns->ln_mul_ref_repaired,
3634 ns->ln_bad_type_repaired,
3635 ns->ln_lost_dirent_repaired,
3636 ns->ln_local_lpf_scanned,
3637 ns->ln_local_lpf_moved,
3638 ns->ln_local_lpf_skipped,
3639 ns->ln_local_lpf_failed,
3640 ns->ln_striped_dirs_scanned,
3641 ns->ln_striped_dirs_repaired,
3642 ns->ln_striped_dirs_failed,
3643 ns->ln_striped_dirs_disabled,
3644 ns->ln_striped_dirs_skipped,
3645 ns->ln_striped_shards_scanned,
3646 ns->ln_striped_shards_repaired,
3647 ns->ln_striped_shards_failed,
3648 ns->ln_striped_shards_skipped,
3649 ns->ln_name_hash_repaired,
3650 ns->ln_success_count,
3655 static void lfsck_namespace_release_lmv(const struct lu_env *env,
3656 struct lfsck_component *com)
3658 struct lfsck_instance *lfsck = com->lc_lfsck;
3659 struct lfsck_namespace *ns = com->lc_file_ram;
3661 while (!list_empty(&lfsck->li_list_lmv)) {
3662 struct lfsck_lmv_unit *llu;
3663 struct lfsck_lmv *llmv;
3665 llu = list_entry(lfsck->li_list_lmv.next,
3666 struct lfsck_lmv_unit, llu_link);
3667 llmv = &llu->llu_lmv;
3669 LASSERTF(atomic_read(&llmv->ll_ref) == 1,
3670 "still in using: %u\n",
3671 atomic_read(&llmv->ll_ref));
3673 ns->ln_striped_dirs_skipped++;
3674 lfsck_lmv_put(env, llmv);
3678 /* namespace APIs */
3680 static int lfsck_namespace_reset(const struct lu_env *env,
3681 struct lfsck_component *com, bool init)
3683 struct lfsck_instance *lfsck = com->lc_lfsck;
3684 struct lfsck_namespace *ns = com->lc_file_ram;
3685 struct lfsck_assistant_data *lad = com->lc_data;
3686 struct dt_object *root;
3687 struct dt_object *dto;
3691 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
3693 GOTO(log, rc = PTR_ERR(root));
3695 if (unlikely(!dt_try_as_dir(env, root)))
3696 GOTO(put, rc = -ENOTDIR);
3698 down_write(&com->lc_sem);
3700 memset(ns, 0, sizeof(*ns));
3702 __u32 count = ns->ln_success_count;
3703 __u64 last_time = ns->ln_time_last_complete;
3705 memset(ns, 0, sizeof(*ns));
3706 ns->ln_success_count = count;
3707 ns->ln_time_last_complete = last_time;
3709 ns->ln_magic = LFSCK_NAMESPACE_MAGIC;
3710 ns->ln_status = LS_INIT;
3712 rc = local_object_unlink(env, lfsck->li_bottom, root,
3713 lfsck_namespace_name);
3717 lfsck_object_put(env, com->lc_obj);
3719 dto = local_index_find_or_create(env, lfsck->li_los, root,
3720 lfsck_namespace_name,
3721 S_IFREG | S_IRUGO | S_IWUSR,
3722 &dt_lfsck_features);
3724 GOTO(out, rc = PTR_ERR(dto));
3727 rc = dto->do_ops->do_index_try(env, dto, &dt_lfsck_features);
3731 lad->lad_incomplete = 0;
3732 CFS_RESET_BITMAP(lad->lad_bitmap);
3734 rc = lfsck_namespace_store(env, com);
3739 up_write(&com->lc_sem);
3742 lu_object_put(env, &root->do_lu);
3744 CDEBUG(D_LFSCK, "%s: namespace LFSCK reset: rc = %d\n",
3745 lfsck_lfsck2name(lfsck), rc);
3750 lfsck_namespace_fail(const struct lu_env *env, struct lfsck_component *com,
3753 struct lfsck_namespace *ns = com->lc_file_ram;
3755 down_write(&com->lc_sem);
3757 com->lc_new_checked++;
3758 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
3759 up_write(&com->lc_sem);
3762 static void lfsck_namespace_close_dir(const struct lu_env *env,
3763 struct lfsck_component *com)
3765 struct lfsck_namespace *ns = com->lc_file_ram;
3766 struct lfsck_assistant_data *lad = com->lc_data;
3767 struct lfsck_instance *lfsck = com->lc_lfsck;
3768 struct lfsck_lmv *llmv = lfsck->li_lmv;
3769 struct lfsck_namespace_req *lnr;
3771 sizeof(*lnr) + LFSCK_TMPBUF_LEN;
3772 bool wakeup = false;
3778 OBD_ALLOC(lnr, size);
3780 ns->ln_striped_dirs_skipped++;
3785 /* Generate a dummy request to indicate that all shards' name entry
3786 * in this striped directory has been scanned for the first time. */
3787 INIT_LIST_HEAD(&lnr->lnr_lar.lar_list);
3788 lnr->lnr_obj = lfsck_object_get(lfsck->li_obj_dir);
3789 lnr->lnr_lmv = lfsck_lmv_get(llmv);
3790 lnr->lnr_fid = *lfsck_dto2fid(lfsck->li_obj_dir);
3791 lnr->lnr_oit_cookie = lfsck->li_pos_current.lp_oit_cookie;
3792 lnr->lnr_dir_cookie = MDS_DIR_END_OFF;
3793 lnr->lnr_size = size;
3795 spin_lock(&lad->lad_lock);
3796 if (lad->lad_assistant_status < 0) {
3797 spin_unlock(&lad->lad_lock);
3798 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
3799 ns->ln_striped_dirs_skipped++;
3804 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
3805 if (lad->lad_prefetched == 0)
3808 lad->lad_prefetched++;
3809 spin_unlock(&lad->lad_lock);
3811 wake_up_all(&lad->lad_thread.t_ctl_waitq);
3816 static int lfsck_namespace_open_dir(const struct lu_env *env,
3817 struct lfsck_component *com)
3819 struct lfsck_instance *lfsck = com->lc_lfsck;
3820 struct lfsck_namespace *ns = com->lc_file_ram;
3821 struct lfsck_lmv *llmv = lfsck->li_lmv;
3828 if (llmv->ll_lmv_master) {
3829 struct lmv_mds_md_v1 *lmv = &llmv->ll_lmv;
3831 if (lmv->lmv_master_mdt_index !=
3832 lfsck_dev_idx(lfsck->li_bottom)) {
3833 lmv->lmv_master_mdt_index =
3834 lfsck_dev_idx(lfsck->li_bottom);
3835 ns->ln_flags |= LF_INCONSISTENT;
3836 llmv->ll_lmv_updated = 1;
3839 rc = lfsck_namespace_verify_stripe_slave(env, com,
3840 lfsck->li_obj_dir, llmv);
3843 RETURN(rc > 0 ? 0 : rc);
3846 static int lfsck_namespace_checkpoint(const struct lu_env *env,
3847 struct lfsck_component *com, bool init)
3849 struct lfsck_instance *lfsck = com->lc_lfsck;
3850 struct lfsck_namespace *ns = com->lc_file_ram;
3854 rc = lfsck_checkpoint_generic(env, com);
3859 down_write(&com->lc_sem);
3861 ns->ln_pos_latest_start = lfsck->li_pos_checkpoint;
3863 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
3864 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
3865 HALF_SEC - lfsck->li_time_last_checkpoint);
3866 ns->ln_time_last_checkpoint = cfs_time_current_sec();
3867 ns->ln_items_checked += com->lc_new_checked;
3868 com->lc_new_checked = 0;
3871 rc = lfsck_namespace_store(env, com);
3872 up_write(&com->lc_sem);
3875 CDEBUG(D_LFSCK, "%s: namespace LFSCK checkpoint at the pos ["LPU64
3876 ", "DFID", "LPX64"]: rc = %d\n", lfsck_lfsck2name(lfsck),
3877 lfsck->li_pos_current.lp_oit_cookie,
3878 PFID(&lfsck->li_pos_current.lp_dir_parent),
3879 lfsck->li_pos_current.lp_dir_cookie, rc);
3881 return rc > 0 ? 0 : rc;
3884 static int lfsck_namespace_prep(const struct lu_env *env,
3885 struct lfsck_component *com,
3886 struct lfsck_start_param *lsp)
3888 struct lfsck_instance *lfsck = com->lc_lfsck;
3889 struct lfsck_namespace *ns = com->lc_file_ram;
3890 struct lfsck_position *pos = &com->lc_pos_start;
3893 rc = lfsck_namespace_load_bitmap(env, com);
3894 if (rc != 0 || ns->ln_status == LS_COMPLETED) {
3895 rc = lfsck_namespace_reset(env, com, false);
3897 rc = lfsck_set_param(env, lfsck, lsp->lsp_start, true);
3900 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep failed: "
3901 "rc = %d\n", lfsck_lfsck2name(lfsck), rc);
3907 down_write(&com->lc_sem);
3908 ns->ln_time_latest_start = cfs_time_current_sec();
3909 spin_lock(&lfsck->li_lock);
3911 if (ns->ln_flags & LF_SCANNED_ONCE) {
3912 if (!lfsck->li_drop_dryrun ||
3913 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3914 ns->ln_status = LS_SCANNING_PHASE2;
3915 list_move_tail(&com->lc_link,
3916 &lfsck->li_list_double_scan);
3917 if (!list_empty(&com->lc_link_dir))
3918 list_del_init(&com->lc_link_dir);
3919 lfsck_pos_set_zero(pos);
3921 ns->ln_status = LS_SCANNING_PHASE1;
3922 ns->ln_run_time_phase1 = 0;
3923 ns->ln_run_time_phase2 = 0;
3924 ns->ln_items_checked = 0;
3925 ns->ln_items_repaired = 0;
3926 ns->ln_items_failed = 0;
3927 ns->ln_dirs_checked = 0;
3928 ns->ln_objs_checked_phase2 = 0;
3929 ns->ln_objs_repaired_phase2 = 0;
3930 ns->ln_objs_failed_phase2 = 0;
3931 ns->ln_objs_nlink_repaired = 0;
3932 ns->ln_dirent_repaired = 0;
3933 ns->ln_linkea_repaired = 0;
3934 ns->ln_mul_linked_checked = 0;
3935 ns->ln_mul_linked_repaired = 0;
3936 ns->ln_unknown_inconsistency = 0;
3937 ns->ln_unmatched_pairs_repaired = 0;
3938 ns->ln_dangling_repaired = 0;
3939 ns->ln_mul_ref_repaired = 0;
3940 ns->ln_bad_type_repaired = 0;
3941 ns->ln_lost_dirent_repaired = 0;
3942 ns->ln_striped_dirs_scanned = 0;
3943 ns->ln_striped_dirs_repaired = 0;
3944 ns->ln_striped_dirs_failed = 0;
3945 ns->ln_striped_dirs_disabled = 0;
3946 ns->ln_striped_dirs_skipped = 0;
3947 ns->ln_striped_shards_scanned = 0;
3948 ns->ln_striped_shards_repaired = 0;
3949 ns->ln_striped_shards_failed = 0;
3950 ns->ln_striped_shards_skipped = 0;
3951 ns->ln_name_hash_repaired = 0;
3952 fid_zero(&ns->ln_fid_latest_scanned_phase2);
3953 if (list_empty(&com->lc_link_dir))
3954 list_add_tail(&com->lc_link_dir,
3955 &lfsck->li_list_dir);
3956 *pos = ns->ln_pos_first_inconsistent;
3959 ns->ln_status = LS_SCANNING_PHASE1;
3960 if (list_empty(&com->lc_link_dir))
3961 list_add_tail(&com->lc_link_dir,
3962 &lfsck->li_list_dir);
3963 if (!lfsck->li_drop_dryrun ||
3964 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent)) {
3965 *pos = ns->ln_pos_last_checkpoint;
3966 pos->lp_oit_cookie++;
3968 *pos = ns->ln_pos_first_inconsistent;
3972 spin_unlock(&lfsck->li_lock);
3973 up_write(&com->lc_sem);
3975 rc = lfsck_start_assistant(env, com, lsp);
3977 CDEBUG(D_LFSCK, "%s: namespace LFSCK prep done, start pos ["LPU64", "
3978 DFID", "LPX64"]: rc = %d\n",
3979 lfsck_lfsck2name(lfsck), pos->lp_oit_cookie,
3980 PFID(&pos->lp_dir_parent), pos->lp_dir_cookie, rc);
3985 static int lfsck_namespace_exec_oit(const struct lu_env *env,
3986 struct lfsck_component *com,
3987 struct dt_object *obj)
3989 struct lfsck_thread_info *info = lfsck_env_info(env);
3990 struct lfsck_namespace *ns = com->lc_file_ram;
3991 struct lfsck_instance *lfsck = com->lc_lfsck;
3992 const struct lu_fid *fid = lfsck_dto2fid(obj);
3993 struct lu_attr *la = &info->lti_la;
3994 struct lu_fid *pfid = &info->lti_fid2;
3995 struct lu_name *cname = &info->lti_name;
3996 struct lu_seq_range *range = &info->lti_range;
3997 struct dt_device *dev = lfsck->li_bottom;
3998 struct seq_server_site *ss =
3999 lu_site2seq(dev->dd_lu_dev.ld_site);
4000 struct linkea_data ldata = { 0 };
4001 __u32 idx = lfsck_dev_idx(dev);
4005 rc = lfsck_links_read(env, obj, &ldata);
4009 /* -EINVAL means crashed linkEA, should be verified. */
4010 if (rc == -EINVAL) {
4011 rc = lfsck_namespace_trace_update(env, com, fid,
4012 LNTF_CHECK_LINKEA, true);
4014 struct lustre_handle lh = { 0 };
4016 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
4017 MDS_INODELOCK_UPDATE |
4018 MDS_INODELOCK_XATTR, LCK_EX);
4020 rc = lfsck_namespace_links_remove(env, com,
4022 lfsck_ibits_unlock(&lh, LCK_EX);
4026 GOTO(out, rc = (rc == -ENOENT ? 0 : rc));
4029 /* zero-linkEA object may be orphan, but it also maybe because
4030 * of upgrading. Currently, we cannot record it for double scan.
4031 * Because it may cause the LFSCK trace file to be too large. */
4032 if (rc == -ENODATA) {
4033 if (S_ISDIR(lfsck_object_type(obj)))
4036 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
4040 /* "la_ctime" == 1 means that it has ever been removed from
4041 * backend /lost+found directory but not been added back to
4042 * the normal namespace yet. */
4043 if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
4044 rc = lfsck_namespace_trace_update(env, com, fid,
4045 LNTF_CHECK_LINKEA, true);
4053 /* Record multiple-linked object. */
4054 if (ldata.ld_leh->leh_reccount > 1) {
4055 rc = lfsck_namespace_trace_update(env, com, fid,
4056 LNTF_CHECK_LINKEA, true);
4061 linkea_first_entry(&ldata);
4062 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
4063 if (!fid_is_sane(pfid)) {
4064 rc = lfsck_namespace_trace_update(env, com, fid,
4065 LNTF_CHECK_PARENT, true);
4067 fld_range_set_mdt(range);
4068 rc = fld_local_lookup(env, ss->ss_server_fld,
4069 fid_seq(pfid), range);
4070 if ((rc == -ENOENT) ||
4071 (rc == 0 && range->lsr_index != idx)) {
4072 rc = lfsck_namespace_trace_update(env, com, fid,
4073 LNTF_CHECK_LINKEA, true);
4075 if (S_ISDIR(lfsck_object_type(obj)))
4078 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
4082 /* "la_ctime" == 1 means that it has ever been
4083 * removed from backend /lost+found directory but
4084 * not been added back to the normal namespace yet. */
4085 if (la->la_nlink > 1 || unlikely(la->la_ctime == 1))
4086 rc = lfsck_namespace_trace_update(env, com,
4087 fid, LNTF_CHECK_LINKEA, true);
4094 down_write(&com->lc_sem);
4095 com->lc_new_checked++;
4096 if (S_ISDIR(lfsck_object_type(obj)))
4097 ns->ln_dirs_checked++;
4099 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4100 up_write(&com->lc_sem);
4105 static int lfsck_namespace_exec_dir(const struct lu_env *env,
4106 struct lfsck_component *com,
4107 struct lu_dirent *ent, __u16 type)
4109 struct lfsck_assistant_data *lad = com->lc_data;
4110 struct lfsck_namespace_req *lnr;
4111 bool wakeup = false;
4113 lnr = lfsck_namespace_assistant_req_init(com->lc_lfsck, ent, type);
4115 struct lfsck_namespace *ns = com->lc_file_ram;
4117 lfsck_namespace_record_failure(env, com->lc_lfsck, ns);
4118 return PTR_ERR(lnr);
4121 spin_lock(&lad->lad_lock);
4122 if (lad->lad_assistant_status < 0) {
4123 spin_unlock(&lad->lad_lock);
4124 lfsck_namespace_assistant_req_fini(env, &lnr->lnr_lar);
4125 return lad->lad_assistant_status;
4128 list_add_tail(&lnr->lnr_lar.lar_list, &lad->lad_req_list);
4129 if (lad->lad_prefetched == 0)
4132 lad->lad_prefetched++;
4133 spin_unlock(&lad->lad_lock);
4135 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4137 down_write(&com->lc_sem);
4138 com->lc_new_checked++;
4139 up_write(&com->lc_sem);
4144 static int lfsck_namespace_post(const struct lu_env *env,
4145 struct lfsck_component *com,
4146 int result, bool init)
4148 struct lfsck_instance *lfsck = com->lc_lfsck;
4149 struct lfsck_namespace *ns = com->lc_file_ram;
4153 lfsck_post_generic(env, com, &result);
4155 down_write(&com->lc_sem);
4156 lfsck_namespace_release_lmv(env, com);
4158 spin_lock(&lfsck->li_lock);
4160 ns->ln_pos_last_checkpoint = lfsck->li_pos_checkpoint;
4162 ns->ln_status = LS_SCANNING_PHASE2;
4163 ns->ln_flags |= LF_SCANNED_ONCE;
4164 ns->ln_flags &= ~LF_UPGRADE;
4165 list_del_init(&com->lc_link_dir);
4166 list_move_tail(&com->lc_link, &lfsck->li_list_double_scan);
4167 } else if (result == 0) {
4168 if (lfsck->li_status != 0)
4169 ns->ln_status = lfsck->li_status;
4171 ns->ln_status = LS_STOPPED;
4172 if (ns->ln_status != LS_PAUSED) {
4173 list_del_init(&com->lc_link_dir);
4174 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4177 ns->ln_status = LS_FAILED;
4178 list_del_init(&com->lc_link_dir);
4179 list_move_tail(&com->lc_link, &lfsck->li_list_idle);
4181 spin_unlock(&lfsck->li_lock);
4184 ns->ln_run_time_phase1 += cfs_duration_sec(cfs_time_current() +
4185 HALF_SEC - lfsck->li_time_last_checkpoint);
4186 ns->ln_time_last_checkpoint = cfs_time_current_sec();
4187 ns->ln_items_checked += com->lc_new_checked;
4188 com->lc_new_checked = 0;
4191 rc = lfsck_namespace_store(env, com);
4192 up_write(&com->lc_sem);
4194 CDEBUG(D_LFSCK, "%s: namespace LFSCK post done: rc = %d\n",
4195 lfsck_lfsck2name(lfsck), rc);
4201 lfsck_namespace_dump(const struct lu_env *env, struct lfsck_component *com,
4204 struct lfsck_instance *lfsck = com->lc_lfsck;
4205 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4206 struct lfsck_namespace *ns = com->lc_file_ram;
4209 down_read(&com->lc_sem);
4210 seq_printf(m, "name: lfsck_namespace\n"
4216 lfsck_status2names(ns->ln_status));
4218 rc = lfsck_bits_dump(m, ns->ln_flags, lfsck_flags_names, "flags");
4222 rc = lfsck_bits_dump(m, bk->lb_param, lfsck_param_names, "param");
4226 rc = lfsck_time_dump(m, ns->ln_time_last_complete,
4227 "time_since_last_completed");
4231 rc = lfsck_time_dump(m, ns->ln_time_latest_start,
4232 "time_since_latest_start");
4236 rc = lfsck_time_dump(m, ns->ln_time_last_checkpoint,
4237 "time_since_last_checkpoint");
4241 rc = lfsck_pos_dump(m, &ns->ln_pos_latest_start,
4242 "latest_start_position");
4246 rc = lfsck_pos_dump(m, &ns->ln_pos_last_checkpoint,
4247 "last_checkpoint_position");
4251 rc = lfsck_pos_dump(m, &ns->ln_pos_first_inconsistent,
4252 "first_failure_position");
4256 if (ns->ln_status == LS_SCANNING_PHASE1) {
4257 struct lfsck_position pos;
4258 const struct dt_it_ops *iops;
4259 cfs_duration_t duration = cfs_time_current() -
4260 lfsck->li_time_last_checkpoint;
4261 __u64 checked = ns->ln_items_checked + com->lc_new_checked;
4262 __u64 speed = checked;
4263 __u64 new_checked = msecs_to_jiffies(com->lc_new_checked *
4265 __u32 rtime = ns->ln_run_time_phase1 +
4266 cfs_duration_sec(duration + HALF_SEC);
4269 do_div(new_checked, duration);
4271 do_div(speed, rtime);
4272 lfsck_namespace_dump_statistics(m, ns, checked,
4273 ns->ln_objs_checked_phase2,
4274 rtime, ns->ln_run_time_phase2);
4276 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4277 "average_speed_phase2: N/A\n"
4278 "real_time_speed_phase1: "LPU64" items/sec\n"
4279 "real_time_speed_phase2: N/A\n",
4283 LASSERT(lfsck->li_di_oit != NULL);
4285 iops = &lfsck->li_obj_oit->do_index_ops->dio_it;
4287 /* The low layer otable-based iteration position may NOT
4288 * exactly match the namespace-based directory traversal
4289 * cookie. Generally, it is not a serious issue. But the
4290 * caller should NOT make assumption on that. */
4291 pos.lp_oit_cookie = iops->store(env, lfsck->li_di_oit);
4292 if (!lfsck->li_current_oit_processed)
4293 pos.lp_oit_cookie--;
4295 spin_lock(&lfsck->li_lock);
4296 if (lfsck->li_di_dir != NULL) {
4297 pos.lp_dir_cookie = lfsck->li_cookie_dir;
4298 if (pos.lp_dir_cookie >= MDS_DIR_END_OFF) {
4299 fid_zero(&pos.lp_dir_parent);
4300 pos.lp_dir_cookie = 0;
4303 *lfsck_dto2fid(lfsck->li_obj_dir);
4306 fid_zero(&pos.lp_dir_parent);
4307 pos.lp_dir_cookie = 0;
4309 spin_unlock(&lfsck->li_lock);
4310 lfsck_pos_dump(m, &pos, "current_position");
4311 } else if (ns->ln_status == LS_SCANNING_PHASE2) {
4312 cfs_duration_t duration = cfs_time_current() -
4313 lfsck->li_time_last_checkpoint;
4314 __u64 checked = ns->ln_objs_checked_phase2 +
4315 com->lc_new_checked;
4316 __u64 speed1 = ns->ln_items_checked;
4317 __u64 speed2 = checked;
4318 __u64 new_checked = msecs_to_jiffies(com->lc_new_checked *
4320 __u32 rtime = ns->ln_run_time_phase2 +
4321 cfs_duration_sec(duration + HALF_SEC);
4324 do_div(new_checked, duration);
4325 if (ns->ln_run_time_phase1 != 0)
4326 do_div(speed1, ns->ln_run_time_phase1);
4328 do_div(speed2, rtime);
4329 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4331 ns->ln_run_time_phase1, rtime);
4333 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4334 "average_speed_phase2: "LPU64" objs/sec\n"
4335 "real_time_speed_phase1: N/A\n"
4336 "real_time_speed_phase2: "LPU64" objs/sec\n"
4337 "current_position: "DFID"\n",
4341 PFID(&ns->ln_fid_latest_scanned_phase2));
4343 __u64 speed1 = ns->ln_items_checked;
4344 __u64 speed2 = ns->ln_objs_checked_phase2;
4346 if (ns->ln_run_time_phase1 != 0)
4347 do_div(speed1, ns->ln_run_time_phase1);
4348 if (ns->ln_run_time_phase2 != 0)
4349 do_div(speed2, ns->ln_run_time_phase2);
4350 lfsck_namespace_dump_statistics(m, ns, ns->ln_items_checked,
4351 ns->ln_objs_checked_phase2,
4352 ns->ln_run_time_phase1,
4353 ns->ln_run_time_phase2);
4355 seq_printf(m, "average_speed_phase1: "LPU64" items/sec\n"
4356 "average_speed_phase2: "LPU64" objs/sec\n"
4357 "real_time_speed_phase1: N/A\n"
4358 "real_time_speed_phase2: N/A\n"
4359 "current_position: N/A\n",
4364 up_read(&com->lc_sem);
4368 static int lfsck_namespace_double_scan(const struct lu_env *env,
4369 struct lfsck_component *com)
4371 struct lfsck_namespace *ns = com->lc_file_ram;
4372 struct lfsck_assistant_data *lad = com->lc_data;
4373 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4374 struct lfsck_tgt_desc *ltd;
4375 struct lfsck_tgt_desc *next;
4378 rc = lfsck_double_scan_generic(env, com, ns->ln_status);
4379 if (thread_is_stopped(&lad->lad_thread)) {
4380 LASSERT(list_empty(&lad->lad_req_list));
4381 LASSERT(list_empty(&lad->lad_mdt_phase1_list));
4383 spin_lock(<ds->ltd_lock);
4384 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4385 ltd_namespace_phase_list) {
4386 list_del_init(<d->ltd_namespace_phase_list);
4388 spin_unlock(<ds->ltd_lock);
4394 static void lfsck_namespace_data_release(const struct lu_env *env,
4395 struct lfsck_component *com)
4397 struct lfsck_assistant_data *lad = com->lc_data;
4398 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4399 struct lfsck_tgt_desc *ltd;
4400 struct lfsck_tgt_desc *next;
4402 LASSERT(lad != NULL);
4403 LASSERT(thread_is_init(&lad->lad_thread) ||
4404 thread_is_stopped(&lad->lad_thread));
4405 LASSERT(list_empty(&lad->lad_req_list));
4407 com->lc_data = NULL;
4408 lfsck_namespace_release_lmv(env, com);
4410 spin_lock(<ds->ltd_lock);
4411 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4412 ltd_namespace_phase_list) {
4413 list_del_init(<d->ltd_namespace_phase_list);
4415 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4416 ltd_namespace_phase_list) {
4417 list_del_init(<d->ltd_namespace_phase_list);
4419 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_list,
4420 ltd_namespace_list) {
4421 list_del_init(<d->ltd_namespace_list);
4423 spin_unlock(<ds->ltd_lock);
4425 if (likely(lad->lad_bitmap != NULL))
4426 CFS_FREE_BITMAP(lad->lad_bitmap);
4431 static void lfsck_namespace_quit(const struct lu_env *env,
4432 struct lfsck_component *com)
4434 struct lfsck_assistant_data *lad = com->lc_data;
4435 struct lfsck_tgt_descs *ltds = &com->lc_lfsck->li_mdt_descs;
4436 struct lfsck_tgt_desc *ltd;
4437 struct lfsck_tgt_desc *next;
4439 LASSERT(lad != NULL);
4441 lfsck_quit_generic(env, com);
4443 LASSERT(thread_is_init(&lad->lad_thread) ||
4444 thread_is_stopped(&lad->lad_thread));
4445 LASSERT(list_empty(&lad->lad_req_list));
4447 lfsck_namespace_release_lmv(env, com);
4449 spin_lock(<ds->ltd_lock);
4450 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase1_list,
4451 ltd_namespace_phase_list) {
4452 list_del_init(<d->ltd_namespace_phase_list);
4454 list_for_each_entry_safe(ltd, next, &lad->lad_mdt_phase2_list,
4455 ltd_namespace_phase_list) {
4456 list_del_init(<d->ltd_namespace_phase_list);
4458 spin_unlock(<ds->ltd_lock);
4461 static int lfsck_namespace_in_notify(const struct lu_env *env,
4462 struct lfsck_component *com,
4463 struct lfsck_request *lr,
4466 struct lfsck_instance *lfsck = com->lc_lfsck;
4467 struct lfsck_namespace *ns = com->lc_file_ram;
4468 struct lfsck_assistant_data *lad = com->lc_data;
4469 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
4470 struct lfsck_tgt_desc *ltd;
4475 switch (lr->lr_event) {
4476 case LE_CREATE_ORPHAN: {
4477 struct dt_object *orphan = NULL;
4478 struct lmv_mds_md_v1 *lmv;
4480 CDEBUG(D_LFSCK, "%s: namespace LFSCK handling notify from "
4481 "MDT %x to create orphan"DFID" with type %o\n",
4482 lfsck_lfsck2name(lfsck), lr->lr_index,
4483 PFID(&lr->lr_fid), lr->lr_type);
4485 orphan = lfsck_object_find(env, lfsck, &lr->lr_fid);
4487 GOTO(out_create, rc = PTR_ERR(orphan));
4489 if (dt_object_exists(orphan))
4490 GOTO(out_create, rc = -EEXIST);
4492 if (lr->lr_stripe_count > 0) {
4493 lmv = &lfsck_env_info(env)->lti_lmv;
4494 memset(lmv, 0, sizeof(*lmv));
4495 lmv->lmv_hash_type = lr->lr_hash_type;
4496 lmv->lmv_stripe_count = lr->lr_stripe_count;
4497 lmv->lmv_layout_version = lr->lr_layout_version;
4498 memcpy(lmv->lmv_pool_name, lr->lr_pool_name,
4499 sizeof(lmv->lmv_pool_name));
4504 rc = lfsck_namespace_create_orphan_local(env, com, orphan,
4507 GOTO(out_create, rc = (rc == 1) ? 0 : rc);
4510 CDEBUG(D_LFSCK, "%s: namespace LFSCK handled notify from "
4511 "MDT %x to create orphan"DFID" with type %o: rc = %d\n",
4512 lfsck_lfsck2name(lfsck), lr->lr_index,
4513 PFID(&lr->lr_fid), lr->lr_type, rc);
4515 if (orphan != NULL && !IS_ERR(orphan))
4516 lfsck_object_put(env, orphan);
4520 case LE_SKIP_NLINK_DECLARE: {
4521 struct dt_object *obj = com->lc_obj;
4522 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4525 LASSERT(th != NULL);
4527 rc = dt_declare_delete(env, obj,
4528 (const struct dt_key *)key, th);
4530 rc = dt_declare_insert(env, obj,
4531 (const struct dt_rec *)&flags,
4532 (const struct dt_key *)key, th);
4536 case LE_SKIP_NLINK: {
4537 struct dt_object *obj = com->lc_obj;
4538 struct lu_fid *key = &lfsck_env_info(env)->lti_fid3;
4543 LASSERT(th != NULL);
4545 fid_cpu_to_be(key, &lr->lr_fid);
4546 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
4547 (const struct dt_key *)key, BYPASS_CAPA);
4549 if (flags & LNTF_SKIP_NLINK)
4553 } else if (rc != -ENOENT) {
4557 flags |= LNTF_SKIP_NLINK;
4559 rc = dt_delete(env, obj, (const struct dt_key *)key,
4565 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
4566 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
4571 CDEBUG(D_LFSCK, "%s: RPC service thread mark the "DFID
4572 " to be skipped for namespace double scan: rc = %d\n",
4573 lfsck_lfsck2name(com->lc_lfsck), PFID(&lr->lr_fid), rc);
4576 /* If we cannot record this object in the LFSCK tracing,
4577 * we have to mark the LFSC as LF_INCOMPLETE, then the
4578 * LFSCK will skip nlink attribute verification for
4580 ns->ln_flags |= LF_INCOMPLETE;
4584 case LE_SET_LMV_MASTER: {
4585 struct dt_object *obj;
4587 obj = lfsck_object_find_by_dev(env, lfsck->li_bottom,
4590 RETURN(PTR_ERR(obj));
4592 rc = lfsck_namespace_notify_lmv_master_local(env, com, obj);
4593 lfsck_object_put(env, obj);
4595 RETURN(rc > 0 ? 0 : rc);
4597 case LE_SET_LMV_SLAVE: {
4598 if (!(lr->lr_flags & LEF_RECHECK_NAME_HASH))
4599 ns->ln_striped_shards_repaired++;
4601 rc = lfsck_namespace_trace_update(env, com, &lr->lr_fid,
4602 LNTF_RECHECK_NAME_HASH, true);
4604 RETURN(rc > 0 ? 0 : rc);
4606 case LE_PHASE1_DONE:
4607 case LE_PHASE2_DONE:
4614 CDEBUG(D_LFSCK, "%s: namespace LFSCK handles notify %u from MDT %x, "
4615 "status %d, flags %x\n", lfsck_lfsck2name(lfsck), lr->lr_event,
4616 lr->lr_index, lr->lr_status, lr->lr_flags2);
4618 spin_lock(<ds->ltd_lock);
4619 ltd = LTD_TGT(ltds, lr->lr_index);
4621 spin_unlock(<ds->ltd_lock);
4626 list_del_init(<d->ltd_namespace_phase_list);
4627 switch (lr->lr_event) {
4628 case LE_PHASE1_DONE:
4629 if (lr->lr_status <= 0) {
4630 ltd->ltd_namespace_done = 1;
4631 list_del_init(<d->ltd_namespace_list);
4632 CDEBUG(D_LFSCK, "%s: MDT %x failed/stopped at "
4633 "phase1 for namespace LFSCK: rc = %d.\n",
4634 lfsck_lfsck2name(lfsck),
4635 ltd->ltd_index, lr->lr_status);
4636 ns->ln_flags |= LF_INCOMPLETE;
4641 if (lr->lr_flags2 & LF_INCOMPLETE)
4642 ns->ln_flags |= LF_INCOMPLETE;
4644 if (list_empty(<d->ltd_namespace_list))
4645 list_add_tail(<d->ltd_namespace_list,
4646 &lad->lad_mdt_list);
4647 list_add_tail(<d->ltd_namespace_phase_list,
4648 &lad->lad_mdt_phase2_list);
4650 case LE_PHASE2_DONE:
4651 ltd->ltd_namespace_done = 1;
4652 list_del_init(<d->ltd_namespace_list);
4656 ltd->ltd_namespace_done = 1;
4657 list_del_init(<d->ltd_namespace_list);
4658 if (!(lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT)) {
4660 "%s: the peer MDT %x exit namespace LFSCK\n",
4661 lfsck_lfsck2name(lfsck), ltd->ltd_index);
4662 ns->ln_flags |= LF_INCOMPLETE;
4668 spin_unlock(<ds->ltd_lock);
4670 if (fail && lfsck->li_bookmark_ram.lb_param & LPF_FAILOUT) {
4671 struct lfsck_stop *stop = &lfsck_env_info(env)->lti_stop;
4673 memset(stop, 0, sizeof(*stop));
4674 stop->ls_status = lr->lr_status;
4675 stop->ls_flags = lr->lr_param & ~LPF_BROADCAST;
4676 lfsck_stop(env, lfsck->li_bottom, stop);
4677 } else if (lfsck_phase2_next_ready(lad)) {
4678 wake_up_all(&lad->lad_thread.t_ctl_waitq);
4684 static int lfsck_namespace_query(const struct lu_env *env,
4685 struct lfsck_component *com)
4687 struct lfsck_namespace *ns = com->lc_file_ram;
4689 return ns->ln_status;
4692 static struct lfsck_operations lfsck_namespace_ops = {
4693 .lfsck_reset = lfsck_namespace_reset,
4694 .lfsck_fail = lfsck_namespace_fail,
4695 .lfsck_close_dir = lfsck_namespace_close_dir,
4696 .lfsck_open_dir = lfsck_namespace_open_dir,
4697 .lfsck_checkpoint = lfsck_namespace_checkpoint,
4698 .lfsck_prep = lfsck_namespace_prep,
4699 .lfsck_exec_oit = lfsck_namespace_exec_oit,
4700 .lfsck_exec_dir = lfsck_namespace_exec_dir,
4701 .lfsck_post = lfsck_namespace_post,
4702 .lfsck_dump = lfsck_namespace_dump,
4703 .lfsck_double_scan = lfsck_namespace_double_scan,
4704 .lfsck_data_release = lfsck_namespace_data_release,
4705 .lfsck_quit = lfsck_namespace_quit,
4706 .lfsck_in_notify = lfsck_namespace_in_notify,
4707 .lfsck_query = lfsck_namespace_query,
4711 * Repair dangling name entry.
4713 * For the name entry with dangling reference, we need to repare the
4714 * inconsistency according to the LFSCK sponsor's requirement:
4716 * 1) Keep the inconsistency there and report the inconsistency case,
4717 * then give the chance to the application to find related issues,
4718 * and the users can make the decision about how to handle it with
4719 * more human knownledge. (by default)
4721 * 2) Re-create the missing MDT-object with the FID information.
4723 * \param[in] env pointer to the thread context
4724 * \param[in] com pointer to the lfsck component
4725 * \param[in] child pointer to the object corresponding to the dangling
4727 * \param[in] lnr pointer to the namespace request that contains the
4728 * name's name, parent object, parent's LMV, and ect.
4730 * \retval positive number if no need to repair
4731 * \retval zero for repaired successfully
4732 * \retval negative error number on failure
4734 int lfsck_namespace_repair_dangling(const struct lu_env *env,
4735 struct lfsck_component *com,
4736 struct dt_object *child,
4737 struct lfsck_namespace_req *lnr)
4739 struct lfsck_thread_info *info = lfsck_env_info(env);
4740 struct lu_attr *la = &info->lti_la;
4741 struct dt_allocation_hint *hint = &info->lti_hint;
4742 struct dt_object_format *dof = &info->lti_dof;
4743 struct dt_insert_rec *rec = &info->lti_dt_rec;
4744 struct lmv_mds_md_v1 *lmv2 = &info->lti_lmv2;
4745 struct dt_object *parent = lnr->lnr_obj;
4746 const struct lu_name *cname;
4747 struct linkea_data ldata = { 0 };
4748 struct lustre_handle lh = { 0 };
4749 struct lu_buf linkea_buf;
4750 struct lu_buf lmv_buf;
4751 struct lfsck_instance *lfsck = com->lc_lfsck;
4752 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4753 struct dt_device *dev = lfsck_obj2dt_dev(child);
4754 struct thandle *th = NULL;
4756 __u16 type = lnr->lnr_type;
4760 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
4761 if (bk->lb_param & LPF_CREATE_MDTOBJ)
4766 if (!create || bk->lb_param & LPF_DRYRUN)
4769 rc = linkea_data_new(&ldata, &info->lti_linkea_buf2);
4773 rc = linkea_add_buf(&ldata, cname, lfsck_dto2fid(parent));
4777 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
4778 MDS_INODELOCK_UPDATE, LCK_EX);
4782 rc = lfsck_namespace_check_exist(env, parent, child, lnr->lnr_name);
4786 th = dt_trans_create(env, dev);
4788 GOTO(log, rc = PTR_ERR(th));
4790 /* Set the ctime as zero, then others can know it is created for
4791 * repairing dangling name entry by LFSCK. And if the LFSCK made
4792 * wrong decision and the real MDT-object has been found later,
4793 * then the LFSCK has chance to fix the incosistency properly. */
4794 memset(la, 0, sizeof(*la));
4795 la->la_mode = (type & S_IFMT) | 0600;
4796 la->la_valid = LA_TYPE | LA_MODE | LA_UID | LA_GID |
4797 LA_ATIME | LA_MTIME | LA_CTIME;
4799 child->do_ops->do_ah_init(env, hint, parent, child,
4800 la->la_mode & S_IFMT);
4802 memset(dof, 0, sizeof(*dof));
4803 dof->dof_type = dt_mode_to_dft(type);
4804 /* If the target is a regular file, then the LFSCK will only create
4805 * the MDT-object without stripes (dof->dof_reg.striped = 0). related
4806 * OST-objects will be created when write open. */
4808 /* 1a. create child. */
4809 rc = dt_declare_create(env, child, la, hint, dof, th);
4813 if (S_ISDIR(type)) {
4814 if (unlikely(!dt_try_as_dir(env, child)))
4815 GOTO(stop, rc = -ENOTDIR);
4817 /* 2a. insert dot into child dir */
4818 rec->rec_type = S_IFDIR;
4819 rec->rec_fid = lfsck_dto2fid(child);
4820 rc = dt_declare_insert(env, child,
4821 (const struct dt_rec *)rec,
4822 (const struct dt_key *)dot, th);
4826 /* 3a. insert dotdot into child dir */
4827 rec->rec_fid = lfsck_dto2fid(parent);
4828 rc = dt_declare_insert(env, child,
4829 (const struct dt_rec *)rec,
4830 (const struct dt_key *)dotdot, th);
4834 /* 4a. increase child nlink */
4835 rc = dt_declare_ref_add(env, child, th);
4839 /* 5a. generate slave LMV EA. */
4840 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4843 idx = lfsck_shard_name_to_index(env,
4844 lnr->lnr_name, lnr->lnr_namelen,
4845 type, lfsck_dto2fid(child));
4846 if (unlikely(idx < 0))
4847 GOTO(stop, rc = idx);
4849 *lmv2 = lnr->lnr_lmv->ll_lmv;
4850 lmv2->lmv_magic = LMV_MAGIC_STRIPE;
4851 lmv2->lmv_master_mdt_index = idx;
4853 lfsck_lmv_header_cpu_to_le(lmv2, lmv2);
4854 lfsck_buf_init(&lmv_buf, lmv2, sizeof(*lmv2));
4855 rc = dt_declare_xattr_set(env, child, &lmv_buf,
4856 XATTR_NAME_LMV, 0, th);
4862 /* 6a. insert linkEA for child */
4863 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
4864 ldata.ld_leh->leh_len);
4865 rc = dt_declare_xattr_set(env, child, &linkea_buf,
4866 XATTR_NAME_LINK, 0, th);
4870 rc = dt_trans_start(env, dev, th);
4872 GOTO(stop, rc = (rc == -EEXIST ? 1 : rc));
4874 dt_write_lock(env, child, 0);
4875 /* 1b. create child */
4876 rc = dt_create(env, child, la, hint, dof, th);
4878 GOTO(unlock, rc = (rc == -EEXIST ? 1 : rc));
4880 if (S_ISDIR(type)) {
4881 if (unlikely(!dt_try_as_dir(env, child)))
4882 GOTO(unlock, rc = -ENOTDIR);
4884 /* 2b. insert dot into child dir */
4885 rec->rec_type = S_IFDIR;
4886 rec->rec_fid = lfsck_dto2fid(child);
4887 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4888 (const struct dt_key *)dot, th, BYPASS_CAPA, 1);
4892 /* 3b. insert dotdot into child dir */
4893 rec->rec_fid = lfsck_dto2fid(parent);
4894 rc = dt_insert(env, child, (const struct dt_rec *)rec,
4895 (const struct dt_key *)dotdot, th,
4900 /* 4b. increase child nlink */
4901 rc = dt_ref_add(env, child, th);
4905 /* 5b. generate slave LMV EA. */
4906 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
4907 rc = dt_xattr_set(env, child, &lmv_buf, XATTR_NAME_LMV,
4908 0, th, BYPASS_CAPA);
4914 /* 6b. insert linkEA for child. */
4915 rc = dt_xattr_set(env, child, &linkea_buf,
4916 XATTR_NAME_LINK, 0, th, BYPASS_CAPA);
4921 dt_write_unlock(env, child);
4924 dt_trans_stop(env, dev, th);
4927 lfsck_ibits_unlock(&lh, LCK_EX);
4928 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant found dangling "
4929 "reference for: parent "DFID", child "DFID", type %u, "
4930 "name %s. %s: rc = %d\n", lfsck_lfsck2name(lfsck),
4931 PFID(lfsck_dto2fid(parent)), PFID(lfsck_dto2fid(child)),
4932 type, cname->ln_name,
4933 create ? "Create the lost OST-object as required" :
4934 "Keep the MDT-object there by default", rc);
4937 struct lfsck_namespace *ns = com->lc_file_ram;
4939 ns->ln_flags |= LF_INCONSISTENT;
4945 static int lfsck_namespace_assistant_handler_p1(const struct lu_env *env,
4946 struct lfsck_component *com,
4947 struct lfsck_assistant_req *lar)
4949 struct lfsck_thread_info *info = lfsck_env_info(env);
4950 struct lu_attr *la = &info->lti_la;
4951 struct lfsck_instance *lfsck = com->lc_lfsck;
4952 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
4953 struct lfsck_namespace *ns = com->lc_file_ram;
4954 struct linkea_data ldata = { 0 };
4955 const struct lu_name *cname;
4956 struct thandle *handle = NULL;
4957 struct lfsck_namespace_req *lnr =
4958 container_of0(lar, struct lfsck_namespace_req, lnr_lar);
4959 struct dt_object *dir = lnr->lnr_obj;
4960 struct dt_object *obj = NULL;
4961 const struct lu_fid *pfid = lfsck_dto2fid(dir);
4962 struct dt_device *dev = NULL;
4963 struct lustre_handle lh = { 0 };
4964 bool repaired = false;
4965 bool dtlocked = false;
4969 bool bad_hash = false;
4973 enum lfsck_namespace_inconsistency_type type = LNIT_NONE;
4976 if (lnr->lnr_attr & LUDA_UPGRADE) {
4977 ns->ln_flags |= LF_UPGRADE;
4978 ns->ln_dirent_repaired++;
4980 } else if (lnr->lnr_attr & LUDA_REPAIR) {
4981 ns->ln_flags |= LF_INCONSISTENT;
4982 ns->ln_dirent_repaired++;
4986 if (unlikely(fid_is_zero(&lnr->lnr_fid))) {
4987 if (strcmp(lnr->lnr_name, dotdot) != 0)
4990 rc = lfsck_namespace_trace_update(env, com, pfid,
4991 LNTF_CHECK_PARENT, true);
4996 if (unlikely(lnr->lnr_dir_cookie == MDS_DIR_END_OFF)) {
4997 rc = lfsck_namespace_striped_dir_rescan(env, com, lnr);
5002 if (lnr->lnr_name[0] == '.' &&
5003 (lnr->lnr_namelen == 1 || fid_seq_is_dot(fid_seq(&lnr->lnr_fid))))
5006 if (lnr->lnr_lmv != NULL && lnr->lnr_lmv->ll_lmv_master) {
5007 rc = lfsck_namespace_handle_striped_master(env, com, lnr);
5012 idx = lfsck_find_mdt_idx_by_fid(env, lfsck, &lnr->lnr_fid);
5014 GOTO(out, rc = idx);
5016 if (idx == lfsck_dev_idx(lfsck->li_bottom)) {
5017 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0))
5020 dev = lfsck->li_next;
5022 struct lfsck_tgt_desc *ltd;
5024 /* Usually, some local filesystem consistency verification
5025 * tools can guarantee the local namespace tree consistenct.
5026 * So the LFSCK will only verify the remote directory. */
5027 if (unlikely(strcmp(lnr->lnr_name, dotdot) == 0)) {
5028 rc = lfsck_namespace_trace_update(env, com, pfid,
5029 LNTF_CHECK_PARENT, true);
5034 ltd = LTD_TGT(&lfsck->li_mdt_descs, idx);
5035 if (unlikely(ltd == NULL)) {
5036 CDEBUG(D_LFSCK, "%s: cannot talk with MDT %x which "
5037 "did not join the namespace LFSCK\n",
5038 lfsck_lfsck2name(lfsck), idx);
5039 lfsck_lad_set_bitmap(env, com, idx);
5041 GOTO(out, rc = -ENODEV);
5047 obj = lfsck_object_find_by_dev(env, dev, &lnr->lnr_fid);
5049 GOTO(out, rc = PTR_ERR(obj));
5051 cname = lfsck_name_get_const(env, lnr->lnr_name, lnr->lnr_namelen);
5052 if (dt_object_exists(obj) == 0) {
5055 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5057 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5058 lnr->lnr_name, lnr->lnr_namelen)) {
5059 type = LNIT_BAD_DIRENT;
5064 type = LNIT_DANGLING;
5065 rc = lfsck_namespace_repair_dangling(env, com,
5074 if (!(bk->lb_param & LPF_DRYRUN) && repaired) {
5077 rc = lfsck_ibits_lock(env, lfsck, obj, &lh,
5078 MDS_INODELOCK_UPDATE |
5079 MDS_INODELOCK_XATTR, LCK_EX);
5083 handle = dt_trans_create(env, dev);
5085 GOTO(out, rc = PTR_ERR(handle));
5087 rc = lfsck_declare_namespace_exec_dir(env, obj, handle);
5091 rc = dt_trans_start(env, dev, handle);
5095 dt_write_lock(env, obj, 0);
5099 rc = lfsck_namespace_check_exist(env, dir, obj, lnr->lnr_name);
5103 rc = lfsck_links_read(env, obj, &ldata);
5104 if (unlikely(rc == -ENOENT)) {
5105 if (handle != NULL) {
5106 dt_write_unlock(env, obj);
5109 dt_trans_stop(env, dev, handle);
5112 lfsck_ibits_unlock(&lh, LCK_EX);
5115 /* It may happen when the remote object has been removed,
5116 * but the local MDT is not aware of that. */
5118 } else if (rc == 0) {
5119 count = ldata.ld_leh->leh_reccount;
5120 rc = linkea_links_find(&ldata, cname, pfid);
5122 (count == 1 || !S_ISDIR(lfsck_object_type(obj)))) {
5123 if ((lfsck_object_type(obj) & S_IFMT) !=
5125 ns->ln_flags |= LF_INCONSISTENT;
5126 type = LNIT_BAD_TYPE;
5132 ns->ln_flags |= LF_INCONSISTENT;
5134 /* If the name entry hash does not match the slave striped
5135 * directory, and the name entry does not match also, then
5136 * it is quite possible that name entry is corrupted. */
5137 if (!lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5138 lnr->lnr_name, lnr->lnr_namelen)) {
5139 type = LNIT_BAD_DIRENT;
5144 /* If the file type stored in the name entry does not match
5145 * the file type claimed by the object, and the object does
5146 * not recognize the name entry, then it is quite possible
5147 * that the name entry is corrupted. */
5148 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type) {
5149 type = LNIT_BAD_DIRENT;
5154 /* For sub-dir object, we cannot make sure whether the sub-dir
5155 * back references the parent via ".." name entry correctly or
5156 * not in the LFSCK first-stage scanning. It may be that the
5157 * (remote) sub-dir ".." name entry has no parent FID after
5158 * file-level backup/restore and its linkEA may be wrong.
5159 * So under such case, we should replace the linkEA according
5160 * to current name entry. But this needs to be done during the
5161 * LFSCK second-stage scanning. The LFSCK will record the name
5162 * entry for further possible using. */
5166 } else if (unlikely(rc == -EINVAL)) {
5167 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5168 type = LNIT_BAD_TYPE;
5171 ns->ln_flags |= LF_INCONSISTENT;
5172 /* The magic crashed, we are not sure whether there are more
5173 * corrupt data in the linkea, so remove all linkea entries. */
5177 } else if (rc == -ENODATA) {
5178 if ((lfsck_object_type(obj) & S_IFMT) != lnr->lnr_type)
5179 type = LNIT_BAD_TYPE;
5182 ns->ln_flags |= LF_UPGRADE;
5187 if (bk->lb_param & LPF_DRYRUN) {
5188 ns->ln_linkea_repaired++;
5194 if (!lustre_handle_is_used(&lh))
5200 rc = dt_xattr_del(env, obj, XATTR_NAME_LINK, handle,
5207 rc = linkea_data_new(&ldata,
5208 &lfsck_env_info(env)->lti_linkea_buf);
5213 rc = linkea_add_buf(&ldata, cname, pfid);
5217 rc = lfsck_links_write(env, obj, &ldata, handle);
5218 if (unlikely(rc == -ENOSPC) &&
5219 S_ISREG(lfsck_object_type(obj)) && !dt_object_remote(obj)) {
5220 if (handle != NULL) {
5221 LASSERT(dt_write_locked(env, obj));
5223 dt_write_unlock(env, obj);
5226 dt_trans_stop(env, dev, handle);
5229 lfsck_ibits_unlock(&lh, LCK_EX);
5232 rc = lfsck_namespace_trace_update(env, com,
5233 &lnr->lnr_fid, LNTF_SKIP_NLINK, true);
5235 /* If we cannot record this object in the
5236 * LFSCK tracing, we have to mark the LFSCK
5237 * as LF_INCOMPLETE, then the LFSCK will
5238 * skip nlink attribute verification for
5240 ns->ln_flags |= LF_INCOMPLETE;
5248 count = ldata.ld_leh->leh_reccount;
5249 if (!S_ISDIR(lfsck_object_type(obj)) ||
5250 !dt_object_remote(obj)) {
5251 ns->ln_linkea_repaired++;
5262 rc = dt_attr_get(env, obj, la, BYPASS_CAPA);
5266 if ((count == 1 && la->la_nlink == 1) ||
5267 S_ISDIR(lfsck_object_type(obj)))
5268 /* Usually, it is for single linked object or dir, do nothing.*/
5271 /* Following modification will be in another transaction. */
5272 if (handle != NULL) {
5273 dt_write_unlock(env, obj);
5276 dt_trans_stop(env, dev, handle);
5279 lfsck_ibits_unlock(&lh, LCK_EX);
5282 ns->ln_mul_linked_checked++;
5283 rc = lfsck_namespace_trace_update(env, com, &lnr->lnr_fid,
5284 LNTF_CHECK_LINKEA, true);
5290 dt_write_unlock(env, obj);
5292 if (handle != NULL && !IS_ERR(handle))
5293 dt_trans_stop(env, dev, handle);
5296 lfsck_ibits_unlock(&lh, LCK_EX);
5298 if (!name_is_dot_or_dotdot(lnr->lnr_name, lnr->lnr_namelen) &&
5299 !lfsck_is_valid_slave_name_entry(env, lnr->lnr_lmv,
5300 lnr->lnr_name, lnr->lnr_namelen) &&
5301 type != LNIT_BAD_DIRENT) {
5302 ns->ln_flags |= LF_INCONSISTENT;
5305 rc = lfsck_namespace_repair_bad_name_hash(env, com, dir,
5306 lnr->lnr_lmv, lnr->lnr_name);
5315 rc = lfsck_namespace_repair_dirent(env, com, dir,
5316 obj, lnr->lnr_name, lnr->lnr_name,
5317 lnr->lnr_type, true, false);
5321 case LNIT_BAD_DIRENT:
5323 /* XXX: This is a bad dirent, we do not know whether
5324 * the original name entry reference a regular
5325 * file or a directory, then keep the parent's
5326 * nlink count unchanged here. */
5327 rc = lfsck_namespace_repair_dirent(env, com, dir,
5328 obj, lnr->lnr_name, lnr->lnr_name,
5329 lnr->lnr_type, false, false);
5338 down_write(&com->lc_sem);
5340 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail to handle "
5341 "the entry: "DFID", parent "DFID", name %.*s: rc = %d\n",
5342 lfsck_lfsck2name(lfsck), PFID(&lnr->lnr_fid),
5343 PFID(lfsck_dto2fid(lnr->lnr_obj)),
5344 lnr->lnr_namelen, lnr->lnr_name, rc);
5346 lfsck_namespace_record_failure(env, lfsck, ns);
5347 if ((rc == -ENOTCONN || rc == -ESHUTDOWN || rc == -EREMCHG ||
5348 rc == -ETIMEDOUT || rc == -EHOSTDOWN ||
5349 rc == -EHOSTUNREACH || rc == -EINPROGRESS) &&
5350 dev != NULL && dev != lfsck->li_next)
5351 lfsck_lad_set_bitmap(env, com, idx);
5353 if (!(bk->lb_param & LPF_FAILOUT))
5357 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant "
5358 "repaired the entry: "DFID", parent "DFID
5359 ", name %.*s\n", lfsck_lfsck2name(lfsck),
5360 PFID(&lnr->lnr_fid),
5361 PFID(lfsck_dto2fid(lnr->lnr_obj)),
5362 lnr->lnr_namelen, lnr->lnr_name);
5365 ns->ln_items_repaired++;
5369 ns->ln_dangling_repaired++;
5372 ns->ln_bad_type_repaired++;
5374 case LNIT_BAD_DIRENT:
5375 ns->ln_dirent_repaired++;
5381 if (bk->lb_param & LPF_DRYRUN &&
5382 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5383 lfsck_pos_fill(env, lfsck,
5384 &ns->ln_pos_first_inconsistent,
5389 ns->ln_name_hash_repaired++;
5391 /* Not count repeatedly. */
5393 ns->ln_items_repaired++;
5395 if (bk->lb_param & LPF_DRYRUN &&
5396 lfsck_pos_is_zero(&ns->ln_pos_first_inconsistent))
5397 lfsck_pos_fill(env, lfsck,
5398 &ns->ln_pos_first_inconsistent,
5405 up_write(&com->lc_sem);
5407 if (obj != NULL && !IS_ERR(obj))
5408 lfsck_object_put(env, obj);
5414 * Handle one orphan under the backend /lost+found directory
5416 * Insert the orphan FID into the namespace LFSCK trace file for further
5417 * processing (via the subsequent namespace LFSCK second-stage scanning).
5418 * At the same time, remove the orphan name entry from backend /lost+found
5419 * directory. There is an interval between the orphan name entry removed
5420 * from the backend /lost+found directory and the orphan FID in the LFSCK
5421 * trace file handled. In such interval, the LFSCK can be reset, then
5422 * all the FIDs recorded in the namespace LFSCK trace file will be dropped.
5423 * To guarantee that the orphans can be found when LFSCK run next time
5424 * without e2fsck again, when remove the orphan name entry, the LFSCK
5425 * will set the orphan's ctime attribute as 1. Since normal applications
5426 * cannot change the object's ctime attribute as 1. Then when LFSCK run
5427 * next time, it can record the object (that ctime is 1) in the namespace
5428 * LFSCK trace file during the first-stage scanning.
5430 * \param[in] env pointer to the thread context
5431 * \param[in] com pointer to the lfsck component
5432 * \param[in] parent pointer to the object for the backend /lost+found
5433 * \param[in] ent pointer to the name entry for the target under the
5434 * backend /lost+found
5436 * \retval positive for repaired
5437 * \retval 0 if needs to repair nothing
5438 * \retval negative error number on failure
5440 static int lfsck_namespace_scan_local_lpf_one(const struct lu_env *env,
5441 struct lfsck_component *com,
5442 struct dt_object *parent,
5443 struct lu_dirent *ent)
5445 struct lfsck_thread_info *info = lfsck_env_info(env);
5446 struct lu_fid *key = &info->lti_fid;
5447 struct lu_attr *la = &info->lti_la;
5448 struct lfsck_instance *lfsck = com->lc_lfsck;
5449 struct dt_object *obj = com->lc_obj;
5450 struct dt_device *dev = lfsck->li_bottom;
5451 struct dt_object *child = NULL;
5452 struct thandle *th = NULL;
5458 child = lfsck_object_find_by_dev(env, dev, &ent->lde_fid);
5460 RETURN(PTR_ERR(child));
5462 LASSERT(dt_object_exists(child));
5463 LASSERT(!dt_object_remote(child));
5465 fid_cpu_to_be(key, &ent->lde_fid);
5466 rc = dt_lookup(env, obj, (struct dt_rec *)&flags,
5467 (const struct dt_key *)key, BYPASS_CAPA);
5470 flags |= LNTF_CHECK_ORPHAN;
5471 } else if (rc == -ENOENT) {
5472 flags = LNTF_CHECK_ORPHAN;
5477 th = dt_trans_create(env, dev);
5479 GOTO(out, rc = PTR_ERR(th));
5481 /* a1. remove name entry from backend /lost+found */
5482 rc = dt_declare_delete(env, parent,
5483 (const struct dt_key *)ent->lde_name, th);
5487 if (S_ISDIR(lfsck_object_type(child))) {
5488 /* a2. decrease parent's nlink */
5489 rc = dt_declare_ref_del(env, parent, th);
5495 /* a3. remove child's FID from the LFSCK trace file. */
5496 rc = dt_declare_delete(env, obj,
5497 (const struct dt_key *)key, th);
5501 /* a4. set child's ctime as 1 */
5502 memset(la, 0, sizeof(*la));
5504 la->la_valid = LA_CTIME;
5505 rc = dt_declare_attr_set(env, child, la, th);
5510 /* a5. insert child's FID into the LFSCK trace file. */
5511 rc = dt_declare_insert(env, obj, (const struct dt_rec *)&flags,
5512 (const struct dt_key *)key, th);
5516 rc = dt_trans_start_local(env, dev, th);
5520 /* b1. remove name entry from backend /lost+found */
5521 rc = dt_delete(env, parent, (const struct dt_key *)ent->lde_name, th,
5526 if (S_ISDIR(lfsck_object_type(child))) {
5527 /* b2. decrease parent's nlink */
5528 dt_write_lock(env, parent, 0);
5529 rc = dt_ref_del(env, parent, th);
5530 dt_write_unlock(env, parent);
5536 /* a3. remove child's FID from the LFSCK trace file. */
5537 rc = dt_delete(env, obj, (const struct dt_key *)key, th,
5542 /* b4. set child's ctime as 1 */
5543 rc = dt_attr_set(env, child, la, th, BYPASS_CAPA);
5548 /* b5. insert child's FID into the LFSCK trace file. */
5549 rc = dt_insert(env, obj, (const struct dt_rec *)&flags,
5550 (const struct dt_key *)key, th, BYPASS_CAPA, 1);
5552 GOTO(stop, rc = (rc == 0 ? 1 : rc));
5555 dt_trans_stop(env, dev, th);
5558 lu_object_put(env, &child->do_lu);
5564 * Handle orphans under the backend /lost+found directory
5566 * Some backend checker, such as e2fsck for ldiskfs may find some orphans
5567 * and put them under the backend /lost+found directory that is invisible
5568 * to client. The LFSCK will scan such directory, for the original client
5569 * visible orphans, add their fids into the namespace LFSCK trace file,
5570 * then the subsenquent namespace LFSCK second-stage scanning can handle
5571 * them as other objects to be double scanned: either move back to normal
5572 * namespace, or to the global visible orphan directory:
5573 * /ROOT/.lustre/lost+found/MDTxxxx/
5575 * \param[in] env pointer to the thread context
5576 * \param[in] com pointer to the lfsck component
5578 static void lfsck_namespace_scan_local_lpf(const struct lu_env *env,
5579 struct lfsck_component *com)
5581 struct lfsck_thread_info *info = lfsck_env_info(env);
5582 struct lu_dirent *ent =
5583 (struct lu_dirent *)info->lti_key;
5584 struct lu_seq_range *range = &info->lti_range;
5585 struct lfsck_instance *lfsck = com->lc_lfsck;
5586 struct ptlrpc_thread *thread = &lfsck->li_thread;
5587 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5588 struct dt_device *dev = lfsck->li_bottom;
5589 struct lfsck_namespace *ns = com->lc_file_ram;
5590 struct dt_object *parent;
5591 const struct dt_it_ops *iops;
5593 struct seq_server_site *ss =
5594 lu_site2seq(dev->dd_lu_dev.ld_site);
5600 parent = lfsck_object_find_by_dev(env, dev, &LU_BACKEND_LPF_FID);
5601 if (IS_ERR(parent)) {
5602 CERROR("%s: fail to find backend /lost+found: rc = %ld\n",
5603 lfsck_lfsck2name(lfsck), PTR_ERR(parent));
5607 /* It is normal that the /lost+found does not exist for ZFS backend. */
5608 if (!dt_object_exists(parent))
5611 if (unlikely(!dt_try_as_dir(env, parent)))
5612 GOTO(out, rc = -ENOTDIR);
5614 CDEBUG(D_LFSCK, "%s: start to scan backend /lost+found\n",
5615 lfsck_lfsck2name(lfsck));
5617 com->lc_new_scanned = 0;
5618 iops = &parent->do_index_ops->dio_it;
5619 di = iops->init(env, parent, LUDA_64BITHASH | LUDA_TYPE, BYPASS_CAPA);
5621 GOTO(out, rc = PTR_ERR(di));
5623 rc = iops->load(env, di, 0);
5625 rc = iops->next(env, di);
5630 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
5632 struct l_wait_info lwi;
5634 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
5636 l_wait_event(thread->t_ctl_waitq,
5637 !thread_is_running(thread),
5640 if (unlikely(!thread_is_running(thread)))
5644 rc = iops->rec(env, di, (struct dt_rec *)ent,
5645 LUDA_64BITHASH | LUDA_TYPE);
5647 rc = lfsck_unpack_ent(ent, &cookie, &type);
5649 if (unlikely(rc != 0)) {
5650 CDEBUG(D_LFSCK, "%s: fail to iterate backend "
5651 "/lost+found: rc = %d\n",
5652 lfsck_lfsck2name(lfsck), rc);
5657 /* skip dot and dotdot entries */
5658 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5661 if (!fid_seq_in_fldb(fid_seq(&ent->lde_fid)))
5664 if (fid_is_norm(&ent->lde_fid)) {
5665 fld_range_set_mdt(range);
5666 rc = fld_local_lookup(env, ss->ss_server_fld,
5667 fid_seq(&ent->lde_fid), range);
5670 } else if (lfsck_dev_idx(dev) != 0) {
5671 /* If the returned FID is IGIF, then there are three
5674 * 1) The object is upgraded from old Lustre-1.8 with
5675 * IGIF assigned to such object.
5676 * 2) The object is a backend local object and is
5677 * invisible to client.
5678 * 3) The object lost its LMV EA, and since there is
5679 * no FID-in-dirent for the orphan in the backend
5680 * /lost+found directory, then the low layer will
5681 * return IGIF for such object.
5683 * For MDTx (x != 0), it is either case 2) or case 3),
5684 * but from the LFSCK view, they are indistinguishable.
5685 * To be safe, the LFSCK will keep it there and report
5686 * some message, then the adminstrator can handle that
5689 * For MDT0, it is more possible the case 1). The LFSCK
5690 * will handle the orphan as an upgraded object. */
5691 CDEBUG(D_LFSCK, "%s: the orphan %.*s with IGIF "DFID
5692 "in the backend /lost+found on the MDT %04x, "
5693 "to be safe, skip it.\n",
5694 lfsck_lfsck2name(lfsck), ent->lde_namelen,
5695 ent->lde_name, PFID(&ent->lde_fid),
5696 lfsck_dev_idx(dev));
5700 rc = lfsck_namespace_scan_local_lpf_one(env, com, parent, ent);
5703 down_write(&com->lc_sem);
5704 com->lc_new_scanned++;
5705 ns->ln_local_lpf_scanned++;
5707 ns->ln_local_lpf_moved++;
5709 ns->ln_local_lpf_skipped++;
5711 ns->ln_local_lpf_failed++;
5712 up_write(&com->lc_sem);
5714 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5718 lfsck_control_speed_by_self(com);
5719 if (unlikely(!thread_is_running(thread))) {
5724 rc = iops->next(env, di);
5728 iops->fini(env, di);
5733 CDEBUG(D_LFSCK, "%s: stop to scan backend /lost+found: rc = %d\n",
5734 lfsck_lfsck2name(lfsck), rc);
5736 lu_object_put(env, &parent->do_lu);
5740 * Rescan the striped directory after the master LMV EA reset.
5742 * Sometimes, the master LMV EA of the striped directory maybe lost, so when
5743 * the namespace LFSCK engine scan the striped directory for the first time,
5744 * it will be regarded as a normal directory. As the LFSCK processing, some
5745 * other LFSCK instance on other MDT will find the shard of this striped dir,
5746 * and find that the master MDT-object of the striped directory lost its LMV
5747 * EA, then such remote LFSCK instance will regenerate the master LMV EA and
5748 * notify the LFSCK instance on this MDT to rescan the striped directory.
5750 * \param[in] env pointer to the thread context
5751 * \param[in] com pointer to the lfsck component
5752 * \param[in] llu the lfsck_lmv_unit that contains the striped directory
5755 * \retval positive number for success
5756 * \retval 0 for LFSCK stopped/paused
5757 * \retval negative error number on failure
5759 static int lfsck_namespace_rescan_striped_dir(const struct lu_env *env,
5760 struct lfsck_component *com,
5761 struct lfsck_lmv_unit *llu)
5763 struct lfsck_thread_info *info = lfsck_env_info(env);
5764 struct lfsck_instance *lfsck = com->lc_lfsck;
5765 struct lfsck_assistant_data *lad = com->lc_data;
5766 struct dt_object *dir;
5767 const struct dt_it_ops *iops;
5769 struct lu_dirent *ent =
5770 (struct lu_dirent *)info->lti_key;
5771 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5772 struct ptlrpc_thread *thread = &lfsck->li_thread;
5773 struct lfsck_namespace_req *lnr;
5774 struct lfsck_assistant_req *lar;
5779 LASSERT(list_empty(&lad->lad_req_list));
5781 lfsck->li_lmv = &llu->llu_lmv;
5782 lfsck->li_obj_dir = lfsck_object_get(llu->llu_obj);
5783 rc = lfsck_open_dir(env, lfsck, 0);
5787 dir = lfsck->li_obj_dir;
5788 di = lfsck->li_di_dir;
5789 iops = &dir->do_index_ops->dio_it;
5791 rc = iops->rec(env, di, (struct dt_rec *)ent,
5792 lfsck->li_args_dir);
5794 rc = lfsck_unpack_ent(ent, &lfsck->li_cookie_dir,
5798 if (bk->lb_param & LPF_FAILOUT)
5804 if (name_is_dot_or_dotdot(ent->lde_name, ent->lde_namelen))
5807 lnr = lfsck_namespace_assistant_req_init(lfsck, ent, type);
5809 if (bk->lb_param & LPF_FAILOUT)
5810 GOTO(out, rc = PTR_ERR(lnr));
5815 lar = &lnr->lnr_lar;
5816 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5817 lfsck_namespace_assistant_req_fini(env, lar);
5818 if (rc != 0 && bk->lb_param & LPF_FAILOUT)
5821 if (unlikely(!thread_is_running(thread)))
5825 rc = iops->next(env, di);
5829 lfsck_close_dir(env, lfsck, rc);
5833 /* The close_dir() may insert a dummy lnr in the lad->lad_req_list. */
5834 if (list_empty(&lad->lad_req_list))
5837 spin_lock(&lad->lad_lock);
5838 lar = list_entry(lad->lad_req_list.next, struct lfsck_assistant_req,
5840 list_del_init(&lar->lar_list);
5841 spin_unlock(&lad->lad_lock);
5843 rc = lfsck_namespace_assistant_handler_p1(env, com, lar);
5844 lfsck_namespace_assistant_req_fini(env, lar);
5846 RETURN(rc == 0 ? 1 : rc);
5849 static int lfsck_namespace_assistant_handler_p2(const struct lu_env *env,
5850 struct lfsck_component *com)
5852 struct lfsck_instance *lfsck = com->lc_lfsck;
5853 struct ptlrpc_thread *thread = &lfsck->li_thread;
5854 struct lfsck_bookmark *bk = &lfsck->li_bookmark_ram;
5855 struct lfsck_namespace *ns = com->lc_file_ram;
5856 struct dt_object *obj = com->lc_obj;
5857 const struct dt_it_ops *iops = &obj->do_index_ops->dio_it;
5858 struct dt_object *target;
5866 while (!list_empty(&lfsck->li_list_lmv)) {
5867 struct lfsck_lmv_unit *llu;
5869 spin_lock(&lfsck->li_lock);
5870 llu = list_entry(lfsck->li_list_lmv.next,
5871 struct lfsck_lmv_unit, llu_link);
5872 list_del_init(&llu->llu_link);
5873 spin_unlock(&lfsck->li_lock);
5875 rc = lfsck_namespace_rescan_striped_dir(env, com, llu);
5880 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan start\n",
5881 lfsck_lfsck2name(lfsck));
5883 lfsck_namespace_scan_local_lpf(env, com);
5885 com->lc_new_checked = 0;
5886 com->lc_new_scanned = 0;
5887 com->lc_time_last_checkpoint = cfs_time_current();
5888 com->lc_time_next_checkpoint = com->lc_time_last_checkpoint +
5889 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5891 di = iops->init(env, obj, 0, BYPASS_CAPA);
5893 RETURN(PTR_ERR(di));
5895 fid_cpu_to_be(&fid, &ns->ln_fid_latest_scanned_phase2);
5896 rc = iops->get(env, di, (const struct dt_key *)&fid);
5900 /* Skip the start one, which either has been processed or non-exist. */
5901 rc = iops->next(env, di);
5906 if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_DELAY3) &&
5908 struct l_wait_info lwi;
5910 lwi = LWI_TIMEOUT(cfs_time_seconds(cfs_fail_val),
5912 l_wait_event(thread->t_ctl_waitq,
5913 !thread_is_running(thread),
5916 if (unlikely(!thread_is_running(thread)))
5920 key = iops->key(env, di);
5921 fid_be_to_cpu(&fid, (const struct lu_fid *)key);
5922 if (!fid_is_sane(&fid)) {
5927 target = lfsck_object_find_by_dev(env, lfsck->li_bottom, &fid);
5928 if (IS_ERR(target)) {
5929 rc = PTR_ERR(target);
5933 if (dt_object_exists(target)) {
5934 rc = iops->rec(env, di, (struct dt_rec *)&flags, 0);
5936 rc = lfsck_namespace_double_scan_one(env, com,
5943 lfsck_object_put(env, target);
5946 down_write(&com->lc_sem);
5947 com->lc_new_checked++;
5948 com->lc_new_scanned++;
5949 ns->ln_fid_latest_scanned_phase2 = fid;
5951 ns->ln_objs_repaired_phase2++;
5953 ns->ln_objs_failed_phase2++;
5954 up_write(&com->lc_sem);
5956 if (rc < 0 && bk->lb_param & LPF_FAILOUT)
5959 if (unlikely(cfs_time_beforeq(com->lc_time_next_checkpoint,
5960 cfs_time_current())) &&
5961 com->lc_new_checked != 0) {
5962 down_write(&com->lc_sem);
5963 ns->ln_run_time_phase2 +=
5964 cfs_duration_sec(cfs_time_current() +
5965 HALF_SEC - com->lc_time_last_checkpoint);
5966 ns->ln_time_last_checkpoint = cfs_time_current_sec();
5967 ns->ln_objs_checked_phase2 += com->lc_new_checked;
5968 com->lc_new_checked = 0;
5969 rc = lfsck_namespace_store(env, com);
5970 up_write(&com->lc_sem);
5974 com->lc_time_last_checkpoint = cfs_time_current();
5975 com->lc_time_next_checkpoint =
5976 com->lc_time_last_checkpoint +
5977 cfs_time_seconds(LFSCK_CHECKPOINT_INTERVAL);
5980 lfsck_control_speed_by_self(com);
5981 if (unlikely(!thread_is_running(thread)))
5984 rc = iops->next(env, di);
5993 iops->fini(env, di);
5995 CDEBUG(D_LFSCK, "%s: namespace LFSCK phase2 scan stop: rc = %d\n",
5996 lfsck_lfsck2name(lfsck), rc);
6001 static void lfsck_namespace_assistant_fill_pos(const struct lu_env *env,
6002 struct lfsck_component *com,
6003 struct lfsck_position *pos)
6005 struct lfsck_assistant_data *lad = com->lc_data;
6006 struct lfsck_namespace_req *lnr;
6008 if (list_empty(&lad->lad_req_list))
6011 lnr = list_entry(lad->lad_req_list.next,
6012 struct lfsck_namespace_req,
6014 pos->lp_oit_cookie = lnr->lnr_oit_cookie;
6015 pos->lp_dir_cookie = lnr->lnr_dir_cookie - 1;
6016 pos->lp_dir_parent = *lfsck_dto2fid(lnr->lnr_obj);
6019 static int lfsck_namespace_double_scan_result(const struct lu_env *env,
6020 struct lfsck_component *com,
6023 struct lfsck_instance *lfsck = com->lc_lfsck;
6024 struct lfsck_namespace *ns = com->lc_file_ram;
6026 down_write(&com->lc_sem);
6027 ns->ln_run_time_phase2 += cfs_duration_sec(cfs_time_current() +
6028 HALF_SEC - lfsck->li_time_last_checkpoint);
6029 ns->ln_time_last_checkpoint = cfs_time_current_sec();
6030 ns->ln_objs_checked_phase2 += com->lc_new_checked;
6031 com->lc_new_checked = 0;
6034 if (ns->ln_flags & LF_INCOMPLETE)
6035 ns->ln_status = LS_PARTIAL;
6037 ns->ln_status = LS_COMPLETED;
6038 if (!(lfsck->li_bookmark_ram.lb_param & LPF_DRYRUN))
6039 ns->ln_flags &= ~(LF_SCANNED_ONCE | LF_INCONSISTENT);
6040 ns->ln_time_last_complete = ns->ln_time_last_checkpoint;
6041 ns->ln_success_count++;
6042 } else if (rc == 0) {
6043 if (lfsck->li_status != 0)
6044 ns->ln_status = lfsck->li_status;
6046 ns->ln_status = LS_STOPPED;
6048 ns->ln_status = LS_FAILED;
6051 rc = lfsck_namespace_store(env, com);
6052 up_write(&com->lc_sem);
6058 lfsck_namespace_assistant_sync_failures_interpret(const struct lu_env *env,
6059 struct ptlrpc_request *req,
6066 * Notify remote LFSCK instances about former failures.
6068 * The local LFSCK instance has recorded which MDTs have ever failed to respond
6069 * some LFSCK verification requests (maybe because of network issues or the MDT
6070 * itself trouble). During the respond gap the MDT may missed some name entries
6071 * verification, then the MDT cannot know whether related MDT-objects have been
6072 * referenced by related name entries or not, then in the second-stage scanning,
6073 * these MDT-objects will be regarded as orphan, if the MDT-object contains bad
6074 * linkEA for back reference, then it will misguide the LFSCK to generate wrong
6075 * name entry for repairing the orphan.
6077 * To avoid above trouble, when layout LFSCK finishes the first-stage scanning,
6078 * it will scan the bitmap for the ever failed MDTs, and notify them that they
6079 * have ever missed some name entries verification and should skip the handling
6080 * for orphan MDT-objects.
6082 * \param[in] env pointer to the thread context
6083 * \param[in] com pointer to the lfsck component
6084 * \param[in] lr pointer to the lfsck request
6086 static void lfsck_namespace_assistant_sync_failures(const struct lu_env *env,
6087 struct lfsck_component *com,
6088 struct lfsck_request *lr)
6090 struct lfsck_async_interpret_args *laia =
6091 &lfsck_env_info(env)->lti_laia2;
6092 struct lfsck_assistant_data *lad = com->lc_data;
6093 struct lfsck_namespace *ns = com->lc_file_ram;
6094 struct lfsck_instance *lfsck = com->lc_lfsck;
6095 struct lfsck_tgt_descs *ltds = &lfsck->li_mdt_descs;
6096 struct lfsck_tgt_desc *ltd;
6097 struct ptlrpc_request_set *set;
6101 set = ptlrpc_prep_set();
6103 GOTO(out, rc = -ENOMEM);
6105 lr->lr_flags2 = ns->ln_flags | LF_INCOMPLETE;
6106 memset(laia, 0, sizeof(*laia));
6107 lad->lad_touch_gen++;
6109 spin_lock(<ds->ltd_lock);
6110 while (!list_empty(&lad->lad_mdt_list)) {
6111 ltd = list_entry(lad->lad_mdt_list.next,
6112 struct lfsck_tgt_desc,
6113 ltd_namespace_list);
6114 if (ltd->ltd_namespace_gen == lad->lad_touch_gen)
6117 ltd->ltd_namespace_gen = lad->lad_touch_gen;
6118 list_move_tail(<d->ltd_namespace_list,
6119 &lad->lad_mdt_list);
6120 if (!lad->lad_incomplete ||
6121 !cfs_bitmap_check(lad->lad_bitmap, ltd->ltd_index)) {
6122 ltd->ltd_namespace_failed = 0;
6126 ltd->ltd_namespace_failed = 1;
6127 spin_unlock(<ds->ltd_lock);
6128 rc = lfsck_async_request(env, ltd->ltd_exp, lr, set,
6129 lfsck_namespace_assistant_sync_failures_interpret,
6130 laia, LFSCK_NOTIFY);
6132 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6133 "to sync failure with MDT %x: rc = %d\n",
6134 lfsck_lfsck2name(lfsck), ltd->ltd_index, rc);
6136 spin_lock(<ds->ltd_lock);
6138 spin_unlock(<ds->ltd_lock);
6140 rc = ptlrpc_set_wait(set);
6141 ptlrpc_set_destroy(set);
6147 CDEBUG(D_LFSCK, "%s: namespace LFSCK assistant fail "
6148 "to sync failure with MDTs, and related MDTs "
6149 "may handle orphan improperly: rc = %d\n",
6150 lfsck_lfsck2name(lfsck), rc);
6155 struct lfsck_assistant_operations lfsck_namespace_assistant_ops = {
6156 .la_handler_p1 = lfsck_namespace_assistant_handler_p1,
6157 .la_handler_p2 = lfsck_namespace_assistant_handler_p2,
6158 .la_fill_pos = lfsck_namespace_assistant_fill_pos,
6159 .la_double_scan_result = lfsck_namespace_double_scan_result,
6160 .la_req_fini = lfsck_namespace_assistant_req_fini,
6161 .la_sync_failures = lfsck_namespace_assistant_sync_failures,
6165 * Verify the specified linkEA entry for the given directory object.
6166 * If the object has no such linkEA entry or it has more other linkEA
6167 * entries, then re-generate the linkEA with the given information.
6169 * \param[in] env pointer to the thread context
6170 * \param[in] dev pointer to the dt_device
6171 * \param[in] obj pointer to the dt_object to be handled
6172 * \param[in] cname the name for the child in the parent directory
6173 * \param[in] pfid the parent directory's FID for the linkEA
6175 * \retval 0 for success
6176 * \retval negative error number on failure
6178 int lfsck_verify_linkea(const struct lu_env *env, struct dt_device *dev,
6179 struct dt_object *obj, const struct lu_name *cname,
6180 const struct lu_fid *pfid)
6182 struct linkea_data ldata = { 0 };
6183 struct lu_buf linkea_buf;
6186 int fl = LU_XATTR_CREATE;
6190 LASSERT(S_ISDIR(lfsck_object_type(obj)));
6192 rc = lfsck_links_read(env, obj, &ldata);
6193 if (rc == -ENODATA) {
6195 } else if (rc == 0) {
6196 fl = LU_XATTR_REPLACE;
6197 if (ldata.ld_leh->leh_reccount != 1) {
6200 rc = linkea_links_find(&ldata, cname, pfid);
6209 rc = linkea_data_new(&ldata, &lfsck_env_info(env)->lti_linkea_buf);
6213 rc = linkea_add_buf(&ldata, cname, pfid);
6217 lfsck_buf_init(&linkea_buf, ldata.ld_buf->lb_buf,
6218 ldata.ld_leh->leh_len);
6219 th = dt_trans_create(env, dev);
6221 RETURN(PTR_ERR(th));
6223 rc = dt_declare_xattr_set(env, obj, &linkea_buf,
6224 XATTR_NAME_LINK, fl, th);
6228 rc = dt_trans_start_local(env, dev, th);
6232 dt_write_lock(env, obj, 0);
6233 rc = dt_xattr_set(env, obj, &linkea_buf,
6234 XATTR_NAME_LINK, fl, th, BYPASS_CAPA);
6235 dt_write_unlock(env, obj);
6240 dt_trans_stop(env, dev, th);
6245 * Get the name and parent directory's FID from the first linkEA entry.
6247 * \param[in] env pointer to the thread context
6248 * \param[in] obj pointer to the object which get linkEA from
6249 * \param[out] name pointer to the buffer to hold the name
6250 * in the first linkEA entry
6251 * \param[out] pfid pointer to the buffer to hold the parent
6252 * directory's FID in the first linkEA entry
6254 * \retval 0 for success
6255 * \retval negative error number on failure
6257 int lfsck_links_get_first(const struct lu_env *env, struct dt_object *obj,
6258 char *name, struct lu_fid *pfid)
6260 struct lu_name *cname = &lfsck_env_info(env)->lti_name;
6261 struct linkea_data ldata = { 0 };
6264 rc = lfsck_links_read(env, obj, &ldata);
6268 linkea_first_entry(&ldata);
6269 if (ldata.ld_lee == NULL)
6272 linkea_entry_unpack(ldata.ld_lee, &ldata.ld_reclen, cname, pfid);
6273 /* To guarantee the 'name' is terminated with '0'. */
6274 memcpy(name, cname->ln_name, cname->ln_namelen);
6275 name[cname->ln_namelen] = 0;
6281 * Remove the name entry from the parent directory.
6283 * No need to care about the object referenced by the name entry,
6284 * either the name entry is invalid or redundant, or the referenced
6285 * object has been processed has been or will be handled by others.
6287 * \param[in] env pointer to the thread context
6288 * \param[in] lfsck pointer to the lfsck instance
6289 * \param[in] parent pointer to the lost+found object
6290 * \param[in] name the name for the name entry to be removed
6291 * \param[in] type the type for the name entry to be removed
6293 * \retval 0 for success
6294 * \retval negative error number on failure
6296 int lfsck_remove_name_entry(const struct lu_env *env,
6297 struct lfsck_instance *lfsck,
6298 struct dt_object *parent,
6299 const char *name, __u32 type)
6301 struct dt_device *dev = lfsck->li_next;
6303 struct lustre_handle lh = { 0 };
6307 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
6308 MDS_INODELOCK_UPDATE, LCK_EX);
6312 th = dt_trans_create(env, dev);
6314 GOTO(unlock, rc = PTR_ERR(th));
6316 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
6320 if (S_ISDIR(type)) {
6321 rc = dt_declare_ref_del(env, parent, th);
6326 rc = dt_trans_start(env, dev, th);
6330 rc = dt_delete(env, parent, (const struct dt_key *)name, th,
6335 if (S_ISDIR(type)) {
6336 dt_write_lock(env, parent, 0);
6337 rc = dt_ref_del(env, parent, th);
6338 dt_write_unlock(env, parent);
6344 dt_trans_stop(env, dev, th);
6347 lfsck_ibits_unlock(&lh, LCK_EX);
6349 CDEBUG(D_LFSCK, "%s: remove name entry "DFID"/%s "
6350 "with type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6351 PFID(lfsck_dto2fid(parent)), name, type, rc);
6357 * Update the object's name entry with the given FID.
6359 * \param[in] env pointer to the thread context
6360 * \param[in] lfsck pointer to the lfsck instance
6361 * \param[in] parent pointer to the parent directory that holds
6363 * \param[in] name the name for the entry to be updated
6364 * \param[in] pfid the new PFID for the name entry
6365 * \param[in] type the type for the name entry to be updated
6367 * \retval 0 for success
6368 * \retval negative error number on failure
6370 int lfsck_update_name_entry(const struct lu_env *env,
6371 struct lfsck_instance *lfsck,
6372 struct dt_object *parent, const char *name,
6373 const struct lu_fid *pfid, __u32 type)
6375 struct dt_insert_rec *rec = &lfsck_env_info(env)->lti_dt_rec;
6376 struct dt_device *dev = lfsck->li_next;
6377 struct lustre_handle lh = { 0 };
6383 rc = lfsck_ibits_lock(env, lfsck, parent, &lh,
6384 MDS_INODELOCK_UPDATE, LCK_EX);
6388 th = dt_trans_create(env, dev);
6390 GOTO(unlock, rc = PTR_ERR(th));
6392 rc = dt_declare_delete(env, parent, (const struct dt_key *)name, th);
6396 rec->rec_type = type;
6397 rec->rec_fid = pfid;
6398 rc = dt_declare_insert(env, parent, (const struct dt_rec *)rec,
6399 (const struct dt_key *)name, th);
6403 rc = dt_declare_ref_add(env, parent, th);
6407 rc = dt_trans_start(env, dev, th);
6411 rc = dt_delete(env, parent, (const struct dt_key *)name, th,
6413 if (rc == -ENOENT) {
6421 rc = dt_insert(env, parent, (const struct dt_rec *)rec,
6422 (const struct dt_key *)name, th, BYPASS_CAPA, 1);
6423 if (rc == 0 && S_ISDIR(type) && !exists) {
6424 dt_write_lock(env, parent, 0);
6425 rc = dt_ref_add(env, parent, th);
6426 dt_write_unlock(env, parent);
6432 dt_trans_stop(env, dev, th);
6435 lfsck_ibits_unlock(&lh, LCK_EX);
6437 CDEBUG(D_LFSCK, "%s: update name entry "DFID"/%s with the FID "DFID
6438 " and the type %o: rc = %d\n", lfsck_lfsck2name(lfsck),
6439 PFID(lfsck_dto2fid(parent)), name, PFID(pfid), type, rc);
6444 int lfsck_namespace_setup(const struct lu_env *env,
6445 struct lfsck_instance *lfsck)
6447 struct lfsck_component *com;
6448 struct lfsck_namespace *ns;
6449 struct dt_object *root = NULL;
6450 struct dt_object *obj;
6454 LASSERT(lfsck->li_master);
6460 INIT_LIST_HEAD(&com->lc_link);
6461 INIT_LIST_HEAD(&com->lc_link_dir);
6462 init_rwsem(&com->lc_sem);
6463 atomic_set(&com->lc_ref, 1);
6464 com->lc_lfsck = lfsck;
6465 com->lc_type = LFSCK_TYPE_NAMESPACE;
6466 com->lc_ops = &lfsck_namespace_ops;
6467 com->lc_data = lfsck_assistant_data_init(
6468 &lfsck_namespace_assistant_ops,
6470 if (com->lc_data == NULL)
6471 GOTO(out, rc = -ENOMEM);
6473 com->lc_file_size = sizeof(struct lfsck_namespace);
6474 OBD_ALLOC(com->lc_file_ram, com->lc_file_size);
6475 if (com->lc_file_ram == NULL)
6476 GOTO(out, rc = -ENOMEM);
6478 OBD_ALLOC(com->lc_file_disk, com->lc_file_size);
6479 if (com->lc_file_disk == NULL)
6480 GOTO(out, rc = -ENOMEM);
6482 root = dt_locate(env, lfsck->li_bottom, &lfsck->li_local_root_fid);
6484 GOTO(out, rc = PTR_ERR(root));
6486 if (unlikely(!dt_try_as_dir(env, root)))
6487 GOTO(out, rc = -ENOTDIR);
6489 obj = local_index_find_or_create(env, lfsck->li_los, root,
6490 lfsck_namespace_name,
6491 S_IFREG | S_IRUGO | S_IWUSR,
6492 &dt_lfsck_features);
6494 GOTO(out, rc = PTR_ERR(obj));
6497 rc = obj->do_ops->do_index_try(env, obj, &dt_lfsck_features);
6501 rc = lfsck_namespace_load(env, com);
6503 rc = lfsck_namespace_reset(env, com, true);
6504 else if (rc == -ENODATA)
6505 rc = lfsck_namespace_init(env, com);
6509 ns = com->lc_file_ram;
6510 switch (ns->ln_status) {
6515 spin_lock(&lfsck->li_lock);
6516 list_add_tail(&com->lc_link, &lfsck->li_list_idle);
6517 spin_unlock(&lfsck->li_lock);
6520 CERROR("%s: unknown lfsck_namespace status %d\n",
6521 lfsck_lfsck2name(lfsck), ns->ln_status);
6523 case LS_SCANNING_PHASE1:
6524 case LS_SCANNING_PHASE2:
6525 /* No need to store the status to disk right now.
6526 * If the system crashed before the status stored,
6527 * it will be loaded back when next time. */
6528 ns->ln_status = LS_CRASHED;
6532 spin_lock(&lfsck->li_lock);
6533 list_add_tail(&com->lc_link, &lfsck->li_list_scan);
6534 list_add_tail(&com->lc_link_dir, &lfsck->li_list_dir);
6535 spin_unlock(&lfsck->li_lock);
6542 if (root != NULL && !IS_ERR(root))
6543 lu_object_put(env, &root->do_lu);
6545 lfsck_component_cleanup(env, com);
6546 CERROR("%s: fail to init namespace LFSCK component: rc = %d\n",
6547 lfsck_lfsck2name(lfsck), rc);